Rationality and Commitment

Rationality and Commitment This page intentionally left blank Rationality and Commitment Edited by Fabienne Peter ...

Author: Fabienne Peter | Hans Bernhard Schmid

136 downloads 1916 Views 6MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Rationality and Commitment

This page intentionally left blank

Rationality and Commitment Edited by

Fabienne Peter and Hans Bernhard Schmid

OX F O RD UN I V E RS I T Y P RE S S · OX F O RD

1

Great Clarendon Street, Oxford ox2 6dp Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Cape Town Dar es Salaam Hong Kong Karachi Kuala Lumpur Madrid Melbourne Mexico City Nairobi New Delhi Shanghai Taipei Toronto With offices in Argentina Austria Brazil Chile Czech Republic France Greece Guatemala Hungary Italy Japan Poland Portugal Singapore South Korea Switzerland Thailand Turkey Ukraine Vietnam Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York  Fabienne Peter and Hans Bernhard Schmid, 2007 The moral rights of the authors have been asserted Database right Oxford University Press (maker) First published 2007 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Data available Typeset by Laserwords Private Limited, Chennai, India Printed in Great Britain on acid-free paper by Biddles Ltd., King’s Lynn, Norfolk ISBN 978–0–19–928726–0 10 9 8 7 6 5 4 3 2 1

Contents List of Figures, Schemata, and Tables List of Contributors Acknowledgments

vii ix xii

Introduction

1

Rational Fools, Rational Commitments Fabienne Peter and Hans Bernhard Schmid

3

Part I. Committed Action

15

1. Why Exactly is Commitment Important for Rationality? Amartya Sen

17

2. Construing Sen on Commitment Philip Pettit

28

3. Sympathy, Commitment, and Preference Daniel M. Hausman

49

Part II. Rethinking Rationality 4. Instrumental Rationality versus Practical Reason: Desires, Ends, and Commitment Herlinde Pauer-Studer 5. The Grammar of Rationality Geoffrey Brennan 6. The Rationality of Rational Fools: The Role of Commitments, Persons, and Agents in Rational Choice Modelling Werner Güth and Hartmut Kliemt 7. Rational Self-Commitment Bruno Verbeek

71 73 105

124 150

vi

contents

8. Rationality and Commitment in Voluntary Cooperation: Insights from Experimental Economics Simon Gächter and Christian Thöni

Part III. Commitment, Intentions, and Identity 9. Beyond Self-Goal Choice: Amartya Sen’s Analysis of the Structure of Commitment and the Role of Shared Desires Hans Bernhard Schmid 10. Cooperation and the We-Perspective Raimo Tuomela 11. Collective Intentions, Commitment, and Collective Action Problems Margaret Gilbert

175

209 211 227

258

12. Theories of Team Agency Natalie Gold and Robert Sugden

280

13. Identity and Commitment: Sen’s Fourth Aspect of the Self John B. Davis

313

Comment

337

Rational Choice: Discipline, Brand Name, and Substance Amartya Sen

339

Index

363

List of Figures, Schemata, and Tables Figures 3.1 3.2 3.3 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 6.9 7.1 8.1 8.2

8.3 8.4 8.5 10.1 10.2 12.1 12.2 12.3 12.4

Prisoner’s Dilemma; game form Prisoner’s Dilemma; game Prisoner’s Dilemma; game form The simple trust game Relative and absolute commitments Take it or leave it in strategic form Take it or leave it in extensive form Modified take it or leave it game An embedded battle of the sexes Sequential Prisoner’s Dilemma Strategic form of sequential Prisoner’s Dilemma Standard Prisoner’s Dilemma with power to commit to strategies as programmes Decision tree The Prisoner’s Dilemma game Average contribution function of types freerider, conditional cooperator, triangle contributor, and ‘others’ Average actual contributions and predicted contributions Average contributions over ten periods Cooperation patterns in the absence and presence of punishment Collective good dilemma Prisoner’s Dilemma The Prisoner’s Dilemma Hi-Lo A Prisoner’s Dilemma with transformed payoffs Hawk–Dove

63 63 68 128 130 130 131 131 134 137 137 138 164 176 186

191 193 196 242 243 281 283 293 305

viii

list of figures, schemata, and tables

Schemata 12.1 12.2 12.3 12.4

Individual rationality Collective rationality Simple team reasoning (from a group viewpoint) Simple team reasoning (from an individual viewpoint) 12.5 Restricted team reasoning 12.6 Circumspect team reasoning 12.7 Mutually assured team reasoning

286 287 288 289 295 297 303

Tables 6.1 Rational choice as maximization and non-maximization 8.1 Overview of the distribution of types in Prisoner’s Dilemma games and Public Goods games

144 188

List of Contributors Geoffrey Brennan is Professor of Social and Political Theory in the Research School of Social Sciences at the Australian National University and holds a regular visiting professorship jointly at UNC-Chapel Hill (in the Philosophy Department) and Duke University (in the Political Science Department). He is author with Nobel Laureate James Buchanan of The Reason of Rules (1985) and The Power to Tax (1980) and most recently of The Economy of Esteem (2004) with Philip Pettit. John B. Davis is Professor of History and Philosophy of Economics at the University of Amsterdam and Professor of Economics at Marquette University. He is author of Keynes’s Philosophical Development (1994) and The Theory of the Individual in Economics (2003), is the former editor of the Review of Social Economy, and currently co-editor of the Journal of Economic Methodology. Simon Gächter is Professor of the Psychology of Economic Decision Making at the School of Economics of the University of Nottingham (UK). His research is in the area of behavioral and experimental economics. His main publications are in Nature, American Economic Review, and Econometrica. Margaret Gilbert holds the Melden Chair in Moral Philosophy at the University of California, Irvine. Her books include On Social Facts (1989), Living Together: Rationality, Sociality, and Commitment (1996), Sociality and Responsibility: New Essays in Plural Subject Theory (2000), and A Theory of Political Obligation: Membership, Commitment, and the Bonds of Society (2006). Natalie Gold is a Lecturer in Philosophy at the University of Edinburgh. She investigates individual and collective decision-making using methods from a variety of disciplines, including economics, philosophy, and psychology. In particular she is interested in the incorporation of developments in cognitive science into social scientific theories. Werner Guth ¨ is currently Director of the Strategic Interaction Group at the Max Planck Institute of Economics in Jena. He has published widely on a number of subjects, including game theory, experimental economics, social psychology, and philosophy which form his main research interests to the present day. Among his

x contributors most recent papers are ‘‘Inequality Aversion in a Variety of Games—An Indirect Evolutionary Analysis,’’ The Economic Journal (2006, with Stefan Napel), ‘‘Bargaining Outside the Lab—A Newspaper Experiment of a Three-Person Ultimatum Game,’’ The Economic Journal (2007, with Carsten Schmidt and Matthias Sutter), and ‘‘On the Co-evolution of Retribution and Trustworthiness,’’ Journal of Institutional and Theoretical Economics (2007, with Hartmut Kliemt, Vittoria Levati, and Georg von Wangenheim). Daniel M. Hausman is the Herbert A. Simon Professor of Philosophy at the University of Wisconsin-Madison. He is a founding editor of the journal Economics and Philosophy and the author or editor of Capital, Profits and Prices, The Inexact and Separate Science of Economics, Causal Asymmetries, and (jointly with Michael McPherson) Economic Analysis, Moral Philosophy, and Public Policy. Hartmut Kliemt has published widely in poltical philosophy, on the foundations of game and decision theory, medical theory, and medical ehtics. At present he is working on a German research foundation project on priority setting in medicine. As an adjunct research associate to the Center for Study of Public Choice, Farifax, Va., he has also been (together with H. Geoffrey Brennan and Robert D. Tollison) one of the co-editors of the collected works of James Buchanan. Herlinde Pauer-Studer is Professor of Philosophy at the University of Vienna. Her publications include Das Andere der Gerechtigkeit (1996), Autonom Leben (2000), ¨ and Kommentar zu David Hume ‘‘Uber Moral’’ (2007), and papers published in journals such as The Monist, Philosophical Explorations, and the Journal of Economic Methodology. Fabienne Peter is an Associate Professor in Philosophy at the University of Warwick. She specializes in political philosophy and the philosophy of economics, with a research focus on the normative foundations of policy evaluation. Her articles have appeared in journals such as Economics and Philosophy, Ethical Theory and Moral Practice, and Politics, Philosophy, and Economics. She has also co-edited a volume on public health ethics with Sudhir Anand and Amartya Sen (Public Health, Ethics, and Equity, 2004). Philip Pettit teaches political theory and philosophy at Princeton University, where he is L. S. Rockefeller University Professor of Politics and Human Values. His recent publications include The Economy of Esteem (2004, with Geoffrey Brennan) and Made with Words: Hobbes on Language, Mind and Politics (2007). Common Minds: Themes from the Philosophy of Philip Pettit was published by Oxford University Press in 2007, edited by G. Brennan, R. E. Goodin, F. C. Jackson, and M. A. Smith.

contributors xi Hans Bernhard Schmid is SNF-Professor of Philosophy at the University of Basel, and Lecturer at the University of St Gallen. He specializes in philosophy of social science, phenomenology, philosophy of action, and social and sociological theory. His main current research project is called ‘‘Collective Intentionality—Phenomenological Perspectives.’’ Publications include Wir-Intentionalität. Kritik des ontologischen Individualismus und Rekonstruktion der Gemeinschaft (2005) and Subjekt, System, Diskurs (2000). Amartya Sen is Lamont University Professor, and Professor of Economics and Philosophy, at Harvard University. Amartya Sen’s books have been translated into more than thirty languages, and include Collective Choice and Social Welfare (1970), On Economic Inequality (1973, 1997), Poverty and Famines (1981), Choice, Welfare and Measurement (1982), Resources, Values and Development (1984), On Ethics and Economics (1987), The Standard of Living (1987), Inequality Reexamined (1992), Development as Freedom (1999), Rationality and Freedom (2002), The Argumentative Indian (2005), and Identity and Violence: The Illusion of Destiny (2006), among others. In 1998 he received the Nobel Prize in Economics. Robert Sugden is Professor of Economics at the University of East Anglia, Norwich. His research uses a combination of theoretical, experimental, and philosophical methods to investigate issues in welfare economics, social choice, choice under uncertainty, the foundations of decision and game theory, the methodology of economics, and the evolution of social conventions. ¨ earned his PhD in 2005 and is currently Assistant Professor in Christian Thoni Economics at the University of St Gallen (Switzerland). His research and teaching are in the field of behavioral and experimental economics. He has published in the Journal of the European Economic Association, the Journal of Economic Behavior and Organization, and Behavioral and Brain Sciences. Raimo Tuomela is Professor of Philosophy at the Department of Social and Moral Philosophy, University of Helsinki, Finland, and Permanent Visiting Professor at the Department of Philosophy, University of Munich. His current field of research is philosophy of social action, collective intentionality. His recent books are The Importance of Us: A Philosophical Study of Basic Social Notions (1995), Cooperation: A Philosophical Study (2000), The Philosophy of Social Practices: A Collective Acceptance View (2002), and The Philosophy of Sociality: The Shared Point of View (2007). Bruno Verbeek is Assistant Professor at Leiden University. He is the author of Instrumental Rationality and Moral Philosophy (2002) and Reasons and Intentions (forthcoming), as well as several papers on the interface of ethics, rational choice theory, moral psychology, and action theory.

Acknowledgments The chapters by Daniel Hausman, Philip Pettit, Hans Bernhard Schmid, and Amartya Sen’s ‘‘Why Exactly is Commitment Important for Rationality?’’ have previously been published in a symposium in the journal Economics and Philosophy (volume sol. 21/1, pages 1–63), guest-edited by Fabienne Peter and Hans Bernhard Schmid.

Introduction


Rational Fools, Rational Commitments F A B I E NNE P E T E R A ND H A NS B E RNH A RD S CHMID

Skepticism about rational choice theory dates back to the subject’s very origins. One of the earliest predecessors of the theory was proposed by Thomas Hobbes. He defended the view that ‘‘all the voluntary actions of men tend to the benefit of themselves; and those actions are most reasonable, that conduce most to their ends’’ (Leviathan xv.4). Hobbes was aware, however, of the possible objection that this conception of practical reason neglects important aspects of human behavior, particularly the demands of morality. In Leviathan, he struggles to answer his foole, who claims that if ‘‘reason dictates to every man his own good,’’ and if ‘‘there could be no reason, why every man might not do what he thought conduced thereunto,’’ it would be irrational to follow the demands of morality when they are detrimental to one’s own good (Leviathan xv.4). Contemporary rational choice theory demands two things of rational agents. First, their actions must be the expression of a preference ranking that fulfills certain axioms of consistency. The second demand is maximization. The agents must always choose the most preferred option available to them. Transported into this context, the question that Hobbes’ foole raises is the following. Is rational choice theory descriptively and normatively adequate or do the two demands clash with certain types of observed behavior and with their normative justification? There are different strategies to address this problem. The defensive strategy is to follow the line of reasoning Hobbes takes in Leviathan. It is to answer the foole by insisting on

4

fabienne peter and hans bernhard schmid

the descriptive adequacy of rational choice theory, and by showing how seemingly differing patterns of behavior—such as norm-oriented action, for example—are in fundamental accord with this notion of rationality. The critical strategy is to side with Hobbes’ foole and to claim that there are types of action which rational choice theory cannot accommodate. This then raises the question of how to conceive of a more comprehensive conception of practical reason.1 In ‘‘Rational Fools’’ (1977), Amartya Sen puts forward a piercing critique of rational choice theory that follows the second, critical strategy. Sen’s rational fool is the counterpart of Hobbes’ foole. Whereas the latter wonders about the limits of a self-interest-based theory of rationality, the former does not realize that there might be such limits. Sen’s argument hinges on the distinction between three types of motivation—(narrow) self-interest, sympathy, and commitment. One acts from self-interest when one aims at maximizing one’s welfare, where welfare is understood as depending exclusively on the goods enjoyed by oneself. Sympathy involves a broader conception of welfare. One acts from sympathy when one’s own welfare is affected by how others are doing, as in the case where helping others makes one feel better. As such, sympathy can easily be accommodated within rational choice theory. Commitment, by contrast, refers to a kind of behavior which is motivationally unrelated to the agent’s welfare, however broadly conceived. The clearest case of action from commitment is when one feels compelled to intervene in a certain matter, even if doing so leaves one worse off. In other cases of committed action, there might not be a negative impact on one’s welfare. What matters, however, is that increasing one’s welfare is not the central motive. The central claim of ‘‘Rational Fools’’ is that committed action cannot be integrated into the standard account of rational choice theory because it drives a wedge between choice and welfare. After ‘‘Rational Fools,’’ Sen has further elaborated and indeed radicalized his diagnosis. In ‘‘Goals, Commitment, and Identity’’ (2002 [1985]), he argues that the possibility of committed action not only breaks the standard model’s link between choice and welfare, but also shows that a person’s choices may not even be based on his or her own goals. Other people’s goals can influence an agent’s choices beyond the extent to which these goals shape the agent’s own goals.

rational fools, rational commitments 5 Since there can be little doubt that committed action plays an important role in human behavior, Sen’s critique challenges, first of all, the descriptive relevance and explanatory power of rational choice theory. In ‘‘Rational Fools,’’ Sen illustrates the importance of commitment with a wide range of examples. These include being truthful when talking to others, voting, sources of work motivation other than income-maximization, cooperation in Prisoner’s Dilemma situations, being polite and other instances of norm-following behavior, and moral action. In addition, the argument challenges the conception of rationality underlying rational choice theory. The rational fool of the essay is rational only in the ‘‘puny’’ sense of rational choice theory; in the eyes of those who take a fuller (and presumably more adequate) view of rationality, it is plain that he is no more than a simple fool. In sum, three consecutive steps can be identified in Sen’s critique of rational choice theory: (i) render plausible the importance of commitments in human behavior; (ii) demonstrate that rational choice theory—at least in the interpretation favored by most economists—cannot accommodate committed action; and (iii) argue for the need to develop a theory of rationality in action that is not marked by this flaw. Although published several decades ago, ‘‘Rational Fools’’ has lost none of its importance. In fact, Sen’s critique meets with important new developments both in economics and in philosophy. In economics, instances of committed action and the questions these raise for economic modeling have been extensively discussed in connection with findings in experimental and behavioral economics. In philosophy, the ‘‘economic’’ or ‘‘Humean model’’ of practical reason, while continuing to be hugely influential, has recently come under increased scrutiny. The question of how to account for the rationality of committed action is thus very much an open issue and, if anything, even more pressing today. In what sense can committed action be called rational? Is the distinction between sympathy and commitment plausible and does commitment indeed drive a wedge between welfare and choice? Are the choices of an individual separable from his or her own goals? Is there a defensive counter-strategy to Sen’s line of critique, or does it convincingly make the case for the need for an alternative theory of practical reason? What would be the elements of such an alternative account? To explore these and related questions, we organized a workshop at the University of St Gallen in May 2004.

6


Economists and philosophers met to discuss the challenge posed by Sen’s ‘‘Rational Fools’’ in light of recent developments in the two disciplines.2 The chapters gathered in this volume were written for this occasion.3 They show impressively just how wide the perspectives opened up by Sen’s critique are. ‘‘Rational Fools’’ has provided a point of reference that, several decades later, continues to be of influence in discussions of practical reason, and has indeed acquired the status of a ‘‘classic’’ in this debate. The chapters in this volume fall into three categories. Those in the first part of the collection, including Sen’s own contribution, revisit the argument of ‘‘Rational Fools,’’ with varying emphasis on the different steps of Sen’s critique. The chapters in the second part of the volume focus on the second and third steps—on the question of whether or not rational choice theory needs to be modified or abandoned in light of the possibility of rational committed action. Lastly, the third set of chapters discusses Sen’s critique from the angle of collective intentionality analysis. The authors use conceptual tools developed in this approach to examine the rationality of committed action. The volume concludes with Sen’s reply to his commentators.

1. Committed action The volume opens with Amartya Sen’s keynote address to the workshop. This chapter provides, first, a brief restatement of the core features of his critique of rational choice theory. Its main emphasis, however, is on arguing for the importance of accommodating commitment in a theory of rational action. Sen gives two reasons for the importance of commitment to rationality. First, commitment necessitates a broader conception of the reasoning underlying people’s choices. Second, commitment is essential to the explanation of actual behavior. Sen ends his chapter by analyzing some of the consequences of taking seriously the possible rationality of acting from commitment. His examples for the importance of commitment range from the role of ethics and morality in human behavior, to pressing policy issues such as sustainable development. This last part of the chapter helpfully situates Sen’s critique of rational choice theory in the broader context of his work. The next two chapters are by Philip Pettit and Dan Hausman. Both authors sympathize with the general thrust of Sen’s writings on rational

rational fools, rational commitments 7 choice theory. In particular, they accept the first step of his critique. They emphasize that commitment is an important category of human action that economists have unduly tended to neglect. They do not, however, endorse the second step of Sen’s argument, i.e. the claim that commitment cannot be accommodated in rational choice theory. Pettit’s chapter is concerned with Sen’s most radical claim: the claim that rational choice theory cannot get a grip on how other people’s goals may influence an agent’s choices beyond the extent to which these goals shape the agent’s own goals. Pettit argues that the ‘‘self-goal choice assumption’’ criticized by Sen is central to the very concept of action, as it is generally conceived, and cannot be violated by commitment. He then goes on to analyze possible reasons which could have led Sen to make such a claim. According to Pettit, the main motive underlying Sen’s critique is to make the concepts of choice and action more deliberative. He distinguishes among three possible views of the role of deliberation and attributes to Sen the weakest version. Pettit credits Sen for his critique of narrow-minded versions of rational choice, but argues that what he seems to have in mind with ‘‘committed action’’ can be accommodated in the goal-maximization schema of rational choice theory. Dan Hausman, too, gives Sen’s concept of committed action a reading that reconciles it with rational choice theory. Discussing alternative interpretations of the concept of preferences, he argues that we should be prescriptive about preferences. Specifically, Hausman advocates the view that preferences should be interpreted as ‘‘all-things-considered rankings.’’ This allows for a clear distinction between the preferences themselves and the factors that influence their development. Commitment, according to the view put forward in this chapter, is one of these factors: our ‘‘all-things-considered rankings’’ will depend, among other things, on our commitments. Committed action can thus be incorporated in rational choice theory. Hausman concludes that Sen’s aim to advocate a more nuanced view of human rationality in general and the role of commitments in particular is better served by a more adequate account of what determines preference than by abandoning a single notion in favor of multiple notions of preference. He thus rejects the second step of Sen’s critique while sharing with Sen the general aim of the project of the ‘‘Rational Fools’’ essay, i.e. to emphasize the complexity, incompleteness, and context-dependence of preferences.

8


2. Rethinking rationality The chapters in this section take up Sen’s challenge of how to account for the rationality of commitment and discuss this problem against the background of debates in both philosophy and economics. Whereas most of the chapters adopt a (partially) defensive strategy, the first chapter, by Herlinde Pauer-Studer, follows Sen’s critical project. Pauer-Studer interprets the ‘‘Rational Fools’’ critique as revealing a tension between two accounts of rationality. A first account would be that of instrumental rationality familiar from rational choice theory. Acting from commitment, by contrast, seems to suggest a more moralized, Kantian account of practical reason. After a discussion of different versions of Kantian practical reason put forward in the current debate, Pauer-Studer concludes that the tension between the two accounts of rationality could be resolved in a moderate Kantianism. In her proposal, instrumental rationality figures as a necessary, but not as a sufficient, condition for practical reason. Pauer-Studer argues that Sen could subscribe to this moderate Kantianism. Geoffrey Brennan’s chapter, too, is concerned with alternative interpretations of rationality, and with the question of how rationality and committed action can meet. In some contrast to Pauer-Studer, however, he adopts a defensive strategy and argues that the ‘‘grammar’’ underlying rational choice theory is—and should be seen as—sufficiently flexible to accommodate a broad range of statements about what is rational. More specifically, Brennan defends two claims. The first is that the rational actor approach should be understood as an ‘‘approach,’’ and not as a theory. Its essential features are structural assumptions about utility functions, not assumptions about the content of these functions. The second is that if one does think of rationality in these terms, Sen falls firmly within the rational actor school. To illustrate the second claim, Brennan discusses two cases that Sen takes to be in conflict with rational choice theory: how to account for the rationality of voting and of following rules of politeness. The next two chapters offer proposals for how to interpret the rationality of commitment in a rational choice context. Werner Güth and Hartmut Kliemt approach the issue from a game theoretic perspective. They first introduce a distinction that is akin to Brennan’s. Güth and Kliemt distinguish between rational choice theory, which contains substantive assumptions about who is a player and what is on that player’s mind, and

rational fools, rational commitments 9 rational choice modeling, which does not contain any such assumptions. They then show how rational choice modeling can accommodate commitments as part of the rules of the game. The reply to the ‘‘Rational Fools’’ critique given in this chapter is thus twofold. The authors argue that Sen is right that explanations generated by rational choice theory for behavior within a given game cannot accommodate non-instrumental forms of rationality. For this reason, they agree with Sen that our understanding of social interaction through (classical) game theory is severely limited. Rational choice modeling, however, can be applied to reasoning about the rules of the game and can thus overcome the neglect of committed action. Bruno Verbeek is also interested in the alternative rationality underlying commitments. He discusses the issue in the temporal dimension. How exactly can we account for the rationality of someone who, at a particular moment in time, sticks to a decision that he or she has made earlier? For Verbeek, the answer to this question lies in our ability to make plans. Plans structure practical reason by imposing certain requirements. A first requirement is that of internal consistency. One cannot plan at time t = I to do two mutually exclusive actions at time t = 2. As such, plans typically impose constraints in the form of ‘filters of admissibility’ on deliberation. A second requirement is means–end coherence. Plans create the context for deliberation and establish standards of relevance for options to be considered in the deliberative process. Verbeek shows how the rationality of commitments is tied to this way in which plans structure deliberation. Simon Gächter and Christian Thöni approach the relation between rationality and commitment from the perspective of experimental economics. They start with a discussion of the finding that people tend to cooperate much more frequently in Prisoner’s Dilemma and public good games than the standard version of rational choice theory predicts. Three different explanations that have been advanced for this empirical result are then examined: (i) cooperation in one-shot games is erroneous and maladaptive; (ii) people’s reasoning differs from the individualistic approach of rational choice theory; and (iii) the Prisoner’s Dilemma or public goods game does not adequately reflect people’s true preferences. Gächter and Thöni favor the third explanation. Empirical research conducted by themselves and others shows that many people have non-selfish preferences that dispose them to cooperate, at least as long as others cooperate as well.

10


3. Commitment, intentions, and identity The chapters in the concluding section of this volume approach the question of how commitment could be incorporated in an alternative theory of rational action from the perspective of collective intentionality analysis. Collective intentionality analysis is a recent development with origins in analytic philosophy of mind. Whereas previous analyses have tended to be limited to individual intentionality and action, the question of what it means to share an intention and act together has come to attract increasing attention over the past two decades. Collective intentionality analysis meshes seamlessly with some features of Sen’s critique of rational choice theory, particularly with Sen’s insistence on the importance of the agent’s social identities, and with the role of shared aims in deliberation. In this concluding section of the volume, special attention is paid to the role of these factors in committed action. In the opening chapter of this section, Hans Bernhard Schmid reexamines Sen’s claim that committed action violates self-goal choice, i.e. that committed agents pursue goals other than their own. According to the critics of Sen’s view, this claim is nonsensical, because even strongly altruistic agents cannot pursue other people’s goals without making them their own. Thus it seems that, contrary to Sen’s claim, self-goal choice is not violated by committed action, but constitutive of any kind of agency. Schmid argues that this objection holds only with respect to a particular kind of goals, while Sen’s claim makes perfect sense with respect to a kind of goals that until very recently has been widely ignored. As opposed to individual goals, the pursuit of shared goals does indeed violate self-goal choice. In his defense of Sen’s claim, Schmid follows up on an argument put forward by Elizabeth Anderson, which links the rationality of committed action to the structure of collective intentions and actions. In the second chapter of this section, Raimo Tuomela introduces some of the key features of the theory of cooperation he has developed over the past two decades, and shows how these conceptual tools can be put to work for the purpose of a better understanding of the role of rationality and commitment in cooperation. The main focus is on the distinction between what he calls we-mode thinking and acting, on the one hand, and I-mode thinking and acting, on the other. Tuomela examines their respective rationality requirements and sheds light on the role commitments may

rational fools, rational commitments 11 play in either of these two modes. In some cases of I-mode thinking and acting, the agents are privately committed to the group’s goals, whereas in the case of we-mode thinking and acting, collective commitments always come into play. We-mode thinking and acting is required for cooperation to be persistent and stable, and can thus be both more rational even from an individual perspective. The topic of Margaret Gilbert’s chapter is the role of commitment in collective action. According to Gilbert, collective actions presuppose collective intentions, and collective intentions originate in joint commitments. In this chapter, Gilbert approaches her topic by distinguishing two meanings of the term ‘‘collective action.’’ In the rational choice sense of the term collective action refers to a mere combination of the actions of participating individuals. By contrast to this, the meaning of the term that concerns her involves a supra-individual unit of agency. Gilbert gives an overview of the central conceptual elements of her theory of collective intentions and addresses collective action problems. By placing commitment at the core of the conceptual analysis of collective action, she develops a proposal for how to meet the challenge posed by Sen’s ‘‘Rational Fools’’ critique, i.e. to take seriously commitment as a part of human behavior. The next chapter, by Natalie Gold and Robert Sugden, examines the same topic from a different angle. The question of the structure of practical reasoning involved in shared agency is approached from the perspective of decision theory. The authors start with an analysis of two basic puzzles of game theory, i.e. the puzzles posed by the Prisoner’s Dilemma and the Hi-Lo game. They argue that solving these puzzles calls for a concept of team reasoning. The authors argue that where the participating individuals team reason, the team can be seen as an agent in its own right. Building on Michael Bacharach’s work, Gold and Sugden provide an outline of a theory of team reasoning. Its basic building block is the concept of a schema of practical reasoning, which describes the way in which explicit premises about the decision environment and about what agents are seeking to achieve enter the deliberative process. Gold and Sugden also show how their theory of team reasoning contributes to a more adequate understanding of collective intentionality. John Davis addresses the much-neglected question of the identity of the agents. From Sen’s critique of rational choice theory the author extracts a notion of the agents’ identity that differs from those recent

12 fabienne peter and hans bernhard schmid accounts that emphasize the individuals’ belonging to a group. According to Davis, Sen’s work on committed action should be interpreted with reference to the personal identity of individuals. Davis reads Sen against the background of Bernard Williams’ concept of commitment and focuses on the reflexive dimension of commitment. He argues that the capacity to engage in reasoning and self-scrutiny is one of the constitutive elements of personal identity. His contribution concludes by exploring possible links between this feature of Sen’s critique of rational choice theory and the capability approach, thus placing this critique in the wider context of Sen’s work.

Notes

1. A third strategy, not discussed in this volume, would be to keep the descriptive and the normative dimensions strictly separate and to develop a conception of human behavior that extends to non-rational elements. This seems to be the view toward which Hobbes tended in parts of his later work Behemoth, where he emphasized the irrational ingredients in human motivation. 2. We co-organized the workshop together with Dieter Thomä. The organizers wish to express their gratitude to the Swiss National Science Foundation, the International Students Committee Foundation (ISCF), the HSG Alumni Association, and the Research Foundation of the University of St Gallen for their generous support of this workshop. 3. A subset of these papers (i.e. the chapters by A. Sen, P. Pettit, D. Hausman, and H. B. Schmid) has been published as a symposium in Economics and Philosophy vol. 21/1 (2005). We wish to thank Cambridge University Press for permission to reprint this material.

References Anderson, Elizabeth. 2001. ‘‘Unstrapping the Straitjacket of ‘Preference’: A Comment on Amartya Sen’s Contributions to Philosophy and Economics.’’ Economics and Philosophy 17: 21–38. Hobbes, Thomas. 1994 [1651]. Leviathan. With selected Variants from the Latin Edition. Edwin Curley (ed.). Indianapolis: Hackett. Hobbes, Thomas. 1990 [1679] Behemoth. Chicago: University of Chicago Press.

rational fools, rational commitments 13 Sen, Amartya. 1977. ‘‘Rational Fools: A Critique of the Behavioural Foundations of Economic Theory.’’ Philosophy and Public Affairs 6: 317–44. Sen, Amartya. 2002 [1985]. ‘‘Goals, Commitment, and Identity.’’ In Amartya Sen, Rationality and Freedom. Cambridge, Mass., Harvard University Press, 2002, pp. 206–24.


PA RT I

Committed Action


1 Why Exactly is Commitment Important for Rationality?1 A M A RT YA S E N

I The idea of rational choice must be founded, in one way or another, on the basic requirement that choices be based on reason. But the interpretation of how reason is to be used can vary so radically between different formulations of rationality that there is frequently very little in common between the different uses of the idea of rational choice. There are differences over the domain of reason: for example, whether it should apply only to the selection of alternatives given one’s preferences over the alternatives, or also be used in the determination of the preferences themselves. Within each domain, there can be diversities in the way reason may be used, for example, whether reasoning can lead to some allowance being made for the preferences of others in addition to reflecting one’s own preferences in choosing over the alternatives. These and other possible variations lead to quite different interpretations of the demands of rational choice. In what is called ‘‘rational choice theory’’ (a misleading name, since it takes such a puny view of reason and rationality), the focus is on characterizing rationality of choice as maximization of self-interest.2 The incorporation of one’s ‘‘self’’ in reasoned choices can be a very difficult issue, since considerations of one’s own ‘‘self’’ may enter the calculations involved in quite different ways (not merely in terms of what can be seen as one’s own ‘‘interest’’). In an earlier paper, published in 1985, I used a distinction between three different ways in which the self may be central to

18

amartya sen

one’s preferences and choices: (1) ‘‘self-centered welfare’’; (2) ‘‘self-welfare goal’’; and (3) ‘‘self-goal choice.’’3 1. Self-centered welfare: A person’s welfare depends only on his or her own consumption, which rules out sympathy or antipathy toward others, as well as the effects of processes and relational concerns on one’s own welfare. 2. Self-welfare goal: A person’s only goal is to maximize his or her own welfare, which rules out incorporating within one’s own objectives other considerations (such as the welfare of others), except to the extent that it influences the person’s own welfare. 3. Self-goal choice: A person’s choices must be based on the exclusive pursuit of his or her own goals, which rules out being restrained by the recognition of other people’s goals, except to the extent that these goals shape the person’s own goals. All three requirements are in fact imposed in the traditional models of rational behavior (for example, those in the ‘‘general equilibrium theory’’ of Arrow and Debreu).4 To illustrate, a person whose own welfare is influenced by the misery of others, because of his or her sympathy for them, certainly does violate the condition of ‘‘self-centered welfare,’’ but this fact tells us nothing about whether the person’s goals will directly include considerations other than the person’s his or her own welfare (thereby violating the self-welfare goal), or whether the person’s choices will depart at all from being based in each case on the pursuit of his or her own goals (thereby violating the self-goal choice). Or, to take another type of case, a person’s goal may include objectives other than maximization of her own welfare, for example, pursuit of her concept of social justice and, while this may violate the axiom of self-welfare goal, it would leave open the question of adherence to self-centered welfare and that of self-goal choice. Similarly, a person’s choice behavior may be constrained or influenced by the goals of others, or by rules of conduct (for reasons that Immanuel Kant and Adam Smith discussed so well), thereby violating self-goal choice, but this need not, on its own, entail any violation of self-centered welfare or self-welfare goal. The influences may work on the person’s conduct, which may be based on the deeply social recognition that there are other people in the society who also are trying to pursue their own goals (and that they should be treated fairly). But this influence

why exactly is commitment important for rationality? 19 on considerate behavior may not require that the person modifies his own goals correspondingly—it is a question of how to act given one’s own priorities and the manifest priorities of others. In a still earlier paper, called ‘‘Rational Fools,’’ I tried to explore a distinction between ‘‘sympathy’’ and ‘‘commitment’’ as possible foundations for other-regarding behavior.5 Sympathy (including antipathy when it is negative) refers to ‘‘one person’s welfare being affected by the position of others’’ (e.g., feeling depressed at the sight of misery of others), whereas ‘‘commitment’’ is ‘‘concerned with breaking the tight link between individual welfare (with or without sympathy) and the choice of action (for example, being committed to help remove some misery even though one personally does not suffer from it).’’6 In this interpretation, sympathy violates self-centered welfare, but not necessarily the other two requirements. Commitment, on the other hand, need not involve a violation of self-centered welfare, but can take the form of modifying the person’s goals, which can include effects on others beyond the extent to which these effects influence her own welfare (thereby violating self-welfare goal). It can also alter the person’s reasoned choice through a recognition of other people’s goals beyond the extent to which other people’s goals get incorporated within one’s own goals (thereby violating self-goal choice). In the important advance made by Gary Becker on the characterization of rational choice, he seized the bull by the horns so far as ‘‘self-centered welfare’’ is concerned.7 But he left the two other bulls of ‘‘self-welfare goal’’ and ‘‘self-goal choice’’ freely roaming the fields without being noticed in the reasoning of the rational agent, and without their getting much attention from Becker himself. The Beckerian utility function that the person maximizes can, therefore, be seen both as the person’s maximand in reasoned choice and as a representation of the person’s own self-interest (rather than some commitment to go, if necessary, against one’s selfinterest). That congruence is quite important for many of the extensive results of the analysis that Becker pursues powerfully. No one, I believe, has done more than Gary Becker—one of the major gurus of the discipline of rational choice theory—in marrying that approach with utility functions that do admit concern for others, accommodating within the concept of self-interest our many non-self-centered concerns. But this does not take any note of the reasons that may lead a person to

20

amartya sen

have different goals from what she sees as her own welfare, or to choose behavior and actions that go beyond pursuing her own goals (making practical reasoning take note of the goals of others). In their illuminating critique of rational choice theory, Jolls, Sunstein, and Thaler also stick largely to accommodating non-self-centered welfare, rather than trying to take fuller notice of the other two departures.8 Jolls, Sunstein, and Thaler would like to go further along the path of reducing the self-centered characterization of self-interest, and the extensions they suggest have conceptual plausibility and explanatory value. But they are not any more hostile to the basic congruence of a person’s own welfare (with sympathies and antipathies taken into account) and the maximand she uses for her choices. The critique that Jolls, Sunstein, and Thaler present is, thus, an important contribution to the debate ‘‘within’’ the basic conception of rationality as formulated in on-going rational choice theory. To go beyond that, we do need to take note of the possibility of committed behavior, so that conduct may not be congruent with the relentless pursuit of one’s own goals, and in particular one’s own welfare (no matter how broadly it is characterized). That is the first answer to the question in the title of this talk: why is commitment important for rationality? If committed behavior is ruled out, then the reasoning that can go into the determination of choices would be correspondingly impaired. That would hardly be the way to give reason its due in the idea of rationality in general and rational choice in particular.

II A second reason for the importance of taking note of commitment is that it can help to explain many patterns of behavior that we actually observe which are hard to fit into the narrow format of contemporary rational choice theory.9 In understanding the relevance of this question, we have to take into account the fact that the idea of rationality is used in economics—and increasingly in the other social sciences too—for two quite distinct purposes, representing two quite different motivations. The direct motivation is to understand what form—or forms—reasoned behavior can take. What makes one choice rational and not another? The indirect motivation, in contrast, is concerned with the description and

why exactly is commitment important for rationality? 21 prediction of actual behavior. The program here is that of understanding, explaining, and prognosticating actual behavior by characterizing, first, rational behavior, and then assuming that actual behavior will coincide with rational behavior. I have already discussed why taking note of commitment may be quite important for the first purpose. It is also important for the second. Indeed, the two purposes are closely linked, especially because the indirect interest cannot but be, in some sense, parasitic on the direct interest. If you cannot characterize rational behavior well, you cannot make good use of that characterization to predict actual behavior, if the strategy is to assume rationality to make actual behavior more understandable. I should, however, note in passing that whereas the indirect interest is contingent on the first, the first (direct) interest may or may not be married to any program of predicting actual behavior. A theorist of rational choice can be content with analyzing what rationality requires (even perhaps advising how to behave rationally), without presuming that people actually act—always or even by and large—according to the dictates of rationality. This elementary point is worth noting straightaway, since the indirect use of rational choice as a method of getting at prediction of actual behavior has been so pervasive in some schools of economic, political, and legal thinking that the direct interest in understanding the nature of rational choice may have, to some extent, been rather overshadowed or clouded. It is, thus, important to remind ourselves that we can be interested in what rational choice requires without wanting to presume that this will tell us how people standardly behave.10 It is also crucial to note that a presumption that people behave rationally does not—and cannot—easily lead to a particular theory of actual choice until the very substantial debates on the nature of rationality of behavior have been adequately addressed. There are many empirical studies that indicate that committed behavior has its actual domain. That demonstration has tended to come from many cases of experimental games, but even in the world of actual practice (such as work motivation and environment-sensitive conduct) understanding the role of commitment helps to explain many cases that would be hard to fit within the narrow limits of self-welfare goal combined with self-goal choice. There is also considerable evidence that the view that is taken of work responsibilities varies greatly from one society to another. These comparisons, which are mainly cross-sectional, can be supplemented by

22

amartya sen

evidence of intertemporal change as a result of movements of social tradition—an old subject that received much attention from Marx, Tawney, and Weber, among many others. Social traditions can indeed be relevant in the formation of objective functions as well as in the specification of action ethics. There has also been much discussion over the last few decades on the extent of committed behavior in Japanese industrial production.11 These theories include alternative attempts at explaining why, as The Wall Street Journal put it (stretching a point somewhat), Japan is ‘‘the only communist nation that works.’’12 One of the observed phenomena in behavioral norms is the tendency to imitate others, much commented on in the 1990s in the context of reasons offered for the pursuit of corrupt activities in business and politics in Italy: ‘‘I was not alone in doing it.’’ The reasoning underlying such behavioral response points to the need to bring in the influence of established norms of behavior in explaining what actions will or will not be chosen, and it is clear from the discussions presented to the Anti-Mafia Commission in Italy that many businessmen felt that they had to depart from their commitment to business ethics, since others—their competitors—were already doing that and they could not be matched without a similar eschewal of business ethics.13 One can consider also many other examples, but the basic point is that commitment is important for characterizing the demands of rationality, and also for explaining behavioral variations between different societies and over time. The admission of committed behavior within a theory of rationality not only enriches our conceptual understanding of rationality, but also helps us to understand actual behavior better, through taking note of the varying role of commitment in different social circumstances.

III I turn finally to ethics and morality. It is easy enough to see that ethics is somehow mixed up with the idea of commitment, even though departures of the committed kind may also come from sheer force of habit. The belief that we, as free and rational agents, are capable of going beyond the exclusive pursuit of our respective well-being was strongly asserted by

why exactly is commitment important for rationality? 23 classical authors such as Adam Smith. Smith, in particular, distinguished actions motivated by ‘‘prudence’’ (taking into account indirect as well as direct advantages) from those influenced by ‘‘sympathy,’’ ‘‘generosity,’’ and ‘‘public spirit,’’ to each of which he gave a distinct and differentiated role.14 The possibility of such ethical motivation was explicitly accepted also by the pioneers of utility theory in economics (such as Mill, Jevons, Edgeworth, and Marshall), though this issue is rarely addressed in modern economic theory. To use a medieval distinction, we are not only ‘‘patients’’ whose needs demand attention, but also ‘‘agents’’ whose freedom to decide what to value and how to pursue what we value can extend far beyond the fulfillment of their needs. A fuller understanding of this issue can, for example, help us to understand better the role of environmental priorities that we recommend or follow. To illustrate, consider our sense of responsibility toward the future of other species, not merely because—nor only to the extent that—the presence of these species enhances our own living standards. For example, a person may judge that we ought to do what we can to ensure the preservation of some threatened animal species, say, spotted owls of some specific kind. There would be no contradiction if the person were to say: ‘‘Our living standards would be largely—or completely—unaffected by the presence or absence of spotted owls, but I do strongly believe that we should not let those owls become extinct, for reasons that have nothing much to do with human living standards (not to mention my own living standards).’’ If such reasoning is permitted within the bounds of rational ethics, as I believe it must be, it has implications for many practical issues, such as the characterization of ‘‘sustainable development.’’15 The need for concerted action for ‘‘our common future’’ was powerfully outlined about a decade and a half ago in a pioneering manifesto, prepared by the World Commission on Environment and Development, led by Gro Brundtland (formerly the prime minister of Norway and later the director general of the World Health Organization (WHO)).16 The Brundtland Report defined sustainable development as meeting ‘‘the needs of the present without compromising the ability of future generations to meet their own needs.’’ Brundtland’s concept of sustainability has been further refined and elegantly extended by one of the foremost economists of our time, Robert

24

amartya sen

Solow, in a monograph called An Almost Practical Step Toward Sustainability.17 Solow’s formulation sees sustainability as the requirement that the next generation must be left with ‘‘whatever it takes to achieve a standard of living at least as good as our own and to look after their next generation similarly.’’ The Solow formulation of sustainability has several attractive features. First, by focusing on sustaining living standards (seen as providing the motivation for environmental preservation), Solow gives more concreteness to Brundtland’s concentration on the fulfillment of needs. Second, in Solow’s neatly recursive formulation, the interests of all the future generations receive attention through provisions to be made by each generation for its successor. The generational coverage is, thus, comprehensive. But does the Solow reformulation of Brundtland’s idea of sustainability incorporate an adequately broad view of humanity? While the concentration on maintaining living standards has some clear merits (there is something deeply appealing in Solow’s formula about trying to make sure that the future generations can ‘‘achieve a standard of living at least as good as our own’’), it can still be asked whether the coverage of living standards is adequately inclusive. In particular, sustaining living standards is not the same thing as sustaining people‘s freedom to have—or safeguard—what they value and to which they have reason to attach importance. Our reason for valuing particular opportunities need not always lie in their contribution to our living standards. Sustainable development has become the ruling theme in much of the environmental literature. The idea has also inspired some significant international protocols for concerted action, for example reducing harmful emissions and other sources of planetary pollution. This is a great achievement, even though it is in danger of some neglect right now given the evident lack of engagement of the Bush Administration in the United States in these priorities even in already accepted obligations. Despite the adversities, the idea of sustainable development is here to stay, and will continue to be used in the future. It is, therefore, particularly important to make the concept do justice to the reach of reasoning involving commitment as well as sympathy. Gautama Buddha used an analogy to make a point that has a direct bearing on the distinction involved. He argued in Sutta Nipata that since we are enormously more powerful than the other species, we have some responsibility toward other species that links with this asymmetry of power.

why exactly is commitment important for rationality? 25 Buddha went on to illustrate the point by an analogy with the responsibility of the mother toward her child, not because she has given birth to the child (that connection is not invoked in this particular argument—there is room for it elsewhere), but because she can do things to influence the child’s life, positively or negatively, that the child itself cannot do. The reason for looking after the children, in this line of reasoning, is not our own standard of living (even though that too will almost certainly be affected), but the responsibility we should acknowledge precisely because of our power. We can have many reasons for our conservational efforts—not all of which are parasitic on our own living standards and some of which turn precisely on our sense of values and of fiduciary responsibility.18 The role of commitment in the ethics of behavior can be investigated in nearly every field of practical reason. The illustration with environmental reasoning and conduct is only one example. The importance of understanding the role of commitment in rationality applies to all behavioral ethics in general, as it does also to the conceptual clarification of the very idea of rationality and the investigation of its implication for predicting behavior and actions. There might be some answer in all this, I would submit, to the question asked in the title of the paper. Notes

1. Paper presented at a Workshop on Rationality and Commitment at the University of St Gallen, May 13–15, 2004. 2. For wide-ranging critiques of this approach from different perspectives, see Anderson (1993), Zamagni (1995), Hausman and McPherson (1996), Ben-Ner and Putterman (1998). 3. Sen (1985). On related issues, see also Frey (1992). 4. See Arrow (1951); Debreu (1959). 5. Sen (1977). 6. Sen, Choice, Welfare and Measurement (1982: 7–8). 7. Becker (1976, 1996). See also his (1981). 8. Jolls et al. (1998). I have discussed their critique in the introductory essay to Rationality and Freedom, Sen (2002). 9. On the difficulties of using standard and unmodified rational choice theory for behavioral explanation, see the pioneering work of George A. Akerlof (1984).

26 amartya sen 10. Among the many contributions made by the far-reaching empirical studies of Daniel Kahneman, P. Slovic, and Amos Tversky are their analyses of the observed departures of actual behavior from the demands of rational behavior. See Kahneman et al. (1982). 11. See Morishima (1982); Dore (1987); Aoki (1989); Ikegami (1995). 12. The Wall Street Journal, 30 January 1989, on page 1. 13. See Camera dei deputati (1993). In the process of allegedly advising the Commission, I learned much more than any enlightenment I could have offered to those fine Italian analysts, led by Senator Violante. 14. See Smith (1790: 187–93, that is, IV.2). 15. These implications I have tried to discuss elsewhere: see Sen (2004). 16. World Commission on Environment and Development (1987). 17. Solow (1992). 18. How rational choice can take full note of fiduciary responsibility, along with other features of committed behavior, is analyzed in my ‘‘Maximization and the act of choice’’ (1997).

References Akerlof, G. A. 1984. An Economic Theorist’s Book of Tales. Cambridge, UK: Cambridge University Press. Anderson, E. 1993. Value in Ethics and Economics. Cambridge, MA: Harvard University Press. Aoki, M. 1989. Information, Incentive and Bargaining in the Japanese Economy. Cambridge, UK: Cambridge University Press. Arrow, K. J. 1951. ‘‘An extension of the basic theorems of classical welfare economics.’’ In J. Neyman (ed.), Proceedings of the Second Berkeley Symposium of Mathematical Statistics. Berkeley: University of California Press, pp. 507–32. Becker, G. S. 1976. The Economic Approach to Human Behavior. Chicago, IL: University of Chicago Press. Becker, G. S. 1981. A Treatise on the Family. Cambridge, MA: Harvard University Press. Becker, G. S. 1996. Accounting for Tastes. Cambridge, MA: Harvard University Press. Ben-Ner, A. and L. Putterman (eds.). 1998. Economics, Value and Organization. Cambridge, UK: Cambridge University Press. Camera dei deputati. 1993. Economia e Criminalità. Rome. Proceedings of the May 1993 conference of the Anti-Mafia Commission of the Italian Parliament.

why exactly is commitment important for rationality? 27 Debreu, G. 1959. Theory of Value. New York: Wiley. Dore, R. 1987. Taking Japan Seriously: A Confucian Perspective on Leading Economic Issues. Stanford: Stanford University Press. Frey, B. 1992. ‘‘Tertium datur: pricing, regulation and intrinsic motivation.’’ Kyklos 45: 161–84. Hausman, D. M. and M. S. McPherson. 1996. Economic Analysis and Moral Philosophy. Cambridge, UK: Cambridge University Press. Ikegami, E. 1995. The Taming of the Samurai: Individualism and the Making of Modern Japan. Cambridge, MA: Harvard University Press. Jolls, C., C. Sunstein, and R. Thaler. 1998. ‘‘A behavioral approach to law and economics.’’ Stanford Law Review 50: 1471–1550. Kahneman, D., P. Slovic, and A. Tversky. 1982. Judgement Under Uncertainty: Heuristics and Biases. Cambridge, UK: Cambridge University Press. Morishima, M. 1982. Why has Japan ‘Succeeded’? Western Technology and Japanese Ethos. Cambridge, UK: Cambridge University Press. Sen, A. 1977. ‘‘Rational fools: a critique of the behavioral foundations of economic theory.’’ Philosophy and Public Affairs 6: 317–44. Sen, A. 1982. Choice, Welfare and Measurement. Oxford: Blackwell, pp. 7–8. Sen, A. 1985. ‘‘Goals, commitment and identity.’’ Journal of Law, Economics and Organization 1: 341–55. Sen, A. 1997. ‘‘Maximization and the act of choice.’’ Econometrica 65: 745–79. Sen, A. 2002. ‘‘Introduction: rationality and freedom.’’ In A. Sen (ed.), Rationality and Freedom. Cambridge, MA: Harvard University Press, pp. 26–37. Sen, A. 2004. ‘‘Why we should preserve the spotted owl.’’ The London Review of Books 26, 3 (February): 10–11. Smith, A. 1975 [1790]. The Theory of Moral Sentiments. Revised edition, reprinted, edited by D. D. Raphael and A. L. Macfie. Oxford: Clarendon Press. Solow, R. 1992. An Almost Practical Step Toward Sustainability. Washington, D.C.: Resources for the Future. World Commission on Environment and Development. 1987. Our Common Future (Brundtland Report). Oxford: Oxford University Press. Zamagni, S. (ed.). 1995. The Economics of Altruism. Aldershot: Elgar.

2 Construing Sen on Commitment P HILIP P ETTIT

One of Amartya Sen’s most distinctive claims, and perhaps also one of his most controversial, is that there is an altruistic attitude toward others that does not make sense within the terms of rational choice theory, however broadly that theory is construed. He describes this attitude as one of ‘‘commitment’’ to others, contrasting it with egoism and also with a distinct altruistic attitude he calls ‘‘sympathy.’’ It involves recognizing the goals of another and, regardless of whether or not this answers to independent goals of one’s own—regardless of whether or not one internalizes those goals in ‘‘sympathy’’ with the other—letting them impact on how one behaves. Committing oneself to another in this way is, on the face of it, a fairly common sort of exercise. Surely we often do take the goals of others into account in the manner proposed. We are bent on pursuit of our own ends, we discover that acting as they require will frustrate someone else’s projects, and then we pause, take stock, and adjust what we were going to do so that the other will not suffer unduly at our hands. But though commitment is phenomenologically plausible in this way, it looks to be architecturally problematic. On Sen’s conceptualization, it characteristically involves putting aside one’s own goals and acting on those of another. But how could one ever fail to act on one’s own goals? The idea offends against a picture of human psychology accepted on almost all sides, whether among professionals or among the folk. This paper attempts to expound Sen’s view, especially as he has presented it in the most recent statements that he has given or reprinted (Sen 2002); and then to explain why he is led to endorse a position that offends, in formulation if not in substance, against our common sense about action.

construing sen on commitment 29 There are two parts to the paper. The first is devoted to the expository task, the second to the explanatory one.

1. Expounding Sen 1.1. Varieties of rational choice theory The problem that leads Sen to make a distinction between sympathy and commitment arises within the context of rational choice theory, as he himself calls it. On a minimal understanding, rational choice theory amounts to nothing more than the claim that human behavior—as it materializes, presumably, in the absence of ‘‘perturbing’’ factors and within ‘‘feasible’’ limits—is regular enough to be representable as ‘‘maximizing behavior with an identifiable maximand’’ (Sen 2002: 30). Maximizing behavior will never lead to choices that are seen as worse than available alternatives, though it does not require the choice of the ‘‘best’’ option, as in ‘‘optimization’’; no ‘‘best’’ option may be definable due to incompleteness in the agent’s preference-ordering (Sen 2002: 181).1 Sen contrasts with this minimal understanding of rational choice theory the interpretation under which the maximand is ‘‘interpretable as the self-interest of the person.’’ Such self-interest, he says, may in turn be understood in either of two ways: in a ‘‘narrowly self-centered’’ sense or in the sense in which my self-interest may encompass the welfare of others, being sensitive to how certain others fare, not just to how well I do myself (Sen 2002: 30–1). The distinction between the two ways of understanding rational choice theory corresponds to a distinction between two ways of understanding utility in the standard theory of preference and choice. In the one approach, corresponding to the minimal version of rational choice theory, ‘‘ ‘utility’ is simply defined as the maximand that a person can be seen as promoting’’; in the other, corresponding to the self-interest versions of the theory, ‘‘ ‘utility’ is used as also representing the person’s self-interest or wellbeing’’ (Sen 2002: 27). The distinction between different ways of understanding rational choice theory also maps onto an account that Sen gives of how the self may be implicated in choice. To believe in the minimal version of the theory

30

philip pettit

is simply to believe that one’s choices are based on one’s own goals; it is to believe in what he calls ‘‘self-goal choice.’’ To believe in the other version is to hold, much more demandingly, that one’s choices are based on the goal of maximizing one’s own welfare—one’s self-interest—whether that be understood narrowly or broadly: whether it be understood just as welfare, no matter how affected, or as self-centered welfare (Sen 2002: 33–4, 213–14).

1.2. Sympathy and commitment With these distinctions in place, Sen characterizes two forms of altruism: sympathy and commitment. Sympathy materializes when one pursues one’s own self-interest but that self-interest is positively sensitive to the welfare of others, say because one is a generous, affectively responsive sort of agent. It stands opposed to antipathy, which occurs when one’s self-interest is negatively sensitive to the welfare of others, say because one is a begrudging, envious type of person, or because one is feeling resentful. ‘‘Sympathy (including antipathy when it is negative) refers to ‘one person’s welfare being affected by the state of others’ (e.g., feeling depressed at the sight of the misery of others)’’ (Sen 2002: 35; cf. 214). Sympathy comes about via enlarging self-interest to the point where it encompasses others too. Thus Sen (2002: 177) says: ‘‘Altruism through sympathy is ultimately self-interested benevolence.’’ It is not self-interested in the sense of being pursued with an instrumental eye to securing some personal benefit. It is self-interested in the sense that the person we favor is someone whose welfare matters to us, intuitively, in the same manner as our own; let them fare well and we feel good, let them fare badly and we feel bad. Altruism through commitment is meant to contrast with altruism through sympathy. The core idea in commitment, endorsed by Sen (2002: 214) over many years, is that it breaks ‘‘the tight link between individual welfare (with or without sympathy) and the choice of action.’’ For example, it may involve ‘‘acting to help remove some misery even though one personally does not suffer from it.’’ Sen (2002: 177–8) traces the idea of commitment to Ragnar Frisch but finds it already present in Adam Smith. ‘‘Doing things for others through commitment may require one to ‘sacrifice some great and important interest of our own’, as Adam Smith puts it in distinguishing ‘generosity’ from ‘sympathy’.’’

construing sen on commitment 31 Where sympathy enlarges a person’s self-interest, commitment transcends it. Where sympathy transforms the motor of self-interest, tuning it to the welfare of others, commitment puts another motor in its place. In sympathy, so the idea goes, one’s sentiments resonate in common with others so that in acting on the basis of those sentiments one naturally takes account of the welfare of others. In commitment one does not resonate in that manner, or at least one may not do so. Rather, what happens is that while one’s sentiments push one in this or that direction, the recognition that others will suffer as a result of going in that direction causes one to alter trajectory. Without relying on the warm stirrings of fellow-feeling, the cold, clear light of reason leads one to change tack. 1.3. Two kinds of commitment How exactly does this happen, according to Sen? He identifies two possibilities and I will describe these respectively as goal-modifying and goal-displacing commitment. He is quite clear about the distinction but does not give the two forms explicit names (Sen 2002: 35, 214). Goal-modifying commitment occurs when I recognize the goals of others, see that they will be negatively affected by what I am about to do, and alter my own goals as a result; in particular, alter those goals without undergoing the transformation of sentiment that sympathy would involve. My modified goals will reflect, not self-interest, however enlarged, but rather ‘‘broader values’’ that bear on how others are to be treated or how common goals are to be promoted. There is no naturalistic mystery involved in thinking that values of this kind may matter to us, according to Sen (2002: 25), since selectional pressures are liable to have induced in us a concern for such things. And it is plausible in any case, he says, that the capacity for ‘‘reasoning and self-scrutiny’’—the capacity with which Sen (2002: 4) ultimately identifies rationality—can transform our goals in this way. Our choices need not relentlessly follow our experiences of consumption or welfare, or simply translate perceived goals into action. We can ask what we want to do and how, and in that context also examine what we should want and how. We might or might not be moved by moral concerns or by social reasons, but neither are we prohibited from entertaining these questions, in shaping our values and if necessary revising our objectives in that light. (Sen 2002: 36)

32

philip pettit

The acknowledgment of goal-modifying commitment does not require a serious departure from rational choice theory, at least on the minimal interpretation of that theory. Even as I become committed in this way, I will conform to rational choice theory in the minimal sense. I will continue to promote my modified goals—I will maximize in the familiar pattern (Sen 2002: 37)—though the goals I come to serve will no longer be the goals of self-interest, even enlarged self-interest. I may not maximize explicitly, since some of the values by which I am moved will take the form of constraints on how to behave (Sen 2002: 214). But acting on such constraints will still count as maximizing so far as the constraints can be ‘‘incorporated into a suitably broadened maximand’’ (Sen 2002: 41).2 Goal-modifying commitment is less radical in this respect, however, than the second, goal-displacing variety of commitment. In this variety, which bulks larger in Sen’s discussions, the recognition of the goals of others does not lead me to modify my own goals but rather to displace them. It leads me to take my guidance, not just from my own aims, but also from the goals that I see those others espouse. Commitment, as Sen (2002: 35) says, ‘‘can alter the person’s reasoned choice through a recognition of other people’s goals beyond the extent to which other people’s goals get incorporated within one’s own goals.’’ Whereas goal-modifying commitment does not require a departure from rational choice theory, as minimally understood, goal-displacing commitment certainly does. According to the minimal version of rational choice theory, people can be represented in action as maximizing an identifiable maximand, or as acting on their own goals: satisfying the assumption, as Sen calls it, of ‘‘self-goal choice.’’ But Sen (2002: 215) maintains that people may be committed to others in such a way that they no longer act in this way on their own goals; ‘‘the pursuit of private goals may well be compromised by the consideration of the goals of others.’’ People may become the executors of a goal-system that outruns the private goals that they endorse in their own name: a goal-system that makes place for the goals of others or for the goals of groupings in which people cooperate with others. This claim is highly implausible, at least on the face of it. Rational choice theory in the minimal sense is close to common sense. It picks up the assumption that when we act intentionally, then we try to advance

construing sen on commitment 33 certain goals in a way that is sensible in light of the apparent facts (Pettit 2002: Essay II.2). The claim that we can be the executors of a goal-system that outruns our own goals is bound to raise a question. Sen (2002: 214), indeed, acknowledges the fact. ‘‘It might appear that if I were to pursue anything other than what I see as my own ‘goals’, then I am suffering from an illusion; these other things are my goals, contrary to what I believe.’’ Sen’s response to this problem is characteristically imaginative, though I don’t think it is very satisfying. He says that apart from acting in a way that reflects my real goals I can also act in a way that reflects imagined or as-if goals; I can act as if the goals of others, or the goals of a group to which I belong, are my own. One context in which this occurs is when I assume the responsibility of a trustee, as it were, seeing myself as charged with furthering the other’s interests (Sen 2002: 179). Another is that in which I see myself as the representative of a group, charged with doing as well as possible by its interests. People, Sen (2002: 214) thinks, might use this representative identity to get out of game-theoretic predicaments where acting on their separate preference orderings would lead to a bad result for each: ‘‘if people are ready to act (individualistically) on the basis of some ‘as if’—more cohesive—orderings, then they can do better than acting individualistically in direct pursuit of their real goals.’’ In stressing these possibilities, Sen suggests that apart from sympathizing with others and escaping from a self-centered version of self-interest, people can also ‘‘identify’’ with one another—adopt the goals they share with others on an as-if basis—and achieve a much more radical liberation. They can escape, not just the rule of self-interest, however sympathetically enlarged, but even the rule of their own private goals. As Sen (2002: 216) himself puts it, ‘‘the sense of identity takes the form of partly disconnecting a person’s choice of actions from the pursuit of self-goal.’’ He believes that this disconnection occurs on a more or less routine basis in any society: One of the ways in which the sense of identity can operate is through making members of a community accept certain rules of conduct as part of obligatory behavior toward others in the community. It is not a matter of asking each time, What do I get out of it? How are my own goals furthered in this way?, but of taking for granted the case for certain patterns of behavior toward others.

34

philip pettit

1.4. The problem with goal-displacing commitment The notion of goal-displacing commitment remains problematic, despite Sen’s attempts to make it plausible by reference to as-if goals and the idea of identifying with another. First of all, I see no real difficulty in making sense of what it is to act in a representative or trustee role—or in any role of the kind—within the terms of self-goal choice. And, second, I find it hard to see how one can seriously envisage giving up on the idea of self-goal choice that goal-displacing commitment is supposed to violate. I am myself quite well disposed toward the idea that we each often act in the name of goals that are endorsed in common with others, as when we represent a group; I hold indeed that we may even have to act on the basis of judgments that we do not ourselves support (Pettit 2003). But the goals endorsed in common with others are still goals we each endorse, and so are in that intuitive sense our individually endorsed goals. I individually have it as a goal that we do so and so, you have that as an individually endorsed goal, and so has each of us in the group. True, I will not be able to realize my individually endorsed goal that we do so and so without the help of others. But that does not mean that it is not my goal. We might as well say that because my success in an archery competition depends on the wind, the goal of hitting the bull’s eye is not my individually endorsed goal when I take aim at the target. What is true in the case of acting for the goals of a group, in the role of representative, is also true of acting for the goals of another as a trustee or advocate or good Samaritan, or acting for a goal that represents a compromise between the other’s goals and my pre-existing objectives. The goals adopted do not mainline my mind, as it were, and take possession of it. Rather they become goals that I take over as my own, even if they are goals such that, like hitting the target in the archery competition, I am not entirely in control of their realization; their realization may depend equally on how the other behaves. These lines of thought suggest that acting as a representative or a trustee or whatever can make perfect sense within rational choice theory, minimally interpreted; it can be a sort of action that fully respects self-goal choice. That alone gives reason to be surprised at the position Sen takes. But what really makes his position quite problematic is that it is very hard to see how one can give up on the idea of self-goal choice that rational

construing sen on commitment 35 choice theory incorporates. For this idea is close to the core of the common-sense psychology of which minimal rational choice theory is an explication (Pettit 2002: Essay II.2). According to that shared folk psychology, intentional action is a form of behavior that is controlled by the agent’s desire to realize a certain condition—the desired condition will count as the agent’s goal—and by the agent’s beliefs about how best to do that. The goal represents the success condition of the action and will be discernible in how the agent is disposed to adjust the behavior as circumstances change and as different interventions are clearly needed for the realization of the condition. To imagine an action that is not controlled by a goal of the agent, by the lights of this approach, will be like trying to imagine the grin on the Cheshire cat in the absence of the cat itself. Let the agent not have a goal and it becomes entirely obscure how the agent can be said to act; to act, or at least to act intentionally, is to act with a view to realizing a goal. When Sen alleges that goal-displacing commitment takes the agent beyond the control of his or her goals, then he is setting himself against our basic common sense about action. What rational choice theory asserts on the minimal interpretation is that rational agents act out of a concern for maximizing the expected realization of their goals. And that is precisely to argue, as in our common-sense psychology, that rational agents aim at satisfying or fulfilling their desires, according to their beliefs. It amounts to nothing more or less than asserting the soundness of the belief–desire schema. Why might Sen be led to question the soundness of that very basic schema? What limitation does he find in the rational-choice representation of action, even when minimally interpreted, that prompts him to think that it is not a comprehensive picture of human decision-making? This is the question to which the remainder of my paper will be devoted.

2. Explaining Sen In discussing the requirements of rationality, and in making room for departures from the narrower assumptions of economics, Sen often talks of answering to the demands of reason, subjecting our choices to reasoned scrutiny, being guided by broader values and objectives, and the like (Sen

36

philip pettit

2002: ch. 1). All of this suggests that in his view the belief–desire schema of folk psychology—equivalently, the goal-maximization schema of rational choice theory—is too restrictive to give us the full truth about human decision-making. It does not do justice to the diversity of ‘‘reasons for what one may sensibly choose’’ (Sen 2002: 6). That is why he is led to recognize the possibility of a goal-displacing commitment in which, as it seems, people altogether transcend the range of possibilities envisaged in folk psychology. There are three possible explanations, I think, as to why Sen might take the view that folk psychology is over-restrictive in this way. Quite independently of how plausible they are as explanations of his line—I shall be suggesting that the first two are implausible—they are of some interest in themselves: they point to different ways in which folk psychology and rational choice theory may be represented and, I would say, misunderstood. I review them, turn by turn, in this second part of the paper. The three explanations are linked respectively with three theses, and my discussion will focus on these. 1. The no-deliberation thesis. The schema of belief and desire—and the rational choice theory that seeks to explicate it (Pettit 2002: Essay II.2)—does not make room for deliberation at all. Thus, when one operates in accord with that schema one cannot be deliberating and when one deliberates—assuming deliberation occurs—one cannot be acting in accord with that schema. 2. The selfish-deliberation thesis. The schema does make room for deliberation but the only deliberation accommodated is of an inherently selfish variety. Thus, when one operates in accord with that schema one can be deliberating but only in the selfish manner; and when one deliberates non-selfishly—assuming this sometimes occurs—one cannot be acting in accord with that schema. 3. The integrated-deliberation thesis. The schema allows for non-selfish deliberation, but only deliberation from a limited basis: that of goals that the agent has internalized and integrated into a standing structure. Thus, when one operates in accord with the schema one can only be deliberating on that limited, integrated basis; when one deliberates otherwise—if this ever happens—one cannot be acting in accord with the schema.

construing sen on commitment 37 2.1. The no-deliberation account According to the belief–desire schema, rational agents act so as to promote the satisfaction of their desires according to their beliefs; this is the fundamental idea preserved in rational choice theory, minimally interpreted. But one striking fact about the schema, so understood, is that it does not require anything on the part of agents that we might be inclined to call deliberation. For all that the picture says, desires are goal-seeking states, beliefs factconstruing states, and what happens in action is that they combine to produce the behavior in question. But they can combine to do this without any extra intentional activity—driven in turn by beliefs and desires—of deliberating over whether the goals are appropriately sought, the facts appropriately construed, or the behavior appropriately selected in light of those beliefs and desires. The goal-driven, fact-sensitive creature may be a more or less autonomic mechanism or organism. It is for this reason that, on most accounts, folk psychology is taken to apply to non-human animals as well as to human; in particular, to non-human animals who show no signs of being able to reason. Under this picture, non-human animals are tuned by evolutionary and experiential pressures so that in appropriate circumstances they will act for the realization of certain goals and, in particular, will act in a manner that makes sense under the way they take the facts to be: under the representations of the environment—the more or less reliable representations—that their perceptions and memories evoke. Such animals will instantiate goal-seeking and fact-construing states and those states will interact in such a way as to produce suitable behavior. The animals will be rational agents in the sense of conforming to the minimal version of rational choice theory. Or that will be so, at any rate, in the absence of intuitively perturbing influences, within intuitively feasible limits: for short, in normal conditions. But if folk psychology is as likely to be true of various non-human animals as it is of creatures like us, there is still a yawning divide between how we and they manage to conform to this psychology (Pettit and Smith 1990). We do not just possess beliefs and desires in the manner of nonhumans, and act as those states require. We can give linguistic expression to the contents of many of those states—we can articulate the goals sought and the facts assumed. We can form higher-order beliefs about those goals

38

philip pettit

and facts; beliefs, for example, to the effect that certain forms of consistency or coherence or mutual support do or do not obtain among them. And we can seek out such higher-order beliefs with a view to maximizing the checks on the overall pattern of attitudes that is going to unfold within us (McGeer and Pettit 2002). The exercise we undertake in seeking out higher-order beliefs with the aim of increasing the checks on our overall attitudes is easily illustrated. Suppose I find myself prompted by perception to take it to be the case that p, where I already take it to be the case that r. While my psychology may serve me well in this process, it may also fail; it may lead me to believe that p, where p is inconsistent with r. But imagine that in the course of forming the perceptual belief I simultaneously ask myself what I should believe at a higher-order level about the candidate fact that p and about the other candidate facts I already believe. If I do that then I will put myself in a position, assuming my psychology is working well, to notice that the alleged fact that p and the alleged fact that r are inconsistent, and so my belief-forming process will be forced to satisfy the extra check of being squared with this higher-order belief—a crucial one, as it turns out—before settling down.3 In this example, I search out a higher-order belief that is relevant to my fact-construing processes and imposes a further constraint on where they lead. But higher-order beliefs, for example higher-order beliefs about the consistency of various propositions or scenarios, can also impose constraints on my goal-seeking states, since it will not make sense to set out to realize simultaneously two inconsistent goals. And by the same token they may impose constraints on combinations, not just of beliefs with beliefs, and of desires with desires, but also of beliefs with desires, and of beliefs or desires with actions. With these points made, I can introduce what I mean by the activity of ‘‘deliberation’’ or ‘‘reasoning’’ or ‘‘ratiocination.’’ Deliberation is the enterprise of seeking out higher-order beliefs with a view to imposing further checks on one’s fact-construing, goal-seeking, and of course decision-making processes. Not only do we human beings show ourselves to be rational agents, as we seek goals, construe facts, and perform actions in an appropriate fashion. We also often deliberate about what goals we should seek, about how we should construe the facts in the light of which we seek them, and about how therefore we should go about that pursuit:

construing sen on commitment 39 about what opportunities we should exploit, what means we should adopt, and so on. We do this when we try to ensure that we will form suitably constraining higher-order beliefs about the connections between candidate goals and candidate facts. The fact that we human beings reason or deliberate in this sense—the fact, in Sen’s (2002: 40) language, that we conduct ‘‘reasoned scrutiny’’ of our beliefs and desires—means that not only can we be moved in the manner of unreasoning animals by goal-seeking and fact-construing states, such as the belief that p or the desire that q. We can also reflect on the fact, as we believe it to be, that p, asking if this is indeed something we should believe. And we can reflect on the goal we seek, that q, asking if this is indeed something that we should pursue. We will interrogate the fact believed in the light of other facts that we believe, or other facts that perceptions and the like incline us to believe, or other facts that we are in a position to inform ourselves about; a pressing question, for example, will be whether or not it is consistent with them. We may interrogate the goal on a similar basis, since the facts we believe determine what it makes sense for us to pursue. Or we may interrogate it in the light of other goals that also appeal to us; in this case, as in the case of belief, a pressing question will be whether or not it is consistent with such rival aims. Nor is this all. Apart from drawing on deliberation to interrogate the facts we take to be the case, and the goals we seek, we can ask after what actions or other responses we ought to adopt in virtue of those facts and goals. Not only can we ask after whether they give us a reliable position at which to stand; we can ask after where they would lead us, whether in espousing further facts or goals, or in resorting to action. We may be rationally led in the manner of non-human animals, for example, to perform a given action as a result of taking the facts to be thus and so and treating such and such as a goal. But we can also reason or deliberate our way to that action—we can reinforce our rational inclination with a deliberative endorsement—by arguing that the facts, as we take them to be, are thus and so, the goals such and such, and that this makes one or another option the course of action to take; it provides support for that response. The first possible explanation of why Sen thinks that the belief–desire schema is overly restrictive could be that he thinks deliberation gives the lie to that schema. It does not represent a means, as in the picture I have just sketched, whereby we human beings might hope to discipline the

40

philip pettit

process of belief-formation, desire-formation, and action-selection, using higher-order beliefs as a source of extra checks on that process. Rather, it represents a whole other enterprise: the ‘‘disciplined use of reasoning and reasoned scrutiny’’ (Sen 2002: 19). And this is an enterprise, so the explanation would go, that transcends entirely the regime described in folk psychology and rational choice theory. Let people instantiate that regime in an act of decision-making, mechanically forming and acting on beliefs and desires, and they cannot be deliberating their way to action. Let them deliberate their way to action and they cannot be instantiating the belief–desire regime; they must be operating under a distinct pilot. But while it is just possible to think that Sen endorses the no-deliberation view of folk psychology, I do not think that this is at all plausible. The thesis is patently false, by my lights, and Sen nowhere shows signs of endorsing it. He never suggests that resort to deliberation and reasoning is inconsistent with our continuing to behave as rational creatures: that is, creatures ‘‘whose behavior is regular enough to allow it to be seen as maximizing’’ (Sen 2002: 30). On the contrary, he envisages deliberation or rational scrutiny as serving often to shape the operation of rational decision-making without disturbing the basic belief–desire regime. The picture in goal-modifying commitment, for example, is that reflection on the needs of others, or something of the kind, can cause us to change the goals we pursue. Such commitment affects how we rationally maximize, it does not undermine maximization itself.

2.2. The selfish-deliberation account Is there any other reason, then, why Sen might be led to countenance the problematic, goal-displacing attitude? A second possibility is worth registering for its inherent interest though, again, I do not think it is likely to be the consideration that moves Sen. The idea here is that the only sort of deliberation that can lead to action, under the belief–desire picture, is deliberation of a self-serving variety. The proposal is not that deliberation has regard only to the agent’s private welfare, whether or not this is sensitive to the welfare of others. Rather it is that the very logic of deliberation in goal-seeking or desire-driven creatures ensures that it has a certain self-serving character. Did Sen accept this sort of view, then it too would explain why he might want to insist that something more is often

construing sen on commitment 41 possible: namely, a sort of committed action that escapes the confines of the belief–desire schema. Suppose that my holding by a certain desire, say that q, makes it rational for me, given the beliefs I hold, to form a further desire or perform a certain action: say to desire that s or to perform B, where each represents an essential means of satisfying the desire that q. And now imagine that I am reflecting on whether there is a deliberative reason why I should hold by that extra desire or perform that particular action. Should I think ‘‘I desire that q; so therefore I should desire that s. I desire that q; so therefore I should perform B’’? Or will that reasoning leave me with the following question. ‘‘Fine, but should I desire that q? Fine, but does this really give me a reason for desiring that s or for performing B?’’ I think it is clear that the formula offered will indeed leave me with that question. For we all allow that our desires may not be well formed—any more than our beliefs—and that we may not always have a reason for responding as they require (Broome 2004). Some desires we naturally regard as pathological, others as the products of a weak will, others as due to a lack of imagination or memory, and so on; pathologies of desire abound. This being so, we cannot think that the proper ratiocinative endorsement for acting on a given desire should simply start from the existence of that desire, putting it into the foreground of deliberation, as if it were something sacred and beyond question. But if we shouldn’t deliberate from the fact of what we happen to desire, what should we take as our point of deliberative departure in decisionmaking (Pettit and Smith 1990)? Presumably, we should enrich the base of deliberation to stipulate that it would be good or desirable or appropriate or whatever to bring about what is desired; holding that this would be good is precisely a way of insisting that the desire is not pathological or wayward in the fashion contemplated. Thus the form of the deliberation that will reasonably take me from what I desire to what else I should desire, or to what I should do, will be: ‘‘It would be desirable to satisfy the desire that q; so therefore I should desire that s. It would be desirable to satisfy the desire that q; so therefore I should perform B.’’ So far so good. But now it is time to notice a clear ambiguity in the formula just offered. What does it mean to think that it would be desirable to satisfy a certain desire? There are three distinctively different readings. An objective reading would say: it means that it would be desirable to

42

philip pettit

fulfill the desire, making the world conform to it, i.e., bringing it about that q. A subjective reading would say: it means that it would be desirable to relieve the desire that q, removing and having the pleasure of satiating the itch that it constitutes, whether or not this means bringing it about that q. And a conjunctive reading would say: it means that it would be desirable to relieve-and-fulfill the desire, with both elements present. The selfish-deliberation thesis, which I can at last introduce, holds that every form of practical deliberation that leads from desire or preference to action always involves one or other of the relief-formulas: that associated with the subjective reading or that associated with the conjunctive. The idea is that when agents who continue to operate in the belief–desire mold deliberate their way to action—or to the formation of an instrumental desire—they reason from the recognition of having this or that desire within themselves and from the desirability of relieving (and perhaps also fulfilling) this desire. Why call this a selfish-deliberation thesis? Well, consider how the process envisaged contrasts with how people will reason if they start from the objective reading mentioned earlier. Under that reading, they will ground their deliberation in the fact that it would be desirable to fulfill a desire that q. This is equivalent to grounding their deliberation, more simply, in the fact that it would be desirable to bring it about that q. The formulations are equivalent, because the fact that it would be desirable to fulfill a certain desire—as distinct from the fact that it would be desirable to relieve a certain desire—does not entail that one currently has that desire.4 Thus people who deliberate in the manner of the objective reading will argue from the desirability of the envisaged state of affairs—that q—without any essential reference to their own state of desire. They will represent that scenario as worth trying to bring about, quite independently of the fact that bringing it about would answer to a desire they themselves happen to harbor. The selfish-deliberation thesis suggests, in contrast to this picture, that all desire-based deliberation represents a prospect as desirable only if it promises to give the agent a distinctive sort of return: to relieve the desire or desires that he or she has for it. The thesis treats all desires as if they were urges or yens, hankerings or yearnings, or pangs: that is to say, itches that exist in consciousness and that naturally demand to be fulfilled and thereby relieved, with all the pleasure that relief would bring (Pettit and Smith 1990).

construing sen on commitment 43 If Sen thought that all deliberation within the belief–desire schema—all deliberation that is consistent with the rational-choice representation of human beings—had to be selfish in this manner, then we could make immediate sense of why he should think that there are ways of acting that escape the shackles of that schema. There certainly are forms of deliberation—deliberation based on the objective reading of the desire formula, I would say—that are not selfish. And if the only way of countenancing these were to say that human beings sometimes act in committed ways that are not representable within the belief–desire schema, then that might be a reasonable thing to say. But though adherence to the selfish-deliberation thesis, like adherence to the no-deliberation thesis, would explain Sen’s line on goal-displacing commitment, there is little or no evidence that he endorses it. True, the economic way of positing preference-satisfaction as the supreme good in social policy suggests that people are assumed to seek the relief of their preferences, not just their fulfillment, and it may seem to presuppose the selfish-deliberation thesis. But that mode of thought is one that Sen above all others has been vociferous in questioning (Sen 1982). It is very unlikely that he himself should be moved by it.

2.3. The integrated-deliberation account The no-deliberation account of Sen’s line says that so far as agents remain in the belief–desire mold—so far, therefore, as they continue to be representable within rational choice theory—they cannot deliberate at all; hence the acknowledgment of the presence of deliberation requires denying that that schema applies to all human decision-making. The selfish-deliberation account of the line holds that so far as agents remain in that belief-desire mold, they cannot deliberate in a non-selfish way; hence the recognition of non-selfish deliberation requires denying, once again, that the schema applies to all forms of action. The integrated-deliberation account of Sen’s line, to which I now turn, maintains that so far as agents remain in the belief–desire mold, and continue to be representable in rational choice theory, they cannot deliberate on the basis of reasons other than those that derive from standing goals that form an integrated system; hence the recognition that people are not just the servants of such standing goals also requires denying that the schema governs everything they do.

44

philip pettit

Under the folk-psychological picture, every agent who performs an intentional action targets some goal, adjusting his or her behavior with a view to realizing that goal. I assumed in presenting that picture earlier that the goal which is targeted in action may be just about any state of affairs on which the agent might hope to have an influence. Putting the matter otherwise, I assumed that for all that was said, there is no limit on the things that an agent may primitively desire and no limit on the ways in which such desires can come and go. Consistently with the bare bones of folk psychology, the agent may mutate from moment to moment in respect of the primitive desires conceived and acted on.5 The integrated-deliberation thesis suggests that the assumption I have been making is mistaken. The idea is that the belief–desire schema, and in particular the rational choice theory that explicates it, posits an image of relatively stable, integrated agents who each come to choice, equipped with a set of dispositions to bring this, that or the other about, and so with a set of corresponding, standing goals. So far as agents act in fidelity to that schema and that theory, so the thesis goes, they will act in the service of those goals. The goals may be entirely benevolent or even self-sacrificing, of course, but they will bear the unmistakeable mark of being goals that belong to the agent involved: goals that have been integrated into the agent’s psychology and are truly his or her property. That the goals envisaged in the thesis are of a standing character does not mean that they have to be long-standing in time. Sen is quite happy to say that goal-modifying commitment can make for a change in someone’s goals, leading immediately to action that is intended to advance those newly minted aims. The integrated-deliberation thesis can fit perfectly well with this. What it says is merely that for any goal that one pursues in conformity with the belief–desire schema, the goal has to have been internalized in a distinct episode, however closely related in time; it has to have been ‘‘incorporated within one’s own goals’’ as an integrated part (Sen 2002: 35). The integrated-deliberation thesis will be vacuous, of course, if the internalization and integration of a goal amount to nothing more than the act of tracking that goal in action. But if they are meant to refer to a distinct psychological episode, one perhaps with a phenomenology of its own, then the thesis will be quite substantive. It will suggest that so far as people conform to the belief–desire schema, they will only act for the furtherance of goals that have been lodged firmly, if recently, within their psychological

construing sen on commitment 45 make-up. It will imply that so far as human beings register and respond to novel demands—demands unconnected with pre-existing goals—they will transcend the bounds of folk psychology and rational choice theory. It should be clear that if Sen endorsed the integrated-deliberation thesis then this would explain why he wants to deny that the belief–desire schema is comprehensive. The cases he invokes in relation to commitment typically involve people’s registering the goals of others and, without any incorporation or integration of those goals within their own goalsystem (Sen 2002: 35), acting with a view to furthering the alien goals, or to furthering some compromise between those goals and their own. Adherence to the integrated-deliberation thesis would make perfect sense of why he claims that in such instances people do not act for the realization of their goals but violate self-goal choice. It is fairly clear that Sen does not espouse either the no-deliberation thesis or the selfish-deliberation thesis and, given the dearth of alternative explanations for his line about goal-displacing commitment, that already makes a certain case for the thought that he must espouse the thesis of integrated-deliberation. The idea would be that by his lights rational choice theory, and the belief–desire psychology it explicates, presuppose an agent who moves from situation to situation with a pre-given set of goals, modifying those goals only occasionally. Hence, so the idea goes, Sen finds the approach incapable of making room for the extreme of commitment that involves people, not in adjusting their behavior in the light of their own standing goals, but in adjusting it in the light of the standing goals of others. According to this interpretation he believes that the way human beings respond to reasons is not limited in the manner allegedly presupposed by the belief–desire schema; he sees goal-displacing commitment as a perfect example of how they can transcend the boundaries of that approach. I do not myself think that the integrated-deliberation thesis is sound. Neither the belief–desire schema in itself, nor the rational choice theory that seeks to explicate it, has to suppose that the goals which agents try to advance are goals internalized or integrated in any substantive sense. As indicated above, the approach says nothing about how far people may vary in their primitive desires from moment to moment and from context to context. But if the integrated-deliberation thesis is false, can we ascribe it with any sense of assurance to Sen? I think that perhaps we can. The economic

46

philip pettit

and social-scientific application of rational choice theory naturally assumes that in principle we can fix relatively stable utility and probability functions on individuals; without that assumption the enterprise of explaining and predicting their behavior would be severely jeopardized. Perhaps it is not entirely implausible that the habit of thinking in this methodologically sensitized way should have led Sen to represent rational choice theory as involving an assumption that agents act on integrated goals that generally remain in place from occasion to occasion. This may be precisely what he has in mind when he speaks of the assumption of self-goal choice. What moves Sen to deny that rational choice theory offers a comprehensive depiction of agency may not be anything so esoteric, then, as the belief that satisfying rational choice is inconsistent with practicing deliberation or that it is consistent only with conducting selfish deliberation. It is perfectly intelligible why Sen might think of the theory as recognizing standing goals only. And if that is how he does think then it should be no surprise that his honesty in recognizing the range of people’s sensitivity to others would lead him to say that rational choice theory—and so, in effect, belief–desire psychology—does not capture the whole truth about how people deliberate. It should be no surprise that he comes to recognize and celebrate a variety of commitment that he takes to be undreamt of in rational choice circles.

3. Conclusion We saw in the first part of the paper that Sen takes an unusual line in arguing, contrary to some core assumptions of rational choice and common sense, that people sometimes act without acting on goals of their own; they act out of a form of goal-displacing commitment. The exploration of why one might take such a view has led us to identify three theses, any one of which would be sufficient to support Sen’s line. I have argued that all three theses are false, holding as I do that Sen is mistaken to think that there is a form of attitude that cannot be accommodated in the belief–desire way of thinking. But it is still interesting to see the grounds on which such theses might be maintained. And it is interesting, in particular, to see that Sen is most likely to have been influenced by the weakest of these three views:

construing sen on commitment 47 the thesis that rational choice only makes room for the form of deliberation in which one argues on the basis of one’s pre-existing, integrated goals. The integrated-deliberation account of why Sen should say what he does about commitment has one merit that I should mention in conclusion. It is a charitable account that does not find anything deeply amiss in his ideas. Consistently with that account, we might even reduce the charge against him to one of terminological infelicity. A perfectly good way of expressing the substance of the account is to say that, for Sen, the word ‘‘goal’’ means ‘‘integrated goal’’ and the name ‘‘rational choice theory’’ refers to a theory as to what it is rational for agents to do in the light of goals in that sense: that perfectly reasonable sense. This forces Sen into countenancing a range of committed behavior in which notions of goal and goal-based choice no longer apply, and it leads him into some quite romantic accounts of our capacity on this front. But once the differences of terminology are taken into account, this divergence from more standard models need not make for any great problem. It will count as an indication of idiosyncrasy, not a sign of intellectual oversight.6 Notes

1. Following Sen (2002: chs 3–4), I shall understand it in a broad sense that allows what he calls ‘‘menu-dependence’’ and related phenomena. Menu-dependence means that an action, A—say, taking an apple from a bowl of fruit offered by another—varies in its identity as an object of preference and choice, depending on context; it will be a polite action if there is an apple left, impolite if there is not (Pettit 1991). 2. The maximand will take account of menu-dependent preferences (Sen 2002: 178). 3. I abstract here from the crucial question of how we come to form concepts such as truth and consistency and the like, and how we come to be able to form the sophisticated beliefs mentioned in the text. For a little on this see McGeer and Pettit (2002). 4. I assume here that when I think that it would be desirable to fulfill the desire that q, I am thinking: it would be desirable to fulfill any desire that q, not that it would be desirable to fulfill this actual desire that q. The latter formula would suggest, counterintuitively, that what makes it desirable to bring about the fulfilling scenario depends on the

48

philip pettit

existence of the particular desire fulfilled. When I say in the text that the selfish-deliberation model presupposes that one of the relief formulas is relevant, I abstract also at that point from this counterintuitive possibility; it too would make deliberation ‘‘selfish.’’ 5. Continuity will be required only so far as it follows from the sort of continuity that unchanging evidence and information imposes on the agent’s beliefs; this will not constrain primitive desires, since they are conceived as nothing more or less than desires that are insensitive to the particular beliefs of the agent (Pettit 2005b). 6. This paper was transformed by a searching discussion in the St Gallen conference on rationality and commitment, May 2004. I am very grateful to the participants there. References Broome, J. 2004. ‘‘Reasons.’’ In R. Jay Wallace, Philip Pettit, Samuel Scheffler, and Michael Smith (eds.), Reason and Value: Themes from the Moral Philosophy of Joseph Raz. Oxford: Oxford University Press, pp. 28–55. McGeer, V. and P. Pettit. 2002. ‘‘The Self-Regulating Mind.’’ Language and Communication 22: 281–99. Pettit, P. 1991. ‘‘Decision Theory and Folk Psychology.’’ In M. Bacharach and S. Hurley (eds.), Essays in the Foundations of Decision Theory. Oxford: Blackwell; reprinted in P. Pettit (2002), pp. 192–221. Pettit, P. 2002. Rules, Reasons, and Norms: Selected Essays. Oxford: Oxford University Press. Pettit, P. 2003. ‘‘Groups with Minds of their Own.’’ In F. Schmitt (ed.), Socializing Metaphysics. Lanham: Rowan and Littlefield. Pettit, P. 2005a. ‘‘On Rule-Following, Folk Psychology, and the Economy of Esteem: A Reply to Boghossian, Dreier and Smith. Contribution to Symposium on P. Pettit, Rules, Reasons, and Norms.’’ Philosophical Studies 124/2: 233–59. Pettit, P. 2005b. ‘‘Preference, Deliberation and Satisfaction.’’ In S. Olsaretti (ed.), Preferences and Well-Being: Royal Institute of Philosophy Supplement 59. Cambridge: Cambridge University Press, pp. 131–53. Pettit, P. and M. Smith 1990. ‘‘Backgrounding Desire.’’ Philosophical Review 99: 565–92; reprinted in F. Jackson, P. Pettit, and M. Smith (eds.) 2004. Mind, Morality and Explanation. Oxford: Oxford University Press. Sen, A. 1982. Choice, Welfare and Measurement. Oxford: Blackwell. Sen, A. 2002. Rationality and Freedom. Cambridge, Mass.: Harvard University Press.

3 Sympathy, Commitment, and Preference DA NI E L M . H AUS M A N ∗

The recent opportunity to reread Amartya Sen’s many writings on preference and choice as a unit increased my appreciation of their depth and intricacy and of the wit and humanity of their author. It also made me realize that I had persistently misread them, mistakenly substituting my own notion of ‘‘preference’’ for Sen’s. While still very much in Sen’s camp in rejecting revealed preference theory and emphasizing the complexity, incompleteness, and context dependence of preference and the intellectual costs of supposing that all the factors influencing choice can be captured by a single notion of preference, I shall contest his view that economists should recognize multiple notions of preference. In my view, Sen’s concerns are better served by embracing a single conception of preference and insisting on the need for analysis of the multiple factors that determine ‘‘preference’’ so conceived. This paper will consequently combine exegesis of what Sen takes preference, sympathy, and commitment to be with an alternative conceptualization that is directed toward Sen’s own ends but is in some ways more conciliatory toward orthodox approaches.

1. Two concepts of preference Like most economists and philosophers, Sen offers no explicit definition of ‘‘preference.’’ Instead, he consistently emphasizes that economists have used the word ‘‘preference’’ to refer to many different things. Among these

50

daniel m. hausman

different concepts of preference, Sen believes that two are most important, and his discussion of preference, especially in his earlier works (Sen 1973, 1977) focuses on these. He writes: Certainly, there is no remarkable difficulty in simply defining preference as the underlying relation in terms of which individual choices can be explained; ... In this mathematical operation preference will simply be the binary representation of individual choice. The difficulty arises in interpreting preference thus defined as preference in the usual sense with the property that if a person prefers x to y then he must regard himself to be better off with x than with y. (1973: 67)

One definition of preference is as ‘‘the underlying relation in terms of which individual choices can be explained,’’ which Sen here identifies with ‘‘the binary representation of individual choice.’’ Call this concept of preference ‘‘choice ranking.’’ Second, there is what Sen labels ‘‘the usual sense’’ of preference. In this sense a person prefers x to y if and only if the person believes that he or she is better off with x than with y.1 Let us call this sense of preference ‘‘expected advantage ranking.’’ Unlike rankings, preferences are attitudes, which can be intense or weak, cool or emotional, but it will simplify this paper to identify preferences with the rankings they imply. Sen has argued again and again that these two notions of preference—these two rankings in terms of choice and expected advantage—should not be conflated. ... the normal use of the word permits the identification of preference with the concept of being better off, and at the same time it is not quite unnatural to define ‘‘preferred’’ as ‘‘chosen.’’ I have no strong views on the ‘‘correct’’ use of the word ‘‘preference,’’ and I would be satisfied as long as both uses are not simultaneously made, attempting an empirical assertion by virtue of two definitions. (Sen 1977: 329)

As this last quotation makes clear, Sen has consistently avoided legislating the meanings of words (see for example Sen 1991a: 588; 1991b). Rather than arguing for some canonical usage for the word ‘‘preference,’’ his message has been that economists need to recognize that the term has many meanings. Yet Sen has substantive objections to the theory of revealed preference and hence to the use of the choice-ranking concept of preference (1973, 1993). That leaves expected advantage as the main concept of preference, and, at least in his earlier works, when Sen uses

sympathy, commitment, and preference 51 the word ‘‘preference’’ without specifying its sense, one should probably interpret it as expected advantage. Sen often shies away from using the word, ‘‘preference,’’ employing instead terms such as ‘‘desires’’ or ‘‘goals.’’ Although expected advantage seems to be Sen’s default concept of preference, particularly in his earlier works, he never argues that this is what economists typically mean by ‘‘preference’’ or that it is what they should mean by preference. He emphasizes that in everyday usage, people have ‘‘preferences’’ over alternatives that are irrelevant to their interests or even contrary to their advantage. Actual usage is multiply ambiguous, and Sen is not prepared to argue for any regimentation of usage. His reason for counseling awareness of ambiguity rather than proposing a cure is, I think, that he fears that a regimentation will encourage an overly simple and one-dimensional view of evaluation and choice.

2. Three other concepts of preference People care about many things that do not bear on their own well-being, ‘‘interests,’’ or advantage, and people may prefer to sacrifice their interests or well-being in order to accomplish something that matters to them more. Such ‘‘preferences,’’ whose satisfaction does not serve self-interest, cannot be conceived of as rankings in terms of expected advantage. So far, the only alternative sense of preference we have considered is choice ranking. But the theory of revealed preference—the view that preference is the ranking of alternatives that is entailed by choice—rarely captures what economists mean by ‘‘preference,’’ let alone ordinary usage. To give some sense of the difficulties, consider, for example, my preference for a state of affairs in which there are no wars on earth in the twenty-second century over a state of affairs in which there are wars (and not because nuclear warfare will have reduced the planet to a cinder). Whether there is war in the twenty-second century or not does not bear on my well-being or self-interest. So one cannot read this as my belief that the absence of war in the twenty-second century will be better for me. But my preference does not appear to be a choice ranking either, because I cannot choose whether there will be wars in the twenty-second century. One might argue that this preference is implicit in my choice of virtually any candidate over George Bush for president in 2004, but presumably many of his misguided

52

daniel m. hausman

supporters have the same preferences as I do with respect to future peace. There is obviously no one-to-one relationship between this preference and any choices that I face. One response is to link preference to hypothetical choice. To say that I prefer that there be no wars in the twenty-second century is to say that if I could, I would choose that state of affairs over one with wars. Indeed, many of those who claim to be defenders of revealed-preference theory implicitly reject the behaviorist revealed-preference theory defended by Paul Samuelson or Ian Little in the 1930s, 1940s, and 1950s, and (misleadingly, in my opinion) regard a hypothetical-choice view of preference as itself a formulation of revealed preference theory. What is most important is that choice rankings not be confused with this third notion of preferences as the rankings implicit in hypothetical choices—not which theory is called ‘‘revealed-preference theory.’’ Although a hypothetical choice view of preference can cope with my preference that there not be wars in the next century, it is inadequate for many economic applications of the notion of preference. When modeling a strategic interaction among individuals as a game, theorists must assign utilities—indices of preference—to the outcomes of their interaction. But the ‘‘outcomes’’ which are the objects of preference are not alternatives among which a player can choose. (They are, in Sen’s terminology, ‘‘comprehensive outcomes’’ (1997b: 745)—paths through games including their results.) What people choose are strategies and these depend on both beliefs about what others will do and preferences as evaluations of outcomes, rather than preferences either in the sense of choice rankings or in the sense of rankings of hypothetical choices. For example, consider a traditional marriage proposal. This can be modeled as a game in which first Darcy proposes or not, and then Elizabeth accepts or refuses his proposal. Darcy prefers the outcome where he proposes and Elizabeth accepts to the outcome where he proposes and she refuses. But Darcy could not have a choice between these outcomes. The utility numbers in the game are not indices reflecting hypothetical choices. What other notions of preference are there? I agree with Sen that there are many, and I shall discuss two more. One of these is especially important, and it has a claim to be the central notion of preference in economics and decision theory. Economists should, I shall argue, be encouraged to regiment their language and reserve the word ‘‘preference’’ for this single

sympathy, commitment, and preference 53 usage. In contrast to Sen, I think there is a great deal to be said for prescribing how the word ‘‘preference’’ should be used in economics.2 This notion of ‘‘preference,’’ which I believe should be the only concept of preference employed in economics and decision theory, is a technical notion that does not conform to ordinary usage of the word ‘‘preference.’’ But it derives from a traditional folk-psychological view of human action. This traditional view holds that human actions can be explained and predicted by the beliefs and ‘‘desires’’ of agents. ‘‘Desire’’ in this context is a catch-all including a diverse array of motivating factors—emotions of all sorts, aversions, appetites, feelings of obligations—basically any mental state that ‘‘pushes’’ an agent. So when one cold Friday night, a hungry student named Ellen takes a frozen pizza out of a refrigerator, unwraps it, puts it in a stove, and turns knobs on the stove, we folk psychologists explain Ellen’s action by Ellen’s beliefs—including especially her beliefs that turning the knobs will cause the stove to heat the pizza—and by her desire to eat hot pizza. This sort of explanation is familiar, but not very satisfactory. Ellen might also like to eat frozen pizza, or she might also have a desire to reheat some left-over meatloaf. Or she might rather skip dinner and keep studying decision theory. What explains her action is not merely desiring to eat hot pizza (plus possessing the requisite beliefs), but desiring to do this as much or more than she wants to do any of the feasible alternatives. One way to tighten up the folk-psychological account of action is to replace the non-comparative catch-all notion of a ‘‘desire’’ with a comparative catch-all notion of ‘‘preference.’’ One can then explain the little interaction between Ellen and the stove in terms of physical constraints, Ellen’s beliefs about the outcomes of the alternative actions she can undertake that Friday night, and her catch-all ranking of those outcomes. One explains the pizza warming by showing that Ellen ranks its expected outcome at least as highly as any feasible alternative. With a bit of mathematical dressing up, this is close to the ‘‘official’’ story of choice and preference in mainstream economics. An agent’s preferences consist of his or her overall evaluation of the objects over which preferences are defined. This evaluation implies a ranking of these objects with respect to everything that matters to the agent: desirability, social norms, moral principles, habits—everything relevant to evaluation. Preferences thus imply all-things-considered rankings. In my view, all-things-considered ranking

54

daniel m. hausman

rather than choice ranking is ‘‘the underlying relation in terms of which individual choices can be explained’’ (Sen 1973: 67). It should be the single correct usage of the term ‘‘preference’’ in economics and decision theory. The links economists draw between preference and advantage and between preference and choice, which Sen takes to be alternative conceptions of preference should, I think, instead be regarded as substantive claims about all-things-considered preferences. In some circumstances, it is a reasonable approximation to maintain that people’s all-things-considered ranking of alternatives match their ranking of alternatives in terms of expected advantage. In that case, one can make inferences about expected advantage from preferences and inferences about preferences from expected advantage. But one need not follow Sen and take expected advantage to be ‘‘the usual sense’’—or indeed any sense at all—of preference either in economics or in ordinary life. Similarly, when George’s all-things-considered ranking is limited to his ranking of the objects among which he is choosing, then, if George’s choice tracks his preferences, his preferences will match the ranking that is implicit in his choices. When one can coherently describe a possibility of George choosing between two alternatives, his all-things-considered ranking of those two alternatives will coincide with how he would choose. But the relevant notion of preference is as an allthings-considered ranking, not as either a choice-ranking or a hypothetical choice ranking. The reason why economists must employ a concept of preference as all-things-considered rankings rather than as choice or hypothetical choice rankings is that they need to relate the ranking of objects of choice to beliefs and to evaluations of things that are not objects of choice. If all that consumer choice theorists could say about why an agent purchased one thing rather than another was that the consumer preferred to make that purchase, rather than relating the action to prices, income, and the consumer’s preferences over the commodity space, there would be no Nobel Prize in economics. If all that game theorists could say about why individuals play one strategy rather than another is that they prefer that strategy, game-theory texts could be very short indeed. Moreover, all-things-considered preferences depend on beliefs as well as on motivating factors. Unlike primitive urges, people’s preferences depend on their beliefs concerning the character and consequences of the objects of their preferences.3 For example, my preference for drinking rather than

sympathy, commitment, and preference 55 discarding a glass of clear liquid in front of me depends on whether I believe it is water or gasoline. This means that preferences can be regarded both as the result of deliberation and as an input into deliberation. Belief thus enters into choice in two ways. Belief and preferences can be inputs that influence choices, as they are when my desire for water and my belief that the liquid in front of me is water lead me to drink. Beliefs and preferences can also influence preferences, as, for example, when my preferences among flavors and textures and my aversion to early death, coupled with my beliefs about the consequences of alternative diets, determine my preferences among alternative foods. When the objects of preferences are the alternative choices—that is actions—themselves, then belief plays its full role in the deliberations that result in ‘‘final’’ preferences among the choices, which coincide with what I called the choice ranking. The notion of preference as an all-things-considered ranking is close to the everyday concept of preference, but not quite the same, because in everyday usage ‘‘preference,’’ like ‘‘desire’’ or ‘‘inclination,’’ is often contrasted with duty or principle. Everyday usage thus often treats preference as a ranking in terms of some, rather than all, relevant evaluative considerations. The distinction between preference and duty does not coincide with the distinction between self-interested motives and motives that are not selfinterested. Self-interested choices can be governed by principles of prudence that restrain impulses, while desires to help others may have no connection to principles. So none of the four concepts of preference discussed above—choice ranking, expected advantage, hypothetical choice ranking, or all-things-considered ranking—match this fifth ordinary-language notion of preference. ‘‘Preference’’ as used by economists is a technical term within economics, and neither the case for taking all-things-considered ranking as the notion of preference economists ought to employ nor Sen’s case for recognizing and distinguishing multiple notions rests on claims concerning ordinary language usage.

3. Sympathy, commitment, and Sen’s two concepts of preference Sen recognizes that when economists speak of preferences, they refer to motivations of all sorts, but he is skeptical of ‘‘the common tendency to

56

daniel m. hausman

make ‘‘preference’’ (or a general-purpose ‘utility function’) an all-embracing depository of a person’s feelings, values, priorities, choices, and a great many other essentially diverse objects’’ (1991a: 589). He sees this tendency as a conflation of different notions of preference—that is, as a failure to draw necessary distinctions. A person is given one preference ordering, and as and when the need arises this is supposed to reflect his interests, represent his welfare, summarize his idea of what should be done, and describe his actual choices and behavior. Can one preference ordering do all these things? A person thus described may be ‘‘rational’’ in the limited sense of revealing no inconsistencies in his choice behavior, but if he has no use for these distinctions between quite different concepts, he must be a bit of a fool. (1977: 335–6)

The last sentence quoted here addresses the question of whether agents need to distinguish between their interests and obligations or between their wants and the claims of others, but the question at issue is whether economists need to draw these distinctions. The argument for the claim that economists need to draw such distinctions rests on Sen’s demonstration that choice rankings do not coincide with expected advantage and that these rankings do not exhaust the factors that influence evaluations and choices. For this reason, Sen suggests that a single all-things-considered ranking should be replaced with a variety of rankings of which choice rankings and rankings in terms of expected advantage are merely two prominent instances. A single distinction between choice rankings and rankings in terms of expected advantage is not nearly fine-grained enough to explain why people may be cutthroats at work, devoted parents at home, liberals at the voting booth, racists at the club, public-spirited at one moment, pious at another, principled before lunch, and utterly selfish afterwards. To make even a first stab at accounting for some reasonable portion of human behavior, economists should, Sen maintains, at the very least, also distinguish between what he calls ‘‘sympathy’’ and ‘‘commitment.’’ Sympathy obtains when ‘‘the concern for others directly affects one’s own welfare’’ (1977: 326). Commitment is in contrast non-egoistic (1977: 326–7) and contrary to preference (at least in the sense of expected advantage) (1977: 327). In Sen’s words, ‘‘If knowledge of torture of others makes you sick, it is a case of sympathy; if it does not make you feel personally worse off, but you think

sympathy, commitment, and preference 57 it is wrong and you are ready to do something to stop it, it is a case of commitment’’ (1977: 326). On Sen’s account of sympathy, helping someone from a simple desire to do so, when the agent neither anticipates nor achieves any (personal) benefit, is not a case of sympathy. Unless I expect to benefit from doing X, my doing X is not a case of sympathy. This is clear and unambiguous, and Sen even cautions the reader not to place too much weight on the particular words chosen. Yet I for one previously misread Sen and identified sympathy with altruistic motivation. One reason is that I also misunderstood what Sen meant by preference. If one interprets preference as all-things-considered ranking, then altruistic motivation accords with preference. But if one interprets preference as expected advantage, then altruistic motivation, unlike sympathy, is typically counter-preferential. The way to think about sympathy is to recognize that among the many sources of George’s expected advantage are states of affairs involving other things than George. For example, when somebody tramples George’s roses, the harm done to the roses diminishes George’s welfare. If roses were people, this would be a case of sympathy. If somebody instead tramples George’s friend, any consequent lessening of George’s welfare counts as sympathy. Sympathy is the way in which benefits and harms to other people register within self-interested preferences. ‘‘Commitment’’ then covers some motivations other than expected advantage. Like altruism, commitment is counter-preferential if preference is identified with expected advantage. It is apparently not counterpreferential if preferences are all-things-considered rankings, because allthings-considered rankings already reflect all those factors that constitute any ‘‘commitments’’ of the agent. The examples Sen gives of commitment suggest that commitment involves adherence to principle, and readers (or at least this reader) have assimilated Sen’s distinction between sympathy and commitment to the everyday contrast between action motivated by altruistic concerns and action motivated by adherence to principle.4 This assimilation is mistaken, both because sympathy must be motivated by an expected benefit to the chooser and because Sen never explicitly restricts what he takes to be ‘‘commitment’’ to cases where principles govern choice. So I am not sure, for example, whether he would regard a sacrifice of my welfare motivated by a simple desire to help some particular person as an instance of commitment, or whether such action involves neither

58

daniel m. hausman

sympathy nor commitment. It is as faithful to Sen’s characterization of commitment to take any non-self-interested motivation as an instance of ‘‘commitment’’ as it is to identify commitment and acting from principle. Recognizing the existence of sympathy in Sen’s sense enables one to square behavior that helps others with a model of individuals as governed by a concern for their own advantage. Commitment adds a recognition that individual choice is not always governed by a concern for one’s own advantage—that choice is sometimes ‘‘counter-preferential.’’5 When Sen speaks of ‘‘counter-preferential’’ choice in ‘‘Rational Fools’’ (1977: 328), it seems that he means non-self-interested choice: choice that is counter-preferential only in the expected-advantage sense of ‘‘preference.’’ Altruistic motives would in this sense be counter-preferential. The fact that choices may be counter-preferential in this sense blocks the inference from choice to welfare and thereby complicates welfare economics. In later works, Sen argues that commitment can involve counterpreferential choice in a stronger sense. In the 1980s, Sen refined his views of sympathy and commitment, distinguishing three theses whose conjunction leads to the identification of people’s preferences with not just expected advantage, but with an exclusive concern with themselves. These are: (1) ‘‘self-centered welfare’’ (the person’s welfare does not depend on benefits or harms to others); (2) ‘‘self-welfare goal’’ (a person’s preferences among alternatives depend only on their upshot for the person’s welfare); and (3) ‘‘self-goal choice’’ (a person’s choices depend only on his or her own goals) (1985a: 213–14; 1987a: 80). Sen then writes: Sympathy does, specifically, violate self-centered welfare, but commitment does not have such a unique interpretation. It can, of course, reflect the denial of self-welfare goal, and indeed it is perhaps plausible to interpret in this way the example of a person acting to remove the misery of others from which he does not suffer himself. But commitment can also involve violation of self-goal choice, since the departure may possibly arise from self-imposed restrictions on the pursuit of one’s own goals (in favor of, say, following particular rules of conduct). (1985a: 214)

When commitment conflicts only with a self-welfare goal, then a person’s choice is still determined by his or her goal (which I here identify with preference). It is just that the person’s preferences are not always addressed to his or her own advantage. But when commitment conflicts with ‘‘self-goal choice,’’ the agent’s self-imposed ‘‘restrictions on the pursuit of [his] own

sympathy, commitment, and preference 59 goals’’ leads to a choice of an option that does not best fulfill the agent’s ‘‘goals.’’ If one can identify ‘‘goals’’ and ‘‘preferences,’’ this is a stronger conception of counter-preferential choice. In what sense of ‘‘preference’’ can one identify preferences with what Sen here calls ‘‘goals’’? It seems to be quite a wide sense. For example, Sen writes of goals as ‘‘including moral objectives’’ (1987a: 81). So the ‘‘self-imposed restrictions on the pursuit of one’s own goals (in favor, of say, following particular rules of conduct)’’ are not moral objectives. But goals must not be all-encompassing, either. If goals are all-things-considered preferences, how could counter-preferential choice ever be rational?

4. Many concepts of preference or just one? Sen’s underlying concern is, I think, to push economists toward a more nuanced view of rationality and rational choice (see particularly his 1985b and the introduction to his 2002). He seeks to avoid a view of rational decision-makers as rational fools who carry around some single ranking of all the objects of choice and, within the constraints of feasibility, simply choose from the top of the ranking. Economists should instead recognize that there are many different ways in which alternatives can be evaluated and hence many different conceptions of preference. Though these need not always be distinguished from one another, any acceptable theory of rational choice will in Sen’s view have to make room for many notions of preference. Those who, like me, believe that economists need only a single notion of preference as an all-things-considered ranking might disagree with Sen in two very different ways. One possibility is that they believe that people are rational fools. As a general view of people, this is a silly view, but it is not obviously silly to maintain that viewing people as rational fools is a reasonable approximation with respect to the phenomena of concern to economists. That is, however, not my view. I am no more in sympathy with modeling people as rational fools than Sen is. A second possibility is that those who believe that preferences are allthings-considered rankings agree with Sen concerning human motivational complexity but disagree with him concerning the best strategy for dealing with this complexity. Rather than capturing this complexity by means of multiple concepts of preference, one might capture it by a nuanced

60 daniel m. hausman account of the many factors that influence preferences. People prefer some things because of their expected benefits, others because of emotional reactions toward other people, others because of adherence to social norms or to moral principles, and others out of mere habit. Depending on the context and the objectives, the account one might offer of the factors that influence preferences might be very simple or extremely complicated. Many economists would go on to argue for a division of labor, whereby psychologists and philosophers study the deliberative processes that give rise to an all-things-considered ranking, and economists investigate the consequences of the choices that arise from those rankings coupled with constraints and individual expectations. Though I think that it is better to locate the complexity in the account of what determines preferences, I reject a division of labor that denies that economists should be concerned with preference formation (Hausman 2005). To attribute to people a consistently articulated all-things-considered ranking of alternatives is a strong idealization, which in many contexts may be extreme and unreasonable. Without thinking about where this ranking comes from, economists will not understand when such an idealization is sensible and when it is not. Nor will they understand how changing beliefs and circumstances will influence both preferences and action. As Sen has shown—though in other terminology—the task of understanding how agents construct their all-things-considered preferences cannot reasonably be left out of economics. In my view, the same considerations that drive Sen to insist on the multiplicity of notions of preference justify instead identifying preferences with all-things-considered rankings and distinguishing sharply between preferences on the one hand and, on the other hand, expected advantage, the ranking implicit in choice, or any other substantive dimension of evaluation. In addition to the expected advantage and choice construals of preferences, Sen notes that economists have taken preferences to refer to ‘‘mental satisfaction,’’ ‘‘desires,’’ and ‘‘values’’ (1997a: 303).6 Because these are often taken to go hand-in-hand, economists have not felt it necessary to draw distinctions between these very different things, and in some contexts that may be a perfectly sensible policy. But, as Sen goes on to argue: the eschewal of these distinctions in the characterization of persons amounts to seeing people as ‘‘rational fools,’’ as undiscriminating and gross thinkers, who

sympathy, commitment, and preference 61 choose one all-purpose preference order ... A theory of human behaviour—even on economic matters—calls for much more structure and many more distinctions. (1997a: 304)

Sen is right to maintain that ‘‘A theory of human behaviour—even on economic matters—calls for much more structure and many more distinctions.’’ But it doesn’t follow that it needs multiple notions of preference. On the contrary, it seems to me that Sen’s concern to draw distinctions can be accommodated at least as well by distinguishing sharply between preference (as all-things-considered ranking) and other things. Rather than taking expected advantage to be one concept of preference, one can distinguish between preference and expected advantage. Instead of taking one concept of preference to refer to ‘‘mental satisfactions,’’ one can distinguish between ‘‘mental satisfactions’’ and the extent to which preferences are satisfied.7 A theory purporting to explain or predict the ways in which agents evaluate states of affairs will need to make distinctions between values and mere tastes, but these can be seen as distinctions among the factors that are responsible for a person’s preferences, not as different conceptions of preference. Sen’s concern with the complexities of rational deliberation and choice can in this way be accommodated by those who take preference to be allthings-considered ranking. That concern consequently provides no strong reason for distinguishing many conceptions of preference. On the other hand, the fact that one can accommodate Sen’s concerns with a single allthings-considered conception of preference is not by itself an argument for making do with this one conception of preference rather than recognizing many conceptions. Why then am I making a fuss? Is the issue just semantics? Does it matter whether one takes ‘‘preference’’ to be multiply ambiguous or whether instead one takes ‘‘preference’’ to be all-things-considered ranking and gives other names to what Sen takes to be other conceptions of preference? There are four reasons why I think it matters. First, to regard choice rankings, expected advantage rankings, hypothetical choice rankings, ‘‘mental satisfaction,’’ ‘‘values,’’ ‘‘tastes,’’ ‘‘all motivational considerations other than principle,’’ and ‘‘all-things-considered rankings’’ as eight different conceptions of preference is an invitation to perpetuate the confusions that Sen has so justly criticized. To mark the distinctions between these different things,

62

daniel m. hausman

economists should use different words. The justification for retaining the word ‘‘preference’’ for the last of these eight concepts is that it matches the ‘‘official’’ notion of preference in basic presentations of the theory of rational choice as well as economic practice, especially in game theory and expected utility theory. Second, treating choice rankings, expected advantage rankings, and so forth as alternative conceptions of preference makes it more difficult to pose questions concerning what things determine preferences. The concept of all-things-considered rankings is the most suitable concept of preference, precisely because it does not settle a priori what those ‘‘things’’ are. That way economists can separate the use of the word ‘‘preference’’ from substantive views about what preferences depend on. Third, several of the supposed conceptions of ‘‘preference’’ fly in the face of everyday understandings of the word.8 I’m not saying that everyday usage is determinative, and indeed I believe that taking preferences to be all-things-considered rankings is not fully in accord with everyday usage. But conforming roughly to everyday usage helps avoid misunderstandings. Taking preferences to imply all-things-considered rankings modestly extends the everyday notion, to serve the purposes of economists and decision theorists. Finally, I shall argue in the next section that only the conception of preference as all-things-considered ranking permits game theory and expected utility theory to serve their predictive and explanatory roles.

5. Game theory and counter-preferential choice In certain contexts, Sen’s ‘‘rational fools,’’ whose behavior is unthinkingly governed by a single all-things-considered ranking, appear very foolish indeed. The foolishness is glaring in the case of finitely iterated Prisoner’s Dilemmas or centipede games. In laboratory circumstances designed to implement these games, people frequently manage to cooperate and to do much better than the rational fools whose strategy choices are studied by game theorists. How should one make sense of this fact? Following Sen’s lead, I shall focus on the one-shot Prisoner’s Dilemma, which has the advantage of being much simpler to present and to analyze than iterated Prisoner’s Dilemmas or centipede games. Consider the

sympathy, commitment, and preference 63 interaction or ‘‘game form’’ shown in figure 3.1, where the first number indicates the dollar payoff to the row player from each strategy combination and the second the dollar payoff to the column player. C

D

C

($4, $4)

($0, $5)

D

($5, $0)

($1, $1)

Figure 3.1. Prisoner’s Dilemma; game form

Assume that the game form is common knowledge. Each player earns a larger monetary payoff by playing strategy D ‘‘defect’’ than by playing strategy C (‘‘cooperate’’), regardless of what the other player does. Note that figure 3.1 does not depict a game. By definition, one does not have a game until preferences are assigned to the outcomes. If in addition each player confronting the situation shown in figure 3.1 cares only about his or her own financial payoff, and this is common knowledge, then one has the Prisoner’s Dilemma game shown in figure 3.2. C

D

C

(3, 3)

(1, 4)

D

(4, 1)

(2, 2)

Figure 3.2. Prisoner’s Dilemma; game

Figure 3.2 is the normal form of a game of complete information.9 The numbers are indices of preference with larger numbers indicating more preferred outcomes. That means that the strategy choices, the outcomes, and the player’s preferences (ordinal utilities) are common knowledge. D is a strictly dominant strategy for both players—that is, each does better playing D rather than C, regardless of what the other does. The reason why two players who both play C do better than two who play D is that each benefits from the other player choosing C. Since this is a one-shot game, in which there is no role for reputation or reciprocation, each player harms himself (in terms of his or her own preference ranking) by playing C. All of this would be clear and uncontroversial if it were clear and uncontroversial what the utility numbers, which represent ‘‘preference,’’ meant. I maintain that they represent all-things-considered rankings. Somewhat tendentiously, I shall call this ‘‘the orthodox interpretation’’ of game theory.

64

daniel m. hausman

If utilities represent all-things-considered rankings, then anyone playing C when faced with the game form in figure 3.1 is either irrational or is not playing a Prisoner’s Dilemma.10 By taking utilities to represent allthings-considered rankings, game theorists are able to predict and explain strategy choices in terms of facts about games, including rankings of their outcomes. ‘‘Outcomes’’ must be understood in this context to be what Sen calls ‘‘comprehensive’’ as opposed to ‘‘culmination’’ outcomes (1997b: 745), since features of the play—that is, of the path through the game tree—may matter to the players in addition to the characteristics of the culmination. Although this is one way to understand the utility payoffs in game theory, it is not the only way, and it has some disadvantages. In particular, it appears to limit game theory proper to an examination of how strategy choices depend on game forms, beliefs, and all-things-considered rankings of comprehensive outcomes. That means that game theory has nothing to say about how individuals construct their all-things-considered rankings of comprehensive outcomes. When faced with the fact that experimenters find high rates of cooperation in interactions that appear to have the structure of Prisoner’s Dilemmas, all the game theorist can say is that the subjects are irrational or, more plausibly, that they are not playing a Prisoner’s Dilemma game. If the subjects are rational, then, in terms of their own all-thingsconsidered preferences, D cannot be a strongly dominant strategy. But the game theorist has nothing to say about why their preferences are like this. The task of figuring out how individuals think about their strategic interactions and how they decide how to rank comprehensive outcomes (which may depend on reasons they have for preferring particular moves or strategies as well as on preferences for the culmination) is ruled out of game theory. The task resides instead in a sort of limbo. It is not governed by any economic theory, but it is not studied by any other discipline either. If one takes preference to be expected advantage or indeed anything short of all-things-considered rankings, then rational strategy choice need no longer be determined jointly by beliefs, knowledge of the game form, and preferences over outcomes. For example, suppose one interpreted the utilities in figure 3.2 as reflecting expected advantage. Given this interpretation of preference, counter-preferential rational choice is clearly possible. The fact that D is a dominant strategy in terms of expected advantage does not imply that rational individuals will play D. Game

sympathy, commitment, and preference 65 theorists would be unable to deduce strategy choices from knowledge of games. This is a serious loss. On the other hand, if one takes utilities to be indices of expected advantage, one opens a space in which to discuss the thought processes that may lead individuals to make choices that are not governed by their expected advantage. So there may be a significant gain here, too. Sen does not adhere to an expected advantage interpretation of preference in his discussion of game theory. Indeed he explicitly recognizes that altruists whose preferences conflict with expected advantage can find themselves in a Prisoner’s Dilemma, too.11 Furthermore, as we saw at the end of section 3, Sen does not limit the possibility of counter-preferential rational choice to cases where individuals make choices that do not serve their expected advantage. He entertains the possibility of (rational) counter-preferential choice, even when preferences or goals include ‘‘moral objectives.’’ Yet preferences cannot be all-things-considered rankings, because then rational choice would have to follow preference. What does he have in mind? Sen writes: The language of game theory ... makes it ... very tempting to think that whatever a person may appear to be maximizing, on a simple interpretation, must be that person’s goal ... There is a genuine ambiguity here when the instrumental value of certain social rules are accepted for the general pursuit of individual goals. If reciprocity is not taken to be intrinsically important, but instrumentally so, and that recognition is given expression in actual reciprocal behaviour, for achieving each person’s own goals better, it is hard to argue that the person’s ‘real goal’ is to follow reciprocity rather than their respective actual goals. (1987a: 86)

I interpret these remarks as follows: suppose individuals cooperate when facing the strategic situation in figure 3.1, and suppose this cooperation is rational. If preferences are all-things-considered rankings, then the individuals are not playing a Prisoner’s Dilemma. But if instead preferences are rankings that are influenced only by what an individual values intrinsically, then the individuals can be playing a Prisoner’s Dilemma game. When the reciprocity the players show is instrumental to pursuit of what they value intrinsically, ‘‘it is hard to argue that the person’s ‘real goal’ is to follow reciprocity rather than their respective actual goals.’’ Sen is right about the importance of modeling the intricate thought processes individuals go through when faced with strategic problems like the one shown in figure 3.1, and his suggestion that those who decide to

66

daniel m. hausman

cooperate may still in some sense ‘‘prefer’’ the outcomes where they defect to the outcomes where they cooperate, is plausible. When one asks game theorists why so many individuals facing the situation shown in figure 3.1 cooperate and thus turn out not to be playing a Prisoner’s Dilemma, they should have something better to say than, ‘‘That’s not our department. Go talk to the psychologists.’’ But it does not follow that economists should reject the conception of preferences as all-things-considered rankings and redefine the notion of a game so that strategy choices are no longer deducible from normal forms such as the one shown in figure 3.2. The costs of doing that are too high. Preserving the determinate relationship between games and strategy choices provides a decisive reason to take preferences to be all-things-considered ranking. A better way to meet Sen’s concerns is to argue that the study of games needs to include the study of how games are constituted as well as the study of strategy choices and equilibria. To argue that economists should seek an explicit theory of how games are constituted, which would include an account of how individuals who are interacting strategically construct their beliefs and preferences, does not require that one break the connection between dominant strategies and rational choices. The way to fight Sen’s battle—and I see myself as his ally, not his opponent—is to argue for an enlargement of the economist’s concerns, from games themselves to the process of modeling of strategic circumstances as games, rather than to argue for a reconceptualization of the concept of a game—which is what rejecting the concept of preferences as all-things-considered rankings would require.

6. Conclusions This paper calls for a reformulation of Sen’s invaluable critique of the ways in which economists conceive of preferences and of the impoverishment of the theory of rational choice that results. While endorsing his criticisms and supporting his call for a more nuanced view of the psychology of rational decision-making, I maintain that it is better to criticize economists for making false claims about what determines preferences, conceived of as allthings-considered rankings, than to criticize them for conflating different notions of preference. The more nuanced theory of rational choice that Sen

sympathy, commitment, and preference 67 rightly looks forward to should in my view make room for many evaluative concepts besides preferences (again conceived of as all-things-considered rankings) rather than making room for many concepts of preference. Theories about rational thought in complex strategic situations are, as Sen argues, badly needed. But they are, I have argued, better supplied by maintaining the view of preferences as all-things-considered rankings, and supplementing game theory with theories of the processes by which actors transform strategic situations into games, than by adopting other notions of preferences and weakening the links between facts about games and conclusions about which strategies rational players choose. Notes ∗

I am grateful to Harry Brighouse, Michael McPherson, the participants in the Workshop at the University of St Gallen, and especially to Geoffrey Brennan for their comments on earlier drafts. 1. In the quotation above, Sen takes expected advantage to be a necessary condition for preference, rather than necessary and sufficient, and one can read his words as making a claim about preference ‘‘in the usual sense’’ (whatever that may be) rather than as defining a concept of preference. But other comments make clear that Sen regards expected advantage as a competing definition of preference. He writes, for example, ‘‘Preference can be defined so as to preserve its correspondence with choice, or defined so as to keep it in line with welfare as seen by the person in question’’ (1973: 73). See also Sen (1980: 442). 2. I am here following Broome (1991) and many others both in suggesting that a single usage be prescribed and in the particular usage I favor. 3. Adapting some useful terminology that Sen introduced in a different context (1970: ch. 5), one might distinguish ‘‘basic’’ preferences, which are independent of beliefs, from ‘‘non-basic’’ preferences that depend on beliefs. It is very difficult to give examples of basic preferences. 4. Some philosophers have read more carefully. See for example Anderson (2001: 22 f.). 5. Non-self-interested motivations might lead to the same choices as expected advantage implies. Yet one would nevertheless have a case of commitment if the person would still have chosen the same action even if expected advantage lay with some other alternative (1977: 327).

68 daniel m. hausman 6. And utility, which is frequently taken to be an index of preference, has an even wider range of meanings. See Sen (1987b: 5–7) and Sen (1991b). 7. A preference is satisfied or not in the same sense that a requirement is satisfied or not, by things being as they are preferred or required to be. If an agent knows that some preference is satisfied (which need not be the case, even if the preference is in fact satisfied), then the agent may feel satisfied. But there is no other connection between preference and ‘‘mental satisfaction.’’ 8. This claim requires qualification, because many people in fact believe that people’s preferences are always dictated by their self-interest, and many hold the psychological hedonist view that whatever people do, they do because of the pleasure they expect. But, as Sen himself emphasizes, these views of preference are false, and they depend on well-known philosophical mistakes. Following ordinary usage, when that usage is confused and mistaken, is not a virtue. 9. Since both players have strongly dominant strategies, they do not in fact need to know the payoffs to the other player in order to arrive at their strategy choices. But their interaction will not, strictly speaking, be a Prisoner’s Dilemma if their information is not perfect. 10. In this I am following Ken Binmore (1994), who argues this point at length. Unlike Binmore, who takes this view to support a revealed preference interpretation of preference, I take this point to be linked to a conception of preference as all-things-considered ranking and to be inconsistent with revealed preference theory. As argued in Hausman (2000) and very briefly above in section 2, ‘‘preferences’’ in game theory are not choice rankings. 11. Altruists whose choices are governed entirely by the payoffs for the other player, and who face the game form in figure 3.1, would not be playing a Prisoner’s Dilemma But suppose the payoffs were those shown in figure 3.3 (below). C

D

C

($4, $4)

($5, $0)

D

($0, $5)

($1, $1)

Figure 3.3. Prisoner’s Dilemma; game form

With these payoffs D would be a dominant strategy for such altruists.

sympathy, commitment, and preference 69 References Anderson, Elizabeth. 2001. ‘‘Unstrapping the Straitjacket of ‘Preference’: A Comment on Amartya Sen’s Contributions to Philosophy and Economics.’’ Economics and Philosophy 17: 21–38. Binmore, Ken. 1994. Playing Fair. Cambridge, MA: MIT Press. Broome, John. 1991. ‘‘Utility.’’ Economics and Philosophy 7: 1–12. Hausman, Daniel. 2000. ‘‘Revealed Preference, Belief, and Game Theory.’’ Economics and Philosophy 16: 99–116. Hausman, Daniel. 2005. ‘‘Consequentialism and Preference Formation in Economics and Game Theory.’’ Philosophy (supplement). Sen, Amartya. 1970. Collective Choice and Social Welfare. San Francisco: Holden-Day. Sen, Amartya. 1973. ‘‘Behaviour and the Concept of Preference.’’ Economica 40: 241–59. Reprinted in Sen (1982), pp. 54–73. Sen, Amartya. 1977. ‘‘Rational Fools: A Critique of the Behavioural Foundations of Economic Theory.’’ Philosophy and Public Affairs 6: 317–44. Sen, Amartya. 1980. ‘‘Description as Choice.’’ Oxford Economic Papers 32: 353–69. Reprinted in Sen (1982), pp. 432–49. Sen, Amartya. 1982. Choice, Welfare, and Measurement. Oxford: Blackwell. Sen, Amartya. 1985a. ‘‘Goals, Commitment, and Identity.’’ Reprinted in Sen (2002), pp. 206–24. Sen, Amartya. 1985b. ‘‘Rationality and Uncertainty.’’ Theory and Decision 18: 109–27. Sen, Amartya. 1987a. On Ethics and Economics. Oxford: Blackwell. Sen, Amartya. 1987b. The Standard of Living. Cambridge: Cambridge University Press. Sen, Amartya. 1991a. ‘‘Opportunities and Freedoms (from the Arrow Lectures).’’ In Sen (2002), pp. 583–622. Sen, Amartya. 1991b. ‘‘Utility: Ideas and Terminology.’’ Economics and Philosophy 7: 277–84. Sen, Amartya. 1993. ‘‘Internal Consistency of Choice.’’ Econometrica 61: 495–521. Sen, Amartya. 1997a. ‘‘Individual Preference as the Basis of Social Choice.’’ In Sen (2002), pp. 300–24. Sen, Amartya. 1997b. ‘‘Maximization and the Act of Choice.’’ Econometrica 65: 745–79. Sen, Amartya. 2002. Rationality and Freedom. Cambridge, MA: Harvard University Press.


PA RT I I

Rethinking Rationality


4 Instrumental Rationality versus Practical Reason: Desires, Ends, and Commitment H E RL I ND E PAUE R- S T UD E R∗

1. Introductory remarks Amartya Sen has repeatedly criticized the standard interpretation of the ‘‘rational economic man.’’ His criticism is directed especially against the assumptions that the only goal of a rational person is to maximize his or her own welfare (understood in a self-centered sense) and that the reasons for the choices a person makes are merely dependent on the person’s own goals, more precisely: her self-centered goals. Sen’s objections to the standard interpretation of economic rationality have a Kantian flavor. Recently, Sen’s thinking seems to go even more in a Kantian direction by the emphasis he places on concepts such as commitment and identity.1 We need, his argument goes, a picture of the self that is more complex than the one that represents the self by a utility function that expresses the subject’s preferences. We should presuppose instead a concept of the self according to which the self is capable of acting out of a commitment and hence on the basis of norms and principles guiding his or her choices. To claim that the concept of commitment is necessary for a notion of the self as having a will and making choices on the basis of critical reflection is to move towards Kantian ethics. In Kantian philosophy we find a very close connection between commitment and

74

herlinde pauer-studer

morality, since morality is closely associated with acting from reasons that often go against our inclinations. In fact, in Kantian ethics being committed means making choices in accordance with moral norms, or more precisely in conformity with the categorical imperative. In my paper I will explore whether Sen’s critique of standard economic rationality commits him to a Kantian position. I will discuss two contemporary readings of Kantian ethics, namely Christine Korsgaard’s and Elizabeth Anderson’s interpretations, and I argue that Sen can defend his theoretical aims without having to accept a strong Kantian position. My thesis will be that we should not subscribe to Kant’s account of practical rationality altogether, because Kant simply identifies the rules of practical rationality with morality. A ‘‘moderate Kantianism’’ that demands critical reflection of our ends in addition to means–end reasoning is sufficient for practical rationality. My conclusion will be that Amartya Sen is a very moderate Kantian—and that this is fine.

2. Amartya Sen’s conception of rationality Among rational choice theorists, there is a tendency to ignore the difference between the ascription of a choice to a certain person and a particular interpretation of the content of the choice. In an unreflected way the step is made from the trivial fact that a person’s goals are her goals (in the sense that the person is making the choice) to the problematic assumption that the content of the goals must be self-centered. There is, as Sen has pointed out, an ambiguity in the idea that a person in making choices pursues her ‘‘own goals.’’ The goals of a person can be her goals, but need not be egoistic. The ‘‘reduction of man to a self-seeking animal’’ is a consequence of the assumption that every choice a person makes reveals a preference of the person whose utility is larger than the alternatives not taken.2 Thus, in making a choice, a person cannot do other than maximize her utility. The restriction to egoism follows from a restricted interpretation of ‘‘utility’’: it is assumed that the preferences of the person reduce to her specific interest in her goods. Rational choice theory blurs, as Sen points out, the distinction between maximizing in general (where it is open what is maximized) and maximization as the fulfillment of a person’s self-interest. The maximand is identified with self-interest, the egoistical interests of a person.3

instrumental rationality versus practical reason 75 Sen considers the distinction between sympathy and commitment as a way to block the step from the rather innocent assumption that a person who chooses rationally is maximizing her maximand (general maximization) to the controversial and highly implausible assumption that the person is always maximizing her own interest or advantage. A commitment is defined as a practical reason that a person has that is independent of the gains and losses for the person in case he or she acts on that reason.4 ‘‘The characteristic of commitment ... is the fact that it drives a wedge between personal choice and personal welfare, and much of traditional economic theory relies on the identity of the two.’5 The notion of commitment allows us to relativize the picture of the rational man as the man who maximizes his welfare, i.e. pursues his self-centered goals as best as he can. Sympathy refers to the case where the welfare of a person is dependent on the welfare of other persons; the pains or pleasures of others reduce or increase one’s welfare. In the case of commitment, persons act out of a sense that something is right or wrong—independently of its impact on their welfare. A commitment in this sense comes close to a categorical normative reason. Commitments are reasons that have force independently of one’s inclinations and that quite often do not allow maximizing one’s personal interest; some commitments demand that we go against our own interest.6 Yet they are compatible with maximization in general—especially if maximizing just means pursuing those ends that seem to be the best ones in the situation at hand, i.e. ends which seem good and appropriate by the standards of critical reflection. Sen links commitment to self-scrutiny and social identification. Commitment is the rational recognition of rules associated with social membership. So action based on commitment is non-egoistic, and what might motivate commitment is a sense of loyalty, duty, and obligation. Commitment demands, as Sen emphasizes, that persons have an identity and must be able to develop a reflective and critical attitude towards the reasons guiding their choices.7 The move to commitments allows, as Sen points out, accounting for a range of different normative phenomena. Commitment helps to explain phenomena such as social bonds, solidarity, work motivation; it is a way to elucidate and understand all those social rules that work in the background of societies and that are so crucial if we want social life to go well.

76


Sen has broadened the picture of rational choice by adding to the rational choice paradigm a form of reasoning that includes values, rules, and principles and that allows us to critically assess our goals and our prima facie commitments. Maximization, as he says, gives us a good understanding of an important part of the discipline of rational choice. Reason in the sense of critical scrutiny, however, extends beyond that. Maximizing behavior is only a necessary, not a sufficient, condition for rationality. If the ends are weird or crazy (Sen mentions the example of a person cutting her toes) then maximizing behavior is ‘‘patently stupid’’ and cannot qualify as rational.8 So it seems that the rational choice paradigm can at best offer some necessary conditions of what practical reasoning means. Maximization in general is quite innocent. It provides us with a good understanding of an important part of the discipline of rational choice. Reason in the sense of critical reflection and endorsement, however, extends beyond that. The crucial thing is, as Sen emphasizes, the interpretation of the maximand—and this demands ‘‘careful assessment and scrutiny’’ and following rules of ‘‘reasonableness’’ in exercising practical rationality.9 One might object that there is a tension in Sen’s position between maximization and instrumental rationality on the one hand and a Kantianbased conception of practical rationality on the other hand. The Kantian conception comes into the picture as a consequence of bringing in the idea of commitment and more generally the demand that we need a deliberative procedure to critically evaluate our ends. Often the Kantian understanding of practical rationality is interpreted as being opposed to the concept of instrumental rationality. Yet maximization and instrumental reasoning are quite plausible conditions of rationality. So the question is: do we have two notions of rationality that are incompatible if we emphasize the importance of instrumental rationality and the importance of rationality as a form of critical deliberation taking into account certain normative standards? But if we are looking for a theory of rationality, then we are certainly looking for a conception that covers all forms of rationality. I think that Sen has not given a definitive answer to this question, though his goal is clearly to have one conception of practical rationality and not several competing ones.

instrumental rationality versus practical reason 77 In the next section I will try to elucidate the connection between instrumental rationality and means–end reasoning. I distinguish between a broad and a narrow sense of instrumental rationality.10 In the broad meaning ‘‘instrumental rationality’’ is just means–end reasoning (without any restrictions in regard to the ends); the narrow meaning of ‘‘instrumental rationality’’ is self-interest rationality—the content of the preferences or the ends is restricted to egoistic interests (pursuing one’s own advantage). Means–end reasoning is part of both senses of instrumental rationality, the broad and the narrow one, but only in the narrow understanding of instrumental rationality (in the sense of maximizing personal advantage) is means–end reasoning tied to a specific interpetation of ends, namely egoistic ends. Then I will look at the Kantian approach to practical rationality. My main thesis will be that we should not follow Kant’s transcendental project, because it identifies the rules of practical rationality with the rules of morality.

3. Humean rationality, means–end reasoning, and instrumental rationality Humean rationality is means–end rationality.11 It focuses on desire fulfillment, since the ends are set by the desires one has. Whenever you have an end, you are rational if you pursue it in an effective way. The ends are arbitrary, merely dependent upon whatever our desires, passions, and inclinations are directed towards. Each end is worthy to be pursued as long as the desire for it is strong enough. For Hume it is possible that ‘‘a trivial good’’ may cause a stronger or ‘‘superior’’ desire than do valuable ends.12 David Hume’s normative neutrality in regard to the ends is best expressed by his provocative remark that it is ‘‘not contrary to reason to prefer the destruction of the whole world to the scratching of my finger.’’13 Whatever a person desires, the person has reason to take the most effective means to realize that end. Reason alone, Hume famously said, is not able to motivate. Reason works only in combination with a desire and a belief that a certain means–end relation holds. Hume makes it clear that when our desires or passions do not choose ‘‘means insufficient for the end, the understanding can neither justify nor condemn’’ them.14

78


The only parameter that holds true for the ends is the subjective intensity of the wish. What Hume presents is skepticism towards reason, not rationality.15 Humean means–end rationality applies to a person driven by her or his impulses and desires, a person who gives in, a person who has no will, a person who has no volitions and resolutions. This is a person who also does not have commitments, either moral or non-moral commitments. The Humean conception falls short of a full-fledged account of practical rationality. Practical rationality presupposes persons who reflect on their aims and attempt to realize them with resolution. They are committed to their aims in the sense that they do not just blindly follow their desires, but reflect on their preferences and ends. Humean rationality is restricted to means–end reasoning and does not judge the ends in regard to their ‘‘goodness.’’ But without such a standard of evaluation it is difficult to assess the rationality of a choice. To just follow the demand that one should take the appropriate means to a certain end does not guarantee overall rationality. If the ends are bizarre or crazy, the person’s choice of the most effective means to realize the end does not save an action from the verdict of irrationality. Even if Sen’s toecutter eventually acts more effectively by using a sharp knife instead of a blunt one, we would not consider his behavior as altogether rational. We must consider the entire action—I do x in order to pursue y (and y is worthwhile)—to be reasonable in order to consider a choice as being rational. We obviously need a standard that allows us to assess the ends as normatively adequate or inadequate in order to judge an action as altogether rational. Hume tells us that our desires can go anywhere. That is correct. Yet in order for our choices to qualify as being rational, there must be some restrictions in regard to what it is that we can desire. We need a normative standard to evaluate the ends, and means–end reasoning cannot provide it, since its normative devices are only valid relative to the acceptance of the ends. And Humean ends depend on the whims and caprices of one’s desires. The restrictions on our ends can come from different sources. They can consist merely in the demand to move from first-order desires to second-order desires, or simply be conventional directives such as the rules of etiquette or politeness, or be the counsels of prudence or the norms of

instrumental rationality versus practical reason 79 morality. In any case, the standard of morality has an especially strong word in regard to the quality of the ends. There are obvious parallels between this account of Humean rationality and its weaknesses and Sen’s criticisms of the deficiencies of rational choice theory. The correction that seems adequate in the case of Humean means–end rationality is similar to the one Sen suggests: to develop a theory of practical reasoning that includes standards that allow us to critically assess the ends. In order to see exactly in what way Sen transcends the notion of rationality directed at maximizing one’s own advantage, it is helpful to take a closer look at the notion of instrumental rationality and the principle of means–end reasoning. The principle of means–end reasoning reads: whenever you have an end, you are rational if you pursue it in an effective way and take the adequate means to realize the end. We know an innumerable number of cases in everyday life that fit this model. You feel hungry and you walk over to a restaurant at the next corner to eat a sandwich. You want to take a trip to India and you start to save money so that you can buy a plane ticket. The structure of this kind of reasoning is: you desire something, you believe that taking x is the adequate means to realize the end, and so you do x. The combination of belief and desire brings about the action. The principle of means–end reasoning seems indispensable for assessing the rationality of our choices. If you are hungry and you do not go to the restaurant at the corner although you still believe that the best thing to do would be to go to that restaurant, then your behavior is not rational. Instrumental reasoning in the sense of means–end reasoning is a part of practical rationality. If we accept, however, that practical rationality cannot be end-neutral, then Humean means–end rationality obviously cannot amount to a complete conception of practical rationality. Exactly this is Sen’s argument in regard to instrumental rationality: ‘‘Rationality cannot be just an instrumental requirement for the pursuit of some given—and unscrutinized—set of objectives and values.’’16 Instrumental rationality and means–end reasoning are often equated: instrumental rationality, it is claimed, is means–end rationality. We should, however, be careful here. In order to assess the scope and status of instrumental rationality within a theory of practical rationality in general it is, I think, important to distinguish between a broader and a narrower

80


understanding of instrumental rationality. The broader understanding of instrumental rationality is the means–end requirement; the means–end principle and the principle of instrumental rationality understood in the broad sense of simply taking the means to an end can indeed be equated. But there is a narrower, more specific understanding of instrumental rationality according to which instrumental rationality is associated with a specific sort of ends, namely utility in the sense of self-interest maximization. In that case the means–end requirement is directed to a very specific end, namely maximizing one’s own advantage in a self-centered way. As a condition of rationality, instrumental rationality, it is often claimed, seems to be self-evident; it amounts, as some Humeans claim, to a condition for identifying a reason for action as a reason at all.17 What is at stake in these versions of reasoning, rightly judged as self-evident, is, however, the means–end principle, not the concept of instrumental rationality in the more specific sense of self-interest rationality. The self-evidence of the means–end principle as such does not support the evidence of instrumental rationality interpreted as self-interest rationality, an interpretation which is often more or less tacitly presupposed in the homo œconomicus paradigm. The difference between these two conceptions of instrumental rationality lies in the kind and status of the ends. Means–end reasoning as such is neutral in regard to the ends. Whenever you have an end, it is rational to take the means. This is not necessarily an invitation for arbitrariness, since means–end reasoning can be connected with a form of rational deliberation and evaluation, with a reflective endorsement of ends.18 The means–end principle as such, though neutral in regard to the ends, can be combined with a form of reflective rationality. The ends must be endorsed by the relevant normative or evaluative standards. The defenders of instrumental rationality in the narrow sense of self-interest rationality often rely on the self-evidence of the means–end principle, but forget that the endorsement of specific economic ends like utility maximization (‘‘utility’’ defined as the person’s personal advantage) needs an additional normative justification. The idea has become quite prominent in current practical philosophy that practical rationality cannot merely build on what we pursue effectively as we desire it or have a preference for it, and has brought an anti-Humean trend along with it. Rationality, it is said, must be directed to the good. We find this idea, for example, in Warren Quinn’s statement that practical rationality ‘‘is special by being the virtue of reason as it thinks about human

instrumental rationality versus practical reason 81 good.’’19 A similar move is made in a recent paper by Houston Smit who claims that ‘‘an agent’s practical reasoning has the good as its formal end, so that an agent acts rationally only if her end in acting is something that she conceives of as belonging to the good that is practicable for her.’’20 So the direction of looking at rationality changes: an action has to be—considered on the whole—good, and then the rationality of the action is established. But what concept of the good is meant here? Is ‘‘the good’’ just a sort of placeholder for various evaluative standards in different social spheres, or does the term refer to the moral good? Often the good to which practical rationality is directed is understood too exclusively in moral terms. This is especially the case with those Kantian accounts of practical rationality which follow Kant’s program closely, since it was Kant who established such a strong connection between practical rationality and morality as a form of the good. But, as I will argue, even if we accept the basic line of these accounts, we should in no way go so far in the Kantian direction as to identify the standards of practical rationality with the rules of morality. Practical rationality qualifies certain ways of acting as rational when the ends meet the normative standards, and in the case of moral reasons these are the criteria and requirements of morality. But then acting in a way that takes the means to the right ends is decisive; that x is the most effective means for y as such is not paramount, since y might turn out to be something entirely bizarre. The special normative force here comes from the fact that something is right, from the normative reason. And since the normative reasons are constituted by a different understanding of rationality than Humean means–end rationality, this means that this conception of rationality is overriding.

4. Instrumental rationality and Kantian practical reason Let me focus next on Christine Korsgaard’s Kantian proposal for how to overcome the limits of the Humean conception of rationality. Korsgaard denies that instrumental rationality is anything like a conception of rationality. Instrumental rationality is, however, not given up or replaced by another account of rationality—it is still considered as being relevant to our being motivated. Instrumental rationality is considered as a principle of

82


reasoning, a principle that gains full justification as a condition of rationality only within a non-Humean conception of rationality. More precisely: it has to be a part of a Kantian account of practical rationality in order to have credibility as a principle of rationality. I will argue that Korsgaard is correct in considering the principle of instrumental rationality (in the means–end sense) as part of a non-Humean conception of practical rationality.21 But I will attempt to show that one’s acceptance of Korsgaard’s account of instrumental rationality does not commit one to accept her specific transcendental justification of the moral standards that she considers as constitutive of practical rationality. Korsgaard argues that the principle of instrumental rationality cannot stand alone. It is dependent upon a generic conception of rationality that contains normative principles that allow one to determine which of the possible aims are justified.22 Only in that case does a demand to pursue the end arise: ‘‘Unless there are normative principles directing us to the adoption of certain ends, there can be no requirement to take the means to our ends.’’23 The instrumental principle tells us merely that we should choose the adequate means to our aims. But it can only be normatively effective if we know which of our aims we have good reasons to pursue. Korsgaard draws a distinction between the instrumental principle and the principle of prudence that advises us to do what is best in regard to our complete good. What holds for the principle of instrumental rationality also holds for the rules of prudence: neither instrumental rationality nor prudence is ‘‘the only requirement of practical reason.’’24 For a person to act rationally, she must be motivated to act by her own recognition of an adequate conceptual connection between a belief and a desire to act. The practical insight that a certain way of acting corresponds to a practical principle can be, as Korsgaard states, part of one’s motivational reasons and cause an action. If we identify the aims of the person with what the person prefers, one is always guided by what one thinks one’s reasons are. But if any wish is giving us a reason, independently of its content, we are always practically rational. Korsgaard thus thinks the concept of practical rationality is undermined, since the alternative of irrational acting is missing. Instrumental rationality restricts the realm of reasons to the determination of the optimal means to reach and satisfy our aims, but reason cannot evaluate these aims. Because the instrumental principle (IP) cannot be

instrumental rationality versus practical reason 83 violated, Korsgaard argues, it cannot be a normative principle. Korsgaard’s example is Jeremy, who wants to study, then takes a walk, meets a friend, goes to a bar, and in the end comes home without having done anything of what he wanted to do with good reasons.25 Jeremy’s problem is that he does not recognize a superior principle which would allow him to distinguish between important aims or reasons and spontaneous impulses and inclinations. Korsgaard’s strategy is to show that the principle of instrumental rationality needs backing by a conception of normativity. Instrumental reasons can function as motivating reasons, but they cannot be normative reasons. The normative force of the IP is, according to Korsgaard, dependent upon the fact that we can say of some of our aims that they have a special normative status. Since each practical reason must be a motive as well as a guiding principle, instrumental rationality can only be a partial condition of practical rationality. There are objections against this way of arguing.26 According to Jonathan Dancy, the principle of instrumental rationality is a meaningful normative principle. Instrumental rationality demands of us to take the most effective means to our ends, but, as Dancy points out, one can be motivated by the recognition that the action x is the most effective means to pursue one’s aims and one can ignore this fact also in a way that one’s irrationality becomes obvious. So instrumental rationality is violated if one does not choose the appropriate means and henceforth the IP is a normative principle.27 Dancy thinks the following statements to be true within the Humean framework: (a) the IP is the only rational requirement; (b) I have an end x; (c) I am taking steps to this end; and (d) the steps I take are acknowledged to be less effective than others available to me. So it follows that the IP can be normative and motivating as well—and therefore Dancy concludes that Korsgaard’s argument fails. Dancy is right to point out that the IP is a normative principle; we can violate it, and doing so amounts to a form of irrationality. In this respect Korsgaard’s claim is wrong; the IP need not be restricted to persons who follow their desires without reflection. In another respect Dancy’s objection misses the point. The additional claim at stake in Korsgaard’s argument is that the IP cannot stand on its own. Dancy’s objection shows that the IP is a normative principle, yet his argument does not justify why the IP should be considered to be the only normative principle.

84


The statement ‘‘You should take the means to the ends’’ is only justified if the ends do have a special normative status. If we would admit the IP to be the only normative principle, then irrationality would just amount to the fact that one does not take the most effective means to one’s ends. But then the character of the ends would have no say in determining the rationality or irrationality of an action. Yet some of our aims might be so grotesque that it would not be irrational if we did not choose the most effective means to realize them. (Remember Sen’s example of the toe-cutter.) The crucial point is that some of our aims are such that means–end rationality alone cannot save an action directed to those aims from being properly judged to be a form of irrationality. So a conception of practical rationality must contain more than merely the instrumental principle that demands the most effective pursuit of one’s ends. This, I take it, is Korsgaard’s point. Not only is Korsgaard’s argument Kantian in spirit, but the way she develops it is in fact Kant’s own position.28 Kant himself thought that instrumental reasoning cannot be the only principle of practical reason, and the difference between hypothetical imperatives and categorical imperatives is a result of this idea. Hypothetical imperatives ‘‘represent the practical necessity of a possible action as a means for attaining something else that one wants (or may possibly want).’’29 Categorical imperatives inform us that certain ways of acting are good in themselves. The hypothetical imperative is either a principle of skill or a principle of prudence. In the case of a principle of skill, it is, for Kant, irrelevant whether the end is reasonable or good, the relevant issue is how we can reach the end. Kant illustrates the end-neutrality by a rather drastic example: the prescriptions that a doctor needs to cure a man and those that a poisoner needs to kill the man are, as Kant claims, of ‘‘equal value’’ since each prescription ‘‘serves to bring about its purpose perfectly.’’30 One’s own well-being or happiness Kant considers to be an aim a priori evident for all rational beings, since it belongs to their essence. The principle of prudence is a hypothetical imperative that tells us to choose the appropriate means to one’s own greatest well-being. The categorical imperative, however, commands certain ways of acting without any reference to a purpose that should be attained by it. The validity of the categorical imperative is not dependent upon any condition, but is due to the insight that it is the law of reason, autonomy, and morality. For Kant, these three different sorts of principles, i.e. rules of skill, counsels of prudence, and laws of morality, correspond to different forms

instrumental rationality versus practical reason 85 of necessitation of the will. Moral laws are objective and universally valid and the corresponding necessity is unconditional. The counsels of prudence are necessary, but their necessity is relative to the subjective condition that a person considers this or that purpose as part of her happiness. And the rules of skill are only necessary in regard to the accepted purpose. Kant considers the imperative of skill as an analytic principle.31 The imperatives of prudence are not analytic, since there is no determinate concept of happiness—it depends on the circumstances of a life whether longevity, being rich, or being successful will bring happiness to it. Therefore, the imperatives of prudence are for Kant more like counsels than commands. Moral imperatives, for Kant, must be justified by reasons that are valid for any rational being. Moral laws should have their origin a priori in reason. If moral laws depend on an empirical cognition, then their justification is contingent. Practical principles for Kant ‘‘should be derived from the universal concept of a rational being in general.’’32 Moral laws cannot be hypothetical imperatives, since then their justification would depend on the justification of their condition. The critique of Kant’s concept of practical reason as being somehow metaphysical or mysterious, since the idea of an unconditional justification seems strange and inadequate, is well-known.33 Kant’s claim that moral imperatives are valid unconditionally has been accused of reducing morality to a system of commands that lack further justification. That, of course, is a misunderstanding. Kant wants to avoid a possible regress of justification, and he tries to do it in such a way that the reasons for stopping are evident. Kant’s transcendental move is to demonstrate the moral law to be a necessary condition of the possibility of practical rationality. To deny it would be to deny the possibility of practical reasoning. Humeans have recently moved much closer to a Kantian position. The criticism of Kant’s conception of categorical imperatives as a form of authoritarian and metaphysical commands is no longer prominent. Many philosophers who still defend a Humean theory of motivation accept that a non-Humean conception of rationality is adequate for the dimension of normativity.34 There are indeed good reasons to follow Kant’s account of practical reason. It is not convincing to draw up a dichotomy between broad instrumental reasoning and a conception of reflective practical reason. Instrumental means–end reasoning is obviously part of practical reasoning, but it is not the whole story. Kant’s explanation of practical reason makes it

86


evident that certain ends deserve a special status and his account undergoes a renaissance in those recent moves that associate the notion of rationality with a standard of the good. The Kantian account amounts to a unifying conception of practical reason: since the principle of instrumental reason is a necessary component of practical reason, the problem of a possible split between the instrumental and the non-instrumental levels of practical reason no longer arises. This allows for a new perspective on the question of normativity and motivation. Both the normative dimension and the motivational dimension of practical reasons can be explained now. A practical reason may be motivating but also obligating at the same time. The insight that it is correct to do x generates a commitment and has a motivating effect. The recognition that from the normative perspective there is a reason to do x creates a desire to do x, provided that a person is rational. The critical question that arises is: even if we accept the Kantian conception of practical reasoning as adequate, does that mean we also have to acknowledge as correct the specific modus of connection that Kant regards as given between reasoning and morality? This is the relevant question in the debate between Humeans and Kantians. James Dreier expresses this point nicely when he writes: ‘‘Certain aspects of the Humean position deserve to be abandoned. We should abandon a hard-line metaphysical position according to which the very idea of practical reason is mysterious. Our skepticism should consist in doubts that the content of practical reason is anything like the content of morality.’’35 I think this is a good question to focus on. For Kant, the rules of rationality are the rules of morality. Kant says that the concept of a moral imperative should be derived from the concept of a rational being. We might be inclined to follow Kant’s explanation of practical reason. But do we also have to share Kant’s conception of how morality is connected to practical rationality? Kant’s answer is clear: since rationality requires autonomy and since autonomy in the sense of self-legislation entails morality, the idea of practical reasoning prompts us to recognize the categorical imperative as the principle of autonomy. The question whether and why we should adopt this specific program of justification requires a separate answer. The reasons cannot just be based on the fact that a Humean conception of rationality per se is not enough

instrumental rationality versus practical reason 87 in order to answer the question of normativity. It is necessary to connect rationality with a standard that evaluates the ends. However, the Kantian answer does not stand alone as the only possible answer to the normative narrowing of the content of normative reasons. Take the approaches to morality that are based on a Humean rationality conception that considers morality to be a system of hypothetical imperatives. They are deficient in their understanding of moral statements and maybe in the way they try to justify moral principles. But they, of course, do not admit complete arbitrariness in regard to the ends. They do try to develop normative restrictions in regard to the aims. Philippa Foot, for example, who explicitly argues against a Kantian position, lays down the ideals of humanity, and the virtues of justice and beneficence as the normative parameters for the appropriateness of the ends.36 She has recently criticized her earlier efforts of restricting the aims we might have as deficient. She acknowledges that just to want to belong to the ‘‘volunteers banded together to fight for liberty and justice and against inhumanity and oppression’’ might be too contingent a basis.37 One also has to address those who do not share those ideals and who have other ends. Her recent strategy is to derive the normative restrictions on what people can do from the necessary facts and basic conditions of human life. Human beings as human beings are life forms, and there are natural necessities that set limits on what people can do.38 Yet natural necessities as such cannot create normative necessities. That there are normative limits that are due to natural necessities has to be supported by considerations of reason. The natural necessities by themselves cannot replace our deliberation about what we should do in a certain case, about what is right and appropriate in certain circumstances. Deliberation about different possibilities of normative standards presupposes rationality in a sense that goes beyond Humean instrumental rationality. However, this does not entail that the standards of morality that we approve of have to be identical with the rules of rationality and that only a transcendental justification of the principles of morality as absolutely necessary is appropriate. It results, indeed, in a beneficial enrichment of our idea of morality if one bases the deliberations about what is morally adequate and correct on the standards of reasonable consideration. We should regard this reasoning as a reflective process of consideration where

88


moral judgments and principles correspond to each other and thus support one another, while being at the same time corrigible.39 And again we are back to a position Sen holds. A strong Kantianism, however, goes further. The moral rules are the rules of rationality. Even more than that: the moral law is necessarily justified because it is part of the concept of a rational being. In my opinion, this strong form of Kantianism is problematic. And I am quite sure Sen does not want to subscribe to it—but as he quite often refers to Kant’s account, it might be important to point out why it seems better to hold only a moderate Kantian position. In the following section I would like to discuss the strong Kantian position more carefully. I will first discuss Korsgaard’s account of practical reason and compare it with Elizabeth Anderson’s pragmatic Kantian extension of Sen’s position.

5. Practical reason, morality, and practical identity Korsgaard is a supporter of such a strong Kantianism.40 Korsgaard follows Kant in claiming that a notion of reflective practical reason necessarily entails the categorical imperative in its first two main formulas: the Formula of Universal Law and the Formula of Humanity. Korsgaard’s argument that brings us to the Kantian laws has two steps: the first part of the argument establishes why we need a law; the second part why this law is the categorical imperative. In order to be persons we cannot simply follow our impulses, but must be able to reflect on what we are doing. But we can only give reasons for our actions and justify them if there are laws guiding our reflective evaluations. We have to ask whether a certain incentive for action can qualify as a law for us. In order to structure our impulses, to tame them and reflect upon them, we have to be a free will, that means to be guided by a self-given law. The principle of a free will is henceforth a law, and this condition, to be just a law, is exactly fulfilled by the categorical imperative in the universal law formulation, which brings in the moral law in the form of a law. This completes the first part. The second part of the argument tries to establish the categorical imperative as the only solution to the given problem.41 Korsgaard brings in a second argument to arrive at a substantive conception of morality: to be a person, we need a normative structure. Normative

instrumental rationality versus practical reason 89 structures are supplied not only by morality, but also by our practical identities, i.e., those normative codes that result from our social roles and social contexts—whether one is a father, mother, mafioso, or philosopher. But we cannot develop practical identities if we do not attribute value to ourselves, if we do not value our humanity. And to value our humanity we have to equally value the humanity of others. That brings us to the second formulation of the categorical imperative, the Formula of Humanity.42 Korsgaard is not merely claiming that the categorical imperatives should guide us. Her argument is stronger: namely that this is necessarily so, inevitably so. Her transcendental argument is that the Kantian imperatives are constitutive standards of our identity. They are constitutive standards of us as unified agents. In order to be persons at all, the categorical imperatives must make up our identity. Action for Korsgaard is self-constitution.43 It is necessary for agency that agents are unified. Otherwise we cannot attribute to persons ‘‘the things that happen because of them’’ and hold them responsible for what they do. The principles of practical reason, hypothetical and categorical imperatives, are principles of the unification of agency. And this explains their normativity. As Korsgaard writes: ‘‘The necessity of conforming to the principles of practical reason comes down to the necessity of being a unified agent.’’44 What makes actions good or bad is how well they constitute you. A good person is one who is good at his or her unification, is ‘‘good at being a person.’’ To sum up: the reflective structure of our consciousness demands that you identify yourself with a law, and this is the source of normativity. This law is a formal principle, and the material side of this formal principle is to value yourself as a human being. Why does Korsgaard make such a strong claim and consider the categorical imperatives as constitutive standards of unified agency and of our identity as persons? Her program is truly Kantian, and she shares Kant’s ambitions to remove all doubts in regard to the claims of normativity and of morality. The idea is clear: if the principles of morality are constitutive of us as agents, as necessarily part of us as the categories of understanding and the intuitions of space and time, then the grounds for moral skepticism are gone. Korsgaard’s point is that we cannot consider the Kantian imperatives as principles that we can choose or reject. They are necessary.

90


Korsgaard offers also another argument to show the necessity of the categorical imperative. This argument starts with the fact that we, as beings with a reflective consciousness, must be normatively structured, we must have a practical identity. This demands that we consider our humanity as valuable and the publicity of reason forces us to consider the humanity of others as valuable. Even if we accept the argument that to be a normatively structured person means that you must value yourself as a person and, because of the publicity of reasons, also other persons, that by itself does not imply that you value the persons around you in the specific and demanding way that is prescribed by Kant’s idea of treating others as ends in themselves. I can take myself as important and value myself also by following the principle that my interests should be simply prior to those of others—and the publicity condition brings us in this case to make concessions, but not to the deep form of respect for others that Kant had in mind. It seems difficult to imagine why I should be denied identity because of making an egoistic strategy my principle. Korsgaard presupposes a morally loaded sense of humanity. Our being human already means to be moral beings—in the Kantian sense. Acting demands, as she points out, unified agents to whom we can attribute actions. Unified agents are constituted by the Kantian imperatives (hypothetical imperative and the categorical imperatives). Hence we are obligated to follow the Kantian imperatives in order to be unified agents. To claim that the categorical imperatives are constitutive standards of our identity might, if at all, be so on our good or best days when the higher ideals inspire and drive us. But what of our bad days when we fail and do not live up to the demands of morality? Yet how can we miss the moral standards if they are constitutive of us? How can we act badly if the Kantian imperatives are constitutive of our identity? The familiar reading of Kant’s theory is to consider the categorical imperatives (CI) as the criteria of a testing procedure. In that case, bad actions are the result of bad maxims—the bad subjective principles of persons. The CI-procedure helps us to find out which of our maxims are good or bad ones. But this is not quite Korsgaard’s point of view. She does not want to suggest the categorical imperatives as possible criteria to test our subjective principles. She wants to show that the categorical imperatives are necessary principles of our actions, our acceptance of them is inevitable

instrumental rationality versus practical reason 91 since they are part of us as unified agents. But then it becomes mysterious to which will our bad actions can be attributed—certainly not to a will whose defining feature is to be made up by the categorical imperative. An obvious way to a solution of this problem seems to be to distinguish between different forms of identity or different senses of autonomy. And we find in Korsgaard’s writings a distinction between practical or social identity on the one hand and moral identity on the other hand. So the way to go would be to argue that someone with the social identity of a gangster does bad acts, but that he would act morally if he were to adopt a moral identity. But in that case there would be a gap: the gangster need not necessarily accept the moral law. Korsgaard sees the problem of bad action, and she tries to solve it. In The Sources of Normativity she gives the following account of bad action: the bad person, e.g. a mafioso, has a practical identity—conforming to the mafioso code. He just does not think things through; he does not reason. If the mafioso were to reflect on his maxims, he would come to see them as wrong.45 This is not specifically Kantian, at least not strong Kantianism—it brings in the CI as a suggestion, not as a necessity. In the Locke Lectures the claim is stronger. Someone who is not constituted and henceforth obligated by the categorical imperatives does not have a practical identity. But this seems to eliminate the possibility of bad action. To distinguish between social identity and moral or constitutional identity does not help here, because the question arises of how the norms making up the social identity can, in the case of a unified agent, be deviant and violate the Kantian imperatives. Maxims are subjective principles of a will, but if maxims are constructed by a will which is constituted by the Kantian imperatives, then that will cannot produce maxims that contradict these imperatives. We also do not dispose of that problem if we distinguish between two senses of autonomy, namely autonomy in the sense of self-legislation and autonomy in the ordinary sense of being free within certain legal and moral boundaries to choose our form of life and to do what we think worthwhile. The problem of bad action comes up again, because if autonomy in the sense of self-legislation is constitutive of our identity, then the acts we can choose and hence our autonomy in the second sense is more limited than we usually do think: it is limited to moral actions. What form of identity does a human being have that does not act in conformity with

92


the categorical imperative? Merely a social or practical identity, so that the content of the maxims is determined by her social or group values? But how can this being have maxims at all since unification is missing? To interpret bad acting as a failure of self-constitution seems strange, because acting, and therefore bad acting as well, already presupposes selfconstitution. When we do criticize the actions of persons as morally deficient, we do not evaluate and talk about whether the identity of these persons is badly constituted or not. And, equally, if we think of really bad actions, for example if someone tortures other persons or kills them, it seems odd to say that he is involved in the same activity as the good person, namely, in the activity of self-constitution, but that he is just doing that activity badly. If the activity of self-constitution is necessarily bound to the Kantian imperatives, we cannot, if we act badly, be involved in the same sort of activity as when we act in a morally good way. Usually we say: the good person who acts well keeps to the standards of morality, the bad person ignores the demands of morality and violates them. Of course we can say that the bad person fails to shape himself or herself into a good person—he or she fails at being good at being a good person. But that means only that the person does not live up to certain standards; it does not mean that he or she falls back from his or her constitutional identity. I think there are two ways of understanding our being bound by the categorical imperatives (and this applies to normative principles in general): A: The agents are, as unified agents, necessarily bound by the CI. B: The agents see, via deliberation and reasoning, the force of the Kantian imperatives; they understand that they have good reasons to accept them. I consider the second to be more plausible. The solution seems to be to give up the ambitious claim of making the Kantian imperatives necessary and inevitable. We should take them as plausible criteria to test our subjective principles of acting, our inclinations and dispositions. This means to come back to a more modest form of reflective endorsement, a form that is less ambitious in the goals of justification but broader in the scope of what can be an object of moral evaluation and justification. Elizabeth Anderson has put forward an interesting proposal for extending Sen’s position in the direction of a Kantian account of practical reason.46 Anderson agrees with Sen that committed action seems a way out of

instrumental rationality versus practical reason 93 PD-situations. She criticizes, however, that Sen does not ‘‘propose an alternative, non-preference-based conception of rationality in terms of which committed action makes sense’’47 and suggests a (Hegelianized) Kantian position as the solution to fill this gap. The first step to evade a PD-situation is, as Anderson points out, to move from the question ‘‘What should I do?’’ to the question ‘‘What should we do?’’ In a PD-situation no common discussion is possible; in a collective deliberation the outcome is the result of a joint strategy. Anderson considers the universalization principle (i.e., to act on principles that it is rational to adopt if one identifies as a member of a collective agent) as the constitutive principle of a collective agent and proposes it as the alternative to the principle of maximizing expected utility. The universalization principle allows, she claims, the step ‘‘from the rationality to the morality of committed action.’’48 Anderson connects identity and rationality. Conceptions of identity she thinks to be prior to rational principles: the sort of rational principles we choose is a function of the self-understanding of the actors. Whether we choose the principle of expected utility or the principle of universalization depends on whether we see each other as isolated individuals or as cooperating agents. As cooperating agents we need, as she emphasizes, to adopt a more general perspective from which we can coordinate and evaluate the different demands we face as members of different groups. This ‘‘requires that we transcend our various parochial identities and identify with a community that comprehends them all.’’49 The universalization principle reflects the deliberative process in such a community as the principle extends to the point of view of each individual. Anderson considers the concept of social identity as important for a reformulation of the concept of practical rationality. She argues that the ‘‘quest for a perspective that can make sense of our experiences and solve our problems leads to more and more expansive, cosmopolitan identifications, in an historical rather than a purely logical process.’’50 She mentions Hegel’s philosophy as a model of such a way to gain a more general practical outlook and she explictly separates her ‘‘dialectical reasoning’’ from Korsgaard’s transcendental form of arguing. Anderson admits that she cannot put forward a systematic justification why we should be rationally required to identify with the universal community of humanity. The argument she offers is an appeal to concrete historical experiences: there are

94


so many grave social and environmental problems that result from human actions that there is need for a system of global cooperation based on an identification with humanity. This way, the informational basis of the economic model of rationality is enriched with the concepts of identity, collective agency, and reasons for action other than egoistic considerations. The test for valid reasons is ‘‘universalizability among those with whom one rationally identifies.’’51 Which of these two approaches is more plausible as an extension and unification of Sen’s ideas of rational choice and rationality? Anderson considers collective action on the basis of group identification as an alternative to PD-situations. However, the appeal to identification has to be seen with caution. Identification with a collective is not necessary to transcend the rational choice paradigm. Anderson moves from the individual to the collective perspective by replacing the principle of maximizing expected utility by the universalization principle. Yet the universalization principle as such does not commit us to an identification with a collective agency. Universalization demands that we consider whether a principle of action can be a principle for all others as well. The Kantian Kingdom of Ends is a society which results if the basic moral principles are adopted by all so that the members respect each other and treat each other not merely as means. But the starting point is the individual decision to adopt the moral principles—a consent that results from insight into the fact that a way of acting is right and not from the identification with a collective and a community. This argument relativizes at least one aspect of Anderson’s claim that we need to adopt a Hegelian instead of a Kantian point of view.52 Anderson, following Sen, considers the concept of commitment as part of a form of rational deliberation that is different from the preference-based account. Again, to take commitments seriously does not presuppose the adoption of a collective standpoint and an identification with a social group or community. A commitment is simply, to use Searle’s phrase, an external (desire-independent) reason: it amounts to an obligation to act from a principle or a reason that is in conflict with our immediate desires.53 Certainly, how we understand ourselves has an important influence on what we feel committed to. But from the normative perspective, commitments are principles or reasons that have force. And the force or validity of a principle does not depend on the social identity we adopt. Persons often

instrumental rationality versus practical reason 95 do have reason to act on certain principles, even if their identity is not in conformity with the principle. Though loud behavior seems to be part of the identity of most soccer fans, we would still say that they should respect the principles of politeness if they use the subway to get to a soccer game. Anderson’s emphasis on identity raises a question in regard to the systematic force of her position. An account of practical rationality, especially if it should be a basis for justified moral decisions, cannot rest on contingent considerations. Anderson is fully aware that we need a justification beyond contingent factors such as concrete historical contexts. This becomes evident when she introduces the notion of practical identity. Unlike ‘‘ascribed social identities of gender, race, caste, ethnicity, nationality,’’ practical identity is an abstract concept of identity that determines which principles of deliberation we choose and consider as adequate.54 For Korsgaard, ‘‘practical identity’’ means the abstract moral identity an agent has as a being able to deliberate rationally and to act from reasons. The moral principles are constitutive of the identity of the rational person, they are indispensable for the structure of a person acting from reasons. In Korsgaard’s account of rational action, the concept of social identity as such is not specifically important. That we have social identities is simply a fact, but the validity of the principles of morality is established independently of our belonging to various social groups. The concept of identity as such does not have justificatory power in Korsgaard’s approach. Decisive is her version of a transcendental argument. A transcendental argument in the Kantian sense tries to justify x by showing that x is a condition of the possibility of y and therefore necessary, given that y has the status of an objective end. So Korsgaard aims to demonstrate that the categorical imperatives are necessary principles of rationality because they are the conditions of possibility of being a person at all. This is a systematic claim whose validity does not depend on social identifications or on the identification with a global community as such. What is crucial in Korsgaard’s account is the claim that the categorical imperatives are constitutive of a rational being. The step from rationality to morality is made by the argument that we need a principle to structure our reasons and that the features of this principle are fulfilled by the categorical imperatives. The systematic point is the functional role the categorical imperatives play as they meet the structural requirements of rational practical deliberation. Practical or moral identity as such is not

96


decisive. Practical identity is just a term to give a name to a being who deliberates in accordance with these principles. On Anderson’s Hegelian account, we need identifications (either with ourselves or with social groups). In the end only a cosmopolitan identity allows us to gain a perspective from which we can come to a solution of the various problems resulting from human actions. So the systematic argument seems to be replaced by a pragmatic one: we need the identification with the ‘‘universal community of humanity’’ because this standpoint enables us to overcome problems that ‘‘can only be solved within a global system of cooperation.’’55 Anderson concedes that she has no argument to justify the identification with the universal community of humanity as a rational necessity. But our historical experiences show that the adoption of such a point of view would make us aware of various possibilities of solving collective action problems. At first sight it looks as if Anderson is offering a too contingent approach. This is supported by the indecision there is in Anderson between the principle of expected utility and the universalization principle. Adopting the universalization principle is more a recommendation than a strict argument. This becomes apparent by the way Anderson uses Sen’s work on gender relations as a test case to answer the question which principle of rationality we should accept. The principle of expected utility does not seem outright wrong, since it often would be better for women to see themselves as rational egoists instead of committed wives and mothers. Rational egoism seems the strategy to escape oppressive commitments. Yet, as Anderson argues, the principle of rational egoism is inadequate, because women most of the time are not in a situation that allows them to bargain with others. Their preferences include the well-being of other people; their actions often are the result of commitments. Rational egoism would recommend, for example, that women seek outside employment that would help them towards autonomy and independence. But outside employment and the various social contacts it offers would allow women to gain a form of collective identity different from the self-conception the economic model presupposes. The background norms, Anderson points out, must be neither those of egoism nor those of collective identification, because women can be disadvantaged by following the principle of expected utility as well as if they follow the principle of collective action. To grant women justice, a Kantian perspective seems adequate that enables women to have

instrumental rationality versus practical reason 97 an autonomously defined self-conception so that they can see themselves ‘‘as a committed member of multiple social groups, among whose claims one must adjudicate in allocating one’s own efforts.’’56 Compared with Anderson’s pragmatic way of reasoning, Korsgaard’s argument seems philosophically much more stringent.57 Yet the way Korsgaard identifies the rules of rationality with the rules of morality is, as I have argued, problematic. In that respect Anderson’s pragmatic justification of different principles of rational action seems more plausible. Anderson’s problem is her ambivalence. She develops a reflective endorsement version of a Kantian justificatory program which entails that the principles of rational choice do not reach all the way to morality; the reflective endorsement procedure has to be guided by additional moral standards and principles to justify choices on moral grounds. But in the end Anderson falls back on the strong Kantian ambitions, namely, to identify the rules of rationality with the rules of morality. Anderson does this by claiming that our identity as social actors demands our identification with the collective action principle of a collective agent (the universalization principle), which turns out to be a basic moral principle. This argument is basically the same as Korsgaard’s argument: rationality analytically entails morality. Seen in abstraction from what Anderson says about the identity of rationality and morality, her method is in fact an instantiation of way B which I have outlined above: agents come to see via deliberation and reasoning the force of moral principles; they understand that they have good reasons to accept them.

6. Concluding remarks Why is all this relevant to the discussion of Sen’s notion of practical rationality? I think that in looking at Humean means–end reasoning and Kantian rationality, we do find obvious parallels to the problems Sen discusses. The Kantian position holds that instrumental rationality in the sense of means–end reasoning is part of practical rationality, but it is only a necessary condition of rationality. So one might be tempted to move to a Kantian account of practical rationality if one thinks that instrumental reasoning and maximization are not sufficient for rationality. But, as I have argued, it is sufficient to associate practical rationality (in addition to the

98


means–end principle) with a form of critical reflection that makes use of different normative standards. Strong Kantianism amounts to an identification of the rules of practical rationality with the principles of morality. That is a step we should not make. Practical rationality should be connected to a pluralism of standards, and morality is one among other criteria that are relevant in a deliberation procedure. The identification of practical rationality with morality is due to the aim of Kantian philosophy to offer a justification of moral rules that cannot be rejected. The idea is that if we identify morality with practical rationality and consider the rules of practical rationality and morality as indispensable for the identity of the person, then the rules of morality are established as inescapable. Personal identity certainly depends on the rules and standards that we accept. Who we are is also a question of the standards that guide our actions. But this does not support the strong claim that the Kantian principles must be constitutive of our identity. Sen equally emphasizes the connection between identity and the principles of rational choice. He distinguishes between four aspects of the self: (1) self-centered welfare; (2) self-welfare goal; (3) self-goal choice; and (4) the self in the form of self-scrutiny and one’s own reasoning.58 I take it that here a rather innocent sense of identity is at stake: what sort of person we are is connected to the values, goals, commitments, and desires we do have. A person who critically reflects on her ends and takes the existence of others into account is a different person from a self-centered maximizer or a dog-like follower of his impulses. And the way the characters of certain persons develop can certainly be a reason to ask for reflection and modification. But this sense of identity does not depend on the assumption that all different forms of social identities must be backed by an identity constituted by principles that are indispensable for rationality and morality. One might object that the identification of practical rationality with means–end rationality and critical reflection and endorsement is too contingent a basis for justification. But I think that coherence between our ends and normative standards is enough. The considerations at stake here are considerations that we can expose and explain to others in a deliberation process where the results are open to modification. Sen is a moderate Kantian when he defines rationality (besides maximizing) as a form of critical scrutiny and assessment—and I think that is

instrumental rationality versus practical reason 99 fine, and he should not be tempted to adopt a stronger Kantian account. One question remains: does Sen’s account of rationality as assessment and scrutiny urge us to transcend the paradigm of rational choice or can it be considered as part of the rational choice approach? I cannot give a full answer to this question. At least I can say: if the antagonism between the (broad) principle of instrumental rationality and rationality as critical scrutiny disappears (and it seems worth mentioning that not even Kant believed in that antagonism) then at least one argument that supports the incompatibility of the rational choice paradigm and rationality as critical deliberation is relativized. Many philosophers claim that instrumental rationality as such is the conception of rationality that underlies rational choice theory. Since maximization is the only standard in rational choice theory, this implies that there is no concern about evaluating or endorsing the ends of action. If, however, we keep separate instrumental rationality in the sense of self-interest rationality, and instrumental rationality in the sense of a mere means–end requirement, then we can become aware of the innocence of a formal system like rational choice theory. It all depends on the semantics and the justification of the normative assumptions underlying certain interpretations of this formalism. Notes ∗

For critical comments on an earlier version of this paper I would like to thank Fabienne Peter, Willem deVries, R. Jay Wallace, and an anonymous referee. 1. See Sen ‘‘Goals, Commitment, and Identity’’ in Sen (2002) and Sen (2004). 2. See Sen (1982: 88). 3. See Sen (2002: 22 ff.). 4. See Sen (1982: 91 ff.). 5. Sen (1982: 94). 6. A person can, however, have a commitment to her own development and to her self-centered goals. Sen has given examples of persons for whom it would be important to concentrate on their self-centered goals, e.g. women who willingly submit to their marginalized position. 7. See Sen (2002: 216 f.).

100


8. Sen (2002: 39). 9. Sen (2002: 41). 10. I would like to thank Fabienne Peter for helpful discussions on this point. 11. One has to be careful here: What Hume says about means–end reasoning is not all he says in regard to practical reflection in the sphere of morality. His provocative and skeptical remarks in regard to reason and the capacities of reason to motivate are developed in Book II (Of the Passions) of the Treatise (esp. Book II, part III, section III, ‘‘Of the influencing motives of the will’’) and in Book III (Of Morals) of the Treatise, section I of part I (‘‘Moral distinctions not deriv’d from reason’’). In part III of Book III, Hume, however, is more positive in regard to the possibilities of evaluative reflection. He develops a method of reflective endorsement in regard to moral judgments (by connecting considered moral judgments to an appeal to general and impartial rules of consideration). Cf. Baier (1991: ch. 12). Korsgaard also attributes to Hume a method of reflective endorsement; (Korsgaard 1996: ch. 2). 12. Hume (1978: 416). 13. Hume (1978: 416). 14. Hume (1978: 416). 15. As I emphasized in FN 11, Hume defends this position in certain sections of the Treatise; it does not represent his position on practical reason altogether. 16. Sen (2002: 39). 17. In defense of Humeanism, James Dreier writes: ‘‘The special status of instrumental reason is due to its being the sine qua non of having reasons at all’’ (Dreier 1997: 98 f.). It is worth mentioning that Dreier puts forward this claim as a direct reaction against the position defended by Korsgaard, namely that the Kantian moral law is a condition for having reasons at all. What is at stake here is a controversy about the identification of the rules of morality with the rules of rationality. 18. As David Schmitz expresses this idea: ‘‘[A] means–end conception of rationality can be made consistent with our intuition that we can be rational in a more reflective sense, calling into question ends we happen to have, revising them when they seem unfit’’ (Schmitz 1994: 226 f.). 19. Quinn (1993: 254).

instrumental rationality versus practical reason 101 20. Smit (2003: 225). I think it to be wrong, however, to attribute this position, as Smit does, to Kant. 21. Korsgaard uses the term ‘‘instrumental rationality’’ in the sense of means–end reasoning. She does not explicitly distinguish between means–end reasoning and instrumental rationality in the self-interest sense. 22. Korsgaard (1997). 23. Korsgaard (1997: 220). 24. Korsgaard (1997: 220, italics in the original). 25. Korsgaard (1999: 19) and Korsgaard (1997: 247, FN 64). 26. Broome (2000) objects that Korsgaard’s way of arguing in regard to the normative force of the IP does not respect the distinction between reasons and normative requirements. Willing an end requires you to take the means, but willing the end is not a reason to take the means. Yet, one might answer, in Korsgaard’s strong Kantianism willing amounts to a requirement; ‘‘to have a reason’’ means being required. Blackburn (1998: 242f.) rejects the assumption that the ends must have a normative status as unconditional and necessary in order that the instrumental requirement to take the means to the ends does have normative force. 27. Dancy (2000: 46 f.). 28. Korsgaard has precisely and impressively interpreted Kant’s position, and she has made explicit the often missed argument in Kant’s work that hypothetical imperatives are part of practical reasoning. 29. Kant (1994: 25; academy edition of Kant’s works: 414). 30. Kant (1994: 25; academy edition: 415). 31. Kant (1994: 27, academy edition: 417). 32. Kant (1994: 23, academy edition: 412). 33. Foot (1978). 34. Cf. Smith (1994). 35. Dreier (1997: 99). 36. Foot (1978: 165–7). 37. Foot (1978: 167) and Foot (2001: ch. 1). 38. Foot (2001: esp. chs 2 and 3). 39. Seen from this perspective, an approach such as Foot’s is indeed interesting, because it extends moral reflection to the normative framework our ways of life presuppose.

102


40. Korsgaard herself does not discuss Sen’s position, but her arguments are relevant to an assessment of Sen’s project of developing an account of practical rationality beyond rational choice theory. This is especially so since Korsgaard argues that the step from instrumental rationality to Kantian practical reason is inevitable. 41. Korsgaard here omits an ingredient aspect of the first two formulas of the categorical imperative, namely the contradiction in conception and the contradiction in willing test. But these are already substantial criteria that cannot be solely justified by means of the argument that we need a law in order to be able to refer to reasons. 42. Behind our practical identities we have a moral identity: We should act only according to those maxims about which all persons can agree in a system of rational cooperation. 43. Korsgaard (2002: lecture one, 1.3.4.). 44. Korsgaard (2002: lecture one, 1.3.4.). 45. Korsgaard, ‘‘Reply,’’ in Korsgaard (1996: 254–8). 46. Anderson (2001). 47. Anderson (2001: 24). Sen mentions various elements that should be considered in the formulation of an alternative principle of rational choice such as understanding people’s identities (their memberships in various groups), their joint strategy, and their ways of discussion. Anderson tries to integrate these ideas into a coherent account. 48. Anderson (2001: 24). Anderson demonstrates the impact of this principle with the example of voting. Even if all members of a political party have agreed that the best outcome would be to elect candidate A, this is not a reason for a certain member to elect A if that member finds voting inconvenient or if the person thinks that his or her voting has no utility because it has no effect on the outcome. But this cannot be a valid reason. The universalization of the maxim not to vote because it is an inconvenience or because the impact is close to zero would be self-defeating, since the practice of voting would be undermined. 49. Anderson (2001: 37). As examples of such a community, Anderson mentions Rawls’s conception of a ‘‘social union of social unions’’ or the Kantian Kingdom of Ends. 50. Anderson (2001: 37, FN 9). 51. Anderson (2001: 38).

instrumental rationality versus practical reason 103 52. But this does not require abolishing the perspective of the individual and identifying with a community of all others so that the perspective of the individual is subordinated to the standpoint of the community. 53. Cf. Searle (2001). 54. Anderson (2001: 31). 55. Anderson (2001: 37). 56. Anderson (2001: 36). 57. Anderson’s approach is pragmatic since she does not put forward a decisive and final justification of the normative principles guiding rational choice. In the end, as she states, it depends on the context whether the principle of expected utility or the universalization principle should have the normative say. 58. Sen (2002: 33 ff.). References Anderson, Elizabeth. 2001. ‘‘Unstrapping the Straitjacket of ‘Preference’: A Comment on Amartya Sen’s Contributions to Philosophy and Economics.’’ Economics and Philosophy 17: 21–38. Baier, Annette C. 1991. A Progress of Sentiments: Reflections on Hume’s Treatise. Cambridge: Harvard University Press. Blackburn, Simon. 1998. Ruling Passions: A Theory of Practical Reasoning. Oxford: Clarendon Press. Broome, John. 2000. ‘‘Normative Requirements.’’ In Jonathan Dancy (ed.), Normativity. Oxford: Blackwell, pp. 78–99. Dancy, Jonathan. 2000. Practical Reality. Oxford: Oxford University Press. Dreier, James. 1997. ‘‘Humean Doubts about the Practical Justification of Morality.’’ In Garett Cullity and Berys Gaut (eds.), Ethics and Practical Reason. Oxford: Clarendon Press, pp. 81–99. Foot, Philippa. 1978. ‘‘Morality as a System of Hypothetical Imperatives.’’ In Foot, Virtues and Vices and Other Essays in Moral Philosophy. Berkeley: University of California Press, pp. 157–73. Foot, Philippa. 2001. Natural Goodness. Oxford: Clarendon Press. Hume, David. 1978 [1739/40]. A Treatise of Human Nature. L. A. Selby-Bigge (ed.), 2nd edition (text revised by P. H. Nidditch). Oxford: Clarendon Press. Kant, Immanuel. 1994 [1785]. Groundwork of the Metaphysics of Morals, translated as Grounding for the Metaphysics of Morals by James Ellington. In Kant, Ethical Philosophy, 2nd edition. Indianapolis: Hackett Publishing Company.

104


Korsgaard, Christine M. 1996. The Sources of Normativity. Cambridge: Cambridge University Press. Korsgaard, Christine M. 1997. ‘‘The Normativity of Instrumental Reason.’’ In Garett Cullity and Berys Gaut (eds.), Ethics and Practical Reason. Oxford: Clarendon Press, pp. 215–54. Korsgaard, Christine M. 1999. ‘‘Self-Constitution in the Ethics of Plato and Kant.’’ The Journal of Ethics 3: 1–29. Korsgaard, Christine M. 2002. Self-Constitution: Action, Identity, and Integrity. The Locke Lectures. . Quinn, Warren. 1993. ‘‘Putting Rationality in its Place.’’ In Quinn, Morality and Action. Cambridge: Cambridge University Press, pp. 228–55. Schmitz, David. 1994. ‘‘Choosing Ends.’’ Ethics 104/2: 226–51. Searle, John R. 2001. Rationality in Action. Cambridge, Mass.: The MIT Press. Sen, Amartya. 1982 [1977]. ‘‘Rational Fools: A Critique of the Behavioral Foundations of Economic Theory.’’ Philosophy and Public Affairs 6; reprinted in Sen, Choice, Welfare and Measurement. Cambridge, Mass: Harvard University Press, pp. 84–106. Sen, Amartya. 2002. Rationality and Freedom. Cambridge, Mass.: Harvard University Press. ´ Sen, Amartya. 2004. ‘‘Social Identitity.’’ Revue de Philosophie Economique 9: 7–27. Smit, Houston. 2003. ‘‘Internalism and the Origin of Rational Motivation.’’ The Journal of Ethics 7: 183–231. Smith, Michael. 1994. The Moral Problem. Oxford: Blackwell.

5 The Grammar of Rationality G E O F F RE Y B RE NNA N

... rational choice theorists do not generally adopt any fixed or narrow self-interest assumption, but leave the question of the content of agent preferences open for investigation. At least, that is what I would call ‘best practice’. Ferejohn (2002: 225)

1. Introduction Since Amartya Sen’s (1977) ‘Rational Fools’ (RF hereafter) was first published, a large literature has emerged around the themes of rationality assumptions in the social science. Sen himself ((2002) for example) has on a number of occasions revisited the various themes first aired in that paper, and has refined and developed his position with respect to a number of those themes. Nevertheless, the original paper is worth careful study—and is, in any event, the reference point for my efforts here. This focus is justified in large measure precisely because the original has acquired the status of a classic—and a classic particularly among those who oppose the use of rational actor methods in the social sciences and who, naturally enough, want to claim a Nobel Laureate in economics as an ally. Moreover, RF shares with many ‘classics’ the fate of being more often cited than read—or, if read, with its messages burdened with prior interpretation. Here, I want to argue for two propositions: first, that it is most useful to think of the rational actor approach as an ‘approach’ rather than as a theory; and, second, that if one does think of rationality in these terms, then Sen falls firmly within the rational actor school. I want to illustrate this

106

geoffrey brennan

latter claim by appeal to an example that Sen mentions (1977: RF, section V), but does not develop—namely the case of individuals voting in large number elections. I choose that example not just because it suggests the sense in which Sen is a rational actor theorist (broadly conceived) but also because it illustrates the advantages of pursuing—as Sen does not—the rational actor reasoning to its logical conclusion. I should emphasize at the outset that my characterization of what Sen actually achieves in this paper is rather different from Sen’s own. As Sen puts it: ‘This paper has not been concerned with the question as to whether human behavior is better described as rational or irrational. The main thesis has been the need to accommodate commitment as a part of behavior’ (Sen 1977: RF, section VIII). I am sceptical as to whether the chief value of the paper is in introducing the concept of ‘commitment’ as Sen defines it—and specifically as distinct from the contrasting notion of ‘sympathy’. I think that that particular distinction neither is as significant as Sen claims, nor does the work that I believe Sen thinks it does. Moreover, while I agree with Sen that there is ‘not much merit in spending a lot of effort in debating the ‘‘proper’’ definition of rationality’ (VIII), I do think it unhelpful to draw the definition too narrowly. And I shall try to indicate why in what follows. I shall develop my own thoughts in three steps. First, I shall set out (in section 2) some of the ‘many alternative conceptions’ of rationality and argue for an inclusive rendering—more or less along the lines that Ferejohn (2002) develops in a different connection, and that is represented in the epigraph. Second, I shall illustrate by appeal to Sen’s voting example how rational actor analysis might plausibly go in cases where problems emerge for ‘rationality’ under a narrow reading. This will occupy section 3. Finally, in section 4, I shall explore Sen’s striking example of the boys and their apples. I shall use this exploration to speculate as to why Sen thinks that commitment is significantly different from sympathy, and to suggest why I don’t think that the distinction does the work that I think Sen thinks it does.

2. Alternative conceptions of rationality One of the reasons why I think the meaning of ‘rationality’ should be broadly interpreted is pragmatic. Rationality can refer to so many different things that it is doubtful whether the ascription of irrationality could

the grammar of rationality 107 carry much meaning without a further specification of what aspect of the rationality apparatus one was rejecting. Some account of rationality is required in order to define irrationality—and that account of rationality had better cover as many bases as possible. Consider the following possibilities. R1: an agent is rational if and only if she chooses that action, A, that best fulfils her desires, D. R2: an agent is rational if and only if she chooses that action, A, that best satisfies her preferences. R3: an agent is rational if and only if given her beliefs, B, she chooses that action that she believes will best satisfy her desires/ preferences. R4: an agent is rational if and only if, given her beliefs, B, she chooses that rule of action (or disposition) N that would maximize expected preference satisfaction. R5: an agent is rational only if she acts on the basis of beliefs that are true (or maximally epistemically justified). R6: an agent has rational beliefs only if she does not increase her degree of belief in response to contrary evidence. R7: an agent is rational if and only if she acts in accordance with the way she believes she should act, all things considered. R8: an agent is rational (if and) only if her preferences are convex, complete, and transitive. R9: an agent is rational if and only if she maximizes her own well-being. R10: an agent is rational if and only if she does that which she has most reason to do. All of these conceptions of ‘rationality’ are in play in various contexts. I have settled on ten not because this exhausts the list of possibilities but because ten is a sufficient number to illustrate the range. R5 and R6 are accounts of rational beliefs; R8 is a definition relating to the structure of preference and is in wide use in economics in the derivation of so-called comparative static results. The remainder are accounts of action, or rather the connection between action and something else (belief, desire, or both). But in R3 the agent doesn’t choose actions directly: she operates instead via a disposition that isolates the action to be taken. This case might be viewed as close to the spirit of Sen’s ‘commitments’, but R7 and R10 are also

108

geoffrey brennan

consistent with the more explicitly normatized account of commitments that Sen seems to have in mind. There is no shortage of fascinating puzzles that arise under each of these definitions. Often such puzzles involve one definition being satisfied when another is not. Precisely what the various differences in definition amount to, and precisely what assumptions are being made when one particular version is chosen—these are interesting and potentially important questions. However, these questions are like family arguments and should not be allowed to obscure the conceptual and analytic connections across the various definitions. Different analyses may emphasize quite different aspects of the relation between desire/preference, belief and action, but the relation itself remains as a central defining feature of the rationality approach. So, for example, a model of rational choice that identifies the objects of choice as different possible sets of preferences over outcomes will obviously share many analytic features with a standard model of choice among actions. Even a model that focuses on the preferences the agent recognizes she ought to have will involve, downstream, the same kind of optimizing apparatus as in the more standard case. In fact, the purposes to which the rationality approach can be put are highly varied—and different purposes might call for rather different conceptions of rationality. But there will, I reckon, remain a reasonably close family resemblance between those applications, and it seems to me perverse not to recognize that resemblance. To put the point in RF terms, I agree with Sen that economists often work with an unnecessarily impoverished motivational apparatus and that for lots of purposes a much richer psychology would offer significant advantages. Part of the difficulty with (some) economists is their propensity to offer a highly restrictive version of agent motivation (narrow economic interest) and then insist that that version defines what rationality (or best practice ‘economics’) involves. A related feature is that in some circles it seems to count as a professional accomplishment to show how phenomena most naturally explained in terms of altruism (or commitment to norms) on the part of relevant agents can, with a measure of ingenuity, be explained as wealth-maximizing. I do not want to rule out ‘wealth-maximizing’ as a significant piece of many agents’ motivational apparatus in many contexts. But I consider it absurdly restrictive to insist that the ascription of rationality be confined to such cases.

the grammar of rationality 109 The defence offered for the use of narrow self-interest often runs in terms of methodological considerations of simplicity. Abstraction from psychological complexity is required, so the argument goes, to allow focus on other things. But think of the design of appropriate techniques of explanation as a ‘rational response’ to the desire for various metatheoretical ends—elegance and simplicity and predictive power and so on—in the face of ineluctable complexity. It is just a general analytic truth that optimization in this domain will typically require us to make trade-offs between those meta-theoretical ends. And it seems highly unlikely that the optimal trade-offs will be entirely independent of context. In particular the acceptable level of abstraction in relation to motivational assumptions will differ according to the nature of the problem addressed. Insistence on a ‘one-size-fits-all’ motivational abstraction seems to me to be just a failure to recognize how the ‘economy of method’ works. I should perhaps emphasize that, as I see it, the primary use of rationality is as a tool in social explanation. Rationality can, of course, also be understood as indicating how one ought to behave. It can be thought of as a directly normative account in that sense. Jon Elster (1990: 19), for example, remarks that, in his view: ‘rational choice theory is first and foremost a normative theory and only secondarily an explanatory approach. It tells people how to act and predicts that they will act in the way theory tells them to.’ I am not sure what determines the priority here—why the ‘normative’ aspect should be ‘first and foremost’. But in any event, I will here focus on rationality in the explanatory context. Having made that distinction, however, one should not think that adopting the assumption of rationality (however exactly interpreted) as a core piece of one’s analytic method carries no normative implication. Here, it is important to draw a distinction between two functions that the rationality assumption can perform. One function is to assist in the prediction of human behaviour and the explanation of social phenomena (or changes in them). The other is as a defence of institutions that respect individual choice. Call the first of these the ‘explanatory’ E function; and the second of these the ‘sovereignty’ S function. The S aspect is, I want to claim, essentially normative. It operates to defend institutional arrangements (such as the ‘free market’ in an appropriately idealized form) that allow individuals to make decisions on their own behalves and produce social outcomes that are consistent with their individual preferences. One

110

geoffrey brennan

can, of course, seek to defend ‘freedom of choice’ on grounds other than that it produces outcomes that are good for the choosers. But the S argument has traditionally played an important role in the defence of ‘liberal institutions’—and in my view properly so. It is important to draw the distinction between S and E functions in the context of the discussion here because they suggest different aspects of the notion of rationality. The E function calls for a conception of rationality rather like R8 and perhaps R9. In this setting, rationality operates as a kind of ordering principle, imposing consistency requirements on choice behaviour. But for this purpose it is not necessary to make any claims about the relation between preference-satisfaction and individual well-being. Consider the following thought experiment. Suppose that each individual has a utility function that accurately represents that individual’s ‘interests’ narrowly defined. This is the utility function that each would maximize were she rationally egoistic in the sense that I take Sen to have in mind. I should perhaps emphasize that ‘egoism’, as I understand it, is a reference to the content of individual preferences—not to the fact that the preferences attach to individuals. Of course, in the standard account, my desires are my desires. And that is true whatever the desires happen to be. But this is a truism and does not make the desires egoistic. Egoism is a reference to what the desire is for—my food, my housing, my health, my education, and perhaps more broadly my esteem.1 So, in the case I have in mind, each person’s preferences will track her own interests in the stipulated sense. The preferences will, however, differ among individuals according to their tastes. Some people love cilantro; others hate it. Some like apples more than oranges; others the opposite. And so on. Now, suppose we perform some permutation of the utility functions across individuals so that individual 1 is made to act according to 2’s preferences; 2 is made to act according to 3’s; and so on. It will remain the case that each individual will act in an essentially predictable fashion. She will respond to changes in relative prices in the way that the economic apparatus predicts. And (possible income effects aside) the aggregate outcome will be much the same as occurs when each individual chooses a consumption bundle consistent with her own preferences. But, by stipulation, there is a flawed connection between the actions and interests of each. I will not be choosing in a manner that maximizes my utility. So we cannot directly induce from mere consistency of choice (i.e. satisfactory fulfilment of the E

the grammar of rationality 111 requirements) an argument for liberal institutions based on the S-principle. The S-principle involves a connection between action and maximal wellbeing (or something else that should be promoted): the E-principle does not. Sometimes, the absence of a connection between action and well-being will not matter. It will not matter, in particular, if you simply want to predict the behavioural responses of the relevant individuals to some exogenous shock. Suppose, for example, that the permutation of preferences had occurred in country W. You, in country Z, want to assess the effect of a tax on your exports of some good X to W. You want to know how much of the revenue that the tax raises will be effectively paid by consumers in W rather than exporters in Z. In answering this question, you will need to predict the responses of W residents. But the prediction does not need you to enquire whether W-residents are reliably pursuing their own interests. As long as W-residents behave predictably, this is all that matters for your purposes. There is then a distinction between definitions of rationality that are focused on E purposes and those that are focused on S purposes. The latter require a connection between individual preference and some magnitude that is of normative significance. Rationality definitions that supply the latter connection (R1 to R4 and R9 arguably) are required. For predictability, all we require is that behavioural responses have the right kind of structure. Suppose one’s purposes are essentially predictive. Within the definitions of rationality relevant to this class of purposes (R8 and R9, say) there is a further useful distinction between restrictions on the structure and the content of preferences. The standard economics textbook will define rationality in terms of certain properties that preferences have to exhibit: completeness, transitivity, and convexity.2 Transitivity serves to rule out preference reversals. Convexity is the property that ensures (roughly) that demand curves slope downwards, and hence permit us to deduce outcome changes from changes in relative prices. These properties can apply whatever the content of preferences—whatever, that is, the particular things are that the chooser regards as goods, and whatever the particular values he places upon them. James Buchanan (2000[1969]) makes a useful distinction in this connection between what he calls the logic of choice and the science of choice. The logic is a matter of deriving general abstract propositions from the axioms

112

geoffrey brennan

of decision theory. The science is a matter of making predictions about human behaviour, and hence about the aggregate consequences of policy actions or changes in exogenous factors. Restrictions on the structure of preference are matters of the ‘logic’ of choice. Specifications of the content of preferences are a matter of the ‘science’. It is, for example, entirely standard in the teaching of economic principles to specify that ‘the demand curve for X slopes downwards’ and indeed to conduct analysis of consumption and production of X’s and Y’s without any further specification of what the X’s and Y’s actually are. ‘The task of filling up the blanks’ as Koko3 might have remarked, ‘I’d rather leave to you.’ In any application, however, the specification of the underlying objects of choice is essential. This is where the scientific element in the Buchanan picture enters. And this specification exercise is sometimes a good deal trickier than it might look. Both in the Sen example of the apples and in the case of voting—both of which I shall examine in more detail below—part of what is at stake is getting precise the specification of the objects between which the actor is making choices. It just isn’t always self-evident how these options should be specified. What I want to insist at this point, however, is just that there is a distinction to be drawn between the content of the utility function and the structure of that function. One can make assumptions about the latter that are completely general. And at least within the economist’s usage, these structural assumptions are usually identified as being more basic than any content assumptions. I take it that this is an important part of what underlies Ferejohn’s statement at the outset of this chapter. These structural assumptions and the ‘logic of choice’ in which they are embedded are a central feature of the ‘rationality approach’ as I would identify it. The assumption that individuals are, for example, income-maximizers is not a central feature. Which is not, of course, to deny that that assumption might be a useful abstraction in certain settings.

3. The rationality of voting I have indicated both in the title of my chapter and in the foregoing section my view that the rational actor/rational choice approach is best interpreted expansively and better understood as a kind of grammar of

the grammar of rationality 113 argument than as a particular ‘theory’. Precisely because rationality is open to a variety of interpretations and because any particular application of rationality assumptions involves a specification of the content of utility functions that is not itself part of the theory, the rationality approach is pretty accommodating. But it will still rule out lots of lines of reasoning as unacceptable. Unless the theory makes appeal to the motivations and beliefs of agents and to the availability of options, and unless it connects explanation of action to those elements, it will not satisfy the grammatical requirements of a rational actor account. One of the functions of that grammatical structure is to suggest ways in which apparent problems might profitably be handled. If some apparent anomaly arises, the grammar tells us where to look for therapy. Suppose, for example, that (some) individuals are observed in experimental situations to contribute to the provision of some public good in large-number social dilemma situations. In doing so, their behaviour flies in the face of any hypothesis that they would act so as to maximize their financial returns. Then we look to their motivations, and/or to their beliefs about the consequences of their actions, to explain why they did not behave as predicted. We attempt to induce their preferences from their actions—something that is a direct application of the rationality approach. We do not shrug our shoulders and declare the individuals in question to be simply ‘irrational’. Equally, suppose that individuals are observed in various contexts to restrict the set of options that will be available in future episodes of choice. The first instinct of the rational actor theorist is to attempt to show how such (apparently preference confounding) action can be made broadly consistent with their preferences/desires. Some commentators speak rather disparagingly of this response strategy as the ‘neo-classical repair shop’. But any such disparagement is surely totally inappropriate. For this strategy is exactly the one we ought to expect rationality theorists to adopt. It is nothing more or less than putting revealed preference logic through its paces. More generally, I regard it as the primary strength of any ‘analytic approach’ that it provides a list of ‘questions to be asked first’ when apparent failures of the approach are observed. This is not ‘saving the theory’ so much as it is using it. Here I want to consider briefly two examples of the ‘neo-classical repair shop’ in action—both based on anomalies that are mentioned in RF. The first involves the question of voting—an example that Sen mentions briefly

114

geoffrey brennan

at the end of section V of RF. The second relates to Sen’s striking example of the boys and the apples, the discussion of which I shall postpone to the next section. In the voting context, Sen queries the logic of ‘instrumental voting’. In an argument somewhat akin to the ‘rational ignorance’ hypothesis that Downs (1957) draws from Schumpeter (1950), Sen observes that: ... in large elections it is difficult to show that the voter has any real prospect of affecting the outcome by his vote, and if voting involves some cost, the expected net gain from voting may typically be negative. Nevertheless, the proportion of turnout in large elections may still be quite high; and I have tried to argue elsewhere that in such elections people might often be ‘guided not so much by maximization of expected utility, but something much simpler, viz. just a desire to record one’s true preference’. (1977: RF, section V)

My own view (see, e.g., Brennan and Lomasky (1993)) is that Sen is basically right to identify problems for the logic of self-interested voting; but I regard the move from ‘self-interested preference’ to ‘true preference’ as much too quick. I shall try to defend this latter claim in what follows. At this point, however, I want to emphasize a more general point—namely, that at each turn in Sen’s argument he is making an appeal to the logic of rationality. First, he identifies a flaw in any account of voting behaviour that depends on the idea that individuals are led to vote by the belief that in doing so they can rationally advance their interests. Second, he observes that large numbers of people nevertheless do vote in large-scale elections, even though they cannot rationally hope to influence the outcome. He could at this point have drawn the conclusion that voters are simply ‘irrational’. But instead, he looks to a motivation that would make their participation a rational act—in this case, a desire to record a preference for its own sake. All this so far seems fine. The only reservation—and it is an important one—that I have with Sen’s claims is the assertion that the preferences so recorded will be the agent’s ‘true’ preferences. Nothing that he says supports this particular claim—and I think there are good reasons to doubt it. Many people take it for granted that it is reasonable to expect, of ‘government’, moral attitudes that we do not expect people to exhibit in their ordinary behaviour. Within a democracy, at least, this is something of a puzzle. What seems to be at stake is an assumed asymmetry between people acting in their roles as voters on the one hand, and in their roles as

the grammar of rationality 115 consumers on the other. The assumption seems to be that individuals wear different ‘hats’ in the arenas of politics and markets—and as public choice scholars have been prone to insist, any such presumed ‘schizophrenia’ demands an explanation. In fact, the tradition within public choice theory has been to take it for granted that no such explanation can be given. The anti-schizophrenia presumption has licensed a direct extrapolation from market to electoral behaviour: voters and market-consumers are taken to be motivated by essentially the same concerns. But suppose we allow as a conceptual possibility that individuals might have ‘two hats’. Suppose we assign them two sets of preferences and then conduct a meta-preference analysis, using standard rational actor methods. If we do this, we can construct reasons within the rational actor account why electoral and market preferences might be different—and indeed, reasons why we ought to expect them to be different. As Sen puts it, ‘it is difficult to show that any voter has any real prospect of affecting the outcome by his vote ... ’ (V). Any rational actor account of voting must therefore reject the idea that voting can be explained as a means of bringing about a particular electoral outcome—unless of course voters nurture wildly implausible beliefs about the probability of being decisive in the election. Rather, voters’ actions are to be explained predominantly in terms of a desire to express a preference for its own sake. There is of course nothing especially odd about people having such a desire. We often express our preferences about states of the world in contexts where that expression has negligible effect on whether that state of the world comes into being or not. Cheering at sporting events, sending ‘get-well’ cards, writing letters to the editor of a newspaper, and just ordinary dinner-party conversation all exemplify. The question is: what can we say about the likely content of the preferences so ‘expressed’, as compared with the preferences ‘revealed’ in market settings? The fact that we have to live with the consequences of choices made in the latter context seems likely to make a difference. Let me elaborate. Most academics are familiar with the idea that preferences ‘revealed’ from behind a veil of ignorance, where one is stipulated to know nothing about one’s own individual location in social outcomes, are likely to be different from preferences revealed when such ignorance is absent. Rawls’ entire enterprise in his Theory of Justice (1971) rests on such a distinction. When each can know the consequences of choice only for everyone in general—and

116

geoffrey brennan

not for himself in particular—then preferences change. In the ballot context, there is a ‘veil of insignificance’ that is not altogether unlike the veil of ignorance in this respect. Each voter really can’t do much to affect the outcome—either for herself or indeed for anyone else. She is instead enrolled in an assessment of alternative social outcomes or policies or candidates or parties, and invited to decide which she will express support for. In such a setting, Sen suggests, people will not maximize expected utility: they will just reveal their ‘true preferences’. However, neither aspect of Sen’s suggestion here seems quite right. In the first place, in doing something other than voting for the policy outcome that would, if put into effect, maximize that person’s utility, the voter is indeed maximizing expected utility over the options that she actually confronts. The voter is not, in fact, choosing among alternative states of the world: she is choosing, rather, among alternative ‘preferences’ over states of the world, which she might express. When the domain of options facing the voter is properly specified, there seems no reason to suppose that she is not actually maximizing utility in choosing among the elements in that domain. One just has to bear in mind that the domain consists of votes not of policies. In the second place, it is much too quick to suppose that the only alternative to expressing a preference for that social state that would be best for you if implemented, is to express your ‘true preference’, whatever exactly that may be. Released from the discipline of consequences, one might express a preference for the candidate who looks nicest or has the nicest voice. Or for the party that one has always voted for before, or that one’s parents always voted for. Or for the name that appears higher on the ballot paper. Or randomly; or according to some whim or fancy of the moment. One might vote for one’s ‘political principles’. Or one might make a careful calculation of the alternatives according to some aggregate assessment of total utility, or total well-being with priority given to the least well-off. Which of these people actually do choose is a matter to be determined by empirical enquiry. The important point in this connection is that enquirers must be free to trawl in the data unencumbered by any necessary presumption that rationality postulates point in one way rather than another. In particular, rationality does not suggest any reason why individuals would vote as egoistic maximizers, even if this is broadly descriptive of the way they behave in market settings.

the grammar of rationality 117 One possibility of course is that, in an attempt to minimize cognitive dissonance, many people just vote for the candidate/party/policy that would if implemented be best for them as individuals—that is, they make the same kind of choice electorally as they would make in market analogues. We do, after all, need some explanation as to why the ‘pocket-book’ account of voting works as well as it often seems to. The pure logic of choice suggests that electoral and market behaviour ought to be very different. Without ceding too much ground to public choice orthodoxy, it may turn out that self-interest remains a reasonably potent motive in electoral contexts. And this could be true, even though other kinds of symbolic considerations obtrude—some of them good, like concerns for justice and aggregate human flourishing; and others, like xenophobia and aggressive nationalism and ancient enmities, much more dubious. My own view is that, though self-interest might well remain a significant determinant of voter behaviour in lots of cases, it also fails to be so in lots of cases—and lots of these latter cases will carry considerable normative clout. At the same time, although electoral preferences and market preferences are likely to diverge, there is no particular reason to think that electoral preferences will not have the same basic structure as market preferences. Voters can be taken to be ‘weakly rational’ in the sense that their individual voting behaviour exhibits transitivity and convexity (‘rational’, that is, in the sense of R8). The situation is a little like the example constructed earlier in which individuals somehow acquire preferences that are not really their own: the connection between preference and own well-being is missing. In the voting context then, the power of the rational actor approach is to declare open a question that might otherwise be closed. But there does not seem any support for the slightly mystical belief that, somehow or other, the ‘voice of the people’ gets it right. Sen’s claim that what people actually do in these circumstances is to reveal their ‘true preferences’ seems just another example of that form of democratic piety—and no better grounded. Public choice theorists, on no better grounds, have tended to model electoral behaviour as a direct extrapolation of market behaviour—with identical propensities to economic self-interest. In my view, public choice orthodoxy has tended to underestimate both the relevance of moral considerations in actual democratic processes and the possibilities for considerable perversity. There is the possibility that the electorate will behave in a morally heroic fashion. But equally, we cannot, on the basis of the logic of rational choice,

118

geoffrey brennan

rule out the possibility that there might be widespread popular enthusiasm for policies that are in the interests of almost no one. My concern here, though, is not to rehearse my ideas about electoral preference. I have done that elsewhere (arguably at excessive length—Brennan and Lomasky (1993) and Brennan and Hamlin (1998)). My main point here is a more general one about the rational actor approach. The ‘expressive’ voting story reveals, in my view, both the power and the complexity of that approach. And note that we do not need a great deal in the way of specific psychological detail to create that complexity. Where the psychological detail will be necessary is in making an overall normative assessment of the status of electoral and ‘market’ preferences. And as I see it, that issue is not one that ‘rationality’ assumptions on their own can answer.

4. Sympathy and commitment I want now to turn to another example—one that plays a rather larger role in RF. This is Sen’s striking story about the two boys and the apples. Recall the case. Boy A offers boy B a plate on which there are two apples, a large one and a small one. B takes the large one. A expresses a certain disgruntlement over the outcome. So B asks him: ‘Well, if you had been offered the plate, which would you have chosen?’ A responds: ‘Why, the small one, of course!’ And B quips: ‘Then what are you complaining about?’ Sen sees this story as ‘illustrating ... the contrast between sympathy and commitment’. He remarks that A would have ‘lost nothing from B’s choice had A’s own hypothetical choice of the smaller apple been based on sympathy as opposed to commitment’; and induces from A’s anger ‘that this was probably not the case’. I am unpersuaded that this is the message that the story reveals. It seems to me that Sen confuses here his own distinction between sympathy and commitment for a different issue—namely, what the precise objects of ‘desire’ are. So at the risk of weighing down a striking example with an excess of analysis, let us dissect the situation in somewhat greater detail. It seems most natural to think that what is at stake in A’s irritation is the failure of B to observe a norm of politeness. That norm states, let us suppose, that when confronted with a choice between a large and a small apple—especially

the grammar of rationality 119 when the other apple will accrue to the offerer—you should take the smaller apple. We can assume that, were A to come across the two apples hanging on a bough in a deserted wood, no inhibition about taking the larger apple would apply. The norm is quite contextual, as norms often are. Note that if this is so, then ‘preference’ between objects x and y is subject to a ‘preference reversal’ depending on context. When offered on the plate, x is preferred to y. In the wood, y to x. There is of course nothing mysterious or unintelligible about this even to an economist! But that is because the preference reversal arises only if we fail to specify the objects of desire accurately. There are two ‘goods’ here—apple size; and politeness. There is a trade-off between politeness and apple size, both of which are objects (we may suppose) of preference. In that trade-off, A would, he assures us, choose ‘politeness’ over apple size; B did not. I say there is no challenge to the ‘rationality’ of agents in this story. So I had better retell it with the objects rightly specified. There are several versions on offer. We might be tempted to suppose that B places no value on politeness at all. But we cannot induce that from what we know. All we can induce is that, whatever value the rational B places on politeness, it is less than the value he places on increased apple size. Perhaps the difference in the apples is very large—or B is especially fond of apples. For A, presuming that we can take his protestations at face value, the opposite is the case. In particular, we can take it that if B had responded to A’s offering the plate: ‘No. You choose’, then A would have taken the smaller apple. But note that this would have been a different outcome from that which occurred, because A would have had the chance of instantiating the norm and exhibiting his politeness. When A chooses, A gets politeness. When B chooses, A does not. In one way, this might seem just to be making Sen’s point—that conceptualizing the outcome as if only apples mattered is a mistake. But I concede that point. Sen is right on this. However, this is not the conclusion that Sen wants to drive to. What Sen wants to say about this case is something else—namely that it would be wrong to model A’s choice in the latter case as if A derived any satisfaction from norm compliance. The difference between sympathy and commitment, on Sen’s account, is that sympathy involves an exercise of preference whereas commitment involves counter-preferential choice. Of course, it would make a difference if we tried to explain A’s behaviour in terms of sympathy for B. Then,

120

geoffrey brennan

to be sure, A’s complaint would seem bizarre. But it would be bizarre whether A enjoyed B’s consumption of the apple more than he enjoyed his own consumption of it or because A had a ‘commitment’ (in Sen’s sense) to B’s flourishing. It seems to me that, in framing this example, Sen switches horses in his account of the difference between sympathy and commitment. Commitment and sympathy are supposed to differ not in terms of the objects of choice, but rather in the way in which the objects of choice connect to preference. Commitment to politeness is of course different from concern for B as such. But they are not different in a way that sustains Sen’s distinction between preferential and counter-preferential choice. Now, I have suggested that the preferential vs counter-preferential distinction makes no difference to behaviour. At least, in this case. There might be a difference of the following kind. Suppose C has a commitment to looking after D and makes transfers to D in amount T. Suppose that C∗ has sympathy for D∗ to an extent that is behaviourally indistinguishable: the transfer T is the same in both cases. Now consider the responses of C and C∗ to an increase of some amount in D (and D∗ s) well-being. One might conclude that in the sympathy case, C∗ will be made better off by the increase in D∗ ’s well-being; whereas in the commitment case, C will not be affected one way or the other. Because C has not brought about the increase in D’s well-being, that increase simply does not bear on C’s commitments. In summary, C∗ might be predicted to respond to the change in D∗ ’s circumstances, while C will not respond to an identical change in D’s. Again, however, whether C is appropriately to be thought of as deriving pleasure/utility/preference satisfaction from acting in accord with her commitments or not seems a side issue. One might of course postulate that C∗ ’s ‘sympathy’ is driven more by the spectacle of D∗ ’s distress than by any pleasure she derives from D∗ ’s flourishing. In short, I am not persuaded that the distinction between commitment and sympathy maps into the distinction between preferential and counterpreferential choice. Nor am I entirely sure what work the latter distinction is supposed to do. I do though have a conjecture in this connection. I suspect that a significant part of the motivation in distinguishing between commitment and sympathy is connected to the ‘equality of what?’ issues that have concerned Sen elsewhere. How, for example, are we to ascribe benefits to agents who make transfers to others? Sen seems concerned that

the grammar of rationality 121 the ‘rationality’ approach answers that question in a particular way—in a way, specifically, that presumes ‘sympathy’ rather than commitment. But is it the case that rationality assumptions imply that anyone who makes transfers, say to children, must be at least as well off by virtue of some kind of warm glow, as another person who makes no such transfers? I do not see that. Nor do I see that the commitment approach solves that issue one way or another. For example, someone who is ‘committed’ to feeding her children but is unable to do so presumably endures a suffering thereby. And does so no less than if she derives a warm glow from her children’s well-being. Simply put, I do not think that the rationality approach on its own can answer questions about what should be the basis of egalitarianism. So I do not see how an interest in this latter question can be grounds for a critique of rationality. What I think rationality in many formulations does suggest is that if, without any other change, we were to prevent the mother from making transfers to her children, we would make her worse off than she was before. But as far as I can see, the question of whether or not she would be worse off than an analogous mother who disregarded the interests of her children—and whether or not she was more deserving of public support—requires additional explicit assumptions. And it is far from clear that the distinction between commitments and sympathies bears at all on the issue.

5. Concluding remarks The object of this paper has been to revisit Sen’s influential and intriguing classic paper, ‘Rational Fools’. My main object in doing so has been to make a claim for a broad conception of the rationality approach—one that recognizes the range of meanings that rationality can claim, the range of contexts in which it can be applied, and the overarching family resemblance that the variants exhibit. I have argued for a conception of the ‘rational actor approach’ as providing not a theory of action so much as a grammar of argument about it. I have sought to indicate the power and range of this conception by appeal to a feature that any broad ‘analytical grammar’ provides—namely, a structuring of ways in which to respond to apparent anomalies. And I have tried to explore two examples of this usage—both of which receive some treatment in Sen’s paper. One is the

122

geoffrey brennan

case of voting in large-scale electoral contexts. The other involves Sen’s fascinating ‘fable’ of the boys and the apples. One moral I draw from both applications is that great care has to be exercised in specifying exactly the ‘objects of desire’ over which preferences are taken to be revealed in action. When those objects of desire are appropriately specified (in each case) the rational approach turns out to yield results that are quite insightful. My only complaint about Sen’s exposition is that he doesn’t follow the logic of the rationality approach quite far enough. That might be a second moral to be drawn from these applications—namely, not to give up on the rationality approach prematurely. But if Sen has not gone far enough, he may also have gone too far. Sen describes his paper as having two central messages. One is ‘the (un)acceptability of the assumption of the invariable pursuit of self-interest in each act’. The other is ‘the need to accommodate commitment as a part of behaviour’. I am entirely sympathetic with both these ambitions but I do not see them—as I think Sen does—as critically related. And I do not think that either message makes quite the assault on rationality that Sen seems to think it does. Moreover, I think the move to conceptualize behaviour-under-commitments as ‘counter-preferential’ is misleading, both in what this move implies about commitments more normally construed, and in what seems to underlie the rationale for such a conceptualization. In my view, it is perfectly proper to treat agents as having preferences for norm-fulfilment. And doing so has no implications one way or the other for the question of measurements of well-being as inputs into redistributive exercises. The rationality approach can do many things. But I do not think it can answer the question of what it is that egalitarians should equalize. Notes

1. Sen’s notion of ‘sympathy’ complicates this picture. As I understand it, an altruistic agent (in my sense, one who gives things away) might do so because he gets a warm glow from doing so, or because he derives esteem from being observed to be a generous fellow, or for some other ultimately self-referential reason. Such an agent evinces ‘sympathy’. 2. Sometimes called ‘concavity’ just to confuse the unwary.

the grammar of rationality 123 3. Koko is the Lord High Executioner in Gilbert and Sullivan’s Mikado, and sings a song about ‘little lists’ that is relevant here. References Brennan, Geoffrey and Loren Lomasky. 1993. Democracy and Decision. Cambridge: Cambridge University Press. Brennan, Geoffrey and Alan Hamlin. 1998. ‘Expressive Voting and Electoral Equilibrium.’ Public Choice 95: 149–75. Buchanan, James. 2000[1969]. ‘Is Economics the Science of Choice?’ In Collected Works of James Buchanan: Economic Enquiry and its Logic, volume 12. Indianapolis: Liberty Fund, pp. 3–21. Downs, Anthony. 1957. An Economic Theory of Democracy. New York: Harper & Row. Elster, Jon. 1990. ‘When Rationality Fails.’ In Karen Cook and Margaret Levy (eds.), The Limits of Rationality. Chicago: University of Chicago Press, pp. 19–46. Ferejohn, John. 2002. ‘Rational Choice Theory and Social Explanation.’ Economics and Philosophy 18: 211–34. Rawls, John. 1971. A Theory of Justice. Oxford: Oxford University Press. Schumpeter, Joseph. 1950. Capitalism, Socialism and Democracy. New York: Harper & Row. Sen, Amartya. 1977. ‘Rational Fools: A Critique of the Behavioural Foundations of Economic Theory.’ Philosophy and Public Affairs 6: 317–44. Sen, Amartya. 2002. Rationality and Freedom. Cambridge, Mass.: Belknap Press of Harvard University Press.

6 The Rationality of Rational Fools: The Role of Commitments, Persons, and Agents in Rational Choice Modelling ¨ H A ND H A RT M UT K L I E M T ∗ W E RNE R G UT

1. Introduction The most basic cognitive distinction a rational actor has to make is that between what does and what does not depend on her or his actions.1 An actor who chooses her act in each and every instance solely in view of the causal consequences of that particular act is ‘opportunistically rational’ if her choice behaviour at each and every single instance of choice-making can be described as maximizing some ‘utility function’ that represents her preferences over the causal consequences of the options of choice at the instance of choice-making. The interaction of opportunistically rational actors with each other can be described in a game model. In such a model ‘the rules of the game’ comprise everything that is beyond the causal influence of choice-making in plays of that (‘given’) game. The choice-making itself concerns the causal influence that can be exerted by opportunistically rational ‘choice-making within plays of the game’. In particular, when exerting their causal influence on the play of ‘a strictly non-cooperative game the players do not have any means of cooperation or coordination which are not explicitly modeled’ (Selten

the rationality of rational fools 125 1975: sec. 2). In a fully specified non-cooperative game model not only the preferences of actors (or the utility functions representing those preferences), the knowledge conditions, the probabilities representing beliefs etc. but also the individual options to establish cooperation or coordination are explicitly represented as part of the rules of the game. Besides the rules of the game that are modelled in the game model there are rules of interpreting game models. The latter fix how the model is to be understood. As far as non-cooperative game theory is concerned the most important such interpretative rule is a seemingly innocuous one: anything that is not explicitly showing up in the model is assumed not to exist when it comes to analysing the non-cooperative game model. This ‘explicitness condition’ also determines what is assumed to be common knowledge and thereby defines an ‘object’ of classical ‘eductive’ game theory (Binmore 1987/88). Only this can be analysed in terms of reasoning about knowledge of or by the players (i.e. as a game in the classical sense). The explicitness condition is the reason why non-cooperative game theory may justly be regarded as more fundamental than (partly) cooperative modelling efforts. For instance, the aforementioned processes of coalition formation that are assumed to be operative behind the scenes in many cooperative game theoretic approaches can always be modelled explicitly. What it means for several individual actors to act as a ‘corporate actor’ can be resolved into inter-individual relations in a game internal to the ‘corporate actor’ (or to a ‘coalition’). In this game all the internal commitments of the corporate actor are explicitly modelled as rules of the game.2 For instance, if the corporate actor decides according to simple majority rule internally then this amounts to an internal commitment of the corporate actor. The internal voting game itself can be explicitly modelled as a way to ‘predict’ the actions of the corporate actor in outside relations. The latter can be ‘explained’ as a result of internal commitments along with the opportunistically rational choices of the voters or internal agents of the corporate player (Brennan and Kliemt 1994). We will drive non-cooperative game theoretic modelling to its extreme by treating persons as a kind of ‘corporate actors’ of opportunistically rational maximizing agents. Thereby we intend to illustrate that and how any ‘commitment ... of a person choosing an act that he believes will yield a lower level of personal welfare to him than an alternative that is also

126

werner g uth ¨ and hartmut kliemt

available to him’ (Sen 1982[1977]: 92) can be expressed within agent-based (but otherwise standard) rational choice modelling (RCM). Sen’s challenge of rational choice theory (RCT) in his ‘Rational Fools’ can at least in part be met by carefully distinguishing between substantive claims of RCT and modelling rules of RCM. Taking recourse to agent-based modelling, it can be shown that personal actors formed exclusively of rational fools need not be rational fools. However, taking the agent-based ‘way out’ comes at the price of giving up the eductive or classical approach to games in that game models can no longer plausibly be analysed in terms of the reasoning of the players themselves.3 So what is a refutation of Sen in one sense is a corroboration of his claim that rational choice approaches and RCT get into trouble when it comes to modelling commitments, though the troubles may be somewhat different from those Sen envisioned.4 The next part of the paper demonstrates how far non-cooperative game theoretic modelling can conceivably be pushed (section 2).5 It is then shown how sticking to the explicitness assumption of non-cooperative rational choice modelling can make us aware of the many ways of smuggling commitment power into the rationality concept (section 3). That the preceding, if true, will also provide a convincing defence of backward induction within RCM (rather than RCT) is almost trivial, but for the sake of completeness we make it explicit again (section 4). By way of conclusion we then reiterate that there is a fundamental tension between the explicitness assumption of non-cooperative game theory and the classical eductive way of interpreting non-cooperative game theoretic reasoning as something that can be on the players’ minds (section 5).

2. Modelling commitments explicitly Distinguishing between rational choice theory (RCT) that contains substantive assumptions about rational choice makers—about who is a player (Güth 1991) and what is on that player’s mind—and rational choice modelling (RCM) that does not, we will focus primarily on the latter here. In what may be seen as our agent-based move towards ‘picoeconomics’ (Ainslee 1992) we show first that in RCM all sorts of personal commitments can be modelled explicitly as part of the rules of the game and that this claim

the rationality of rational fools 127 is quite independent of any substantive assumptions of specific versions of RCT. 2.1. Utility in RCM and in RCT As long as utility was a quality in itself—for example, measuring levels of well-being—it could figure among the reasons for preferring one option over another one. As such it could conflict with other reasons for positioning options in a hierarchy of relatively better or worse options. But, according to the modern concept of utility as used in RCM, an option is never preferred because it has higher utility. The higher utility is assigned to the option because it is preferred (for whatever reasons). In forming the preferences over options all the reasons (whether material well-being or some high ideals, etc.) are taken into account and the utility function ‘afterwards’ represents these preferences all things considered. What appears as a kind of (rational) foolishness of opportunistically rational actors is a direct consequence of relying on ‘representative utility all things considered’. Whatever can make human action intelligible and reasonable in a substantive sense is concealed from our view by representing preferences through utility in RCM. At least if it is correctly interpreted utility just indicates the fact that an option has a certain relative rank in a set of options after all the reasoning and, in particular, the ‘weighing of goods’ has taken place.6 What shows up at the end nodes of a game tree is a ranking information without substantive content, a pure ranking or a measure without dimensionality (though ‘cardinality’ is needed to deal with risk). The ranking may or may not be reasonable as evaluated in substantive terms, and therefore behind ‘the veil of the utility representation’ the wise can look exactly like the fool (on how that relates to the concept of revealed preference, including very telling citations of the original behaviourist interpretation of the concept by, e.g., Hicks and Little, see Sen 1982[1973]). As has often been stated the only remaining substantive assumption is ‘consistency’.7 Rationality as consistency requires that evaluations must be such that we avoid becoming victims of a money pump or a Dutch book. Otherwise utility is a stenographic device for presenting orders (over lotteries) as formed after taking into account all dimensions of value (whatever they may be). As such it gives us hardly any information about why and how the inclusion of these dimensions of value has affected

128


the ordering that is represented. Neither does it—beyond the ranking itself—inform us about intentions ‘behind’ utility functions. 2.2. Commitments in RCM Commitment power can show up in one of two ways in the rules of the game: payoff modification or modification of the move structure of games (elimination or addition of moves). In the most simple and pure cases commitment manifests itself either through the choice of one sub-game from a class of sub-games with identical game form8 but different payoffs, or through the choice of one of a class of sub-games that can all be derived by removing some of the moves from one ‘master game form’.9 Take the example of Ulysses. If he employs some sailors to crack the whip he is using the first commitment technique: since he expects to be beaten should he make the slightest move towards going overboard when hearing the Sirens, his payoffs are modified by the expectation of pain. If Ulysses orders to be bound to the mast he uses the second technique since the option of jumping overboard is ‘cut off’ the game tree so to say. In the first case, Ulysses in a move preceding the sub-game chooses between one sub-game with the original and one with modified payoffs; in the second case he is choosing between a sub-game based on a full tree and one of which at least one branch is cut off. 2.2.1. Commitments in simple trust games or credible promises To illustrate, consider commitment options for simple trust games. A paradigm case of such a game is presented in figure 6.1 with payoffs in the order of (A, B).

Figure 6.1. The simple trust game

If the second mover B is to move at all his preferences are such that he should choose to move down, DB . The first mover, A, foreseeing that, should choose down, DA , as well. Among opportunistically rational actors the obvious solution of the simple trust game therefore is (DA , DB ). Though both actors could have been better off had they both chosen to ‘continue’

the rationality of rational fools 129 according to CA , CB , in the simple trust game the Pareto superior result corresponding to the strategy combination (CA , CB ) is closed off. It is closed off by (commonly known) opportunistic rationality itself since the fact that the second mover B is known to be uncommitted to the move CB makes it impossible for A to trust by moving on. With an additional commitment option for player B a Pareto superior result could be reached. It could be reached in ways compatible with opportunistically rational choice-making and the common knowledge thereof if the additional option could be chosen opportunistically rational. This would be the case if the player moving second in the original trust sub-game could, by a prior commitment move, influence which ‘modified’ trust sub-game is to be played. First, he could make moving down less attractive for himself than to continue (corresponding to Ulysses engaging the sailors to crack the whip) and inform the other player about that move. This first possibility is depicted in the first game of figure 6.2 as ‘relative commitment’. By payoff modification in the sub-game reached after an initial choice to commit by B it is less attractive for B to choose DB than CB . Player A who knows that B has chosen to become relatively committed foresees this and therefore a play of (CA , CB ) emerges once the sub-game after an initial ‘commitment’ is reached. Since the solution of the sub-game after B chose to commit is superior to the solution of the original trust sub-game also in B’s terms, B obviously should choose to commit and the Pareto superior result should be reached due to the presence of relative commitment options. Second, the Pareto superior result could be reached in opportunistically rational ways also if by a prior move player B—the second mover of the original trust sub-game—could move first to become absolutely committed (corresponding to Ulysses being bound to the mast) and could perfectly inform player A about that commitment. This second possibility is depicted as the second game tree in figure 6.2. Both trees include the original or basic trust game as reached if the actor does not, and the modified game if he does, commit by his initial move. 2.2.2. Commitments in simple ultimatum games or credible threats The following example in figure 6.3 of ‘take it or leave it’, which contains both relative and absolute commitment possibilities, can illustrate in more detail

130

werner g uth ¨ and hartmut kliemt A

CA

CB

B

do not commit

B commit

DA (0,0)

A

(1,1)

CA

B

CB

(1,1)

CB

(1,1)

do not commit DA

DB (-1,2)

(0,0)

DB (-1,2)

B commit A

DA (0,0)

CA

CB

B

A

(1,1)

CA

B

DA

DB (-1,0)

(0,0)

Figure 6.2. Relative and absolute commitments

how flexible the tools of RCM are. So let us assume that two personal players have to play the game shown in figure 6.3. It is obvious that take it or leave it in strategic form has two equilibria in pure strategies, namely (+, +) and (−, −). However, if we go on to its extensive form in figure 6.4 it becomes immediately clear also that only (+, +) is plausible in terms of opportunistically rational choice-making. The second-moving player, B, will give in to the ultimatum of the first mover, A. This is anticipated by B himself as well as by A. If the second-moving player commands relative or absolute internal commitment power he may, however, issue a credible threat. So assume that as a matter of fact B has such powers. As a personal actor he may, for instance, be intrinsically motivated to execute threats as uttered in pre-play communication. +

-

+

1,0

-2,-2

-

0,1

0,1

B A

Figure 6.3. Take it or leave it in strategic form

Again this can be modelled quite simply. Like a corporate actor the personal player possibly may be dissolved into several agents who play a game internal to the personal actor. To say that this is conceivable does not amount to claiming that it must be the case. As there may be ships with

the rationality of rational fools 131

Figure 6.4. Take it or leave it in extensive form

and without a mast to which Ulysses could be bound there may or may not be commitment power internal to a personal player. But if there is such commitment power then it should and indeed can—as we claim—be modelled explicitly in non-cooperative game models of RCM. After such efforts a game like ‘take it or leave it’ with explicitly modelled commitment options internal to the second-moving personal player might emerge. For the sake of specificity we split the first-moving player, A, of the original game into three agents (corresponding to the three possible decisions to be taken in the sub-games of the enlarged game) and the second-moving player into four agents (corresponding to the four possible decisions of the personal player B in the new sub-games). The ‘splitting’ yields two ‘classes’, A and B, of agents representing the two personal players A=(Aii , Aiv , Avi ) and B=(Bi , Biii , Bv , Bvii ). With payoff vectors representing the payoffs according to (Bi , Aii , Biii , Aiv , Bv , Avi , Bvii ), we arrive at the ‘modified take it or leave it’ game as depicted in figure 6.5:

Figure 6.5. Modified take it or leave it game

The nodes corresponding to the seven agents in figure 6.5 are numbered by Roman numerals. At the first node, agent Bi of player B must decide whether to play the old (sub-)game played by agents Avi , Bvii . This old subgame is reached when no commitment is made by choosing −c. Making an absolute, c, or relative, c∗ , commitment, two modified sub-games are reached. The first agent of B has an incentive to choose either c or c∗ since

132


in both cases the backward induction solution of the sub-game would yield 1 for that agent (as well as all other agents of the personal player). For instance, if the agent Bv of B were to move then she would use ‘−’ since −3 < −2. But then the agent Aiv of A would have to opt for ‘−’ as well since 0 < −2. Likewise we would see that the sub-game that could be reached by the choice of the absolute commitment option c yields the corresponding result. We have assumed here that all agents of a personal player would evaluate end-results in the same way and at each move of a play would be motivated only according to preferences over the whole path. The assumption that a personal player at different decision nodes is endowed with ‘identical’ preferences is thereby explicitly modelled. It is an empirical assumption about the decision-making process of a personal player which may or may not be true.10 RCM does not rule out in principle that the utilities could be different for different agents of the same personal player at different locations of the tree. Since in the ‘agent form’ representation of the game a separate agent for each information set is included, this is obviously possible.11 By virtue of the agent form RCM allows for an explicit modelling of commitments, changing preferences and so forth of a personal player. At the same time it does not imply the strange view that persons of the real world are split into as many real players. Relying on the agent form representation does not force us to accept an empirical assumption about multiple or split selves in a real-world meaning of those terms (see also Ainslee 1992 and Elster 1987). Quite to the contrary, in the commonsense meaning of ‘player’ there is only one ‘personal player’ still. Only for theoretical purposes of making explicit all assumptions about personal identity, the internal commitment technology and the decision-making process of players, each player is split into several choice-making agents.12 At each information set of the tree, we must know the value order relevant for the decision at that node. The lists containing one payoff for each agent of each player, at the ends of the game tree, show how each agent of each player is affected by any decision (and in particular as seen from the point of view of the node where she is to move).13 Those game models in which personal players who make several decisions are represented by one agent, implicitly make an empirical assumption which may be true or false. We have to read that assumption as the requirement that evaluations or preferences of a personal player remain constant independently of how the

the rationality of rational fools 133 game is played. In those cases in which the assumption applies not much is lost by the simplification. However, there is a tendency to forget that an empirical assumption has been made. Using the agent form, rational choice modelling, RCM, can take into account deviations from the specific theory of personal identity which requires that all agents of a personal player always have identical evaluations of all results of the process of choice-making. Vice versa, if we assume all evaluations to coincide in this sense by not splitting the personal choice maker into agents then we implicitly assume that all agents (or at least all who still have to make decisions) have the same evaluations independently of the path through a tree. Nothing in the rules of non-cooperative game modelling precludes that we dissolve personal actors into agents. Analysing the interaction situations fully into what, respectively, is beyond and what is subject to causal influence in separate acts of choice, it seems rather natural to identify each instance of choice-making with a separate agent.14 For each instance of choice we have a separate utility function representing the preferences of the agent who is active in that particular instance of choice-making. It should also be noted that this view is almost implied, at least very strongly suggested, by the modern interpretation of the utility function as representative of preferences ‘all things considered’.15 For, if utility represents the preferences at a specific node all things considered, then the assumption of having everything considered at that node separates that decision from decisions at other nodes. This applies even if decisions at different nodes are taken by the same person. All that is relevant at any node is included in the evaluation at that node and only that.

3. Mixing up RCM and RCT? Nobody intends to maximize a utility function. Intending other things and having reasons other than utility per se the actor behaves as if maximizing utility. Starting from representative utilities sets economics free of the necessity to consider motives and intentions explicitly (they have all been considered in forming preferences as represented by the utility index). Economists are led to endorse the illusion that (cognitive) psychology is irrelevant for economics since everything that this science could contribute is ‘considered’ in the utilities. Whatever it is that motivates humans and

134


what they intend to do, they behave as if they are maximizing their utility function. With the utility function in hand the economist seems to have all she needs to know about choice makers, their motives and intentions. However, though the elimination of human motives and intentions is at the root of the specific style of modern rational choice modelling, many economists ‘want to have it both ways’. They try to bring in motives and intentions again, thereby mixing up RCM and RCT. Again efforts to modify game theoretic rational choice modelling can serve as an instructive example. In game theory, intentions could conceivably show up in moves during a play of the game or in the players’ strategies of which such moves are a part. We think that in both cases ‘commitment’ power is smuggled in. To see how let us discuss the two closely related possibilities in turn. 3.1. Forward induction and signalling of plans In game theory there is a long tradition of assuming ‘omnipotent players’ (von Neumann and Morgenstern 1944; Kohlberg and Mertens 1986). ‘Forward induction’ becomes viable among such actors. By their prior choices—that form first steps of a strategy that potentially contains later ones—such actors can allegedly signal how they intend to go on in a later choice-making.16 Consider the game shown in figure 6.6. 1 O*

O

2 L

R

1 l (2,0)

(1,4)

1 r

l

(0,0) (0,0)

r (3,1)

Figure 6.6. An embedded battle of the sexes

In the game of figure 6.6 there are two pure strategy sub-game perfect equilibria, namely ((O, l), L) and ((O∗ , r), R). The sub-game starting at the single instance of choice of player 2 contains two sub-game perfect pure strategy equilibria: ( l, L) and ( r, R). By choosing O∗ the choice maker 1

the rationality of rational fools 135 indicates that he intends to go for the second sub-game perfect pure strategy equilibrium. According to the standard argument the choice of O∗ would not make sense17 unless player 1 were to have the firm intention to choose r along with the expectation that player 2 would understand this and act accordingly by in fact choosing R. Therefore, according to those accepting forward induction, the equilibrium selection problem for the ‘battle of the sexes’ sub-game that is starting with the decision node of player 2 is solved in favour of (r, R), although whatever speaks for (r, R) speaks even more for ( l, L) when considering only the sub-game. If equilibrium selection in the preceding or a fully symmetric battle of the sexes game is regarded as impossible (see Sugden 1991) then, if we accept that there is a solution in the larger game, it seems to follow that the past matters. For, if the game were not embedded in a larger game tree no such selection could be made. According to adherents of forward induction, after O∗ a selection can be made for a good reason. Moreover, since in the sub-game no violation of such principles as backward induction is committed, the selected equilibrium solution of the sub-game is fully in line with forward-looking rational choice in that sub-game. We do not deny that the preceding argument, if made in terms of objective rather than subjective payoffs, is psychologically compelling. But note that in standard game theory such an argument is not made in objective payoff terms. It is formulated in payoffs which represent preferences that include all considerations relevant with respect to the future. If the future is indeed all that matters and if we take the notion of a ‘saturated’ utility function seriously then the sub-game that starts with the decision of player 2 must have a solution independently of any preceding history of the play. Whether that game would be analysed all by itself or as embedded in a game as in figure 6.6 should, once the sub-game is reached, not influence the choice-making of players who choose in view of the causal consequences of their acts in that sub-game. Therefore the presence of the option external to the sub-game ensuing after O∗ should not have any influence on the solution behaviour in that sub-game. Still there is the argument that the strategy (O∗ , l) is dominated by both (O, l) and (O, r) and should therefore be eliminated. So, if move O∗ is observed and if a rational player will not choose dominated strategies then this observation shows that player 1 must have adopted (O∗ , r). But this is true only if the player is omnipotent in the sense of being able to choose

136


whole strategies. As we will illustrate in some more detail now, the latter is contrary to basic assumptions of opportunistically rational choice-making (or, for that matter, incompatible with modelling choice behaviour in such terms). 3.2. Choosing strategies as actions and as plans Strategies can be ‘chosen’ only as plans for the game (being complete in specifying a move for each contingency that might conceivably arise in the game). Strategies are not choices that can be made in the game. If intentions are ascribed to another player by means of strategies then it must be possible to choose strategies in a stronger sense than forming a plan for choices to be made. It must be possible to choose strategies ‘in’ the game. Note that in that case there must exist corresponding moves in the game (after all, by assumption, in a non-cooperative model all possible moves show up explicitly). Such moves show up as ‘commitment options’ in the tree. The language of RCM as such does not commit us to any substantive assumption about whether or not such commitment options exist. Yet, according to the explicitness condition, the ways to signal intentions in the game must be modelled explicitly if they exist. A personal player may or may not have these capacities. But if he can commit to a conditional strategy and thereby signal his intentions then we should model this option explicitly as an additional choice to be made in the game. If we do not model it explicitly, then, according to the rules of interpreting models of RCM, it does not exist when analysing the model—and that is the end of it. Again a specific example may be helpful. Consider in figure 6.7 what may be called a ‘sequential PD’ with perfect information. Strategies are plans that specify an action for each contingency that might arise in any play of the game. If an actor in the role of a second mover in a sequential PD has the option of actually choosing the behavioural programme for each contingency before the game is played, the normal form bi-matrix game shown in figure 6.8 is an adequate representation of the strategic situation. It contains four rather than two choice options for the second mover. This strategic form of the sequential PD with perfect information is categorically different from the conventional tabular representation of the original PD in strategic form with ordinal payoffs as preference representations. It shows clearly how easily we are led astray by commonly used

the rationality of rational fools 137 A CA

DA B

B CB

DB

(3,3)

(1,4)

D‘B

C‘B (4,1)

(2,2)

Figure 6.7. Sequential Prisoner’s Dilemma

Actor A

Actor B CB/CA C’B/DA

CB/CA D’B/DA

DB/CA C’B/DA

DB/CA D’B/DA

CA

3, 3

3, 3

1, 4

1, 4

DA

4, 1

2, 2

4, 1

2, 2

Figure 6.8. Strategic form of sequential Prisoner’s Dilemma

phrases like ‘choosing a strategy’. Strictly speaking strategies are plans. We can choose to form one plan rather than another for a game. But we cannot choose the execution of the plan in one single act in the game. This is not among the options of the original sequential PD as presented in the extensive form. To choose a strategy in one act rather than merely planning to make several choices consecutively presupposes that there are ‘higher-order’ options to make such choices. Vice versa, the strategy as plan states how B intends to react to both CA and DA . It is, however, impossible to actually react to both, whereas a reaction function that contains pre-programmed responses to both can be selected only if the option to make that choice (the choice of a reaction function or behavioural disposition) in fact exists. The extensive game representation of the preceding strategic form representation would be the one shown in figure 6.9 (leaving out the ‘primes’ distinguishing equivalent moves CB and C’B etc.): Owing to the presence of the commitment option the relevant choices of B are made independently of A’s choice. After that, B does not make any choice any more. Once the information about A’s choice transpires, B’s programmed response follows. In figure 6.9 player B cannot inform player A about his commitment choices. All decision nodes of player A are included in one information

138


Figure 6.9. Standard Prisoner’s Dilemma with power to commit to strategies as programmes

set. If contrary to that, A’s information partition were to consist of four separate singleton sets (i.e., if in figure 6.9 the connecting lines between A’s nodes were to be removed), then B could make his strategy choices and inform A about them. Under such conditions of perfect information the commitment power of player B becomes sufficient to solve the problem of avoiding the Pareto dominated result and the strategy combination (CA , (CB /CA &DB /DA )) would be chosen.18 It may well be that individuals command the faculty to modify their future behaviour by simply planning and intending to do certain things in the future. It may also be that conventional rational choice modelling does not provide the best or most adequate tools to model such phenomena. But if we use the signs and the semantics that rational choice modelling supplies we should stick to the rules that go along with that ‘language game’. If we do so we must model commitments as part of the rules of the game explicitly. Therefore, if ‘commitments to plans’ affect the future then this must be modelled as a causal effect.19 Payoffs are construed for players who do intend to discriminate in their view of the world between what is and what is not subject to their causal

the rationality of rational fools 139 influence at each instance of choice. So if game theoretic logic ever does apply then it should apply also with respect to the rest of the game tree. If it is assumed that players, by reaching nodes, signal an intention of further play, what can that mean if the payoffs at any future node express the preferences operative at that node? In a theory that is based on forwardlooking opportunistic choice according to given preferences (operative at the very instance of choice-making), signalling an intention requires that players had the option of choosing a strategic commitment. But once that option is explicitly modelled the game tree is changed either in the ways of figure 6.9 or by introducing additional players in ways akin to figure 6.5.20 Analogous arguments quite straightforwardly apply to the notorious backward induction problem. If we stick strictly to the requirement that for the rational choice maker only the expected future matters, that his expectations are fully represented in his subjective probabilities/utilities, and that all influences on the future must be causal and as such be explicitly modelled, backward induction seems rather obviously implied. For the sake of completeness, let us turn to the somewhat confusing debate. Even though from our point of view everything necessary has been said already it may be useful to apply it to this ‘hard case’.21

4. Having it backward According to a common view, violations of backward induction on preceding stages of games should lead to the conclusion that backward induction should not be used for analysing the remainder of the game. The standard argument being, if you reach a node at all that you should not have reached according to backward induction arguments, how can you then still assume backward induction to apply from where you are to the ‘remaining’ future? In discussing backward induction it seems reasonable to ask how a game tree is to be derived in the first place. If it was written down correctly in agent form under the explicitness condition then the preferences and branches of the tree at each decision node must have been fixed factoring in the fact that there was a preceding history that may have influenced the evaluation of the future at that node. The payoffs for the future as fixed in construing the tree take into account that the node can be reached only by a play of the game that violates backward induction. If that violation does not

140


affect the rules of the game, in particular the payoffs, at later rounds of play all things considered then the payoffs are what they are ‘all things considered’. Assumptions of forward-looking rational choice seem to separate any informationally closed sub-game of a larger game from its preceding history. Edward McClennen, who is presumably the philosopher who has been objecting to backward induction most strongly over the years (McClennen 1990), is clearly right in insisting that a requirement of ‘separability’ is crucial for backward induction to emerge. He rejects separability as intuitively implausible. But separability is almost implied in RCT if we take seriously the view that rationality requires the intention to distinguish between what is and what is not a causal effect of an act (along with the desire to improve one’s situation). Nevertheless, since it is only ‘almost’ implied one might conceivably doubt that RCT implies separability. But it is certainly implied by the way RCM proceeds. Without separability the whole enterprise of game theoretic analysis by ‘separate’ parts of a game would become non-viable. To see more specifically what is involved here let us introduce a somewhat coarsened form of the separability principle (McClennen 1998: section 3, in particular pp. 21–2): Let T be a decision or game tree. Consider a node s in a singleton information set and an informationally closed sub-tree T/s that emerges after all nodes preceding node s are cut off while s and all its subsequent nodes remain as a complete sub-tree. Then, according to separability, the solution of T/s remains the same: (a) if T/s is considered separately as a game (b) if s is reached in T when playing the larger game. Backward induction appears so paradoxical because we intuitively tend to forget that according to the semantic rules of RCM, utility is representing preferences all things considered (including causal influences on preferences by going through a game tree along a specific path). We always tend to relate to utilities as if they were objective payoffs and to treat them as if they were reasons for action. But once we argue that we prefer a higher utility to a lower one we apply the concept of preference to the representation of preferences and thereby are already using categories akin to metapreferences. We treat utility as if it was money and thus as something we have preferences about. In the next step we are deceived into thinking that

the rationality of rational fools 141 we prefer an act ‘because’ it leads to higher utility. But utility represents preference and is not itself an object of preference. If we make it an object of preference then we are reasoning about the tree as if we preferred to get the more preferred results in the tree. According to the foes of backward induction we signal this alleged preference for higher utilities by violating backward induction. But this is exactly what an opportunist cannot signal under conditions of perfect information about utility functions that represent preferences all things considered. We cannot have it both ways: we cannot treat utility as representing preferences after all reasons for the ranking of alternatives have been considered—thereby avoiding ‘psychology’ and the necessity to speak of human motives—and then treat utility as merely one consideration among all others. We cannot meaningfully assume that preferences describe completely what is preferred at each decision node of the tree—because in writing down the tree we have fully analysed what it is like to be at that node all things including the path leading to the node considered—and then renege on this assumption if a decision node is in fact reached. Either our modelling tool is RCM or it is not. If it is RCM then backward induction seems to apply for the almost trivial reason that the game tree must be set up in ways that factor the past into the utilities relevant at each node. Then the causal influence of the past is ‘in’ the payoffs representing preferences relevant for forward-looking choice-making at any node of the tree. There is no backward induction puzzle in RCM. Backward induction must rather form an integral part of that kind of modelling. We have illustrated that and how RCM can be applied. Still one may wonder whether it is worth it. Or, to put it more bluntly, what are the merits of going to the extremes of RCM described before? Let us finally turn to some answers to such queries and to future avenues of research that seem open from where we stand.

5. Concluding remarks on eductive analyses, objective success, and psychology Economics always had a clear perception of opportunism and was willing to insist on its presence. This gave economics the edge over the wellintentioned normativism of much of social science and social philosophy.22

142


Non-cooperative game theory has been instrumental to spelling out all the implications of the traditional economic focus on opportunistic rationality. But in spelling out the implications it became completely clear, too, that the traditional (in particular also the traditional philosophical) criticisms of the model of opportunistic rationality as applied to personal players were justified. As a matter of fact human personal actors are not rational fools and a theory that makes an empirical claim to that effect is patently wrong. However, at the other extreme, a theory that completely eliminates the human faculty to act opportunistically will not lead to adequate accounts of human social interaction either. Rational choice modelling as such cannot tell us anything about the real world but can possibly help us to state more clearly the theories and views that we may hold about the world. Again the test of the pudding will be in the eating and it would be very interesting to see whether RCM, besides its obvious philosophical merits, has some empirical merits, too.23 But quite apart from such possibilities of using rational choice modelling as a tool for formulating empirical theories about the world, whenever we observe phenomena as being fully in line with assumptions of opportunistically rational choice we had better acknowledge that we have an explanandum not an explanans. We are provoked to ask how it was possible that things appear as if brought about by opportunistically rational choice. Convincing answers to this question, such as Armen Alchian’s model of market selection or Vernon Smith’s experiments on how markets clear, are formulated in terms other than rational choice—or other than Homo œconomicus (Alchian 1950; Smith 2000). Markets are substitutes for individual rationality and work even with rational fools or, for that matter, so-called ‘zero intelligence players’ (Gode and Sunder 1993). The assumption that preferences are ‘given all things considered’ (rather than opportunism as possibly conceptualized in terms of objective payoffs) is behind the impression that economics perceives the world as if populated by rational fools. It simplified economics and rational choice theory greatly. But the widely shared hope that economics, relying on the concept of representative utility, might not need a foundation in (cognitive) psychology is completely mistaken. It seems surprising that theorists of social interaction could ever have believed that they could understand the interaction of rational persons by reducing them to a utility cum probability function. If any real progress is to be made, economists will have to reconsider their

the rationality of rational fools 143 basic explanatory strategy and will have to look behind the veil of the preference and belief representation by ‘utility cum probability’. The ability to distinguish between what is among the causal consequences of our choices and what is not, as well as our faculty to seize opportunities, is clearly related to those higher faculties of the mind that we commonly associate with ‘human reason’. Without the ability to act as opportunistically rational beings it would hardly be conceivable to speak of human rationality as we know it. Saying this, we are quite willing to concede that there may be other forms of rationality besides ‘opportunistic rationality’ than those that are suggested by too na¨ıve an interpretation of RCM or RCT. These other forms of rationality may include all sorts of rule-following behaviour. Since the latter exhibit certain types of ‘boundedness’ they should, however, be classified as ‘boundedly rational’ behaviour. Calling them boundedly as opposed to opportunistically rational we are not implying that they are inferior forms. They may well be superior in leading to superior outcomes as measured in objective terms. In any event, a theory of social interaction in terms of the cognitive psychology of boundedly rational choice-making will lead to insights superior to any theory starting from the rational fools’ assumption of given preferences. The weaknesses of the latter false start are obvious. First, if all commitments are explicitly modelled and preferences are ‘given’ all things considered then the analysis of a game is basically complete after writing down the tree. Second, the whole notion of representative utility was from the beginning unsuitable for any kind of eductive analysis. Strategic thinking is much more naturally invoked if the questions are formulated in terms of objective or material payoffs that are desired by an actor and may show up among the reasons for action. Third, it is not clear at all whether we can separate the specification of the rules of the game, in particular the fixing of the preferences, from the strategic analysis of the game in terms of ‘given’ preferences. Obviously ‘all things considered’ must not include the strategic considerations that lead to preferring an action x over some action y at some node. But, then, how exactly can we draw the line between formulating and analysing the game tree? Fourth, starting from given preferences that are merely represented in a stenographic way by a utility function leaves in the dark practically everything that is of explanatory interest. Why care about such a rational fools’ world at all?

144


Table 6.1. Rational choice as maximization and non-maximization Rational choice as Maximization

Non-maximization

Local maximization ‘opportunism’

Global maximization

Dual maximization global and local

Rule-guided bounded behaviour

Advantages: 1. Human faculty to seize opportunities is prominent feature

Advantages: 1. No backward induction and chain store paradoxes

Advantage: Cognitive science psychology etc. can be used to explicate ‘rational choice’ as close to real choice

2. One-off PD solution in dominant strategies stands 3. Clear statement of all assumptions about commitments

2. Personal players represent persons

Advantages: 1. Fundamental intuitions that involve some trade-off between local and global maximization are respected 2. Opportunism and commitment in the same model

Disadvantages: 1. Backward induction and chain store paradoxes emerge

Disadvantages: 1. Elimination of sub-game perfectness problems

Disadvantage: There is no criterion as to when global should dominate local considerations and vice versa

Disadvantage: There is no way to find a standard of ideally rational behaviour that is not open to revision by the facts of choice

2. Dissolution of persons into agents eliminates eductive interpretation of game theory

2. One-off PD solution in dominant strategies does not stand 3. Unclear distinction between committed and uncommitted choice

the rationality of rational fools 145 Though we do think that RCM and conventional RCT were in fact a tremendous success as instruments of reflection on and understanding of interaction in strategic situations, the preceding account leaves us with the riddle why that was so and what exactly social theory in general and economics in particular learnt from going through that extended exercise. By starting from ‘given’ preferences all things considered it becomes practically impossible to account for human action in terms of real processes operating in the human psyche. Casting over human motivation the veil of the utility representation we abstract from too much that is relevant for motivating human behaviour. Introducing such constructs as the agent form, we can capture some of the inner constraints of choicemaking without giving up the (as if) maximization of utility. But in the end the only way out of living in a rational fools’ world is to be found in realistic (cognitive) psychology. Rational individuals who are not rational fools are boundedly rational rather than commanding some higher form of unbounded rationality. Notes ∗

Kliemt is most grateful to the Center for Study of Public Choice and his friends there for providing a pleasant, inspiring, and interesting environment when working on the paper. We have profited greatly from comments by Geoffrey Brennan, Erik Davis, Dan Hausman, Ron Heiner, David Levy, Alex Tabarrok, and Bruno Verbeek, and even more so from extended written comments by Susanne Hahn and Bernd Lahno, two anonymous referees, and last but not least Fabienne Peter and Bernhard Schmid. The conventional disclaimer applies. 1. From now on we follow the convention of using ‘he’ and ‘she’ as substitutes of each other. 2. Including for instance also joint randomization and so on. 3. Those who are in disagreement with non-cooperative game theoretic modelling as based on opportunistic rationality can legitimately suggest competing forms of modelling. But it is illegitimate not to acknowledge that they are doing something other than conventional non-cooperative game modelling. We classify non-opportunistic forms of rationality as ‘bounded’. But we do so without implying any judgement to the effect that ‘boundedly rational’ is inferior to ‘fully or opportunistically

146


rational’ choice-making. Bounded rationality is simply different from opportunistic rationality. 4. One should not forget here that they were in most serious trouble ever since 1961 when Herbert Hart revived classical insights of the British Moralists, according to which government is founded on ‘opinion’—including commitments to standards of legitimacy—and demonstrated that an adequate account of the workings of a legal order could not be given without taking recourse to commitments to an internal point of view to rules; see Hart (1961). 5. There may be, and presumably should be, a revival of partially cooperative game theoretic modelling that leaves certain rules implicit to choice-making itself or to the concept of a player, rather than making them explicit as part of the rules of the game (which, of course, include the preferences). But there is no justification for confusing non-cooperative with cooperative modelling and, for that matter, opportunism with non-opportunism. 6. Of course, we intend here to allude to Broome (1991). 7. One might want to be careful here to acknowledge that so-called time consistency is a substantive assumption of RCT going beyond the basic consistency assumption. 8. Basically the same tree. 9. There are other possibilities but we confine ourselves to these paradigm cases here since they suffice to make all crucial points about commitment in RCM. 10. In fact the same results of analysing the game would emerge if any two agents were to have identical preferences over the plays of the game still possible for both of them. 11. Note that we are not speaking of the ‘agent normal form’ but of the ‘agent form’ here. The former as opposed to the latter would indeed by definition impose substantive assumptions about ‘personal identity’. 12. More traditionally speaking, by the agent form Homo œconomicus is dissolved into (a team of) homunculi œconomici. 13. We could go even further here and modify information conditions so as to include agent forms without perfect recall as well as games with incomplete information (see Harsanyi 1967/68). In the latter case, as a matter of fact, only one of the types of an agent exists when the game is actually played. The multiplicity of types is a consequence of closing the

the rationality of rational fools 147 game informationally by a fictitious random move. As such the types are assumed to be on the players’ minds when analysing the game. 14. After all, the choices are modelled as ‘made’ separately from each other rather than ‘caused’ by each other. 15. Daniel Hausman is particularly willing to take the ‘all things considered’ concept as seriously as it deserves; see his chapter in this volume (pp. 49–69). 16. See also Verbeek’s chapter in this volume (pp. 150–74). 17. More rigorously, all strategies prescribing O dominate the strategy prescribing O∗ and l for the omnipotent player 1. 18. The case of the original Prisoner’s Dilemma with imperfect information would be as follows. If actors—before that game is played—can choose a behavioural disposition to cooperate then the Pareto dominated result can be avoided by rational choice makers if the commitment to cooperate can be chosen such that it will be ‘binding’ if and only if the other one is committed likewise. If players cannot make their commitments contingent on each other the PD problem will not go away but rather resurface on the commitment stage. 19. Obviously we could rely on additional tools like creating a type distribution by forming a plan. The plan would affect the future by creating a set of more or less committed or uncommitted agents with different utility functions, one of which in a fictitious random move would be chosen. A prior agent conceiving a plan could then causally affect the type composition and probability distribution over the set. It seems unnecessary to go into these complications since the basic approach seems obvious and an analysis of such complications, though showing how flexible RCM as a ‘language game’ is, would not add much of an insight. 20. As indicated, such players could be selected either by nature in a fictitious random move or strategically by a choice to modify payoffs. 21. It is basically said once the modern utility notion as representing all things that are relevant at a decision node is taken seriously. 22. Of course, not all social theorists were ‘normativists’. Hobbes (1968 [1651], see in particular chapters 10–17), and Spinoza (1951 [1670], see in particular chapter 16), are, of course, more hard-nosed adherents of Homo œconomicus than most economists. 23. Alex Tabarrock raised the issue whether the models formulating the internal commitment structure of an actor might lead to empirical

148 werner g uth ¨ and hartmut kliemt predictions that could be tested. They might, but then we would presumably have to put in so much empirical knowledge from cognitive psychology that we perhaps should just go for that kind of modelling entirely.

References Ainslee, George. 1992. Picoeconomics. Cambridge: Cambridge University Press. Alchian, Armen A. 1950. ‘Uncertainty, Evolution, and Economic Theory’. Journal of Political Economy 58: 211–21. Binmore, Ken. 1987/88. ‘Modeling Rational Players I & II’. Economics and Philosophy 3: 179–214 and 4: 9–55. Brennan, H. Geoffrey and Hartmut Kliemt. 1994. ‘Finite Lives and Social Institutions’. Kyklos 47/4: 551–71. Broome, John. 1991. Weighing Goods. Equality, Uncertainty and Time. Oxford: Blackwell. Elster, Jon (ed.). 1987. The Multiple Self. Cambridge: Cambridge University Press. Gode, Dhamanjay K. and Shyam Sunder. 1993. ‘Allocative Efficiency of Markets with Zero Intelligence Traders: Markets as a Partial Substitute for Individual Rationality’. Journal of Political Economy 101: 119–37. Güth, Werner. 1991. ‘Game Theory’s Basic Question: Who is a Player? Examples, Concepts, and their Behavioral Relevance’. Journal of Theoretical Politics 3/4: 403–35. Harsanyi, John C. 1967/68. ‘Games with Incomplete Information Played by Bayesian Players’. Management Science 14: 159–82, 320–34 and 486–502. Harsanyi, John C. and Reinhard Selten. 1988. A General Theory of Equilibrium Selection in Games. Cambridge, Mass.: MIT Press. Hart, Herbert. 1961. The Concept of Law. Oxford: Clarendon Press. Hausman, Daniel M. 2007. ‘Sympathy, Commitment, and Preference’. In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 49–69 (this volume). Hobbes, Thomas. 1968 [1651]. Leviathan. Harmondsworth: Penguin. Kohlberg, Elon and Jean-Francois Mertens. 1986. ‘On the Strategic Stability of Equilibria’. Econometrica 54/5: 1003–37. McClennen, Edward F. 1990. Rationality and Dynamic Choice—Foundational Explorations. New York / Port Chester / Melbourne / Sydney: Cambridge University Press. McClennen, Edward F. 1998. ‘Rationality and Rules’. In Peter A. Danielson (ed.), Modeling Rationality, Morality and Evolution. New York and Oxford: Oxford University Press, pp. 13–40.

the rationality of rational fools 149 Selten, Reinhard. 1975. ‘Reexamination of the Perfectness Concept for Equilibrium in Extensive Games’. International Journal of Game Theory 4: 25–55. Sen, Amartya K. 1982 [1973]. ‘Behaviour and the Concept of Preference’. In Sen, Choice, Welfare and Measurement. Oxford: Basil Blackwell, pp. 54–73. Sen, Armatya K. 1982 [1977]. ‘Rational Fools’. In Sen, Choice, Welfare and Measurement. Oxford: Blackwell, pp. 84–106. Smith, Vernon L. (ed.). 2000. Bargaining and Market Behavior. Cambridge: Cambridge University Press. Spinoza, Benedikt de. 1951 [1670]. A Theologico-Political Treatise. A Political Treatise. New York: Dover. Sugden, Robert. 1991. ‘Rational Choice: A Survey of Contributions from Economics and Philosophy’. The Economic Journal 101: 751–85. Verbeek, Bruno. 2007. ‘Rational Self-Commitment’. In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 150–74 (this volume). von Neumann, John and Oskar Morgenstern. 1944. Theory of Games and Economic Behavior. Princeton: Princeton University Press.

7 Rational Self-Commitment B RUNO V E RB E E K ∗

1. Ulysses and the Sirens After a long day at the office in Troy, Ulysses passes his favorite bar, the Sirens. From far away he can hear the laughing and singing of his friends. Back home, in the suburb of Ithaca, Penelope is waiting with dinner and he could really use a good night’s sleep. Right now, all things considered, he would prefer to go home and leave beer and Sirens for another night. However, if he comes closer to the bar, his friends will notice him and call him over to have a beer or two with them. He also knows that if he accepts their invitation, he will never be home in time. Dinner will be cold, Penny will be cross, and the good night’s sleep will be too short. So Ulysses decides to go home and decides that if his friends call him over he will wave at them and tell them that he will see them some other night. Ulysses’ decision seems sensible and nobody would be terribly surprised when Ulysses indeed passes the Sirens and declines the invitation of his friends. It is the rational thing to do given his decision. This last observation—that it is rational for Ulysses to go home given his decision—is the one I want to discuss in this paper. Exactly why is Ulysses rational in this case? To understand why, we need to know more about the circumstances and the nature of his deliberations. Consider the following three scenarios. First, it could be the case that Ulysses prefers to be home with Penelope, both now (t = 1) and in the (near) future (t = 2). So that when his inebriated pals at t = 2 call out to him, he still prefers to be home. In this case, Ulysses’ decision at t = 1 is rational, as is his going home at t = 2. However, in this scenario it is unclear why Ulysses needed to

rational self-commitment 151 make a decision before he receives the invitation. He could simply stroll along and without any trouble decline his friends’ invitation. Second, it could be the case that Ulysses makes a prediction about himself. He could realize that, without any measures on his part, he will succumb to temptation when passing the bar. Therefore, he decides to go home and does go home. However, if this is correct, the role of Ulysses’ prior decision again is unclear. Whereas in the former scenario the decision seems superfluous in accounting for the rationality of Ulysses’ actions, in this scenario it is unlikely that the decision alone can account for Ulysses’ actions, let alone their rationality. The standard way of understanding weakness of the will is as a lapse in rationality.1 If Ulysses is tempted, he is not fully rational. If Ulysses is weak-willed, he will accept his friends’ invitation against his better judgment. Therefore, at the time of action, he would not be rational. How could his decision at t = 1 to go home at t = 2, counteract the perverting influence of his friends’ invitation on his rational powers. It seems then that merely deciding to go home will not explain how a weak-willed Ulysses manages to go home. A lot more can be said about weakness of the will and how to go about when one anticipated one’s weakness. In particular, what sorts of external self-management one can use to deal with such lapses of rationality.2 However, I will leave the suggestion that Ulysses predicts that he will be tempted when passing the bar here and move to the third scenario, which is the one that really interests me in this essay. In this third scenario, Ulysses does prefer now (at t = 1) to go to Ithaca, but he anticipates that when his friends call out to him (at t = 2), he will change his mind about the attractions of Penelope, dinner, and the benefits of a good night’s sleep in comparison to a few hours in the company of his beer-drinking friends. In this scenario, Ulysses anticipates no (temporary) loss of his rational powers. Instead, he foresees that his preferences will change. On the orthodox theory of rational choice, giving in to the lure of the Sirens then would be rational: it would satisfy Ulysses’ strongest preference at the time of action. Acting otherwise would involve a form of counter-preferential choice. How can Ulysses’ decision and subsequent actions be rational in this scenario? Jon Elster has argued that, in such cases, a rational agent has to treat her predicament like a case of anticipated weakness of the will.3 In such situations, Elster argues, a rational agent should resort to commitment devices.

152

bruno verbeek

The fundamental reason for Elster’s recommendation is his skepticism about the feasibility of counter-preferential choice. According to Elster, an agent prefers x to y if and only if she chooses x when offered the choice between x and y.4 Given this conception of preference, genuine counter-preferential choice is not feasible. Therefore, Ulysses can only expect to go home if he uses a commitment device. According to Elster, an agent commits herself if she acts at t = 1 in such a way that the set of her available alternatives at t = 2 is a proper subset of her alternatives at t = 2 if she did not act at t = 1, where it is the case that the act of binding is easier to perform and less costly than the preferred action at t = 2 without commitment.5 Therefore, Ulysses should manipulate the situation in such a way that the undesired course of action is no longer available. For example, he could make it impossible to enter the bar by taking an alternative route home.6 According to Elster, Ulysses needs to do something in addition to the decision that he will not enter the bar. In other words, on Elster’s theory the only type of commitment that is feasible for Ulysses is a form of causal commitment. Ulysses needs to do something which subsequently exercises a causal—not a rational—influence on the available option. Rational agents are sometimes in a position that they have to protect themselves against their own future rationality through such methods. Causal commitment, then, is a potentially costly way of avoiding future rationality. Making alternatives unavailable is expensive in terms of the freedom you might otherwise enjoy. To some this is sufficient reason to reject it.7 However, the fundamental problem with Elster’s suggestion is that it assumes that Ulysses’ decision to continue on his way home does not carry any weight itself. His decision does not justify in any way his going home, while this seems to be exactly what is going on in this scenario. How is this possible? How can it be that simply by deciding to go home, Ulysses justifies his going home?

2. Decisions as causal commitments The difficulty with interpreting Ulysses’ actions as the result of a causal commitment is that the role of his earlier decision is negligible in the justification of his subsequent actions. However, it is certainly possible to

rational self-commitment 153 build in this feature in an account of causal commitment. We can include the decision at t = 1 in the justification of his actions at t = 2 if Ulysses’ earlier decision is itself a commitment device. Scott Shapiro has argued that if Ulysses genuinely decides to go home it must be the case that the earlier decision removes the option of entering the bar.8 The decision to go home makes entering the bar no longer feasible. The idea is that once an issue has been decided there is no opportunity for choice any more: the issue is settled. So if Ulysses decides to go home, he can do only one thing and that is to go home. Trivially, this makes Ulysses’ action rational. His changed preference for entering the Sirens is a mere wish, not a preference that can be satisfied. Shapiro’s theory has several advantages. First, it stays well within the traditional boundaries of rational choice theory. No appeal to counterpreferential choice is necessary. Second, it can make sense of the observation that Ulysses’ going home is rational given his earlier decision. Finally, it can explain why Ulysses is successful in going home. He has—after all—no alternative. In spite of these advantages, I am not convinced by Shapiro’s theory at all. The most fundamental problem with it is that its answer to the question as to why Ulysses is rational in going home responds to a different question. The issue is not why Ulysses cannot enter the bar but why he should not enter the bar. This becomes clearer when we look at what would happen on Shapiro’s theory if Ulysses were not rational when he passes the bar. If it is really true that because of his earlier (rational) decision entering the bar is not an available option any more, Ulysses would continue to go home even if the drunken singing of his friends in the bar is enough to drown out all sense in him. In other words, if Shapiro were right, we would have a wonderful panacea for weakness of the will. Simply decide, genuinely decide, to do the right thing and you will see that you always do the right thing. The mistake Shapiro makes is that he does not treat Ulysses’ case as a normative issue. Suppose that Ulysses, in spite of his earlier decision, spends the night boozing with his buddies at the Sirens. We would blame Ulysses in that case for his inconsistency. However, Shapiro’s theory makes it impossible to find fault with Ulysses’ actions in that case. The only way he could make room for the possibility of Ulysses entering the bar would be by arguing that Ulysses apparently did not really decide to go home after

154

bruno verbeek

all. However, in that case his entering the bar would be rational and his earlier decision would be irrational. In other words, on Shapiro’s theory it just can never be the case that a person makes a genuine and rational decision and subsequently does not execute that decision.9 This shows that Shapiro sees the relation between the decision at t = 1 and the actions at t = 2 as a causal relation and not at all as a normative one.10 From this brief discussion of Shapiro’s proposal we can infer the following desiderata for any successful analysis of the case of Ulysses and the Sirens. First, it must be such that Ulysses’ decision at t = 1 is rational. Second, it must be such that Ulysses’ action at t = 2 is rational. Third, that (part of) the reason his action is rational is his earlier decision to go home. That is, the decision to go home is normatively—and not (just) causally—relevant for his going home.

3. Plans We are rational agents—but only up to a degree. In our everyday lives we face constraints that limit the extent to which we can exercise rational choice. We have limited resources. Some limitations are internal in nature. We often do not have the acumen to oversee all the relevant information, nor can we determine which of the information is relevant. Furthermore, our reasoning capabilities are not perfect. We make mistakes and draw invalid conclusions. These limitations are further aggravated by external constraints. We are often, if not always, under time pressure to conclude our deliberations. As result of this pressure, we do not have the opportunity to gather, let alone inspect, all the available information. This pressure further compromises our ability to process the information correctly in our practical reasoning. Given these constraints, we need a way to deal with these constraints on our deliberation. The solution is that we are planning agents. Rather than make our decisions on the spot each and every time we face a choice, we tend to structure our deliberations through plans. This is the fundamental insight of Michael Bratman.11 In order to fulfill this role in our deliberations, plans have certain characteristics. First, plans are typically partial. When I plan to go to a conference, I make some general decisions about when I will go. However,

rational self-commitment 155 I do not yet decide which sessions I will attend, or what I will wear or have for breakfast. Those decisions I leave for later deliberation. In this way, plans relieve the burdens that time pressure puts upon us in an important way. By planning at t = 1 to do X at t = 2, I do not leave it up to the very last minute whether or not to do X. Instead, by settling on this plan now, when there is sufficient information and I am not under as much stress to come up with a decision, I can determine the more important things now, and leave trivial things (like what to wear or what to have for breakfast) for a later date. Secondly, because of this partiality, plans are hierarchical. Plans concerning ends embed plans concerning the means. Once I have determined to attend the conference, I can settle on a plan about the means of transportation. Note that the decision for means of transportation only makes sense within the framework of this higher-order plan. The features of partiality and hierarchy enable us, planning creatures, to deal with the limitations of our rationality. Through planning we can coordinate among the opportunities we have for deliberation. Big complex decision problems can be tackled by planning the necessary moves one at a time. Further, the limitations necessitate that prior deliberation can shape later conduct. We simply do not have the time to make reasonable (let alone rational) decisions at the time of acting. Thus when we act upon a plan, the action is non-deliberative in the sense that it is not the immediate result of some practical deliberation. The decision-making process is in the past. These two features of plans place demands on our plans. Plans need to be internally consistent. That is, it should be possible for my entire plan to be successfully executed given all my beliefs about my abilities and the world.12 In addition, plans need to satisfy means–end coherence. Plans that satisfy these demands structure practical reasoning in the following ways. First, they create the context and problem for further deliberation and establish standards of relevance for options considered in deliberation. Once you have the plan to attend the conference, you face the problem of how to get there. Second, plans typically constrain further deliberation by placing ‘‘filters of admissibility’’ on it.13 My decision to attend the conference rules out the option of attending a faculty meeting at the same time. Two final remarks about plans conclude our discussion. First, it is not conceptually impossible to deviate from a plan (unlike Shapiro’s theory). Rather, once I have settled on the plan to attend the conference, it is

156

bruno verbeek

no longer rational—but not impossible—for me to consider attending the faculty meeting at the same time. In other words, the ‘‘filters of admissibility’’ are normative not causal.14 Second, plans can only function in these ways if they are inert. It is not impossible to reconsider one’s plan, but in the absence of new information, plans resist reconsideration. They do so typically in a non-reflective way: once you have settled upon a plan, that is it; problem solved—no need to think about it any more.15 Only when you encounter new information that was not available to you before, there may be grounds for reconsideration (provided there is time to deliberate). Therefore, plans come with certain commitments. Some of these commitments concern the sort of options I can reflect upon; others concern the circumstances in which I can reconsider my plan.

4. Rational commitment Now that we have more or less fixed the notion of plans and the place plans have in our deliberative economy, we can explain the rationality both of Ulysses’ decision and his actual carrying through of his decision. At t = 1, Ulysses considers his options. He could go home or he could go to the bar. He determines the reasons for either option and weighs them. The company of Penelope, warm dinner, and a good night’s sleep weigh heavier than the joys of a beer or two with his pals at the Sirens. So in light of what he now knows and values about each option, going home is the better course of action. Therefore, Ulysses decides to go home and in doing so, settles on a plan, the execution of which involves declining the lure of the Sirens. At t = 2, his earlier plan to go home constrains his deliberation in that it makes entering the bar inadmissible. That is, his plan makes it irrational to enter the bar and rationalizes going home. Ulysses is committed, rationally committed, to go home. What about Elster’s worries about counter-preferential choice? These worries originate in a behaviorist notion of preferences that Amartya Sen has criticized, where preference and choice are synonymous.16 Sen is correct in rejecting this view. The planning theory of commitment does not proceed from this view. On this theory, preferences are complex states that relate judgments of value to motivation, but not in the way the orthodox view suggests. There is no one-on-one relation between preferring A to B and

rational self-commitment 157 choosing A when confronted with this pair of options. Space does not permit me to elaborate here. For purposes of the argument it suffices to think of preferences as comparative judgments of value.17 At this point, there are two objections. First, it is far from clear that Ulysses’ decision at t = 1 is in fact rational. Ulysses takes his preferences at t = 1 to determine the choice of plan. However, the plan is to be executed at t = 2, so why should Ulysses consider his preferences at t = 1 instead of those at t = 2 in settling on a plan? We can distinguish between two views on the rationality of plans in general.18 On the first view, the rationality of a plan is completely determined by the reasons for the intended action (apart from his decision). Thus, Ulysses’ plan to go home is rational if and only if at t = 2 it is indeed rational to go home independent of his plan. This is the primacy of action view. This view is implicit in the suggestion that Ulysses’ decision to go home is not rational. The alternative view—the view I endorse—is the primacy of planning view, which says that the rationality of a plan is not exclusively determined by the reasons for the intended action. On this view, the reasons for settling on a plan can include considerations other than those that obtain at the time of action. The truth of the latter view is obvious when we reflect on the pragmatic need for plans. Plans play such an important role in our deliberative psychology because we face time pressure and other constraints. In order to avoid last-minute mistakes; in order to cut down complex decision problems into manageable sub-problems; in order to coordinate intra- and interpersonally, we make plans. All of these reasons do not obtain at the time of action, but at the time of decision-making. In order to avoid the risk of last-minute mistakes at t = 2, we can settle on a plan at t = 1. Therefore, the rationale for a decision need not be confined to reasons for the intended action, but can also be found in other factors. Still we need to establish that there are additional reasons to form the plan to go home that do not obtain at the time when Ulysses’ drinking buddies request his company in the bar. One factor to consider is the regret Ulysses expects afterwards if he spends the evening with his friends instead of with Penelope.19 This, however, is a consideration that holds at t = 2 as well. It is not a reason for his decision at t = 1 that is not available at t = 2. My suggestion is different. At t = 1, Ulysses prefers going home over entering the bar. Also at t = 1, Ulysses believes that at t = 2 he will prefer the Sirens to the domestic bliss that awaits him in Ithaca. Ulysses has this

158

bruno verbeek

belief at t = 1 and yet prefers to go home. Therefore, Ulysses considers his future preferences already at t = 1 and gives them some weight in his decision. So the question really is ‘‘why don’t his future preferences have all the weight?’’ The reason is that even though Ulysses believes that he will value entering the bar over going home, he cannot be sure that his future preferences will reflect the value of both options correctly. The possibility of last-minute mistakes looms. Note that this is not the same as weakness of the will. Ulysses believes that without a decision now, he will act at t = 2 on his best judgment of the reasons that obtain. However, Ulysses is skeptical that his preferences at t = 2 reflect the reasons for either option as well as his preferences now. This is why he makes a decision now at t = 1 and why it is rational to do so. Furthermore, this explains why his decision to go home is rational. It is based, among other things, on his lack of trust in his future judgment. Note that Ulysses does not believe that he will necessarily be wrong at t = 2. It might be true that spending the evening at the Sirens is better for him than going home. As far as he can tell now at t = 1, this is not the case, but he might be mistaken. However, given the nature of the situation, the possibility of last-minute mistakes, etc., there are reasons for settling on a course of action now, rather than leaving it up to last-minute deliberations. Therefore, there are reasons for deciding to go home at t = 1 that cannot be reduced to reasons for the intended action at t = 2. This argument for the rationality of Ulysses, decision at t = 1 explains some other things as well. Consider the case of Mark, a typical 11-year-old boy. He firmly believes that girls are totally yucky, but at the same time he observes how boys not much older than him suddenly lose their cool and go completely gaga over girls. Being somewhat precocious, he expects that in due time he too will be drawn to these yucky creatures. So we have a similar shift in his preferences to that of Ulysses. At t = 1, Mark prefers to avoid his female classmates, whereas he will actively seek their company (at least some of them) at t = 2. Mark believes that this shift will not be the result of a collapse in his rational abilities (such as they are) but simply reflect a change in values. Should Mark decide to lead a chaste life and spurn any interaction with these alien creatures? The difference between Ulysses and Mark is that, unlike Ulysses, Mark has no reason to be skeptical about his future judgment in the same way as Ulysses is. Ulysses distrusts his future judgment on the merits of the Sirens versus Penelope. Mark

rational self-commitment 159 has no reason to distrust his judgment in that way. He can anticipate a genuine change in what is good for him (what is good for an 11-year-old boy is not always good for a 16-year-old boy). Therefore, Mark should not make any decision at t = 1 and simply ‘‘go with the flow.’’ That is, rational commitment of the type I have been discussing is a means to deal with the constraints agents face in their deliberations. If Mark is not expecting such constraints (at least not relevant ones), there is no need to commit. On the other hand, if he does expect that his judgment might be clouded, he should commit. And, finally, if he has reason to expect that he will not be able to act on his best reasons or his prior decision, he should take causal measures to deal with that situation. Thus, this rationale for settling on a plan allows for a nice threefold distinction about the rationality of such commitments. We can distinguish situations where there is a need for rational commitment, situations where there is a need for causal commitment, and situations where there is no need for commitment at all. This concludes the discussion of the first objection against the idea that Ulysses is rational when he decides to go home to Ithaca at t = 1. There is a second objection. Suppose that Ulysses is rational in deciding at t = 1 to go home. What prevents him from reconsidering at t = 2? After all, no plan is sacred. We can, and often do, change our plans. Why should Ulysses not change his plan? To answer this worry, we need to return to the pragmatic rationale for forming a plan in the first place. As I argued above, the reason we form plans and make future-oriented decisions are tied up with the limited nature of our rationality. We need plans to reduce the possibility of lastminute mistakes and to reduce complex decisions to several smaller, simpler decisions. Ulysses’ decision at t = 1 to go home and not enter the Sirens at t = 2 is such a plan. Plans can only have this function if they are inert. They should not come up for reconsideration arbitrarily. Plans have a default stability. Once you have decided, you have no reason to revise the plan, unless you discover a reason to revise it. That is to say, Ulysses is committed to his decision, unless there is new information between t = 1 and t = 2. However, everything that was relevant to his decision at t = 1 remains the same, the beliefs about his own abilities for rational decision-making, the beliefs about the nature of his options, the beliefs about his shifting preferences, and so on. In other words, there is no reason for Ulysses to

160

bruno verbeek

revise his plan. Since there is no reason for reconsideration, Ulysses is not free to reconsider. Furthermore, suppose that we change the story a bit. Suppose that his shift in preferences is not expected at t = 1 and that as he strolls along the bar, Ulysses discovers that his friends are there. It still would not follow that Ulysses necessarily has good reasons to start deliberating about his earlier decision and determine whether he should reconsider. Because at t = 2, the pressure is on: he needs to make a decision now and the chance that he would make a mistake is considerable. That fact in itself provides a reason to stick with the earlier decision, even though there is new information.20 Therefore, if we interpret Ulysses’ decision at t = 1 as a plan to go home and not enter the bar at t = 2, we have a genuine case of rational commitment. Ulysses’ decision at t = 1 is rational and so is his execution of his decision at t = 2. The reason why going home at t = 2, despite his shift in preferences, is rational is his earlier decision. Finally, the influence of his earlier decision on his actions at t = 2 is not a form of causal commitment. It is not the case that entering the bar at t = 2 is no longer an option as result of his decision at t = 1. Rather, entering the bar is no longer rational. It seems, therefore, that our account fulfills all desiderata we formulated at the end of section 2.

5. Rational commitment or bootstrapping? In sections 1 and 2, I argued that rational commitment is different from causal commitment. However, at this point it might be argued that this is not really the case. I have argued that by forming a plan at t = 1, Ulysses has a reason at t = 2 to go home. Is that not just saying that by forming a plan at t = 1, the option of going home has become relatively more attractive than staying at the Sirens? It seems that Ulysses, through his decision at t = 1, has started a causal process that resulted in making the option of going home more attractive. If that is the case, Ulysses’ plan at t = 1 looks like a form of causal commitment after all. Strictly speaking, however, this is not the case. By deciding to go home, Ulysses does not change anything about either home or the Sirens. He does not alter the relative desirability of these options. Ulysses, in forming his plan, puts himself in a state in which it is no longer rational to act on

rational self-commitment 161 his strongest preference at t = 2. This state, the state of having-decided-togo-home provides Ulysses with a reason to ignore his current preference (that is, his judgment at t = 2 about the relative values of home versus the Sirens), since he has reason to doubt that judgment.21 Obviously, this has a causal effect on his subsequent actions. However, this causal effect is not what commits Ulysses; his decision at t = 1, on the other hand, does. Be that as it may, the objection shows us an obvious further objection to the analysis of rational commitment. If the analysis were correct, agents can create their own reasons for action. Simply by deciding to go home at t = 2, Ulysses has created sufficient reasons to go home. That seems wrong. We either have reasons to do something or we don’t—we cannot create these reasons. Of course, we can and often do create reasons by changing things in the world. For example, if I decide to go to a conference and I buy a non-refundable ticket, then, arguably, I have changed the balance of reasons for going to the conference. However, that is not the case with Ulysses. If my suggestion is correct then right after deciding so at t = 1 Ulysses has a decisive reason to go home at t = 1 + ε before anything else has changed in the world. The only thing that is different is that Ulysses has decided. How could that difference create a genuine reason? Arguing that it does, constitutes, so the argument goes, an unacceptable form of bootstrapping.22 Bootstrapping seems to spell problems for the feasibility of rational commitment. If decisions do not create reasons for a course of action, decisions do not rationalize the future action. Therefore, if bootstrapping is impossible, rational commitment is not feasible and we have to revert to a version of the causal theory of commitment. So it seems we face something of a dilemma: either we insist that Ulysses’ decision justifies his future choice, in which case we have to accept that we can bootstrap our own reason into existence, or we avoid bootstrapping but then we have to give up on the feasibility of rational commitment altogether. Let me state at this point the conclusion that I will reach after the next two sections. I accept the dilemma. However, I do not believe that it is vicious. Intentions bootstrap reasons for action into existence that did not exist prior to the decision. In order to reach this conclusion, I will discuss two recent attempts to circumvent the dilemma. Both John Broome and Govert den Hartogh have argued, on different grounds, that the dilemma is false. Both authors try to account for the intuition that Ulysses’ decision justifies his

162

bruno verbeek

future conduct, without accepting bootstrapping. In the remainder of this paper I will discuss their attempts at avoiding the dilemma. I will argue that they are unsuccessful. The dilemma is unavoidable, but it is unclear if accepting bootstrapping is as objectionable as the critics make it seem.

6. Broome on reasons and requirements In several publications, John Broome has argued that the exclusive focus on reasons obscures an important distinction in the normative landscape.23 Two types of normative considerations, so he argues, govern rational deliberation. On the one hand, there are considerations that can best be labeled ‘‘pro tanto reasons,’’ or ‘‘reasons’’ for short.24 On the other hand there are ‘‘normative requirements,’’ or ‘‘requirements’’ for short.25 What distinguishes reasons from requirements is that the former have ‘‘weight’’ whereas the latter are ‘‘strict.’’ Let me explain. Reasons have weight. A reason to F is a consideration in favor of F-ing that could result in the conclusion to F, other things being equal. For example, the sunny weather is a reason to go to the beach. If there were no other reasons to the contrary (such as the deadline for this paper), I would go. In determining what to do, reasons enter deliberation like weights on a scale: they could tip the balance one way or the other. That is, reasons ‘‘add’’ to your decision. Reasons continue to play this role even when they are outweighed by alternative considerations. As Broome puts it, reasons are ‘‘slack’’ (they do not necessarily determine the outcome of rational deliberation) and ‘‘absolute’’ (they reside in fact and continue to exist even when outweighed).26 Requirements, on the other hand, have no weight. The requirement to F does not admit of degrees: either you are required to F or not. For example, you are required to choose the necessary means to your ends. This requirement cannot be outweighed by other considerations. If you have the end q and p is the necessary means to q, you ought to p. It does not mean that you have a reason to p. In fact, it could be the case that there are good reasons against p. Let q be the end of achieving world domination and let us suppose that the necessary means for that is killing 5,000 people; it does not follow that you have a reason to commit mass murder. However, that does not remove the requirement. In this case you ought to give up

rational self-commitment 163 the end to achieve world domination. In general there are two ways to deal with this particular requirement: choose the necessary means or give up the end. Requirements, therefore, are ‘‘strict’’ (they cannot be outweighed) and as a result not absolute since they do not remain ‘‘on the scene’’ even when defeated. This highlights a further distinction that Broome makes in this context, that of scope. Both reasons and requirements can vary in their scope. For example, there is a reason to go to sleep in time this evening. It may not be a particularly strong reason for some, but that is not relevant right now. What is relevant is that it has a very narrow scope: it concerns just one action. Contrast this with the requirement that you ought to p if your end is q and p is the necessary means to q. This requirement has a wider scope, for it holds between q and p. It is ‘‘relational.’’ With these distinctions in place, we can now discuss Broome’s solution to the problem of bootstrapping. According to Broome, decisions do not generate reasons. However, once Ulysses has decided to go home, he is required to see to it that he goes home. Thus, Ulysses makes a rational mistake if he were to enter the bar, even though the balance of reasons at t = 2 favors going to the Sirens. It is not the case that by deciding to go home he has given himself a reason to go home. Instead, having decided to go home, he is required to go home or to reconsider his original decision. Note that it is not the case that Ulysses has created this requirement. It was there all along. By deciding to go home at t = 2, it became applicable to Ulysses. According to Broome, Ulysses’ decision does not bootstrap a reason into existence that was not there before; however, Ulysses would be irrational if he did not go home (or repudiate his decision). Reasons and requirements play a different role in the deliberative economy of rationality. If we appreciate these differences, so Broome claims, we will see that the bootstrapping dilemma is false. This is an elegant and plausible proposal. However, I am not convinced. There are two problems with Broome’s solution. First, although we avoid bootstrapping, we return to one of the problems that I disposed of above. There is nothing in the concept of requirements that looks like the necessary inertia of plans. Perhaps we could solve this by arguing that decisions create requirements as well as some (weak) reasons for non-reconsideration. However, this would get us back to the bootstrapping that Broome wanted to avoid altogether.

164

bruno verbeek

There is a second, more serious problem with the proposal. Sometimes, decisions create strong reasons for a course of action. Consider the following example.27 Like most mothers, Mom loves her two children, Peter and Jane. She would like to give them both a treat. Unfortunately, she can give only one of them a treat. Since both children are equally deserving, needy, and desirous, she is indifferent which of the two should get it. Therefore, she is indifferent between the outcome in which Peter receives the treat (P), and the outcome in which Jane receives the treat (J). However, she prefers to flip a fair coin and let it decide who gets the treat. Note that this is fair: giving the treat to one of her children using this device is better (since fair) than giving it to one of them straightaway. Note further that Mom has two possible plans at her disposal that would do the trick. She could give Peter the treat if ‘‘heads’’ comes up and give Jane the treat when ‘‘tails’’ comes up. Alternatively, she could decide to give Jane the treat if ‘‘heads’’ comes up and Peter if ‘‘tails’’ comes up. We can put this scenario in a schematic representation (see figure 7.1). (A square node represents a point in the tree where a choice has to be made, whereas a circular node represents a coin flip.) J tails

n1

n2

P J

n3

P

heads

J P

Figure 7.1. Decision tree

Suppose that Mom decides for the first plan. She decides she will give the treat to Jane if ‘‘tails’’ and to Peter if ‘‘heads.’’ Suppose ‘‘tails’’ comes up. She is now in n3 . Again she faces the choice between giving the treat to Peter or Jane. They are equally deserving, needy, and keen on the treat, so it seems that the reasons in Broome’s sense are equally strong and that it is just Mom’s earlier decision that requires her to give the treat to Jane.

rational self-commitment 165 However, that is incorrect. Having made the decision to give the treat to Jane if ‘‘tails’’ makes it the case that it is simply fair to give the treat to Jane. Mom’s decision has thus created a reason of fairness to give the treat to Jane. In other words, in this situation Mom’s earlier decision has created a strong reason. She has bootstrapped this reason into existence simply in virtue of her decision. In conversation about this example, Broome argued that what makes it the case that it is only fair that Jane gets the treat is that Peter would have received it if ‘‘heads’’ had come up. In other words, it is not Mom’s decision that creates the reason, but the fact—a counterfactual fact so to speak—that Peter would have received it had the world been otherwise. So it is not the decision that creates the reason but the world (i.e., the world that would have been the case if ‘‘heads’’ had come up). At the time I did not know how to reply, but now I do. It is true that Jane’s receiving the treat if ‘‘tails’’ is fair because Peter would receive it had things turned out differently. However, we should ask ourselves why Peter would receive the treat if ‘‘heads’’ had come up. The answer is, obviously, Mom’s earlier decision, her plan, in n1 to give the treat to Peter if ‘‘heads’’ comes up. In other words, Mom’s decision is what made it the case that it is only fair that Jane gets the treat when ‘‘tails’’ comes up. So this really is a case where decisions bootstrap reasons into existence. Therefore, the distinction between requirements and reasons does not help us to avoid the complaint of bootstrapping in every case of rational commitment to a future course of action. This is not to say that the distinction is not a useful one. On the contrary; it elucidates some very important features of rational deliberation. However, I doubt that it avoids the dilemma that the bootstrapping objection seems to pose to the idea of rational commitment.

7. Den Hartogh on the authority of intentions A different way of avoiding the bootstrapping objection has been proposed by Govert den Hartogh.28 Den Hartogh makes an interesting analogy between rational self-commitment and the exercise of authority. If Ulysses’ earlier decision at t = 1 creates reasons to act in a certain way at t = 2, it is as if his earlier self executes authority, rational authority, over his later self. Just

166

bruno verbeek

as the decisions of an authority carry weight in the subsequent deliberations of those subjected to the authority, so it seems that the decision of the self at t = 1 carry weight in the subsequent deliberations of those subjected to the authority of the self at t = 1, and this can only be the later self of t = 2. Den Hartogh thinks that the similarities are so striking that he proposes to adopt the analysis of interpersonal authority for the analysis of intrapersonal authority. According to Den Hartogh, rational authority—as opposed to all other (that is, irrational) forms of authority—comes in two types only. First, there is coordinative authority. This is the sort of authority that helps to solve coordination problems. Imagine a busy road with cars coming from both ends. Suppose that there is no rule as to what side of the road cars should pass. Obviously, if others will keep to the right, you have a reason to keep to the right as well. However, the same holds for the left. What is more, nobody has any independent reasons to stay on one side of the road. So there is a coordination problem. Suppose that a person, say, a traffic warden, makes a clear signal to the effect that all cars should stay on the right-hand side of the road. Now all car drivers have a reason to stay on the right-hand side of the road. The warden’s signal functions as a focal point for the relevant expectations of all drivers. Prior to the signal, we had no special reason to expect that others would drive on any side of the road, but now we do. I believe that you have seen the signal and I believe that you believe that I believe that you have seen the signal. What is more, I believe that you believe that I have seen the signal, etc., etc. Our relevant beliefs refer to each other. Because we now have these nested interdependent mutual expectations, we have a reason to drive on the right-hand side of the road. In short, the signal of the traffic warden to stay on the right-hand side of the street has created a new reason to do so that did not exist before.29 He has coordinative authority. Second, there is epistemic authority.30 This is the sort of authority where the subject takes the command of the authority as a reason because the authority is better placed than she is to oversee and weigh the relevant reasons. One turns to the authority for his or her judgment because the judgment of the authority is more likely than one’s own to be correct. For example, a medical doctor has authority (and I defer to his authority) when he orders me to take antibiotics. I surrender my judgment in this case. Why would it be rational to do so? Obviously, because the doctor is better

rational self-commitment 167 placed than I am to oversee the merits of the case. He is in a better position than I am to see what is in my interest and what is not. And as long as the doctor’s orders are not obviously off (e.g., when he would prescribe huge quantities of cocaine for a common cold), it is perfectly rational to accept his decisions and act accordingly.31 Den Hartogh argues that future-directed decisions only have epistemic authority. Thus Ulysses’ decision to go home creates a reason for his actions at t = 2, but these reasons depend crucially on the question whether Ulysses at t = 1 is better situated to determine the merits of a visit to the Sirens compared to the journey home. This characterization of the sort of reasons that decisions give is very helpful. It explains, first, why prior decisions have inertia. Just like a rational agent does not continue to question the doctor’s orders, provided she trusts his judgment, a rational agent does not reconsider her earlier decision. Unless, of course, there is new information available that gives reason to doubt the earlier judgment. Furthermore, such reasons can be ignored if the agent has no grounds for assuming she is likely to make a better decision than the earlier self. So even if Ulysses’ preferences change as result of some new, unanticipated information, he can still be rationally committed to go home. Finally, the proposal makes clear when it is rational to reconsider: this is when the earlier decision loses its authority. This will be the case, first, when the reasons for referring to the authority are no longer valid. That will be the case when one is relatively sure that one will not commit any last-minute mistakes. Second, when one has reason to doubt the superior epistemic position of the authority (relative to one’s own), the prescriptions of the authority lose all their force. This could happen when one receives new information about the alternatives, which would have altered the initial decision. Den Hartogh’s position implicitly makes a distinction between vicious and benign bootstrapping. It avoids the sort of unwarranted bootstrapping that would occur if the self at t = 1 were to have coordinative authority as well as epistemic authority. So it is not the case that Ulysses creates entirely new reasons; however, by deciding to go home, he specifies a reason that is there, which is his reason to avoid last-minute mistakes. In other words, rather than accepting bootstrapping entirely, or trying to avoid it completely, Den Hartogh accepts a very limited sort of bootstrapping. The sort of reason that Ulysses’ earlier decision creates is a function of his finite, constrained rational abilities. If the danger of last-minute mistakes

168

bruno verbeek

is minimal, if there is more than sufficient time to oversee the merits of a visit to the Sirens relative to the comforts of home, Ulysses need not, perhaps even should not, make a decision about what to do. I agree with Den Hartogh in this case. This is exactly my defense for the rationality of Ulysses’ decision at t = 1.32 The reasons to decide at t = 1, rather than postponing his decision until t = 2, are determined by the time pressure and the danger of last-minute mistakes. However, I am not convinced by Den Hartogh’s claim that the reasons that decisions create are only of this epistemic type. Two obvious problems for this view are, first, Buridan’s Ass cases (where the agent is absolutely indifferent between the two alternatives), as well as cases of incomparability. In both cases, the reasons for either alternative do not settle the issue. In such cases, Den Hartogh argues, we do not bootstrap any reasons into existence by deciding for one of the available alternatives. That would be a case where decisions have more than epistemic authority. Rather, in such cases the prior decision simply creates the causal ‘‘umpf’’ necessary to prevent indecision. In those cases, decisions just have causal impact on the agent, but no rational weight. The argument, in a nutshell, is the following. Suppose you face two indifferent options A and B. Suppose at t = 1 you decide for A, but at t = 2 you take B instead. Den Hartogh denies that in such cases you have made a rational mistake.33 I have doubts about this particular argument similar to my reservations about Shapiro’s theory. However, there is a more fundamental objection to the entire theory. There is a counter-example to Den Hartogh’s claim that future-oriented decisions only have epistemic authority. The example of Mom and her two children from the previous section is a case where decisions do more than relieve the burdens of decision-making at the last minute. On Den Hartogh’s view, Mom would not have created any reason to give the treat to Jane. Mom’s decision to flip a coin is merely a device to create the push that the whole machinery needed. However, this seems to miss something. Mom’s decision to flip the coin is not just a way to break the deadlock between the reasons for Jane and Peter. Rather, flipping the coin and letting its outcome determine who gets the treat is fair—and that is a decisive reason in favor of coin flipping. Thus having flipped the coin, there really is a reason and not just a causal push to give the treat to Jane. So this seems to be a case where a decision has more than epistemic authority.

rational self-commitment 169

8. Conclusion I conclude that neither Broome’s nor Den Hartogh’s attempt at avoiding bootstrapping works in all cases. Sometimes we can and do create our own reasons, just like the rational commitment model I suggested predicts. Therefore, I seriously doubt whether bootstrapping is a problem for the analysis of rational commitment. The unavoidability of bootstrapping at least suggests that it is not a problem in the first place. As planning agents we do it all of the time. The dilemma I introduced above is not vicious. We can safely take the first horn and accept that rational agents sometimes create reasons for their actions simply in virtue of their earlier decisions. However, this does not settle the matter in the specific case of Ulysses. First, both Broome and Den Hartogh could try to argue that in the central case of Ulysses and the Sirens, there is no objectionable bootstrapping. Broome then still has to account for the inertia of intentions. Den Hartogh’s theory of the epistemic authority of intentions provides such an account in terms of the circumstances of choice at t = 2 (time pressure and last-minute mistakes). So maybe all of Ulysses’ reasons to ignore his preference for the Sirens and continue his course home are reducible to his inferior epistemic position at t = 2. Though I tend to agree with this analysis of Ulysses, I doubt that it holds for all cases of rational self-commitment. Regardless of this, though, I conclude that rational self-commitment is not only feasible, it is also advisable for limited rational agents like us. Without it, the Penelopes of this world would be very lonely indeed. Notes ∗

This paper is a revised version of ‘‘The feasibility of rational selfcommitment’’ which was presented at the Workshop on Rationality and Commitment at the University of St Gallen, 13–15 May 2004. I want to thank the participants of the workshop, John Broome, Govert den Hartogh, and Luc Bovens, for their helpful comments. 1. For example, Davidson (1970). 2. Elster (1979) is still the locus classicus. 3. Elster (1979). 4. Elster (1979: 6 ff.).

170

bruno verbeek

5. Elster (1979). This is not a good definition of commitment, even in Elster’s sense. The central feature of a commitment according to Elster is that certain options at t = 2 are made impossible. It should be irrelevant whether the ‘‘act of binding’’ is easier or less costly than the most preferred action at t = 2, for the act to be efficacious in restricting the options at t = 2 to a proper subset of those at t = 1. Secondly, Elster’s statement seems to rule out that actions which affect the options at t = 2 in such a way that these partly overlap with those at t = 1 as a form of commitment. For example, if Ulysses attends a lecture in the evening, rather than walking home, he has excluded the option of the Sirens from his options at t = 2 but he also has new, other options that he would not have without this action. Now imagine that Ulysses attends the lecture in order to avoid his friends hailing him from the Sirens. Should this not count as a form of commitment? 6. Elster discusses two additional techniques of ‘‘self-management’’ that could be relevant for Ulysses-like cases (Elster 1979: ch.2). First, Ulysses could manipulate the situation in such a way that his future preferences will continue to favor going home to Penelope. He could, for example, deposit a large sum of money with a friend and tell this friend that he is free to keep it, should he enter the Sirens. Second, Ulysses could attempt to tinker with his rational decision-making powers at the time of action, in such a way that he will not be completely rational and, as a result, go home. For example, Ulysses could undergo hypnotherapy or take special drugs, which make him unreceptive to the lure of the Sirens. (Note that this is the mirror image of the scenario of weakness of the will discussed above.) 7. For example, McClennen gives such pragmatic arguments to abandon ‘‘sophisticated choice’’ and argue for the rationality of ‘‘resolute choice’’ (McClennen 1990). 8. Shapiro (forthcoming). Similar suggestions have been made by Isaac Levi (1994). 9. This assumes, of course, that there is no external interference. 10. There are other problems with this approach as well. In Verbeek (2002a) I argue that this position assumes an incoherent notion of feasibility which implies (if correct) that at the time of choice one has only one option. That is, there is no choice in the first place, which makes the requirements of rational choice empty.

rational self-commitment 171 11. Bratman (1987). 12. There are some complications here concerning self-prediction. If I can predict that I would not do X at t = 2, would a plan to do X be feasible? Could it be rational to plan to X under such circumstances? In Verbeek (2002a) I argue that there are good reasons to resist certain types of self-prediction in the determination of what plans are feasible. Furthermore, there can be situations where (part of) my plan cannot be successfully executed given my beliefs about my abilities and the world yet it is rational to adopt such a plan. The standard example is that of a conditional plan (I will do X if C obtains), where one is certain that the condition C will not obtain, for example, the plan to leave one’s spouse if he or she is unfaithful. 13. To a large extent, the debate about rational self-commitment is a debate about the question of how these ‘‘filters’’ are to be characterized. 14. Obviously plans also have a causal impact on our deliberation. Such impact does not rationalize the future choice. 15. This is not the only type of non-reconsideration: it is possible to resist reconsidering plans in a reflective way. However, the typical way plans function is such that you only start worrying about your plan if you have new reasons to doubt its rationality. See also Bratman (1987: esp. 64–72). 16. Sen (1977). 17. I am fully aware of the many questions this view might raise. For example, what to say to the type of objection that David Lewis (1988, 1996) has raised? In addition, there is the following puzzle. If preferences are belief-like states about the comparative value of options, then it looks like Ulysses is in the following predicament. He now believes (p) going home to be better than entering the Sirens, but he also believes that he will come to believe that ∼ p, and this seems paradoxical. See also Van Fraassen (1984). I have to leave all these problems for a future occasion. 18. For a similar distinction, see Robins (1995). 19. Bratman (1998) takes up this suggestion in the context of the so-called ‘‘toxin puzzle.’’ 20. Raz (1978) makes a similar point in his discussion of legal authority.

172 bruno verbeek 21. Remember that I suggested that the best way to think of preferences is as comparative judgment with motivational effect. 22. See Bratman (1987: 24–7, 86–7). 23. Broome (1999, 2001a, 2001b, 2002). 24. I follow the name that Broome gives to this type of considerations in Broome (2004). 25. This term comes from Broome (2001a). 26. Broome (2001a). 27. The example is due to Diamond (1967). I discuss this version of it at length in Verbeek (2001). 28. Den Hartogh (2004). 29. This is a widely shared analysis that can be traced to Lewis (1969), Schelling (1960), Sugden (1986). I discuss their ideas in Verbeek (2002b). Den Hartogh (2002) gives a detailed discussion of this and other types of authority. 30. Raz (1985). 31. According to Den Hartogh (2002), what makes these two phenomena forms of authority is that both issue so-called content-independent reasons. That is, it does not matter what the authority commands, only that he or she commands it. However, both coordinative and epistemic authority provides content-independent reasons only within a certain range. It is rational to accept an authority (i.e., the other person has rational authority over the subject) when his decisions are sensitive to the reasons at hand, including the reason why you refer to him. For example, a coordinative authority who commands the car drivers to pick a side of the road at random has no authority, since his command obviously does not do anything to improve the coordination problem. However, whether he commands us all to drive on the right side or the left side of the road makes no difference to the reasons he issues for driving on the right side or the left side respectively. Both are equally authoritative commands. 32. See section 4. 33. Note that Broome’s proposal that decisions fall under a requirement to execute them or reconsider them can explain why such an agent would be inconsistent and, therefore, make a mistake.

rational self-commitment 173 References Bratman, Michael E. 1987. Intention, Plans, and Practical Reason. Cambridge, MA: Harvard University Press. Bratman, Micheal E. 1998. ‘‘Toxin, Temptation, and the Stability of Intention.’’ In Christopher W. Morris and Jules Coleman (eds.), Rational Commitment and Social Justice. Cambridge: Cambridge University Press, pp. 59–83. Broome, John. 1999. ‘‘Normative Requirements.’’ Ratio 12/4: 398–419. Broome, John. 2001a. ‘‘Are Intentions Reasons?’’ In Arthur Ripstein and Christopher Morris (eds.), Practical Rationality and Preference: Essays for David Gauthier. Cambridge: Cambridge University Press, pp. 98–120. Broome, John. 2001b. ‘‘Normative Practical Reasoning I.’’ Aristotelian Society 75: 175–93. Broome, John. 2002. ‘‘Practical Reasoning.’’ In Jose Luis Bermudez (ed.), Reason and Nature: Essays in the Theory of Rationality. Oxford: Oxford University Press, pp. 85–111. Broome, John. 2004. ‘‘Reasons.’’ In R. Jay Wallace, Philip Pettit, Samuel Scheffler and Michael Smith (eds.), Reason and Value: Themes from the Moral Philosophy of Joseph Raz. Oxford: Oxford University Press, pp. 28–55. Davidson, Donald. 1970. ‘‘How is Weakness of the Will Possible?’’ In Joel Feinberg (ed.), Moral Concepts. Oxford: Oxford University Press, pp. 93–113. Den Hartogh, Govert Arie. 2002. Mutual Expectations: A Conventionalist Theory of Law. Law and Philosophy Library. Dordrecht: Kluwer Academic Publishers. Den Hartogh, Govert Arie. 2004. ‘‘The Authority of Intentions.’’ Ethics 115/1: 6–34. Den Hartogh, Govert Arie. Forthcoming. ‘‘Intending for Autonomous Reasons.’’ In Bruno Verbeek (ed.), Reasons and Intentions. Diamond, Peter. 1967. ‘‘Cardinal Welfare, Individualistic Ethics, and Interpersonal Comparisons: Comment.’’ Journal of Political Economy 75: 765–6. Elster, Jon. 1979. Ulysses and the Sirens. Cambridge: Cambridge University Press. Elster, Jon. 1982. ‘‘Sour Grapes—Utilitarianism and the Genesis of Wants.’’ In Bernard Williams and Amartya Sen (eds.), Utilitarianism and Beyond. Cambridge: Cambridge University Press, pp. 219–38. Levi, Isaac. 1994. ‘‘Rationality and Commitment.’’ In Carol C. Gould (ed.), Artifacts, Representations and Social Practice. Dordrecht: Kluwer, pp. 257–75. Lewis, David. 1969. Convention: A Philosophical Study. Cambridge, MA: Harvard University Press. Lewis, David. 1988. ‘‘Desire as Belief.’’ Mind 97 (387): 323–32.

174

bruno verbeek

Lewis, David. 1996. ‘‘Desire as Belief II.’’ Mind 105 (418): 303–13. McClennen, Edward F. 1990. Rationality and Dynamic Choice: Foundational Explorations. Cambridge: Cambridge University Press. Raz, Joseph (ed.) 1978. Practical Reasoning. Oxford, New York: Oxford University Press. Raz, Joseph. 1985. ‘‘Authority and Justification.’’ Philosophy and Public Affairs 14: 3–29. Robins, Michael H. 1995. ‘‘Is it Rational to Carry Out Strategic Intentions?’’ Philosophia (Israel) 25/1–4: 191–221. Schelling, Thomas. 1960. The Strategy of Conflict. Cambridge, MA: Harvard University Press. Sen, Amartya K. 1977. ‘‘Rational Fools: A Critique of the Behavioral Foundations of Economic Theory.’’ Philosophy and Public Affairs 6: 317–44. Shapiro, Scott. Forthcoming. ‘‘The Difference that Rules Make.’’ In Bruno Verbeek (ed.), Reasons and Intentions. Sugden, Robert. 1986. The Economics of Rights, Co-operation and Welfare. Oxford: Blackwell. Van Fraassen, Bas. 1984. ‘‘Belief and the Will.’’ Journal of Philosophy 81/5: 235–86. Verbeek, Bruno. 2001. ‘‘Consequentialism, Rationality, and the Relevant Description of Outcomes.’’ Economics and Philosophy 17/2: 181–205. Verbeek, Bruno. 2002a. ‘‘Feasible Intentions.’’ Unpublished manuscript. Verbeek, Bruno. 2002b. Instrumental Rationality and Moral Philosophy: An Essay on the Virtues of Cooperation. Dordrecht: Kluwer Academic Publishers.

8 Rationality and Commitment in Voluntary Cooperation: Insights from Experimental Economics ¨ NI S I M O N G A¨ C H T E R A ND C H RI S T I A N T H O

1. The rationality of voluntary cooperation Cooperation problems arise when individual incentives and social optimality diverge. This tension has intrigued social scientists and philosophers for decades.1 In this chapter we look at the cooperation problem from the viewpoint of experimental economics, a subfield of economics which studies decision-making under controlled laboratory conditions and under real monetary incentives.2 Years of careful experimentation have led to a body of results, which may shed new light on old philosophical questions, and on the foundations of the behavioural sciences. In particular our results will shed light on selfishness as one important foundational assumption of the behavioural sciences. The selfishness assumption has long been criticized (by Sen 1977, for instance, in a highly influential article). Yet only recently experimentalists have started to systematically scrutinize the selfishness assumption. We will discuss some selected evidence in this chapter on how people solve the cooperation problem and to what extent people’s cooperation behaviour can be explained by their (non-)selfish preferences. We refer the reader to Fehr, Fischbacher, and Gächter (2002), Camerer (2003), Fehr and Fischbacher (2003), and Hammerstein (2003) for broader discussions and surveys.

176

¨ simon g a¨ chter and christian th oni

Our discussion will focus on two games, the Prisoner’s Dilemma game and the public goods game. The widely known Prisoner’s Dilemma game (PD from now on) is the prototype game in which this tension between individual and collective rationality arises. Figure 8.1 depicts a game between two players, which illustrates the issue. Column Player cooperate

defect

cooperate

R;R

S;T

defect

T;S

P;P

Row Player

Figure 8.1. The Prisoner’s Dilemma game (if T > R > P > S)

When both players cooperate, both receive a payoff of R (for ‘reward’’). If the column player cooperates and the row player defects, then the row player receives the ‘temptation payoff’ T, and the cooperating column player the ‘sucker’s payoff’ S. Payoffs are reversed if the row player cooperates and the column player defects. If both defect both receive the ‘punishment payoff’ P. This game is a PD, if T > R > P > S.3 It is now easy to see why the PD depicts a prototypical cooperation problem: if both cooperate both would be better off than when they both defect.4 Yet, irrespective of the choice of the opponent each player always has a higher payoff (of either T or P) if he or she defects than if he or she cooperates. Thus, if the payoffs in the game of figure 8.1 obey T > R > P > S, then it is a PD. Rational players will therefore defect, since defection is a dominant strategy for both of them. They will defect even if they fully understand that mutual cooperation would collectively yield them a higher payoff than mutual defection, which is the only Nash equilibrium in this game. The PD has intrigued researchers for decades (see Poundstone 1992 for an interesting discussion). It is probably one of the most extensively investigated games, both theoretically and experimentally. The empirical results from many experiments are equally as stark as the theoretical prediction of mutual defection. In a series of early experiments on the PD, Rapoport and Chammah (1965) found mutual cooperation in 30 to 50 per cent of all cases. This result has been replicated many times by now (see, e.g., Dawes 1980; Andreoni and Miller 1993; Ledyard 1995; Cooper

rationality and commitment in voluntary cooperation 177 et al. 1996). Oberholzer, Waldfogel, and White (2003) report a particularly striking result. They analyse behaviour in a television show called Friend or Foe. In this show two subjects play a PD-like game for high stakes (between $200 and $16,400). If both players play ‘friend’ then they share the stake at hand equally. If one player plays ‘foe’ and the other plays ‘friend’ then the former receives the whole pie while the latter receives nothing. In case both players play ‘foe’ they both earn nothing. The participants of this television show choose ‘friend’ in slightly more than half of the cases. Though very insightful, the PD models a two-person cooperation problem. Yet, in reality, many interesting cooperation problems involve many people. The public goods game is a suitable tool for studying cooperation in groups of more than two players. It can be seen as an n-person version of a PD. In the public goods game (PG from now on) a number of players form a group and each player is endowed with e tokens. The players decide simultaneously how many of their tokens (gi ) they want to contribute to the public good. The tokens not contributed count automatically as private income. All individual contributions in the group are summed up to G = gi . A player’s payoff results as πi = e − gi + aG

(1)

where the parameter a is the marginal per capita return (MPCR). This parameter measures the private return from contributing to the common good. For the game to be a PG game this parameter has to be within 1 > a > 1/n. The first part of this condition ensures that the players have a dominant strategy to contribute nothing to the public good. The second part of the condition ensures that it is socially beneficial to contribute. The solution of the PG is straightforward under the assumption that (1) represents the players’ preferences. Contributing to the public good yields a return of a. This is less than what could be earned when keeping the tokens for oneself. Therefore, independent of the others’ actions, each player has an incentive to choose the lowest possible contribution. However, since every member of the group profits from a player’s contribution the social return is na, which is larger than unity. Therefore, joint payoff is maximized when all players contribute their full endowment. Like the PD this game belongs to the most extensively studied games in experimental economics. Ledyard (1995) reviews the literature and reports that in a typical public goods game subjects contribute on average between

178


40 and 60 per cent of their endowment. When the experiment is repeated, contributions decrease over time to very low levels. In summary, in many cases people manage to achieve mutual cooperation despite the fact that defection would have led to higher earnings for them individually. Thus, we have an empirical puzzle: there is much more cooperation than is compatible with the stark theoretical predictions of defection and freeriding. As a consequence, people overall are much better off than with the ‘rational choice’ of defection and freeriding. In the everyday sense, for many people voluntary cooperation rather than freeriding seems the ‘rational’ thing to do. How can we explain this? We look at three different explanations that have been advanced: (i) cooperation in one-shot games is erroneous and maladaptive; (ii) people’s reasoning may differ from the individualistic approach applied above; and (iii) the PD or public goods game does not adequately reflect people’s true preferences. These possibilities have important conceptual consequences to which we will return in the final section. Our focus in the remainder of this chapter will be mainly on the last hypothesis. We will therefore only briefly sketch the first two explanations. (i) According to the maladaptation hypothesis, one may argue that most games in real life are in fact repeated games. From the theory of repeated games it is well known that if ‘the shadow of the future’ is important, i.e., if players interact for an unknown length of time, and if people are not too impatient and therefore care for the future, then strategic cooperation becomes possible, because defection can be punished by withholding future cooperation and even more complicated punishment strategies (e.g., Fudenberg and Maskin 1986). The most famous idea is probably reciprocal altruism (Trivers 1971) and the related strategy of ‘tit-for-tat’, which turned out to be a very successful strategy in an ‘evolutionary contest’ where strategies played against each other in a computer simulation (Axelrod and Hamilton 1981). Its essence is the idea that favours are reciprocated (‘I’ll scratch your back if you’ll scratch mine’) and that unhelpful behaviour is reciprocated by withholding future help. Thus, in indefinitely repeated games even selfish individuals have an incentive to cooperate. Why, then, do people cooperate in one-shot games, where there is no future interaction? One explanation is just errors and confusion. A more

rationality and commitment in voluntary cooperation 179 refined explanation in terms of errors is that people adopt behavioural rules that are beneficial in repeated cooperation games to the artificial oneshot game they are in (see, for instance, Binmore 1994, 1998). A related argument, advanced by some evolutionary theorists, is the ‘maladaptation hypothesis’ (e.g., Johnson, Stopka and Knights 2003), according to which ‘human brains apply ancient tendencies to cooperate that persist in newer environments, even if they are maladaptive (heuristic rules that violate expected utility often make sense for common tasks in our evolutionary history).’ One problem with this argument is that in experiments people immediately change their behaviour in repeated games with the same opponent. For instance, in the ten-period repeated PDs of Andreoni and Miller (1993) and Cooper et al. (1996) cooperation rates were two times higher than in the ten one-shot games against different opponents. Keser and van Winden (2000) and Fehr and Gächter (2000) got similar results in the repeated vs one-shot PG (see figure 8.5 on page 196 below). We will come back to the maladaptation explanation in section 5. (ii) The reasoning that has led to the theoretical prediction of mutual defection in the PD and full freeriding in the PG is based on the standard approach of rational choice analysis: given a decision-maker’s preference over certain outcomes, each decision-maker chooses the outcome that maximizes his or her preference. Thus, the decision-maker looks at the problem from his or her own individual perspective. Yet people may reason differently in that they see themselves as being team members and therefore ask ‘what should we as a team do?’ People who apply the team perspective think about the actions they choose from the team perspective. People who apply ‘team-directed reasoning’ (Sugden 2000) will cooperate in the PD and the PG. Sugden (1993, 2000), and Bacharach (1999) have formalized this psychologically intuitive idea.5 To our knowledge, there is no systematic evidence on team reasoning. We will sketch below some arguments used by experimental subjects that are consistent with team reasoning. (iii) A third explanation for observed cooperation is that many people’s preferences in the PD or PG are not adequately described by the payoffs depicted in figure 8.1 or equation (1). To appreciate this argument, one has to notice that game theory assumes that the payoff numbers in figure 8.1 reflect people’s preference ordering of all possible strategy combinations. Recall that the game of figure 8.1 is only a PD if preferences obey

180


T > R > P > S. If this is the case, and if people apply individual instead of team-directed reasoning, then there is no way around the conclusion that rational people will defect in the PD and freeride in the PG (see Binmore 1994 for an extensive discussion of this issue). If one neglects explanations (i) and (ii) for a moment, the stylized fact that many people cooperate in such simple games like the PD and the PG suggests that the actual utilities that people derive do not correspond to the specified payoffs. To see this, notice that all experiments require the specification of payoffs for the subjects. In virtually all experiments people receive monetary payments that induce the incentive structure that gives rise to a PD or PG, i.e., monetary payoffs obey T > R > P > S, or payoff function (1) in the PG. From a revealed preference approach, the observation of a cooperative choice may reflect that people actually prefer cooperation over defection. Yet, this implies that the utility of cooperation exceeds the utility of defection. Assume this is true for both players. Then the game of figure 8.1 actually is an ‘assurance game’’, where mutual cooperation is an equilibrium. In other words, the material incentives may not fully reflect people’s preferences. Elements other than people’s own material well-being might be relevant as well. For instance, people might have other-regarding preferences and simply care for the well-being of others. They might feel guilty if they do not cooperate or they may feel committed to reciprocate if they believe that others cooperate. This line of reasoning is not without difficulty from a methodological point of view. Without further discipline, one can ‘rationalize’ any outcome by specifying the appropriate preferences. For this reason, theorists have resisted opening ‘Pandora’s box’ by specifying preferences that rationalize outcomes. This argument is correct in our view in the absence of empirical tools to measure (or infer) people’s preferences. We believe (and hope to demonstrate in this chapter) that the tools of experimental economics (and some further instruments like neuroscientific methods) may allow us to learn about the structure of people’s motivations. This information may then guide theory-building by putting empirically disciplined structure on preference assumptions. We will return to this issue in section 6. In the remainder of this chapter we will first focus on measuring motivations that might explain why people cooperate. Our purpose is twofold. We

rationality and commitment in voluntary cooperation 181 demonstrate some methods how one can learn about people’s motivations beyond the pecuniary payoffs they receive as a function of their choices. We will also show that all methods produce substantial evidence against the selfishness assumption frequently invoked by behavioural scientists, most notably by economics. We show that rather than being selfish many people are ‘conditional cooperators’ who are committed to cooperation if others cooperate as well. However, freeriders, who never contribute to the public good, exist as well. In other words, there is substantial heterogeneity in people’s cooperative attitudes. We will present evidence on the consequences of such heterogeneous motivations in section 3. We will show that observed patterns of cooperation can be explained by preference heterogeneity but not easily by errors. We will then look at the role of emotions in cooperation in section 4. Emotions are interesting because it has been argued that they serve as a ‘commitment device’ (Hirshleifer 1987; Frank 1988). Specifically, freeriding may trigger feelings of anger in the cheated person, who may then be disposed to punish the freerider. If sufficiently many people are prepared to punish the freeriders, then freeriding may not pay off any more and may induce even selfish people to cooperate. The evidence presented in sections 2 to 4 shows that many people apparently have unselfish preferences. This begs an explanation. In section 5 we will therefore sketch some recently advanced arguments by evolutionary theorists (Boyd et al. 2003) that provide some ultimate account for observed preferences. Section 6 provides some concluding remarks on possible methodological consequences of the findings presented in this chapter.

2. Measuring motivations We will discuss some methods in this section on how to learn about people’s motivation to (not) contribute to public goods. We will start by presenting qualitative evidence from verbal protocols and will then discuss tighter methods to infer motivations. We will demonstrate that all methods yield the same qualitative conclusions: a majority of people are non-selfishly motivated. In particular, they are prepared to cooperate if others cooperate. An important minority is best described as being selfish.

182


2.1. Reasoning A natural first way to explore people’s motivations is to ask them about their motives. Gächter and Fehr (1999) did PG experiments where the subjects had to explain their contribution decision.6 In the following we discuss some of the answers the subjects gave when choosing their contribution in the first period of a repeated PG game. The subjects who chose to contribute their full endowment (of 20 tokens) provided the following statements: A. ‘By my decision I expect to motivate my team mates to high contributions.’ B. ‘Trial to achieve ‘‘safely’’ the maximum. I try to convince the others.’ C. ‘For maximal payoff, 20 to the group account.’ D. ‘This way we earn the most as a group.’ Statements A and B speak for the notion of rational cooperation, i.e., the subjects try to encourage other subjects to contribute by providing a good example. On the other hand, statements C and D rather point in the direction of team-directed reasoning. Casual inspection of all answers favours the notion of rational cooperation since approximately two-thirds of the cooperative subjects provide some sort of ‘motivating others’ argument. Subjects with intermediate contributions often say that they face a tradeoff between securing their own income and motivating other subjects to contribute, for example: E. ‘No full risk. Signal disposition to contribute to the group.’ F. ‘Do not want to put in everything before I know how the rest of the group will act.’ Subjects with low contributions are either on the very cautious side or plain freeriders. G. ‘I don’t invest that much because I don’t know yet whether the others are pro-social or egoistic. If the others are egoistic, I have a loss.’ H. ‘Most will contribute to the project. Maximal earnings for me.’ Statements in later periods naturally depend on the course of the game. We have seen in the previous section that contributions typically erode throughout the experiment. Three exemplary statements from later periods are the following:

rationality and commitment in voluntary cooperation 183 I. ‘The average contribution is high. I will try to keep the level of the group’s contribution for a while and rip off thereafter.’ J. ‘I will contribute the average of the others’ contributions in the previous round.’ K. ‘Enough is enough, from now on I will keep everything for myself. Everyone profited from my contributions, now I have to think about myself.’ Statement I is again a nice indication for rational cooperation. On the other hand, statements J and K hint to a type of behaviour that will be of special interest in the next section, namely conditional cooperation. In fact, most of the statements contain some sort of conditional cooperation argument. Contributing to the public good is clearly seen as desirable. Yet, for the majority of subjects, the reaction of other group members is crucial. Statements J and K can also be seen as supporting evidence for the notion of inequality aversion or reciprocity. Subjects obviously do not like their income to fall short of the others’ incomes. 2.2. Eliciting beliefs about others’ contributions While the verbal reasoning statements are insightful and suggestive of underlying motivations, the statements do not allow drawing tight conclusions about motivations. Specifically, we do not know what this subject expects others to contribute. A direct way around this is of course to simply ask the subjects about their belief about other group members’ contributions. A subject who contributes nothing and expects a positive contribution of the other subjects might be seen as a freerider. If the contribution and the belief are positively correlated we would call the subject a conditional cooperator. If the contribution is high irrespective of the belief we would call the subject an unconditional cooperator or an altruist. Croson (2007) was among the first to elicit beliefs and to correlate them with subjects’ contribution behaviour. She found a very high and statistically significant correlation of beliefs and contributions: subjects who expected others to contribute a lot were more likely to contribute high amounts than subjects who expected others to freeride. Croson (2007) did not look at individual behaviour. Her observation is that, on average, people behave conditionally cooperatively in that their contributions and beliefs are positively correlated. Fischbacher and Gächter

184


(2006) also elicited beliefs and replicated Croson’s findings of a positive correlation between beliefs and contributions. At the individual level they find subjects who do show a positive correlation between beliefs and contributions, whereas other subjects contribute zero even if they believe that others contribute positive amounts. 2.3. Inferring preferences from sequential decisions The beliefs data reported above provide the first systematic account of heterogeneous motivations. However, beliefs are not fully conclusive for inferring underlying motivations. Here is why. Consider we observe a subject contributing zero in the PG game. The subject might be classified as egoistic. Yet the subject might also be a conditional cooperator with pessimistic beliefs about the other group members’ contributions. The problem of the belief dependency of a conditional cooperator’s contribution decision can be solved by a very simple trick. Instead of letting the subjects decide simultaneously one can conduct the PD or PG game sequentially. In such a game one can observe the decisions of players who know the contributions of the other team members rather than just have a belief about it. Fehr, Kosfeld and Weibull (2003) conducted the PD game as shown in figure 8.1 in the sequential mode. Subjects received monetary payoffs that ensured that the incentive structure induced a PD. But, as explained above, monetary incentives might not coincide with preferences. To elicit actual preferences, Fehr et al. applied the following procedure: the row players have to indicate whether they cooperate or defect for both cases where the first-moving column player has cooperated and where he or she has defected. Fehr et al. now take a ‘revealed preference approach’, i.e., an individual’s preference is derived from observed choices. To see this, notice that if a player chooses to cooperate, when he or she could have also chosen to defect, then she apparently has a preference for cooperation. Since there are four possible outcomes, four possible preference types can be inferred: (1) a row player who chooses defect for both choices of the first mover is a selfish freerider; (2) a row player who chooses ‘freeride’ in case the column player chooses ‘freeride’ and contributes if the column player does so too is a reciprocating conditional cooperator; (3) a row player who contributes in any case can be classified as an altruistic unconditional cooperator, and (4) a row player who does the opposite of the column player (i.e., behaves anti-reciprocally) is called ‘other’.7

rationality and commitment in voluntary cooperation 185 The data reported by Fehr et al. (2003) show that the first two types are clearly the most important; 47 per cent of the subjects act like freeriders and 38 per cent show the pattern of conditional cooperators.8 Unconditional cooperators make up about 9 per cent of the population and the remaining 6 per cent of the subjects prefer to choose the opposite action. The elicited preferences can now be used to answer the question of which game the players really are playing. To appreciate this question, recall that the game in figure 8.1 is only a PD if both players in the game of figure 8.1 have preferences such that for both of them T > R and P > S, i.e., both are selfish. In all other cases, the game they really play is not a PD. Thus, if types were randomly and independently matched to play the PD, then they would play a PD in 22 per cent of the matchings, given the results of Fehr et al. (2003). 2.4. Eliciting ‘contribution functions’ Fischbacher, Gächter and Fehr (2001) and Fischbacher and Gächter (2006) use a similar revealed preference method to infer people’s contribution preferences in a PG as a function of other group members’ contributions. Therefore, the subjects in their experiment do not choose one contribution but a contribution as a function of other group members’ average contribution. The PG game is played in groups of four subjects and the payoff function is again the same as in (1). The game is played just once to avoid confounds with strategic considerations. Every subject has to indicate a contribution conditional on the average others’ contribution, i.e. for each of the 21 possible values of the average others’ contribution subjects have to enter the number of points they want to contribute. Fischbacher et al. classify their subjects according to their contribution function. A subject is called a freerider if and only if he or she contributes zero in all 21 cases. A subject is called a conditional cooperator if the contribution schedule is a clearly positive function of the others’ average contribution. A somewhat peculiar type is the triangle contributor whose contribution is increasing in the others’ contributions for low values and decreasing for high others’ contributions. Figure 8.2 illustrates the (average) contribution function of the different types in the experiments by Fischbacher and Gächter (2006). More than half of all subjects are conditional cooperators and 23 per cent are freeriders. The rest are either triangle contributors, or ‘others’. Fischbacher et al. (2001), and Herrmann

186

¨ simon g a¨ chter and christian th oni 20

Own contribution according to the 'Contribution table'

20 18 16 14

Conditional Cooperators: 55.0% Free Riders: 22.9% Triangle Contributors: 12.1% Others: 10.0%

18 16 14

12

12

10

10

8

8

6

6

4

4

2

2 0

0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Average contribution level of other group member

Figure 8.2. Average contribution function of types freerider, conditional cooperator, triangle contributor, and ‘others’. Observations on the diagonal would correspond to the type of a perfect (i.e. one-to-one) conditional cooperator. Source: Fischbacher and Gächter (2006).

and Thöni (2007), who replicated this experiment in Russia, got very similar distribution of types and even of average contribution patterns. 2.5. Further methods There are further studies that try to understand preference heterogeneity. They use versions and/or combinations of the methods described above. Bardsley and Moffatt (2007), for instance, study sequential decisions in the PG game. The game is played in groups of seven people who face a similar payoff function to (1). Unlike the typical PG experiment, the game is played sequentially, i.e., the subjects choose their contributions consecutively. The authors are in particular interested in the way a subject’s contribution decision is affected by the contributions of subjects deciding earlier. For a subject choosing early in the row there are several motives to choose a high contribution. On the one hand the subject might simply be cooperative. On the other hand, even a freerider might find it worthwhile to choose a high contribution if this induces other, later deciding subjects to contribute more. In other words, if a subject believes

rationality and commitment in voluntary cooperation 187 that the subsequent subjects are conditionally cooperative then there is a strategic reason to choose a high contribution. Bardsley and Moffatt use econometric techniques to classify their subjects and find that 25 per cent are conditional cooperators, 25 per cent are freeriders, and the remaining 50 per cent contribute strategically. Since the latter contribute only strategically the authors conclude that, in a one-shot situation, they should count as freeriders as well. Therefore Bardsley and Moffatt characterize one-quarter of their subjects as conditional cooperative and the remaining subjects as freeriders. Kurzban and Houser (2005) report results from a similar PG experiment where the subjects first choose their contribution. Then the subjects are given the chance to change their contribution in a circular way. At every step, one of the subjects learns the actual group average and has to reconsider their own contribution decision. Kurzban and Houser classify 20 per cent as freeriders, 63 per cent as conditional cooperators, and 13 per cent as unconditional cooperators. Burlando and Guala (2005) combine four different methods to assess a subject’s type.9 They find 32 per cent freeriders, 35 per cent conditional cooperators, and 18 per cent unconditional cooperators. The remaining 15 per cent cannot be classified. Finally, the subjects in Muller et al. (2005) play a two-stage public goods experiment (using a variant of the strategy method). Muller et al. (2005) classify 35 per cent as selfish subjects who give nothing in the second stage irrespective of the first stage contribution of the other players; 38 per cent are conditional cooperators who condition their second stage contribution positively on the first stage contribution of the other players. 2.6. Summary Table 8.1 summarizes the results of the studies discussed in sections 2.3 to 2.5. Comparing the results across studies reveals that the distribution of types varies considerably. Clearly, the distribution of types is sensitive to the experimental tool used and the classification scheme. Some studies do not mention the unconditional cooperator as a special type. However, it is encouraging that, when using the same methodology, the numbers hardly differ. This is obvious when comparing Fischbacher et al. (2001), Fischbacher and Gächter (2006), and Herrmann and Thöni (2007) (in the latter study at least the fraction of conditional cooperators is similar to

188


the other studies). In addition to that, in the PG game where degrees of cooperativeness are allowed (as opposed to the PD game) it seems that the fraction of pure freeriders is between one-fourth and one-third of the population (with the exception of the study by Bardsley and Moffatt). The fraction of conditional cooperators seems to be substantially larger. While numbers differ between studies, the significance of the findings summarized in Table 8.1 is that there is considerable heterogeneity in subjects’ cooperative motivations. The results from the systematic preference elicitation experiments are consistent with the findings from the verbal protocols and the belief elicitation methods. A particularly noteworthy observation from this synopsis is the fact that freerider types are not ubiquitous. Many subjects have preferences that commit them for conditional cooperation. From the perspective of revealed preference theory, therefore, the game that subjects really play may not be the PD game, or the public goods game as induced by the material incentives. The game

Bardsley and Moffatt (2007)

Kurzban and Houser (2005)

Fischbacher et al. (2001)

Fischbacher and Gächter (2006)

Herrmann and Thöni (2007)

Burlando and Guala(2005)

Muller et al. (2005)

N Freeriders (%) Conditional Cooperators (%) Unconditional Cooperators (%) Others / Unclassified (%)

Fehr et al. (2003)

Table 8.1. Overview of the distribution of types in Prisoner’s Dilemma games (Fehr et al. 2003) and Public Goods games (all other studies)

96 47

98 75

84 20

44 30

140 23

148 7

92 32

60 35

38

25

63

50

55

57

35

38

9

-

13

2

1

2

18

3

6

-

4

18

21

34

15

24

rationality and commitment in voluntary cooperation 189 that subjects actually play may well have multiple equilibria, in which cooperation may be an equilibrium outcome. In the following section we discuss some consequences of this preference heterogeneity for the prospects of successful cooperation. We will show that conditional cooperators, who know that others are like-minded cooperators as well, manage to maintain very high and largely stable cooperation rates. By contrast, when groups consist of a mixture of types, cooperation almost inevitably unravels.

3. The consequences of heterogeneous motivations Two immediate testable consequences of preference heterogeneity are that (i) in groups where group members are randomly selected cooperation is bound to be fragile, and (ii) in groups that are composed of ‘like-minded types’ (i.e., groups composed of either cooperators or freeriders) we should see starkly different cooperation patterns. The reason for the first conjecture is that because freeriding types will not contribute, conditional cooperators will withdraw their cooperation and therefore cooperation is bound to collapse. The rationale for conjecture (ii) is that conditional cooperators who know that the other group members are ‘like-minded’ cooperators as well should be able to cooperate at a higher level than if they must fear the freeriders in their group. Groups composed solely of freerider types should not cooperate at all. In the following we discuss these two implications of preference heterogeneity in turn. 3.1. The instability of voluntary cooperation We provide evidence in this section that heterogeneous motivations in randomly composed groups will lead to fragile cooperation. The reason is that freeriders presumably do not contribute to the public good while the conditional cooperators’ contributions might be non-minimal, depending on their belief about other group members’ contributions. Subjects learn the contributions of the other team members during the repeated interaction. The freeriders have no reason to react to that information. The conditional cooperators on the other hand will update their beliefs. Given that the average conditional cooperator does not fully match the others’ contribution, the reaction will most likely be a reduction of contributions. There is no

190


reason to expect that the remaining types (triangle contributors and ‘others’) will behave in a way that offsets the negative trend. To rigorously test this argument, Fischbacher and Gächter (2006) combined the elicitation of contribution functions described above with a standard ten-period public goods game played in the stranger mode, i.e., in every period the groups of four are formed randomly out of all subjects in a session. As predicted, contributions actually fall over time (from initially 40 per cent to 10 per cent by the last period). Is this decline really due to the interaction of heterogeneously motivated types? A first hint is that the types (as identified by their contribution schedules) really contribute differently. The conditional cooperators contribute on average 28 per cent of the endowment while the freerider’s average contribution is only 12 per cent. Surprisingly, too, the freeriders contribute in the repeated game. However, looking at individual data Fischbacher and Gächter report that 70 per cent of the freeriders never choose a contribution above zero during the ten periods. Therefore, the majority of the subjects classified as freeriders do indeed freeride all the time. Among the conditional cooperators this fraction of subjects who always choose the minimal contribution during the ten periods is much lower at 25 per cent. A second and more stringent support for the conjecture comes from using the elicited contribution functions for predicting contributions. Recall that the strategies asked subjects to indicate how much they are prepared to contribute to the public good for all feasible average contribution levels of the other group members. In the standard ten-period public goods game Fischbacher and Gächter (2006) also elicited in each period each subject’s belief about the other group members’ contributions. Therefore, we can—given a stated belief about other group members’ average contribution—predict what a subject should contribute to the public good if he or she were perfectly consistent with his or her elicited contribution function. Figure 8.3 depicts the actual average contributions in the ten rounds of the public goods game and the predicted contributions as a result of stated beliefs and contribution schedules. Although average predicted contributions are too low compared with actual contributions, we find that predicted contributions, which are derived from the contribution functions and the elicited beliefs, actually decline. Therefore, this result supports the argument that preference heterogeneity, rather than solely learning and reduced errors, leads to unstable

rationality and commitment in voluntary cooperation 191 14 Actual contribution Mean contribution

12

Predicted contribution derived from beliefs and schedules

10 8 6 4 2 0 1

2

3

4

5 6 Period

7

8

9

10

Figure 8.3. Average actual contributions and predicted contributions. Source: Fischbacher and Gächter (2006).

cooperation. A further conceptually important implication of this result is that the interaction of heterogeneously motivated subjects may lead to freeriding behaviour despite the fact that not everyone is motivated by selfishness. 3.2. Voluntary cooperation among like-minded people We have seen that a mixture of conditional cooperators and freeriders is unfavourable for reaching cooperation in the PG game. According to our second conjecture, conditional cooperators would presumably prefer to play the game with like-minded cooperators. ‘Team-directed reasoning’ and subsequent cooperation should be easy if the team players know that they are among like-minded group members. Similarly, if the ‘true game’ subjects are playing a game where cooperation is one of the equilibria (freeriding being another one), then knowing that others are like-minded cooperators should make it easy for subjects to coordinate on cooperation and to prevent freeriding. Likewise, if freerider types know that they are among other freeriders, freeriding should be paramount. Gächter and Thöni (2005) conducted an experiment where the subjects play in groups of ‘like-minded’ people. Thereby, like-mindedness refers to the type of a subject according to a classification whether one is a

192


freerider or a conditional cooperator. The experiment starts with a oneshot PG game. When all subjects have chosen their contribution the subjects are ranked according to their contribution. Then the subjects are reassigned to new groups. The reassignment works as follows. The three subjects with the highest contribution in the one-shot PG game constitute a first group. The subjects with the fourth to sixth highest contribution are in the second group and so on. Finally, the three least cooperative subjects find themselves in the last group. The subjects are informed about the reassignment procedure only after they have finished the first game. Then the subjects learn the contributions their new group members chose in the one-shot PG game. In the new group subjects play a ten-period PG-game. The reassignment mechanism adopted in this experiment sorts the subjects according to their contribution in a one-shot PG game. We have seen above that from the mere contribution decision we cannot determine the type of a subject unambiguously. This is because the type of conditional cooperator is compatible with all levels of contribution. We believe nevertheless that the reassignment mechanism as adopted by Gächter and Thöni provides a useful classification of the subjects along the dimension ‘uncooperative–cooperative’. In addition, the mechanism is easy to understand from the subjects’ point of view, which is crucial for the experiment. It is also important to note that the subjects do not know the reassignment mechanism when choosing their contribution in the one-shot PG game. A high contribution in this game therefore credibly reveals a cooperative attitude. How do subjects play the PG game when they know they are among like-minded people? Gächter and Thöni (2005) report the results from eighteen groups of three subjects. The left panel of figure 8.4 shows the results of the main treatment. In this game the maximal contribution is 20. For expositional ease the groups are divided into three classes (TOP, MIDDLE and LOW) according to their average contribution in the oneshot PG game. The three graphs show the average contribution during the ten periods separated by class. The unconnected dots in period zero show the average contribution in the one-shot PG game that determines the group composition. The classes remain clearly separated over all periods. The groups in the TOP class consist to a large degree of subjects who contributed their entire endowment in the one-shot PG game. These

20 18 16 14 12 10 8 6 4 2 0

Sorted

Contribution

Contribution

rationality and commitment in voluntary cooperation 193

0 1 2 3 4 5 6 7 8 9 10 Period TOP MIDDLE

20 18 16 14 12 10 8 6 4 2 0

Random

0 1 2 3 4 5 6 7 8 9 10 Period LOW Average

Figure 8.4. Average contributions over ten periods. Left panel: Average contributions over the ten periods for the TOP, MIDDLE, and LOW class in the Sorted treatment. The unconnected dots in period zero are the average contributions in the Ranking treatment. Right panel: Average contribution of the most, intermediate, and least cooperative groups over the ten periods. Source: Gächter and Thöni (2005).

groups manage to maintain almost full cooperation until the penultimate period. The contributions of the MIDDLE class (consisting of subjects with intermediate contributions in the one-shot PG game) show a similar pattern on a somewhat lower level. Surprisingly, the subjects in the LOW class also, who almost all chose a contribution of zero in the one-shot PG game, manage to reach a certain level of cooperation in the repeated game. The right panel of figure 8.4 shows the results from a control experiment. Groups are formed randomly as usual in this experiment, i.e., there is no reassignment according to cooperativeness. In order to make the two treatments comparable the data are still separated into the three classes. The separation now merely reflects the fact that there is variance in the contributions. What does the comparison between the left and the right panel of figure 8.4 tell us about the effect of grouping like-minded subjects? First of all, cooperation in the TOP class of the sorted treatment is much higher than the average contribution in the random treatment (dotted line in the right panel). However, the real value of the sorting mechanism becomes clear if we compare the TOP class with the most cooperative third of the groups in the random treatment. The average contribution of the TOP class of like-minded groups is significantly higher than the average contribution of

194


the most cooperative third of the groups in the random treatment. The fact that even the groups in the LOW class contribute somewhat more if they know they are among like-minded people is surprising at first sight. However, if uncooperative subjects know that they are among themselves then it is clear that there are no cooperative subjects to freeride on. This presumably motivates even uncooperative subjects to contribute some of their endowment in order to encourage the other freeriders to contribute as well. These groups might engage in ‘rational cooperation’ in the sense of Kreps et al. (1982). The fact that contributions drop to zero in the last period supports this hypothesis. These results are hard to reconcile with an error hypothesis but are consistent with social learning (i.e., learning about the behaviour of others) by heterogeneous types. The reason is that an error hypothesis would not easily predict that group composition effects matter for cooperation behaviour. Since people are heterogeneous with respect to their attitudes to cooperation, the results suggest that the dynamics of cooperation as produced by social learning will depend very strongly on the extent to which group members are ‘like-minded’. The results also confirm that social norms of cooperation are quite easy to sustain in homogeneous groups of people who are aware that others share their attitudes.

4. Altruistic punishment and negative emotions as a commitment device The experiments discussed in the previous section have shown that the cooperation problem can be solved if the ‘right’ people are grouped together. However, the cooperation problem is thereby only solved for the groups consisting of very cooperative subjects. In mixed groups cooperation is bound to collapse, since conditional cooperators will reduce their contributions once they realize that others freeride on them. Stopping cooperation is the only way to punish defectors. What if targeted punishment were possible? In this section we will discuss a slight change in the PG game that allows for targeted punishment of group members. This game (as introduced by Fehr and Gächter 2000, 2002) has two stages. The first stage is identical to the usual PG game. In the second stage the subjects learn the contributions

rationality and commitment in voluntary cooperation 195 of the other group members. They then have the possibility to punish each other by assigning ‘negative points’ to other group members. Each negative point costs one money unit to the punisher and reduces the income of the punished subject by three money units. Why should such a mechanism change the behaviour in the PG game? According to standard economic theory (i.e., under the joint assumptions of rationality and selfishness) it would not. The reason for this lies in the fact that punishment is costly for both parties involved. Even if it is possible to ‘educate’ other team-members with the stick, the cooperative subjects of a group still have to solve a cooperation game. Punishing other subjects is itself a public good. However, standard economic theory neglects a potentially influential factor, namely the subjects’ emotions. Being ‘suckered’ is presumably a negative experience for most of us. Such negative emotions might trigger revenge. The PG game with punishment gives the subjects a much more precise measure to seek revenge than just to withhold cooperation. People can use the punishment option to eliminate the freerider’s payoff advantage. Panel A of figure 8.5 (adopted from Fehr and Gächter 2000) shows that the possibility of using informal sanctions indeed leads to significantly higher contributions relative to the PG without punishment opportunities. This is true both for repeated interactions (the so-called ‘Partners’-treatment, where group composition stays constant) and for one-shot situations (the ‘Strangers’-treatment, where group composition changes randomly from period to period). In the Partner-condition, contributions in the presence of punishment approach almost full cooperation. Notice also that cooperation—both with and without punishment—is substantially higher among partners than among strangers. This holds already from the first period. We see this result as evidence against the maladaptation hypothesis discussed in section 1. Panel B of figure 8.5 depicts the average punishment a subject has received for a given deviation of that subject’s contribution from the average contribution of his or her group. The figure makes clear that more freeriding leads to more punishment. This holds for both partners and strangers. There is also no important difference in punishment between the two treatments, despite the fact that cooperation levels differ strongly. This suggests that the same deviation from a given group average is punished equally and punishment seems not to be used strategically.10

Mean contribution rate

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

with punishment

Partners Strangers 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Period

B Average punishment points

no punishment

A

7 6 5

Partner Stranger

4 3 2 1 0 [-20,-14) [-14,-8) [-8,-2) (-2,2] (2,8] (8,14] (14,20] Deviation from the average contribution of the other group members

Figure 8.5. Panel A: Cooperation patterns in the absence and presence of punishment and in stable (‘Partner’) and randomly changing (‘Stranger’) groups. Panel B: Mean received punishment as a function of one’s deviation from the group average contribution. Source: Fehr and Gächter (2000).

rationality and commitment in voluntary cooperation 197 Subjects’ cooperation behaviour looks quite rational given punishment. Subjects cooperate in accordance with the group norm to avoid punishment. Yet, why do people punish freeriders in a one-shot context although this is costly? Emotions may play a decisive role here (Fessler and Haley 2003), and negative emotions, in particular, may provide a proximate explanation. Freeriding may cause strong negative emotions among the cooperators and these emotions may in turn trigger the willingness to punish the freeriders. Theorists like Hirshleifer (1987) and Frank (1988) have argued that emotions may serve as a commitment device that induces people to retaliate if they feel cheated since an important property of emotions is that they imply an action tendency (see, e.g., Elster 1998). If the conjecture is correct that freeriding triggers negative emotions, we should observe particular emotional patterns in response to freeriding. To elicit these patterns the participants of the Fehr and Gächter (2002) experiments were confronted with the following two hypothetical scenarios after the final period of the second treatment. (The numbers in square brackets relate to the second scenario.) You decide to put 16 [5] francs into the project. The second group member puts 14 [3] and the third 18 [7] francs. Suppose the fourth member puts 2 [20] francs into the project. You now accidentally meet this member. Please indicate your feeling towards this person. After they had read a scenario subjects had to indicate the intensity of their anger and annoyance towards the fourth person (the freerider) on a seven point scale (1 = ‘very angry’, 4 = ‘neither angry nor happy’, 7 = ‘very happy’). The difference between scenario 1 and 2 is that the other three persons in the group contribute relatively much in scenario 1 and relatively little in scenario 2. Subjects report that they are angry if the fourth group member contributes less than they did. This effect is certainly more pronounced in the scenario where they contributed 16 than in the scenario where they contributed 5. The difference is highly significant. When the fourth group member contributes more than the pivotal subject, then people report to be quite happy. Surprisingly, subjects are equally happy about the contribution of 20 of the fourth member both when they have contributed 5 or 16 tokens. In other words, the gain in happiness seems not to depend on the

198


own contribution, whereas the intensity of the negative emotions strongly depends on the own contribution. Overall, the results suggest that freeriding causes negative emotions. Moreover, the emotional pattern is consistent with the hypothesis that emotions trigger punishment. First, the majority of punishments are executed by above-average contributors and imposed on below-average contributors. Second, recall from figure 8.5B that punishment increases with the deviation of the freerider from the other members’ average contribution. This is consistent with the observation that negative emotions are the more intense the more the freerider deviates from the others’ average contribution. Third, evidence from neuroscientific experiments supports the interpretation that emotions trigger punishment. For instance, Sanfey et al. (2003) had their subjects play the ultimatum game while the subjects’ brains were scanned (using fMRI). The ultimatum game is a two-player game in which player 1 is asked to split an amount of money, say ¤10, between him- or herself and player 2. Player 2 can only accept or reject the proposal. The offer is implemented if he accepts; both get nothing if he rejects. A rejection of a positive offer in the ultimatum game is also an instance of punishment. The brain scans showed that in the recipients who received an unfairly low offer by a human player 1, areas in the brain lit up that are related to negative emotions. When the unfair offer came from a computerized player 1, recipients were much less negatively aroused. De Quervain et al. (2004) also studied neural activations of punishing subjects. They found that punishment activates the ‘reward centre’ of the brain, i.e., to punish is rewarding. Hence, the proverb ‘revenge is sweet’. They were also able to show that subjects for whom punishment was more rewarding, actually punished more. Taken together, these regularities are consistent with the view that emotions are an important proximate mechanism in the process that triggers punishment. In the next section we look at ultimate explanations for cooperation and punishment.

5. Ultimate explanations The evidence presented above shows that many people—but not all—behave reciprocally. They cooperate if others cooperate and they punish freeriders. Since this takes place even in one-shot games, this kind

rationality and commitment in voluntary cooperation 199 of reciprocity has been termed ‘strong reciprocity’ (e.g. Gintis 2000; Fehr, Fischbacher and Gächter 2002) to distinguish it from reciprocal altruism that occurs in repeated games. Reciprocal altruism is strategic reciprocity that can also be exhibited by a completely selfish individual, who would never cooperate or punish in a one-shot context. What explains the existence of strong reciprocity? Specifically, if sufficiently many people punish freeriders sufficiently strongly, freeriders have no incentive to freeride any more. Yet, why should anyone punish and not freeride on other’s punishment, since altruistic punishment is just a second-order public good? The answer will probably be found in the evolutionary conditions of the human species that caused a propensity for strongly reciprocal behaviour among a significant fraction of the population. The evidence presented suggests that strong reciprocity cannot easily be explained by kin selection (Hamilton 1964), reciprocal altruism (Trivers 1971; Axelrod and Hamilton 1981), indirect reciprocity (e.g. Nowak and Sigmund 1998), or by costly signalling theory (Zahavi and Zahavi 1997). One explanation, already mentioned above is the ‘maladaptation hypothesis’. According to this account, strong reciprocity is merely a by-product of reciprocal altruism, indirect reciprocity, or signalling. Humans evolved in small and mostly stable groups and thereby acquired the psychology and emotions needed for sustaining cooperation. Thus, the human brain applies ancient cooperative heuristics even in modern environments, where they are maladaptive. Humans did not evolve to play one-shot games and therefore, when they are in a novel environment like a one-shot lab experiment, they behave as if they were in a repeated game. This argument is problematic in our view for two reasons. First, it is obvious that people did not evolve to play one-shot lab experiments and the strong reciprocity observed there does not represent adaptive behaviour. Yet, laboratory experiments allow us to test to what extent people distinguish between one-shot and repeated games and to what extent they think strategically. People cooperate substantially more with ‘partners’ than with ‘strangers’ as demonstrated above (see figure 8.5B). People also report stronger negative emotions when they are cheated by a ‘partner’ than by a ‘stranger’ (Fehr and Henrich 2003). Second, anthropologists have shown that group dispersal and migration, and thereby the possibility to meet strangers, were quite common (see Fehr and Henrich 2003, in particular pp. 69–76). Thus, vigilant individuals who are able to

200


distinguish whether they are dealing with a ‘partner’ or a ‘stranger’ should have a fitness advantage. An alternative and, in our view, quite promising approach is ‘geneculture co-evolution’ (e.g., Boyd, Gintis, Bowles, and Richerson 2003; Boyd and Richerson 2004). One line of reasoning (e.g., Boyd et al. 2003) goes as follows. Assume that in a population there are two behavioural types: cooperators and defectors. The cooperators incur a cost c to produce a benefit b that accrues to all group members. Defection is costless and produces no benefit. If the fraction of cooperators is x then the expected payoff for cooperators is bx − c, whereas defectors get bx. Thus, the payoff difference is c, independent of the number of cooperators. Cooperators would always be at an evolutionary disadvantage under such circumstances. Now assume that there is a fraction y of ‘punishers’ who cooperate and punish defectors. Punishment reduces the payoff of the punished defector (by p) but also of the punishing subject (by k). The payoff of cooperators who cooperate but don’t punish (‘second-order freeriders’) is b(x + y) − c; the punished defectors get b(x + y) − py, and the punishers earn b(x + y) − c − k(1 – x– y). If the cost of punishments exceed the costs of cooperation (i.e., if py > c), then cooperators have a higher fitness than defectors and the fitness disadvantage of punishers relative to the second-order freeriders is k(1 – x – y). Thus, punishment is altruistic and the cooperation and punishment game can have multiple equilibria. This line of reasoning reveals two things. First, there is an important asymmetry between altruistic cooperation and punishment. In an environment without punishment, cooperators are always worse off than defectors, irrespective how numerous they are. Second, by contrast to the first observation, the cost disadvantage of altruistic punishment declines as defection becomes infrequent because punishment is not needed any more. The selection pressure against altruistic punishers is weak in this situation. This latter observation suggests that within-group forces, such as copying successful and frequent behaviour, can stabilize cooperation. Boyd et al. (2003) formally investigated another mechanism: cultural group selection. Recall that in the presence of strong reciprocators the cooperation game may have multiple equilibria, equilibria which imply cooperation, and defection equilibria. Different groups may settle at different equilibria. Here, cultural group selection may come into play. The main idea is that groups with more cooperators are more likely to win inter-group conflicts

rationality and commitment in voluntary cooperation 201 and are less likely to become extinct, because they may survive during famine, manage their common resources better, and so on. Therefore, this kind of group selection will tend to increase cooperation because groups who arrived at a cooperative equilibrium are more likely to survive. Moreover, cooperative groups will tend to have more punishers. Since the within-group selection effect is weak if there is a lot of cooperation, cultural group selection can support the evolution of altruistic punishment and maintain it, once it is common. Boyd et al. developed a simple model to test this intuition rigorously. They simulated the model for important parameters, like group size, migration rates between groups, and the cost of being punished. The parameters were chosen to mimic likely evolutionary conditions. The simulation results are very interesting because they show that cultural group selection can support altruistic punishment under a wide range of parameters. First, in the absence of punishment, group selection can only sustain cooperation in very small groups, whereas in the presence of punishment high and stable cooperation rates can be achieved even in large groups. Second, higher migration rates between groups decreases cooperation rates. Cooperation breaks down if the cost of being punished is small. The significance of this and related models is to show that individual selection and cultural factors, such as conformism and group selection, may coexist (and not be incompatible as in purely gene-based models) and can explain why strong reciprocity may survive and why we see the preferences we see.

6. Concluding discussion In this essay we have demonstrated that carefully controlled laboratory experiments allow for a systematic investigation of how people actually solve cooperation problems that by definition involve a tension between collective and individual interests. We have also shown that according to several different experimental instruments apparently many people have non-selfish preferences that dispose them to cooperate if others cooperate as well. Under appropriate interaction structures (being among like-minded cooperators or having punishment opportunities) these preferences allow people to realize much higher gains from cooperation than would be

202


possible if all were selfish. However, all models have also shown that a non-negligible fraction of subjects behave selfishly and freeride in one-shot games with no punishment. We believe that these results have important methodological consequences. First, the experiments have demonstrated that one can collect empirical information on the structure of people’s preferences. In other words, it is not necessary to treat preferences as a black box any more. Second, the results challenge the ubiquity of the selfishness assumption that underlies many models in the social sciences, in particular in economics, and in biology. In addition to the material payoff, many people are apparently also motivated by other-regarding considerations, and/or by reciprocity. In fairness, one should notice that abstract economic theory does not invoke selfishness. Preferences only need to obey consistency axioms and can otherwise encompass any sort of motivation. Yet, in practice, selfishness is often invoked. Third, while most experimental facts that we and others have interpreted as strong reciprocity are now undisputed because they have been often replicated, the ultimate explanation of strong reciprocity is still open to debate. We believe that the issue of providing an ultimate account of strong reciprocity is of considerable theoretical interest for all behavioural scientists who hitherto have based their models on selfishness as being the only evolutionary reasonable assumption. Fourth, the observed results do not contradict rational choice theory. They only contradict the universal selfishness assumption. The experimental evidence from many different games, including those discussed here (see Camerer 2003), has led to the development of rational choice models of ‘social preferences’, which take others’ well-being into account or model their taste for reciprocity. These models are standard game-theoretic models that put structure on people’s utility functions and otherwise apply standard solution concepts like subgame perfect Nash equilibrium. For instance, Fehr and Schmidt (1999) or Bolton and Ockenfels (2000) propose a utility function with inequality aversion. In addition to own material payoff, people experience some disutility whenever there is disadvantageous or advantageous inequality in the payoff allocation. In the former, one receives a smaller payoff than the comparison partner; in the latter one earns more than the comparison partner. Inequality-averse subjects might choose non-minimal contribution in equilibrium if they believe that the other subjects do so as

rationality and commitment in voluntary cooperation 203 well. This is because they do not like their income to be higher than the others’ incomes, i.e., they dislike advantageous inequality. Other models like Rabin (1993), Dufwenberg and Kirchsteiger (2004), and Falk and Fischbacher (2006) propose reciprocal preferences. Reciprocally motivated subjects might choose a non-minimal contribution if they believe the other subjects do so as well, in order to return the favour.11 Fifth, one may criticize these models because people’s reasoning may not be individualistic as typically assumed in the rational choice approach but may be ‘team-directed’ (Bacharach 1999; Sugden 2000). The casual evidence that we have sketched above suggests this to be a psychologically plausible explanation of how people reason in games (although the approaches by Bacharach and Sugden are more philosophical than psychological). Sixth, a further criticism is that people’s rationality may fall short of the high rationality required in these sorts of rational choice models because people are boundedly rational (see, e.g., Brandstätter, Güth and Kliemt 2003 for a recent encompassing discussion). We agree with this criticism but note that these models were first steps in demonstrating that the relaxation of the selfishness assumption has led to many important insights that appeared puzzling before the new models were available. Notes

1. See, e.g., Binmore (1994, 1998) for an extensive treatment. 2. Laboratory experiments are probably the best tool for studying cooperation. The reason is that in the field many factors are operative at the same time. The laboratory allows for a degree of control that is not feasible in the field. In all the laboratory experiments that we will discuss below participants earned considerable amounts of money depending on their decisions. Thus, the laboratory allows observing real economic behaviour under controlled circumstances (see Friedman and Sunder 1994 for an introduction to methods in experimental economics; Kagel and Roth 1995; Camerer 2003 for overviews of important results; and Guala 2005 for a thorough discussion of the methodology of experimental economics). 3. The original story goes as follows. Two arrested criminals are interrogated separately and have to decide whether to confess or not to confess. If

¨ 204 simon g a¨ chter and christian th oni both criminals do not confess they have to stay in jail for a short time. If one confesses while the other does not then the confessor can leave the prison while the other stays in jail for a long time. If both criminals confess they both stay in jail for an intermediate time. 4. Since the PD and an n-person version of the PD, the public goods game introduced below, highlight the tension between individual and collective rationality they have been used to analyse such diverse areas as warfare, cooperative hunting and foraging, environmental protection, tax compliance, voting, the participation in collective actions like demonstrations and strikes, the voluntary provision of public goods, donations to charities, teamwork, collusion between firms, embargos and consumer boycotts, and so on. 5. We regard team-direct reasoning as psychologically plausible because team-directed reasoning is consistent with the observation that group identity (‘‘we-feelings’’) are important for cooperation (see Dawes 1980 and Dawes, van de Kragt and Orbell 1988). 6. Gächter and Fehr (1999) did not report the statements for lack of space. 7. In general, utilities in games are von-Neumann–Morgenstern utilities. To infer them, one would have to elicit more than just ordinal rankings as Fehr et al. do. For an insightful discussion see Weibull (2004). Yet, for the purposes of solving the game with the concept of strict Nash equilibrium, ordinal preferences are sufficient. 8. Clark and Sefton (2001) also study a sequential PD and find that between 37 and 42 per cent of the subjects cooperate conditionally on others’ cooperation. 9. The four methods are: (i) the strategy method, similar to one described in section 2.4; (ii) a value orientation test devised by psychologists; (iii) a repeated public goods game; and (iv) a post-experimental questionnaire. 10. By now, these results have been replicated many times. For a survey of the most important results see Kosfeld and Riedl (2004). 11. Sugden (1984) was one of the first to argue for the importance of reciprocity in the voluntary provision of public goods.

References Andreoni, J., and Miller, J., 1993. ‘Rational Cooperation in the Finitely Repeated Prisoner’s Dilemma: Experimental Evidence’. Economic Journal 103: 570–85.

rationality and commitment in voluntary cooperation 205 Axelrod, R., and Hamilton, W. 1981. ‘The Evolution of Cooperation’. Science 211: 1390–6. Bacharach, M. 1999. ‘Interactive Team Reasoning: A Contribution to the Theory of Co-operation’. Research in Economics 53: 117–47. Bardsley, N., and Moffatt, G. 2007. ‘The Experimetrics of Public Goods: Inferring Motivations from Contributions’.Theory and Decision 62: 161–93. Binmore, K. 1994. Game Theory and the Social Contract Vol. 1: Playing Fair. Cambridge, MA: MIT Press. Binmore, K. 1998. Game Theory and the Social Contract Vol. 2: Just Playing. Cambridge, MA: MIT Press. Bolton, G., and Ockenfels, A. 2000. ‘ERC: A Theory of Equity, Reciprocity, and Competition’. American Economic Review, 90/1: 166–93. Boyd, R., and Richerson, P. J. 2004. Not by Genes Alone: How Culture Transformed Human Evolution. Chicago: University of Chicago Press. Boyd, R., Gintis, H., Bowles, S., and Richerson, P. J. 2003. ‘Evolution of Altruistic Punishment’. Proceedings of the National Academy of Sciences 100/6: 3531–5. Brandstätter, H., Güth, W., and Kliemt, H. 2003. ‘The Bounds of Rationality: Philosophical, Psychological and Economic Aspects of Choice Making’. Homo Oeconomicus 20/2-3: 303–56. Burlando, R., and Guala, F. 2005. ‘Heterogeneous Agents in Public Goods Experiments’. Experimental Economics 8/1: 35–54. Camerer, C. 2003. Behavioral Game Theory. Princeton: Princeton University Press. Clark, K., and Sefton, M. 2001. ‘The Sequential Prisoner’s Dilemma: Evidence on Reciprocation’. Economic Journal 111: 51–68. Cooper, R., DeJong, D., Forsythe, R., and Ross, T. 1996. ‘Cooperation without Reputation: Experimental Evidence from Prisoner’s Dilemma Games’. Games and Economic Behavior 12: 187–218. Croson, R. 2007. ‘Theories of Commitment, Altruism and Reciprocity: Evidence from Linear Public Goods Games’. Economic Inquiry 45: 199–216. Dawes, R. M., 1980. ‘Social Dilemmas’. Annual Review of Psychology 31: 169–93. Dawes, R. M., van de Kragt, A. J. C., and Orbell, J. M. 1988. ‘Not Me or Thee, but We: The Importance of Group Identity in Eliciting Cooperation in Dilemma Situations—Experimental Manipulations’. Acta Psychologica 68: 83–97. de Quervain, J.-F., Fischbacher, U., Treyer, Y., Schellhammer, M., Schynyder, U., Buck, A., and Fehr, E. 2004. ‘The Neural Basis of Altruistic Punishment’. Science 305: 1254–8. Dufwenberg, M., and Kirchsteiger, G. 2004. ‘A Theory of Sequential Reciprocity’. Games and Economic Behavior 47: 268–98.

¨ 206 simon g a¨ chter and christian th oni Elster, J. 1998. ‘Emotions and Economic Theory’. Journal of Economic Literature 36: 47–74. Falk, A., and Fischbacher, U. 2006. ‘A Theory of Reciprocity’. Games and Economic Behavior 54: 293–315. Fehr, E., and Fischbacher, U. 2003. ‘The Nature of Human Altruism’. Nature 425: 785–91. Fehr, E., and Gächter, S. 2000. ‘Cooperation and Punishment in Public Goods Experiments’. American Economic Review, 90/4: 980–94. Fehr, E., and Gächter, S. 2002. ‘Altruistic Punishment in Humans’. Nature 415: 137–40. Fehr, E., and Henrich, J. 2003. ‘Is Strong Reciprocity a Maladaptation? On the Evolutionary Foundations of Human Altruism’. In P. Hammerstein (ed.), Genetic and Cultural Evolution of Cooperation. Cambridge, MA: MIT Press, pp. 55–82. Fehr, E., and Schmidt, K. 1999. ‘A Theory of Fairness, Competition, and Cooperation’. Quarterly Journal of Economics 114: 817–68. Fehr, E., Fischbacher, U., and Gächter, S. 2002. ‘Strong Reciprocity, Human Cooperation, and the Enforcement of Social Norms’. Human Nature 13/1: 1–25. Fehr, E., Kosfeld, M., and Weibull, J.W. 2003. ‘The Game Prisoners (Really) Play’. Mimeo, Institute for Empirical Research in Economics, University of Zurich. Fessler, D., and Haley, J. K. 2003. ‘The Strategy of Affect: Emotions in Human Cooperation’. In P. Hammerstein (ed.), Genetic and Cultural Evolution of Cooperation. Cambridge, MA: MIT Press. Fischbacher, U., and Gächter, S. 2006. Heterogeneous Motivations and the Dynamics of Freeriding in Public Goods. CeDEx Discussion Paper No. 2006–1, University of Nottingham. Fischbacher, U., Gächter, S., and Fehr, E. 2001. ‘Are People Conditionally Cooperative? Evidence from a Public Goods Experiment’. Economics Letters 71: 397–404. Frank, R. 1988. Passion Within Reason. The Strategic Role of the Emotions. New York: W.W. Norton & Company. Friedman, D. and Sunder, S. 1994. Experimental Methods. A Primer for Economists. Princeton: Princeton University Press. Fudenberg, D. and Maskin, E. 1986. ‘The Folk Theorem in Repeated Games with Discounting or with Incomplete Information’. Econometrica 54: 533–56. Gächter, S. and Fehr, E. 1999. ‘Collective Action as a Social Exchange’. Journal of Economic Behavior and Organization 39: 341–69.

rationality and commitment in voluntary cooperation 207 Gächter, S., and Tho¨ ni, C. 2005. ‘Social Learning and Voluntary Cooperation Among Like-minded People’. Journal of the European Economic Association 3/2-3: 303–14. Gintis, H., 2000. ‘Strong Reciprocity and Human Sociality’. Journal of Theoretical Biology 206: 169–79. Guala, F., 2005. The Methodology of Experimental Economics. Cambridge: Cambridge University Press. Hamilton, W. 1964. ‘Genetical Evolution of Social Behavior I, II’. Journal of Theoretical Biology 7/1: 1–52. Hammerstein, P. (ed.). 2003. Genetic and Cultural Evolution of Cooperation. Cambridge, MA: MIT Press. Herrmann, B., and Thöni, C. 2007. ‘Measuring Conditional Co-operation’. Mimeo, University of Nottingham. Hirshleifer, J. 1987. ‘On the Emotions as Guarantors of Threats and Promises’. In J. Dupré (ed.), The Latest on the Best. Essays on Evolution and Optimality. Cambridge, MA: MIT Press. Johnson, D., Stopka, P., and Knights, S. 2003. ‘The Puzzle of Human Cooperation’. Nature 421: 911–12. Kagel, J., and Roth, A. E. (eds). 1995. Handbook of Experimental Economics. Princeton: Princeton University Press. Keser, C., and van Winden, F. 2000. ‘Conditional Cooperation and Voluntary Contributions to Public Goods’. Scandinavian Journal of Economics 102/1: 23–9. Kosfeld, M., and Riedl, A. 2004. ‘The Design of (De)centralized Punishment Institutions for Sustaining Cooperation’. Tinbergen Institute Discussion Paper TI 2004-025/1. Kreps, D., Milgrom, P., Roberts, J., and Wilson, R. 1982. ‘Rational Cooperation in the Finitely Repeated Prisoners’ Dilemma. Journal of Economic Theory 27: 245–252. Kurzban, R., and Houser, D. 2005. ‘An Experimental Investigation of Cooperative Types in Human Groups: A Complement to Evolutionary Theory and Simulations’. Proceedings of the National Academy of Sciences 102/5: 1803–7. Ledyard, J. 1995. ‘Public Goods: A Survey of Experimental Research’. In J. Kagel. and A. E. Roth (eds.), Handbook of Experimental Economics. Princeton: Princeton University Press. Muller, L., Sefton, M., Steinberg, R., and Vesterlund, L. 2005. ‘Strategic Behavior and Learning in Repeated Voluntary-Contribution Experiments’. CeDEx Working Paper No. 2005-13, University of Nottingham. Nowak, M., and Sigmund, K. 1998. ‘Evolution of Indirect Reciprocity by Image Scoring’. Nature 393: 573–7.

208


Oberholzer-Gee, F., Waldfogel, J., and White, M. 2003. ‘Social Learning and Coordination in High-Stakes Games: Evidence From Friend or Foe’. NBER Working Paper 9805. Poundstone, W. 1992. Prisoner’s Dilemma. New York: Anchor Books. Rabin, M. 1993. ‘Incorporating Fairness into Game Theory and Economics’. American Economic Review, 83/5: 1281–1302. Rapoport, A., and Chammah, A. M. 1965. Prisoner’s Dilemma. A Study in Conflict and Cooperation. Ann Arbor: The University of Michigan Press. Sanfey, A. G., Rilling, J. K., Aronson J. A., Nystrom L. E., and Cohen, J. D. 2003. ‘The Neural Basis of Economic Decision-making in the Ultimatum Game’. Science 300: 1755–8. Sen, A. 1977. ‘Rational Fools: A Critique of the Behavioral Foundations of Economic Theory’. Philosophy and Public Affairs 6: 317–44. Sugden, R. 1984. ‘Reciprocity: The Supply of Public Goods Through Voluntary Contributions’. Economic Journal 94: 772–87. Sugden, R. 1993. ‘Thinking as a Team: Toward an Explanation of Nonselfish Behaviour’. Social Philosophy and Policy 10: 69–89. Sugden, R. 2000. ‘Team Preferences’. Economics and Philosophy 16: 175–204. Trivers, R. L. 1971. ‘The Evolution of Reciprocal Altruism’. Quarterly Review of Biology 46: 35–57. Weibull, J. 2004. ‘Testing Game Theory’. In S. Huck (ed.), Advances in Understanding Strategic Behavior. Game Theory, Experiments, and Bounded Rationality. Houndmills: Palgrave Macmillan, pp. 85–104. Zahavi, A. and Zahavi, A. 1997. The Handicap Principle: A Missing Piece of Darwin’s Puzzle. New York: Oxford University Press.

PA RT I I I

Commitment, Intentions, and Identity


9 Beyond Self-Goal Choice: Amartya Sen’s Analysis of the Structure of Commitment and the Role of Shared Desires H A NS B E RNH A RD S C H M I D

1. Two concepts of commitment In the current debate on economic rationality, Amartya Sen’s analysis of the structure of commitment plays a uniquely important role (Weirich 2004: 387 ff.). However, Sen is not alone in pitting committed action against the standard model of rational behavior. Before turning to Sen’s analysis in section 2 of this paper, I shall start with an observation concerning some of the other relevant accounts. It seems that the concept of commitment plays a key role in two opposing views on what is wrong about the classical model. On the first view, commitment epitomizes everything that transcends those egoistic preferences, inclinations, and desires on which Homines œconomici are usually taken to act. What is needed in order to accommodate committed action is, first of all, to widen the concept of the subjective motivational base of actions, and perhaps to allow for a less static conception, which gives more room for deliberation. On this first view, the talk about ‘‘desires’’ as being the motivational base of action has to be taken in something like the formal sense in which Bernard Williams uses this term. As Williams puts it, the ‘‘subjective motivational set,’’ is not limited to egoistic impulses or desires, but ‘‘can contain such things as dispositions of evaluation, patterns

212

hans bernhard schmid

of emotional reaction, personal loyalties, and various projects ... embodying commitments of the agent’’ (Williams 1979: 20; my emphasis). According to the second account, however, ‘‘commitment’’ stands for the necessity of much more radical changes in our understanding of practical reason. On this view, it is not enough to widen our concept of the motivation on which individuals act in order to accommodate commitments. If commitments are a reason for action, this is not because these commitments somehow express what the agent wants. Commitments are not based in the agent’s desires. The opposite holds. If an agent wants what she does when she acts on a commitment, she wants it because she believes she has a reason to do so, and not the other way around. Thus on this second view on committed action, reasons and not motivations are metaphysically basic (cf. McNaughton and Rawling 2004: 117). In this sense, commitment plays a key role in those theories of practical reason which are radically skeptical of the understanding of rationality in action that is usually called ‘‘Humean’’ (even though it has perhaps not much to do with David Hume’s actual views). Robert B. Brandom describes his anti-Humean turn in the following words: ‘‘The concepts of desire and preference are ... demoted from their position of privilege ... Endorsement and commitment are at the center of rational agency ... and inclination enters only insofar as rational agents must bring inclination in the train of rational propriety, not the other way around’’ (Brandom 2000: 30).1 Most prominently, John Searle has sketched a non-Humean account of rationality in action, in which an analysis of the structure of commitment plays a key role (Searle 2001). On his view, commitments do not fit into an account of rationality in action, which bases the reasons for action in the subjective motivations of the agent. Rather, commitments create, as Searle puts it, desire-independent reasons for action. In Searle’s example, one does not have to have any (egoistic or altruistic) desire to have reason to pay for the beer one has ordered. The fact that one has ordered the beer is quite reason enough. Searle’s analysis of the structure of commitments runs about as follows: commitments are created with the use of language; by means of some ‘‘semantical categorical imperative,’’ as Searle calls it, ordering a beer in a bar results in the creation of a reason to pay for the beer, a reason which is independent of whatever the agent in question does or does not have in her or his subjective motivational set (Searle 2001: 167 ff.).

beyond self-goal choice 213 As opposed to the first, Humean or internalist, account of commitment, the second one is the Kantian or externalist one. I do not want to go further into the details of either of these accounts here, but limit myself to the most obvious problems of both views. The problem with the Humean view of commitment seems to be that it blurs the distinction between two different cases of reasons for action. From the agent’s point of view at least, it seems important to distinguish the case in which he or she believes they have reason to do x because he or she wants to do so, from the case in which he or she wants to do x because he or she believes they have a reason to do so. Sometimes, there are even cases of conflict. One sometimes feels bound by commitments against one’s ‘‘subjective motivations,’’ however wide these motivations are (one might even feel bound by commitments against one’s altruistic motivations). It seems that the Humean view cannot do full justice to these cases. The existing Kantian or externalist accounts of commitment, in turn, have their own problems. If one takes reasons for action, but not motivation, to be metaphysically basic, especially if one accepts the creation of reasons for action through the semantic categorical imperative, the old question imposes itself, of how those reasons, all in themselves, should move us to act, without the aid of some desires such as the one to be a rational agent.2 In his book on rationality in action, Searle tries to answer this question with what he calls ‘‘secondary desires,’’ which are desires created by the recognition of some prior desire-independent reason (Searle 2001: 168 ff.). In other words, those secondary desires play the decisive role of ensuring that one really wants to do what one ought to do. As such, secondary desires are simply too good to be true. In Searle’s story, these secondary desires play the dubious role of the deus ex machina, who suddenly puts in an appearance on the scene to save Searle’s externalist account. And indeed it seems hard to see why we should worry about the semantic categorical imperative were it not for some prior desire such as the one to be consistent in our views, or the desire to be a trustworthy person and not to erode the base of mutual trust, or some other desire of this type. Both accounts of committed action have their relative strengths and weaknesses. Perhaps the problem with finding out what side to take has to do with the way the line between the two camps is drawn. Looking at this constellation from afar, I think it is plausible to assume that there might be something wrong with this whole controversy. Maybe the whole

214


question concerning the relation between motivation and commitment is wrongly put. Even though I do not know what Sen’s own position on the controversy between internalist and externalist accounts of commitment is,3 I think that some elements in his analysis of committed action point the way to leaving that constellation behind. In the following, I shall turn to Sen’s analysis (sections 2 and 3), before coming back to the controversy between internalist and externalist accounts of commitment at the end of the paper (section 4).

2. Self-goal choice It seems that in his papers on the topic, Sen’s analysis of the structure of committed action revolves around two main ideas, one of which is widely accepted, while the other, as far as I can see, has not met with much approval so far. The first, less controversial point concerns the ‘‘wedge between choice and welfare’’ driven by committed action, which Sen postulates in his paper ‘‘Rational Fools.’’ Committed action requires us to go beyond narrow standard models of preference. ‘‘Preferences as rankings have to be replaced by a richer structure involving meta-rankings and related concepts’’ (Sen 1977: 344). In his paper on ‘‘Goals, Commitment, and identity,’’ Sen further analyzes this by saying that committed action violates both the assumption that a person’s welfare depends only on her or his own consumption (goal-self-regardingness), and the assumption that a person’s only goal is to maximize his or her welfare (self-welfare goal), including satisfaction of sympathy. Both assumptions are implicit in the standard economic model of rational action (Sen 1985: 213). Whereas these two points can be seen as a refinement of the earlier statement made in ‘‘Rational Fools,’’ Sen now goes one step further by saying that there is yet another standard assumption that is violated by committed action. It is selfgoal choice. According to the more radical of Sen’s two statements of the self-goal choice assumption (Sen 2002: 34), it basically says the following: ‘‘a person’s choices must be based entirely on the pursuit of her own goals.’’ (In a slightly softer version, self-goal choice is taken to mean that ‘‘each act of choice is guided immediately by the pursuit of one’s own goals’’ (Sen 1985: 214; 1987: 80, my emphasis).) Since, in Sen’s view, committed action violates this assumption, the wedge driven by commitment is not between

beyond self-goal choice 215 the agent’s choice and her or his welfare, as it was in ‘‘Rational Fools.’’ Rather, it is between the agent’s choice and her or his goals. The claim is that committed agents do not pursue their (own) goals. As Sen knows well, this claim sounds rather extreme. Indeed it seems that in spite of its appeal to some everyday phrases, it is not even understandable. In everyday parlance, we might say of strongly altruistic or heteronomous people that they do not pursue their own goals, but the goals of other people instead. Yet in the proper sense, self-goal choice is not violated even in the most extreme cases. For the whole clue of such strongly altruistic or perhaps heteronomous behavior seems to be that the agent makes the other’s goals his own. As Sen, who is well aware of this problem, puts it: ‘‘it might appear that if I were to pursue anything other than what I see as my own ‘goals’, then I am suffering from an illusion; these other things are my goals, contrary to what I might believe’’ (Sen, 2002: 212). Perhaps the problem of Sen’s claim becomes clearer if we take a closer look at the role of goals in agency. I take it that in a basic sense, goals are something like the conditions of satisfaction of intentions. ‘‘Conditions of satisfaction’’ is meant in Searle’s sense (Searle 1983), and it has nothing to do with any kind of psychological enjoyment. The claim that goals are the conditions of satisfaction of intentions simply means the following: goals are whatever has to be the case for somebody to have done what she or he intended to do. In order to attain my goal to close the door, I simply have to close the door. As compared with other, more elaborated accounts of goals and their roles in agency, this approach might seem overly simplistic. More than that, it might appear that this reading draws intentions and goals too close together. Especially, it seems that to identify goals with conditions of satisfaction of intentions wrongfully excludes such cases in which somebody may be said to have a goal without actually intending to do something about it. I might have the goal to close the door, and yet not the intention to close the door, because my more important goal is to eat the ice cream.4 Against this objection, one might argue that the intention to do something about it is what distinguishes an actual goal from a mere wish, or desired state of affairs. However, we need not settle this issue here, because in the present context the role of goal interests us only insofar as goals pertain to intentionality and action (or, in the parlance of the economic model of behavior: to choice). Thus we need

216


not claim that there are no goals without intention, or no intentions without goals, for that matter (even though I conjecture that the use of the term ‘‘goal’’ in these cases is widely equivocal). All that is claimed is that the role of goals in action is that of conditions of satisfaction of the corresponding intentions. I assume that something similar must be included in any account of the role of goals in agency. And this claim seems especially fit to shed light on the trouble with Sen’s critique of self-goal choice. The example mentioned above may serve to illustrate the point. In order to attain my goal to close the door, I simply have to close the door. This, however, I have to do myself, because the mere fact that the door is shut is not enough to satisfy my intention. If you pre-empt me and close the door for me, or if the draft does the job before I could get around to doing it, this might fully satisfy some other intentional state of mine such as my long-standing desire that the door be closed. However, it does not satisfy my intention to close the door (which might have been prompted by that desire). This well-established fact directly pertains to what is at stake in Sen’s claim that self-goal choice is violated in committed action. In a manner of speaking, one can transcend one’s own aims in all sorts of ways, for example by intending to do something on behalf of others, or for the benefit of others. Also, one can intend to influence other people so as to prompt them to act according to one’s own wishes. However, one cannot directly intend the other’s actions, because one can intend only what one takes oneself to be able to do (cf. Baier 1970). I can intend to make it the case that you close the door, but I cannot intend your closing the door (Stoutland 1997). In continental philosophy, this basic feature is sometimes called the ‘‘mineness’’ or ‘‘ownness’’ of intentionality.5 Just as one cannot die the death of others, even though in some cases, one can die for them, one cannot pursue the other’s goals without making these goals one’s own. This is an essential fact about our intentionality. Thus it seems that what Sen believes to be violated by committed action is nothing less than a basic trait of what makes an agent an agent—at least if we take intentionality as constitutive of agency, and if we take goals to be the conditions of satisfaction of intentions.6 Or, to put it negatively: no agency without self-goal choice. In this sense, the claim that the structure of committed action (or any action, for that matter) violates self-goal choice seems to be a contradictio in adjecto.

beyond self-goal choice 217 Should we therefore simply forget about Sen’s second claim, taking it as a condonable excess of his righteous fury at the annoyingly persistent small-minded idea of agency in economic theory? Should we just return to the first feature of Sen’s analysis of the structure of committed action, the wedge between choice and welfare, which is less controversial, and still an important contribution to the theory of rationality in action? Or is there any way to make sense of the idea of a violation of self-goal choice by a committed agent? I suggest that we start by taking a closer look. In ‘‘Rational Fools,’’ Sen already emphasized the role of group membership for committed action. In ‘‘Goals, Commitment, and Identity,’’ as well as in other papers, Sen further elaborates this idea. On a first line of thought, Sen introduces ‘‘as if ’’ goals to explain the violation of self-goal choice by committed action.7 However, Sen is well aware that ‘‘as if ’’ goals offer no more than a formal equivalent, which does not capture the real structure of the phenomenon.8 Just the fact that committed action can sometimes be accommodated in ‘‘as if ’’ objective functions (Sen 2002: 41), in itself, does not shed light on the structure of committed action. The question is: what do people actually do when their behavior violates self-goal choice? In addressing this question, Sen introduces the concept of identity. As Sen puts it, ‘‘the pursuit of private goals may well be compromised by the consideration of the goals of others in the group with whom the person has some sense of identity’’ (Sen 2002: 215). It is, as he says, this ‘‘sense of identity’’ which ‘‘partly disconnects a person’s choice of actions from the pursuit of self-goal’’ (ibid.: 216). One might wonder what this ‘‘sense of identity’’ which drives a wedge between choice and self-goal is. In some passages, Sen seems to suggest a reading according to which the agent identifies himself so thoroughly with another person that the goals he pursues are no longer his own goals. The assumption that one can pursue other people’s goals without making them one’s own, however, flies in the face of our understanding of agency as analyzed above; taken in this sense, identification amounts to some paradoxical self-elimination. If the object of identification is taken to be some other person, any attempt to go beyond self-goal choice by means of identification amounts to nothing but the futile attempt to stop being oneself by taking on somebody else’s identity (cf. Charlie Kaufman’s Being John Malkovich for a vivid illustration). In this self-eliminative sense, identification with others is simply self-defeating.

218


The harder one tries to get rid of one’s own identity by identifying with somebody else, the more it becomes apparent that it is all about oneself trying to be another, and not another. In this sense, identification is self-defeating, because the very act of identification presupposes the very difference in identity that the agent in question tries to eliminate. On this line, there is no way to go beyond self-goal choice, because no matter how far one goes in making somebody else’s goals one’s own, it is still invariably one’s own goals that one pursues. However, this self-eliminative sense is not the only reading of the role of identification that Sen suggests. The predominant line is quite a different one: here, identification is not with others, taken as single agents. It is not a matter of any I–Thou relation, but between agents and groups—a matter of the I–We relation, as it were. In this sense, identification is not self-eliminating (which would be self-defeating). Rather, it is self-contextualizing. This kind of identification is not about trying to be somebody else with whom one identifies, but simply about not just being oneself, but one of us. This second concept of identification is the one put forth in Sen’s talk on ‘‘Reason before Identity,’’ where Sen develops an understanding of belonging that avoids the pitfalls of the communitarian critique of liberalism (Sen 1999; cf. also Sen 2004). On this second line, the claim that committed action violates self-goal choice takes on a different meaning. If identification with a group lies at the heart of the structure of commitment, an agent does not have to perform the paradoxical task of choosing someone else’s goal without making it his own in order to qualify as truly committed. In a sense, committed action is neither about one’s own goals, nor about anybody else’s goals. The point seems to be that in committed action, the goals in question are not individual goals, but shared goals. If the scandal of the self-goal choice assumption is that it implies too narrow a conception of goals, this is not because it excludes some form of altruism, but because it wrongfully limits goals to individual goals, thereby banning shared goals from the picture. What is needed in order to correct the shortcomings of the self-goal choice assumption is not an account of other-goal choice, but an account of the pursuit of shared goals, or of collective agency.9 As Sen puts it: ‘‘ ‘We’ demand things; ‘our’ actions reflect ‘our’ concerns; ‘we’ protest at injustice done to ‘us’ ’’ (Sen 2002: 215).

beyond self-goal choice 219

3. Shared goals This ‘‘self-contextualizing’’ notion of identification, however, has its own problems. How does the claim that collective agency violates own-goal choice square with the earlier thesis that self-goal choice is a defining feature of any kind of agency? If the earlier considerations on the status of goals in intentional behavior are correct, it seems that departing from self-goal choice amounts to endorsing one of the following two equally repellent alternatives. Either it requires denying that the individuals taking part in collective actions are proper agents, or it requires making a category mistake of the most basic Rylean type. The first of these alternatives seems implausible because whatever one takes collective action to be, it is clear that the individuals involved in shared activity are agents, not just, say, organs in some collective body. There is no reason to doubt that it is legitimate to demand that an account of collective agency be consistent with the notion that individuals do act when they act together. If one accepts this assumption, however, it appears that the only reason left to believe that collective agency violates self-goal choice is a category mistake. For the only alternative then seems to be to understand collective action as something different from the actions of the participating individuals. This, however, is in direct conflict with the predominant view, according to which it is not only the case that individuals act when they act together, but that the actions of the participating individuals are what collective agency is. There is no collective agent, no macro-subject, that acts in addition to the participating individuals, when individuals act jointly. To adapt the Rylean example to the given case, it seems that whoever contests this makes a mistake similar to the spectator watching some soccer game for 90 minutes, before saying ‘‘I have had enough of those twenty-two people running about on the field in some coordinated way. I just wonder when, finally, the teams will start playing’’—because individuals, running about on the field in some coordinated way is what teamplay is. Therefore, it appears that collective agency does not violate self-goal choice: all that is chosen in collective action is individual goals, namely the goal to contribute to the attainment of some shared aim. As it was put in an earlier contribution to the theory of shared goals: if a team has goal x, then each individual member has goal x (cf. Levesque and Cohen 1991)—or,

220


more precisely, some contributive goal y—which conforms to self-goal choice. Thus it seems that any attempt to depart from self-goal choice faces a dilemma. It amounts to ending up either in some massively collectivist conception, which flies in the face of even our most basic understanding of intentional autonomy (cf. Pettit 1996: 117 ff.), or in a conception that is based on a simple category mistake. Since both alternatives appear equally unacceptable, it seems that we should not depart from self-goal choice. I think, however, that the argument concerning the second alternative is not sound. In the following, I shall argue that even though the participants act when they act jointly, there is no category mistake in assuming that joint action violates self-goal choice. The thesis I would like to put forward is not that agents violate self-goal choice when they act together (this claim would lead directly into some of the nonsense we have encountered before). Rather, my claim is that the self-goals which individuals choose when they act together cannot be adequately represented within an account which takes all goals to be self-goals, because these self-goals presuppose shared goals. The argument is the one put forward by those advocating a nonreductivist reading of collective agency. Participative intentions and goals are, to use a term coined by Wilfrid Sellars, ‘‘we-derivative’’ (Sellars 1980: 99). If we play a duet together, my aim is not just to play my part while you play yours (such cases may occur, but they do not constitute genuine cases of shared agency). Instead, it is as a part of our shared activity that you and I do what we do individually when we play together (cf. Searle 1990). In order to account for our contributive self-goal choices, an observer needs to understand that what she or he observes is something the agents are doing together (for more arguments for the non-reductivist view see Schmid 2003). Some current accounts of shared agency and collective intentionality are accused of circularity, because their analysis of what individuals do when they act together presupposes what should be explained. From a non-reductivist perspective such as the one I just have taken, this is not surprising, but simply reflects the ontological structure of participative intentions or participative goals. In the sense of the ‘‘we-derivativeness’’ of participatory intentions and goals, togetherness is irreducible; or, to use Sen’s term of the ‘‘privateness’’ of goals: shared goals are not simply a

beyond self-goal choice 221 combination of private goals. There is a difference between goals that individuals just somehow happen to have in common, on the one hand, and goals which individuals have individually only because they have this goal in common, on the other.10 An account of agency that is unable to see beyond the limits of self-goal choice cannot account for the latter kind of goals, i.e., the case of genuinely shared agency. Paradoxically, the self-goal choice assumption renders action theory blind for one special, but important kind of self-goal choice, namely, contributive self-goal choice. There is yet another argument for a non-reductivist account of collective agency that I would like to mention, which brings me into some tension with Raimo Tuomela’s account of collective agency. As Annette Baier (1997a: 26; 1997b: 37) has pointed out, there are some rare cases in which individuals fail to form an appropriate we-derivative individual intention, even though, in a sense, they still can be said to share an intention (for a differing view cf. Tuomela 1991: 271 ff.; 1995: 135 ff.). Take the case of some spontaneous and transitory collective action, such as the one of a couple of passers-by joining their forces in order to push a car. As a participant in that activity, I might suddenly feel estranged from my role and lack the aim to provide my contribution, even though I might still think of our goal to push the car as our goal, and not merely as their, the other people’s, goal. In such cases, it seems to make perfect sense to speak of collective goals or collective intentions in a sense that does not refer to corresponding individual contributive goals or intentions. An account that is based on self-goal choice seems to be blind for such cases. Admittedly, these are rare and perhaps even pathological cases. But in the light of such deviant cases, normality reveals some of its basic traits. If I think of some goal as our goal, I can be expected to have a corresponding individual contributive goal, or some other kind of pro-attitude. In the absence of overriding reasons, I should choose to do my part. The relation between shared goals and individual contributive goals (i.e., between shared goals and self-goal choice) is a normative one. This, however, points against a constitutive relation between individual contributions and shared goals of the kind at work in reductivist accounts of collective agency. Normativity entails contingency. That I should choose my contributive goal in our collective project presupposes the possibility that I decide not to contribute to the attainment of what is our goal. The possibility (perhaps more than the fact) of dissidence, as well as of other kinds of failures to do one’s part, is an essential

222


part of shared agency. It is what makes the relation between shared goals and individual choices normative. And again, an account that is limited to self-goal choice seems to be blind to the fact that some self-goal choices normatively depend on shared goals. In short, the self-goal choice assumption is incompatible with a nonreductivist account of collective agency.11

4. Commitment: a third account As was pointed out early on in the collective intentionality debate, shared intentions or projects provide us with a standpoint from which we critically measure and evaluate our individual plans and aims (Rosenberg 1980: 159). As normative sources, shared intentions, aims, goals, and projects provide us with reasons for individual action. This brings me back to the initial point concerning the controversy between internalist and externalist accounts of commitment. For these special reasons, which are based in shared intentions and projects (in short: shared desires), have an interesting status. They are neither internal nor external reasons. In some sense, they are independent of us as single individuals, or, more precisely, they transcend our ‘‘subjective motivational set’’—that is why they can serve us as a critical standpoint for our self-evaluation. In this sense, reasons that accrue from shared desires are not internal. On the other hand, these reasons are not external either. They are not disconnected from the sphere of ‘‘desires’’ (in Williams’ formal sense of the word). If and insofar as the reasons for committed action are ultimately based in shared desires, the distinction between internal and external reasons does not apply. Because shared desires are neither internal to one’s motivational set, nor external. Instead, they transcend one’s subjective motivational set. An account of the structure of commitment that has neither ‘‘subjective motivations’’ nor ‘‘metaphysically basic’’ reasons, but shared desires playing the leading part in committed action, seems to avoid the two problems I have mentioned at the beginning of the paper. It avoids both the ‘‘Humean’’ inability to conceive of the agent’s power to transcend their individual desires, and the old ‘‘Kantian’’ problem of first throwing motivation out with some great gesture of depreciation and then having to beg it in again through the back door. In the rich literature on Williams’ internalism about practical reason, it seems that Martin Hollis’ view is closest to the one developed here. In

beyond self-goal choice 223 spite of his externalist bias, Hollis comes close to an account of shared desires, when he discusses the relation between ‘‘interest’’ and community (Hollis 1987). If we move from interest to shared desire, the problem with Williams’ internalism is not that it bases reasons in motivation. Instead, it is the way in which Williams conceives of human motivation. Not all our motives are part of our ‘‘subjective motivational set.’’ Some are intersubjective. I believe that this insight is part of what makes Sen’s invitation to look beyond the limits of self-goal choice so important.12 Notes

1. For another non-Humean account of practical rationality based on an analysis of the structure of commitment see Benn and Gaus (1986). 2. Or, to put it in Amy Peikoff’s words: ‘‘Rational action entails rational desire’’ (Peikoff 2003). 3. In a footnote on the relation between his own ‘‘external reference’’ approach and Williams’ internalism, Sen claims to be in line with Williams, because unlike Williams’ internalism, ‘‘external reference’’ externalism is about choice, not about persons (Sen 1995: 30). 4. The example is by courtesy of Peter Vallentyne, to whom I am grateful for pointing out the problem. 5. ‘‘Mineness’’ translates such terms as Martin Heidegger’s ‘‘Jemeinigkeit’’ (Heidegger 1996 [1927]). 6. The last clause is of special importance. Clearly, there is no problem involved in pursuing other people’s goals where goals are simply desired states of affairs, rather than conditions of satisfaction of intentions. Concerning the decision for an intentionality related concept of goals, see the above remarks. 7. ‘‘Consider a pair of individuals whose real goals are those as in the Prisoner’s Dilemma, but whose actual behavior violates goalpriority (and self-goal choice). The ‘revealed preference’ relation of their respective choice functions may place the cooperative outcome on top, that is, they may behave ‘as if ’ they would favor that particular outcome most of all’’ (Sen 2002: 217). 8. In ‘‘Maximization and the Act of Choice,’’ Sen states with regard to the phenomenon of Japanese employees working themselves literally to death: ‘‘The as if preference works well enough formally, but

224 hans bernhard schmid the sociology of the phenomenon calls for something more than the establishment of formal equivalences’’ (Sen 2002: 191). 9. For an analysis of the link between Sen’s concept of identification and the demand for a robust concept of collective agency, see Anderson (2001). In her reflections on collective agency, Carol Rovane clearly distinguishes projection into other individuals’ points of view from orientation on common ends: ‘‘these activities do not require that persons project themselves all the way into another person’s own rational point of view so as to take up that person’s perspective. These activities require rather that persons project themselves into a rational space that is generated by the ends which they hold in common ... When persons project themselves into this common rational space, they can reason and act together from the perspective of their common ends’’ (Rovane 1998: 138). 10. Jay Rosenberg calls the former type of ends ‘‘common’’ and the latter ‘‘communal.’’ ‘‘A communal end ... will be one which is collective without being conjunctive. It will be an end which is mine and hers and his by virtue of the fact that it is ours and that each of us represents himself/herself as one of us. It will, in other words, be a genuinely plural end, attributable to all of us collectively and therefore univocally to each of us severally and to all of us conjunctively’’ (Rosenberg 1980: 160). 11. Thus I assume that the self-goal choice assumption is ultimately equivalent to what Margaret Gilbert (1989: 418–25) criticizes under the label ‘‘singularism.’’ 12. I am grateful to the participants of the Workshop on Rationality and Commitment, held at the University of St Gallen on May 13–15, 2004 (especially to Raimo Tuomela and Philip Pettit), to Peter Vallentyne, and to the two referees for Oxford University Press for their criticism and comments on this paper.

References Anderson, E. 2001. ‘‘Unstrapping the Straitjacket of ‘Preference’: A Comment on Amartya Sen’s Contributions to Philosophy and Economics.’’ Economics and Philosophy 17: 21–38. Baier, A. 1970. ‘‘Act and Intent.’’ Journal of Philosophy 67: 648–58. Baier, A. 1997a. The Commons of the Mind. Chicago: Open Court.

beyond self-goal choice 225 Baier, A. 1997b. ‘‘Doing Things with Others: The Mental Commons.’’ In L. Alanen, S. Heinämaa, and Th. Wallgren (eds.), Commonality and Particularity in Ethics. London: Macmillan, pp. 15–44. Benn, S. I. and G. F. Gaus. 1986. ‘‘Practical Rationality and Commitment.’’ American Philosophical Quarterly 23: 255–66. Brandom, R. B. 2000. Articulating Reasons. An Introduction to Inferentialism. Cambridge, Mass.: Harvard University Press. Gilbert, M. 1989. On Social Facts. London: Routledge. Heidegger, M. 1996 [1927]. Being and Time. A Translation of Sein und Zeit. Translated by Joan Stambaugh. Albany: State University of New York Press. Hollis, M. 1987. ‘‘External and Internal Reasons.’’ In Hollis, The Cunning of Reason. Cambridge: Cambridge University Press, pp. 74–94. Levesque, H. J. and P. R. Cohen. 1991. ‘‘Teamwork.’’ Nous 25: 487–512. McNaughton, D. and P. Rawling. 2004. ‘‘Duty, Rationality, and Practical Reasons.’’ In A. R. Mele and P. Rawling (eds.), The Oxford Handbook of Rationality. Oxford: Oxford University Press, pp. 110–31. Peikoff, A. 2003. ‘‘Rational Action Entails Rational Desire. A Critical Review of Searle’s Rationality in Action.’’ Philosophical Explorations 4: 124–38. Pettit, P. 1996. The Common Mind. An Essay on Psychology, Society, and Politics. Oxford: Oxford University Press. Rosenberg, J. 1980. One World and Our Knowledge of It. Dordrecht: D. Reidel. Rovane, C. 1998. The Bounds of Agency. An Essay in Revisionary Metaphysics. Princeton: Princeton University Press. Schmid, H. B. 2003. ‘‘Can Brains in Vats Think as a Team?’’ Philosophical Explorations 6: 201–18. Searle, J. R. 1983. Intentionality. An Essay in the Philosophy of Mind. Cambridge: Cambridge University Press. Searle, J. R. 1990. ‘‘Collective Intentions and Actions.’’ In P. Cohen, M. Morgan, and M. Pollack (eds.), Intentions in Communication. Cambridge, Mass.: MIT Press, pp. 401–15. Searle, J. R. 2001. Rationality in Action. Cambridge, Mass.: MIT Press. Sellars, W. 1980. ‘‘On Reasoning about Values.’’ American Philosophical Quarterly 17: 81–101. Sen, A. K. 1977. ‘‘Rational Fools: A Critique of the Behavioral Foundations of Economic Theory.’’ Philosophy and Public Affairs 6/4: 317–44. Sen, A. K. 1985. ‘‘Goals, Commitment, and Identity.’’ Journal of Law, Economics, and Organization 1/2: 341–55. Reprinted in A. K. Sen, 2002, Rationality and Freedom. Cambridge, Mass.: Harvard Universtiy Press, pp. 206–25. Sen, A. K. 1987. On Ethics and Economics. Oxford: Blackwell.

226


Sen, A. K. 1995. ‘‘Is the Idea of Purely Internal Consistency of Choice Bizarre?’’ In J. E. J. Altham and J. Harrison (eds.), World, Mind, and Ethics. Essays on the Ethical Philosophy of Bernard Williams. Cambridge: Cambridge University Press, pp. 19–31. Sen, A. K. 1999. Reason Before Identity. The Romanes Lecture for 1998. Oxford: Oxford University Press. Sen, A. K. 2002. Rationality and Freedom. Cambridge, Mass.: Harvard University Press. Sen, A. K. 2004. ‘‘Social Identity.’’ Revue de philosophie ećonomique 9: 7–27. Stoutland, F. 1997. ‘‘Why are Philosophers of Action so Anti-Social?’’ In L. Alanen, S. Heinämaa, and Th. Wallgren (eds.), Commonality and Particularity in Ethics. London: Macmillan, pp. 45–74. Tuomela, R. 1991. ‘‘We will Do It: An Analysis of Group-Intentions.’’ Philosophy and Phenomenological Research 51: 249–77. Tuomela, R. 1995. The Importance of Us. A Study of Basic Social Notions. Stanford: Stanford University Press. Weirich, P. 2004. ‘‘Economic Rationality.’’ In A. R. Mele and P. Rawling (eds.), The Oxford Handbook of Rationality. Oxford: Oxford University Press, pp. 380–98. Williams, B. 1979. ‘‘Internal and External Reasons.’’ In R. Harrison (ed.), Rational Action. Studies in Philosophy and Social Science. Cambridge: Cambridge University Press, pp. 17–28.

10 Cooperation and the We-Perspective RA I M O T UO M E L A

1. Introduction The ability to cooperate is one of the key elements in the success of the human species and it can even be claimed that this ability is an innate property of humans, because they have evolved in a group context.1 Cooperation will easily come about in situations in which it is intrinsically rewarding and does not involve an incentive to freeride. Simple coordination situations are of this kind. In contrast, there are collective action dilemma cases in which collective and individual rationality conflict, and rationally achieving cooperation will be notoriously difficult in such cases. In them, cooperation will be costly and defection will give a higher payoff. Nevertheless, as experimental results and common-sense knowledge show, people tend to cooperate much more both in single-shot and long-term situations than most available accounts of rational cooperation, such as that given by standard game theory, would predict. This kind of cooperation need not be based on kin or direct reciprocity, for cooperation also takes place with total strangers. I have elsewhere advocated a ‘‘we-ness’’ approach to explain cooperation.2 According to it, people often act as group members, respecting the cooperative requirements of the group, and this results in cooperation that is for the benefit of their group (‘‘we’’ for them). This view can also often explain why acting as a private person leads to defection in dilemma situations but not when acting as a group member. To explain cooperation in large societies a considerably extended ‘‘we’’ is needed, and in these cases achieving cooperation is more difficult. However, in this

228

raimo tuomela

paper I will not say much about why and when people in fact cooperate but rather focus on the conditions of rational cooperation in various social conditions, including the case with explicit reference to a ‘‘we-group.’’ In my approach, also groups of agents can instrumentally be taken to have beliefs, goals, and preferences enabling agents to make decisions based on these group attitudes. The ‘‘I-mode’’ perspective (the individualistic perspective) and the ‘‘we-mode’’ perspective (the group perspective), to speak in broad terms, are conceptually and psychologically different. Thus, we-mode attitudes are derivable from, and justifiable by, one’s group’s attitudes (e.g. I intend to participate in X because we intend to perform X together) whereas I-mode attitudes are not so based.3 We-mode and Imode concepts are not only different, but the we-mode is irreducible to the I-mode and these modes may be in conflict.4 Basically, the irreducibility follows from the fact that the we-mode depends on group reasons, on viewing a group’s attitudes or views to be one’s central reasons, while the Imode is based on private reasons. As I have argued elsewhere, the we-mode is needed, for example, to account for social groups and their properties, social practices, social institutions.5 The we-mode accordingly lies at the bottom of especially the ‘‘hermeneutic’’ dimension (the dimension of understanding), but neither can the explanation of what exists and happens in the social world be adequate without the use of we-mode concepts and the we-perspective. Both modes are needed to account for an agent’s psychology, namely, his mental states and processes. There are also differences concerning actions. Even so, in actual practice both perspectives often lead to the same actions, although the actions are based on respectively different reasons, namely, we-mode and I-mode reasons. In this paper I will argue for the importance of the we-mode on functional grounds in the case of rational cooperative action, especially action in a group context and show concretely that there are some interesting cases of rational activity that are best accounted for in we-mode terms. While cooperative order can be created either in the I-mode way or in the we-mode way, it will thus be argued that we-mode decision-making may lead to beneficial cooperation also in situations where individualistic, or I-mode, decisionmaking fails. In addition, cooperation may be more persistent, because we-mode cooperation is based on the group members’ collective commitment to cooperate. Both types of decision-making, or more generally

cooperation and the we-perspective 229 thinking and acting, are needed for capturing the essential features of social life. There are many ways to take into account the various social factors that influence cooperation. A person may be prosocial, e.g. altruistic, and cooperate on that basis with other people. This is not yet a group reason. On the other hand she may strive to maximize the benefit of a group in an action situation in some kind of egalitarian way. This can be regarded as a group reason of the kind I will call an I-mode ‘‘pro-group’’ reason. A stronger kind of group reason is one based on a person’s identification with the group and adopting a group perspective (‘‘we-perspective’’) and properly acting as a group member. This is a we-mode reason that intrinsically is a pro-group reason.6 Below I will consider both I-mode and we-mode reasons for cooperation and assume that we are dealing with a social group that has the participants of cooperation as its members. In general, the three central characteristics of the we-mode concerned with a reason, attitude, or action, here assumed to be a distributive attitude (namely, each participant is collectively committed or, as I will also say, we-mode committed), are as follows. The content in question must be (i) collectively accepted with (ii) collective commitment (iii) for the (benefit of the) group. Collective acceptance is reflexive: if, for instance, the group members collectively accept a goal that means that they accept the goal as their group’s goal.7 In this paper I mean by an agent’s being committed to an item (e.g. an action) simply that he has bound himself to that item or that he has become bound to this item (e.g. to an action, typically because of his intention to perform it). The basis of such, possibly non-reflective commitment can be psychological affection, and in the group-case ‘‘wefeeling’’ or feeling of solidarity. In reflective cases there may be more cognitive content. Thus, if a person has a goal, then, roughly speaking, he ‘‘technically’’ ought to do what it takes for him to reach the goal. In more demanding, properly normative cases there may be a social or moral obligation involved to back the commitment (cf. ‘‘I promised to do X, therefore I am committed to the promisee to doing X’’). Rationality and private versus collective commitment is considered throughout the paper. Here rationality is not only instrumental but takes into consideration also the rationality of the agents’ goals. For simplicity’s sake, expected utility maximization will typically be used as a criterion of rational decision-making. Conditions for rational cooperation are given

230

raimo tuomela

in both I-mode and we-mode cases. One central purpose of the present paper is to present concrete implementations of the we-perspective and the I-perspective and to compare them at least in some interesting cases.8 The main technical results are of the form ‘‘there are cases such that ...’’ and these claims are shown to be right by producing singular examples. The role of commitment to cooperation is central here. In the I-mode case only private commitment is involved but in we-mode case collective commitment is crucial. Section 2 discusses I-mode cooperation and we-mode cooperation, the two basic kinds of cooperation. Section 3 discusses group preferences. Section 4 considers a simple case of a proper group context in which the participants act as members of a social group—either in the I-mode sense or in the we-mode sense. There is also a brief concluding section on the broader questions related to the I-mode versus we-mode debate.

2. Two kinds of cooperation Cooperation will below be discussed primarily in terms of I-mode and we-mode goals, preferences, intentions and intention-based commitments, beliefs, and actions of the participants, assumed in sections 4 and 5 to be members of the same—externally and internally autonomous—social group. Maximization of expected utilities will mostly be used as a rational criterion of decision-making and intention formation. However, my approach is not intimately geared to this kind of maximization view, at least in some cases ‘‘satisficing’’ (or some other ‘‘bounded rationality’’ criterion) will be rational enough. I will take the participants’ preferences to be (rational) preferences about goals and the means actions leading to them. To illustrate, suppose an agent has two goals, G1 and G2 , in a certain situation. By preference I now mean, for instance, that he prefers a sequence of actions X1 , ... ,Xm leading to G1 (or the sequence X1 , ... ,Xm , G1 ) to another sequence of actions leading to G2 . In other words, he prefers sequences of actions ending with a goal state to another comparable sequence.9 When maximizing utility, goal-rationality will thus be included in addition to instrumental rationality (namely, mere comparison of means actions). For simplicity’s sake, I will assume that a quantitative representation of these preferences is available

cooperation and the we-perspective 231 and that some preliminary screening concerning the means actions for a goal has taken place with the result that we only need to consider one ‘‘best’’ action sequence connected to each goal. When a rational criterion of decision-making and intention formation (e.g. maximization of expected utility) has been applied, then an intention to act is formed and the agent will therefore be committed (at least subjectively bound) to performing the action. In I-mode decision-making the agents have beliefs and desires (or goals) and they aim at selecting the best action sequences, those with the largest (expected) utility. Although this mode of decision-making is self-interested in the sense that the agents base their action on their private goals or private utilities, it is not necessarily selfish. For example, an agent may have as a personal goal to satisfy a goal of another agent, and speaking in terms of utilities an agent’s utility function may be affected by another agent’s utility function. All proper cooperation is necessarily intentional. The participants accordingly cannot cooperate without the intention to cooperate. This applies both to I-mode and we-mode cooperation. Consider this account of I-mode cooperation:10 (CIM) A1 and A2 (intentionally) cooperate in the I-mode with respect to their (proximate or further) goals G1 and G2 if and only if (1) A1 and A2 have the goals G1 and G2 , respectively, as their intended private goals (namely, goals had as private persons in contrast to goals had as full-blown members of a social group); (2) they intentionally perform respective dependent means actions x1 and x2 believed by them to be (at least indirectly, via the other’s action) conducive to their respective goals so that each takes himself thereby to have adjusted his acting and goal to the other’s action and goal with the purported result that the other’s achievement of his (possibly adjusted) goal is furthered and that, by the other’s analogous acting, also the achievement of one’s own (possibly adjusted) goal is furthered. To get a notion of (functionally) rational I-mode cooperation from the above conceptually minimal notion, we require at least the truth of the beliefs in (2) and add two clauses saying that the information in clauses (1)

232

raimo tuomela

and (2) must be mutually believed (clause (3)) such that in addition the mutual belief is not idle (clause (4)): (3) A1 and A2 mutually believe that (1), (2). (4) (2) in part because of (3). Goals can be dependent or made dependent in a variety of ways, and I can here only make some remarks concerning simple cases. First we note that G1 and G2 may—e.g. in cases of spontaneous cooperation with no further individual goals—be just the proximate goals conceptually inherent in intentional actions, namely, the so-called result-events of x1 and x2 , respectively. As to the action dependence clause (2), we already noted that action dependence might be due only to goal dependence. However, for some situations sharper and stronger formulations of action dependence may be used. Thus one may require optimal satisfaction of a participant’s goal Gi (i = 1, 2), entailing that he has to choose his action xi so that it is optimal for Gi in the sense of a suitable criterion (be it maximization of expected utility or something weaker). A still stronger dependence requirement is that, relative to goal satisfaction, x1 can be (optimally) performed if and only if x2 is performed, and conversely. Here the goals may be only parallel (be they the same or different) and satisfied by different event (or state) tokens or they may be the same in the stronger sense of being satisfiable only by the same token, ‘‘divided’’ by the participants through their actions. In clause (3) of (CIM) the agents are required to have a mutual belief concerning the relevant pieces of information. Here is an argument for clauses (3) and (4). Suppose that the participants’ goals and actions are dependent in the sense of clause (2), be the dependence antecedent or voluntarily created on the spot. Then the participants will rationally need the other’s performance at least for an optimal result (namely, goal achievement) for themselves. This rationally requires that each agent Ai , i = 1, 2, believes (and, in addition, as will be seen in section 4, trusts) that the other one will adjust his goal and perform his means action as required by clause (2).11 In addition, A1 must of course believe that A2 believes that A1 will similarly participate, for otherwise A2 would not have good reason to participate; and similarly for A2 . Furthermore, they must lack the belief that the negation of the relevant higher-order belief contents are true.12 To

cooperation and the we-perspective 233 justify clause (4), we note that the resulting mutual belief cannot be idle: it must serve as the participants’ partial reason for participation. In I-mode cooperation there need be no collective goal involved, but the agents have their private goal-involving preferences which are compatible in the sense that satisfying the goal of one agent does not make it impossible to satisfy the goal of the other. What is central here is that they intend to act in order to achieve their private goals, while at the same time adjusting their goals and means actions to the others’ goals and means actions so as to facilitate both their own and others’ goal achievement. The agents mutually believe that they are all acting similarly. All this enables coordination of action. This kind of cooperation where a shared collective goal is not required and where the participants act for an I-mode reason is I-mode cooperation. Standard game theory deals with I-mode cooperation. In the full, we-mode sense, cooperation is acting together as members of a group to achieve a shared collective goal.13 This is acting for a wemode reason. We-mode cooperation requires that agents do not try to maximize their own benefit (preferences involving goals) but contribute to the joint effort, which usually involves a cost to the agent. Typical cases are collective action dilemmas where there are two distinct and incompatible actions C and D such that D is the individually rational action and C is the collectively rational action. Individual rationality is usually taken to involve utility-maximization whereas collective rationality requires Pareto preference or efficiency or maximization of group’s benefit (in I-mode treatments often regarded as the sum of individual utilities). In this paper, I will attribute preferences to groups; collective rationality then means maximizing group utility (assumed to have been derived from the group’s preferences).14 In we-mode decision-making, in contrast to the I-mode case, an agent is supposed to view himself as part of a social group or collective and acts as a full-fledged member of the group. The group here can even be a temporary task group consisting of some agents who face the task of carrying a heavy table upstairs. Having formed a joint intention to do it, they already form a group in a minimal we-mode sense. Given the joint intention, each group member is supposed to infer his own part action that contributes to that joint goal. Then he, typically together with other group members, acts so as to contribute to the group’s goal (intention content). It should also be required that it is not only the particular group

234

raimo tuomela

member’s belief that the goal should be pursued by the group members but the goal has been collectively accepted by the group members as the group’s goal. Typically, adoption of the group’s point of view will reduce conflict between the members. Thus, when a group faces a PD situation it tends to transform it into a milder, coordination type of problem (e.g. into an Assurance Game). This is understandable also from a social psychological point of view. The ingroup will fail to be functional if there are strong conflicts between the members. After all, they are all supposed to be promoting the ‘‘ethos’’ of the group, and that must involve a considerable amount of harmony. We-mode talk requires having joint attitudes as a group and acting as a group. Thus the concept of group is referred to and relied upon here. However, this does not entail that groups must exist as entities. It suffices that they have ‘‘intentional inexistence’’ (as Brentano put it), namely, occur in intentional mental contents and thus lead people to act in relevant ways vis-à-vis each other and the rest of their environment. What is ontically required to exist is the ‘‘jointness level,’’ namely, joint actions, joint intentions, joint preferences, mutual beliefs, etc. and the individual group members’ we-mode states (weintentions, etc.) of which the joint states are ontically made up. Group members’ functioning in the we-mode requires that goals and standards (etc.) to be attributable to groups, although they in my account do not literally, in an ontic sense, have them: groups are not persons in a literal sense.15 Here is my account of we-mode cooperation as a joint or collective action in the two-person case:16 (CWM) A1 and A2 (successfully) cooperate with each other in the we-mode (namely intentionally act together as a group) in bringing about goal G if and only if (1) G is a collective goal type, namely, an ‘‘achievement-whole’’ the achievement of which can be divided—either ex ante actu or ex post actu—into A1 ’s and A2 ’s parts; (2) A1 and A2 jointly intend to achieve G by acting together, and they achieve G together in accordance with and partly because of this joint intention of theirs to achieve G together.

cooperation and the we-perspective 235 To go from the above conceptually minimal notion to a notion of rational we-mode cooperation, we add two clauses saying that the information in clauses (1) and (2) must be mutually believed (clause (3)) such that the mutual belief is not idle (clause (4)): (3) A1 and A2 mutually believe that (1), (2); (4) (2) in part because of (3). I will not here discuss (CWM) except for pointing out that the joint intention of clause (2) is the most central element here. A joint intention consists of the participants’ shared we-intentions to cooperate in the meant way. A group’s intention can basically be accounted for in terms of joint intention of its members.17 Clauses (3) and (4) can be justified as in the I-mode case, mutatis mutandis, and I will not here comment on this matter.18 Problematic cases occur when the joint effort increases the utility of all of the agents and freeriding agents gain more because they get the benefit without paying the cost. As a typical example, consider a simple two-person two-choice case of a Prisoner’s Dilemma with the familiar choice alternatives C and D, where the row player’s preference ranking is DC, CC, DD, CD and the column player’s symmetric ranking is CD, CC, DD, DC. The agent may choose either C or D when he acts for an I-mode reason. Thus, considering the single-shot case, if the agent—here the row player, ‘‘I’’—thinks strategically and intends to maximize his value or utility he can reason thus: I prefer the joint outcome DC to all the other joint outcomes; however, I realize that if you are reasoning similarly and planning to go for CD, we will end up in DD. Still, wanting to avoid the worst outcome CD, I cannot rationally aim at the Pareto preferred CC (namely, CC is better DD both for you and me). So I choose D and rationally expect the equilibrium outcome DD to result in the single-shot case. In this case the agent’s reason for choosing D is to secure at least the third best alternative given his beliefs of the nature of the game and the other player’s rationality. Thus, as long as change of the game structure is not allowed, in a single-shot PD mutual defection is the rational outcome, and this result holds independently of how egoistic or altruistic the participants’ preferences and utilities are. The agents would do better both individually and collectively by both choosing C, and they can rationally jointly arrive at the cooperative joint outcome by acting as a group. In that case a switch

236

raimo tuomela

from single to collective agency takes place, and an agent acting as a group member will choose C, assuming that the group is a rational one choosing C over D. Experimental results and anecdotal evidence indicate that normal socialized agents do have a disposition to cooperate, to act as group members, even in single-shot collective action dilemmas.19

3. Group preferences I will assume below that groups can be treated as social systems, as relational systems consisting of group members, their relevant interrelations and dependencies, as well as (possibly) some artifacts. Roughly, social groups are formed by agents who have similar goals or interests (concerning some topic) and who are mutually believed to be members of the group. Their central interests and goals related to the group’s ‘‘topics of concern’’ serve to constitute what can be called the group’s ‘‘ethos.’’ The ethos consists of the group’s central, constitutive goals, standards, and norms collectively accepted by the group members. Its mere existence gives a certain coherence and unity to the group. Thus cooperation is more likely to emerge in social groups at least with respect to matters concerning the ethos-related topics because the agents’ preferences are likely to be positively correlated.20 When a group member acts as a group member he must respect the ethos (E) of the group, and this involves that the participants in question act for the group and, at least in full-blown cases, purport to satisfy (and, to some extent, uphold) the ethos, being collectively committed to it. There can also be I-mode functioning as a group member based on the private adoption of the same E as one’s private ‘‘goal’’ and being privately committed to it.21 Acting as a group member in the we-mode sense amounts to cooperation with the other group members towards the satisfaction and, perhaps, maintenance of the ethos.22 When discussing we-mode preferences underlying group intentions and joint intentions we need a notion of group preference in the sense of a preference attributed to a social group, g. This is because individual group members’ we-mode preferences conceptually derive from the preferences of the group (from what they prefer as a group) and must in general be ethos-respecting. We can say that the ethos of g will generally determine

cooperation and the we-perspective 237 how group-relevant states in the world are to be ordered in terms of preference. This preference ranking is complete only with respect to the topics of concern of g. It should be kept in mind that an individual can have both a preference ranking as a private person (the I-mode case) and a preference ranking as a group member and based on the group’s preferences (the we-mode case). Arrow’s impossibility result tells us that in many cases aggregation cannot in the general, unrestricted case rationally yield group preferences. However, while Arrow’s and other similar impossibility results are relevant to aggregative (thus I-mode) groups, the situation is the opposite in the case of organized groups capable of action. They amount to we-mode groups. Group preferences are artifacts analogous to other group attitudes in the we-mode case and are thus based on collective acceptance (not necessarily normative acceptance) by the group members qua group members. Such acceptance may be preceded by group discussion, negotiation, bargaining, etc.23 We may also speak of group utilities qua the group’s utilities (that must be reflexively collectively accepted as the group’s utilities). They can be based on group preferences analogously with how individual utilities are represented on the basis of an individual’s preferences and thus come to represent the (operative) members’ collective or joint preferences. A rational group will choose as its collective goal the outcome that maximizes the group utility, and the group members can then use the utility matrix to infer their required part actions (action sequences). In a two-person Prisoner’s Dilemma the best joint outcome will not rationally be DD but typically the Pareto-preferred CC. In general, a group’s preferences (namely, what the group members have collectively formed as the group’s preferences) can be attributed to group members when they are functioning as group members. Thus if the group has G as its intended goal, its members will be disposed to accept and endorse ‘‘We intend to achieve G.’’ This kind of distribution principle seems valid for all group preferences. Thus we can discuss group preferences in our PD in terms of simple stylized practical inferences as follows:24 (1) We intend to achieve collective goal G. (2) We mutually believe that to achieve G we must bring about CC by our collective action. Therefore:

238

raimo tuomela

(3) We intend to bring about CC by our collective action. Each group member is then normally assumed to reason in the following manner: (a) We intend to bring about CC by our collective action. Therefore: (b) I intend to perform my part of our bringing about CC by our collective action. Note that here even in clause (b) we are dealing with a we-mode intention, because the participation intention that it concerns is conceptually dependent on the we-mode joint intention expressed by (a). Group members are assumed to be collectively committed to make (perhaps only implicitly) such practical inferences when appropriate. More generally, when the group has accepted a view for the group the members are assumed to be able to acquire mutual knowledge of this acceptance and thus to be able to act on the belief content in question, when needed. (This is of course a nontrivial assumption, and the group’s monitoring that its members indeed appropriately participate may be a hard task.) Collective acceptance also in general creates collective commitment for the group members to act toward the accepted goal or goals. Especially this can be assumed to be the case when the group is one that is voluntarily entered, because when signing up for group membership one is thereby obligated to act as a group member and to be we-mode committed to the group’s ethos. This we-mode collective commitment involves that the agents as group members are also committed to other group members contributing to the goal achievement. In addition to collective acceptance and collective commitment, we-mode also requires ‘‘forgroupness’’ which entails that the collectively accepted goal is taken to be for the use of the group and to further the group’s basic purposes (its ethos).25 Consequently, it can be argued that we-mode cooperation will in general be more persistent than I-mode cooperation. There are two basic ways to act while taking other agents into account. One can operate on the basis of one’s other-regarding I-mode goals, as seen in section 2, or on the basis of a full-blown collective goal. These considerations can sometimes be combined. The two kinds of preferences

cooperation and the we-perspective 239 under discussion can also be discussed in quantitative terms as utilities. In self-explaining symbols, an agent Aj will have I-mode utilities of the kind uij (X) and we-mode utilities of the kind ugj (X) concerning an action-goal sequence X, and these utilities often are different and not reducible to each other. (We can here alternatively use ugj (X) for Aj ’s quantitative I-mode evaluation of g’s preference, which would mean that both utilities here would be I-mode utilities.) Rational action can be based on one of these or on some kind of optimal way of taking both kinds of utilities into account (cf. below). An individual’s we-mode and I-mode utilities need not be closely connected, e.g. a group may even prefer something that none of its members prefer in the I-mode. (Of course, the closer these two kinds of utilities are the more incentive there is for cooperation in the long run.) Let us suppose that X is an action in a group context but not one which our agent is obligated to perform when acting as a group member. In such a case a linear combination may give a feasible way to define a combined resultant utility that the agent will maximize or satisfice when deciding which action X to rationally perform:26 utj (X) = wi uij (X) + wg ugj (X) Here uij represents I-mode utility (utility from the individual’s point of view) and ugj we-mode utility (or in some cases pro-group I-mode utility, and in any case utility from the group’s perspective but as evaluated by Aj ). Intersubjective differences can be allowed not only with respect to individual reasons but also with respect to (evaluated) group reasons. To consider some special cases, when wi = 0 and wg = 1 unconditional pro-group cooperation is entailed, and when wi = 1 and wg = 0 we get action on mere private preferences or goals (which may be other-regarding, of course). When the unconditional pro-group condition holds in the case of all members, they share and are committed to the goal G (here: the ethos) and can be said to have at least a non-normative agreement to cooperate (namely, to act together toward G). However, a member may take group reasons into account in a weaker sense, while also respecting individual reasons. This may happen, for instance, if he has not fully committed himself to a collective goal. Basically, the weights are determined by the interaction situation and the individual’s dispositions to value individual and group reasons. They are not usually up to the individual to strategically choose.

240

raimo tuomela

In a PD situation, if the group (dyad) has selected CC as its goal (or takes it to be entailed by its goal), then wi = 0 for both players, and thus the expected utility of C will exceed that of D for all probability assignments, and we accordingly get a kind of solution to a collective action dilemma. In some other cases, however, the above kind linear combination account does not work well, if at all. These cases include some of those where an agent is acting as a group member in a full sense.

4. Cooperation in a group context To make our discussion more concrete, I will in this section discuss a simple example related to the acquisition and provision of a collective good, E, in a group, g. Assume first that E is an ‘‘indivisible’’ and non-excludable collective good, namely, a good whose use or consumption does not reduce the amount available to others and which is made available to all the members of the collective in question. I will also assume that E is a step good and that there are only two participants with two possible choices (C and D). Furthermore, I will begin by assuming that E can be produced even alone and that there is a fixed total collective cost, the individual cost of producing it alone being higher than when it is produced together. Let u be the gross reward (utility) that each participant receives when E is produced. Utility u might include a component which stands for the utility of being a group member, reflecting both the ethos of the group and also the sociality aspect of being and acting together with the other members. I will assume that utilities are intersubjective and that objectively existing states of affairs (or processes) are being evaluated here. In my technical treatment below I will denote the increase in ‘‘social capital’’ (‘‘jointness’’ factors and interaction gains) due to collective action by e. For the purpose of illustration we assume that j = i + e − c/2. Thus, when both participants contribute they will each receive the payoff u + i + e − c/2, and this is the same as u + j. Consider now the possibility that one of the agents defects. Supposing that E can be brought about alone and that the other participant’s contribution here amounts to bringing about E alone, the freerider will then gain a certain amount f by his defection, namely, from switching from C to D. Alternatively, f can be regarded as a freerider effect or a freerider

cooperation and the we-perspective 241 incentive.27 What happens when A contributes while B defects? Then A has to produce E alone, and that may of course be much more costly for him than performing it jointly with B. The sole contributor has to pay cost c rather than c/2, but A may lose more since he may also psychologically and socially lose something from the very fact that he (correctly believes he) is being cheated or is a ‘‘sucker.’’ I will denote by s the total loss (namely, the extra effort plus the sucker effect) accruing to A from his being the sole contributor. In principle, factor s can also have a positive value in that the sole contributor may view herself as a morally good person—or something of the kind. There might also be other higher-order effects while A’s utility may be affected by his thought that he is being cheated by B, this fact may in turn affect B’s utility, and so on. Typically there is a social norm barring freeriding.28 Especially, any full-blown collective goal, due to its being based on collective acceptance, can be taken to presuppose that the goal-holders socially (‘‘group-socially’’) ought to contribute. Informed and conforming participants accordingly will think that they ought to participate in the performance of X. If they do not, they face social sanctions (disapproval and even being thrown out of the group) and often feel guilt or shame. Thus, there is a negative normative effect from defection, given that there was an obligation or at least a normative expectation to contribute. I will denote the effect of social sanctions (and the related psychological consequences of shame as well as guilt) by m. In the case of mutual defection this social sanction may be nonexistent or at any rate may have a different value. Let the mutual defection utility, including the social sanction effect, be m∗ . Let me summarize the elements to be included in the technical analysis: (c) cost of producing E; (j) jointness effect when E is produced collectively; j consists of two components, namely a change i in effectivity and a social capital component e measuring the increase in trust, networking, skills, and knowledge gained from cooperative action (here: action as member of the group); to simplify exposition it is assumed that j = i + e − c/2, thus when both participants contribute they will each receive the payoff u + i + e − c/2, which is the same as u + j; (f) a freerider effect gained by defection; (s) the increase in the complying agent’s cost when the other defects;

242

raimo tuomela

(m) (negative) normative effect from defection (the effect of social sanctions, guilt, and shame); in the case of mutual defection this social sanction may be nonexistent or have a smaller value, denoted by m∗ . I will now write out the various sources of utility making two assumptions: (i) the utility factors can be linearly totaled as net utilities, and (ii) these factors can have negative values. What we get is shown in figure 10.1, assuming that the participants are in a symmetrical position with respect to their payoffs and writing out only participant A’s (the row player’s) payoffs. C D

C u+j u+f+m

D u+s m∗

Figure 10.1. Collective good dilemma

Here E can be taken to amount to the joint outcome CC or at least E is taken to entail CC. When should a rational agent A contribute (cooperate, choose C) rather than defect or act separately? This question can be answered by comparing the expected payoffs for C and D accruing to A: EU(C) = p(CB /CA )(u + j) + p(DB /CA )(u + s) EU(D) = p(CB /DA )(u + f + m) + p(DB /DA )(m∗ ) If A acts on his preferences he will cooperate rationally (in an instrumental sense) in the production of E by choosing C if and only if EU(C) > EU(D). If both players view the situation in similar terms we get mutual cooperation, and it is plausible to regard the analysans of (CIM) of section 2 to be satisfied. This is because, other things being equal, the jointness factor j then will have to be sufficiently large and the sucker effect will have to be sufficiently small as compared with the case in which the agents had not yet entered the interaction situation and reasoned about taking action. Consider now a Prisoner’s Dilemma defined by the preference ordering DC, CC, DD, CD for A and by CD, CC, DD, DC for B. For the purpose of illustration we might assign the numerical values shown in figure 10.2 to the participants’ preferences (utilities). Our parameters might now get the following values: u = 2, j = 0.50, e = 0, s = −3, f = 1, m = 0, m∗ = 0. Then, under full uncertainty, EU(C) = 0.75 and

cooperation and the we-perspective 243 EU(D) = 1.50. Hence the rational thing to do for both participants is to defect. The situation can be changed e.g. by changing the values of m and m∗ . Thus, we may assume that m = −1 and m∗ = −1. Then EU(D) = 0.50. Thus EU(C) > EU(D) and cooperation is rational. Note, however, that the transformations have changed the preferences in the PD situation. Thus the present solution to the dilemma is ‘‘external’’ rather than ‘‘internal.’’ C D

C 2.50, 2.50 4,-1

D -1,4 0,0

Figure 10.2. Prisoner’s Dilemma

Let us next consider in more detail under what circumstances it is rational to act cooperatively as a group member and thereby promote the ethos of the group, discussing the situation mainly from a single member’s point of view. This member will be assumed to ground his actions at least in part on what the ethos requires. There are two ways of viewing the situation. The first view is to take a group member to be well socialized so that the we-mode goals, preferences, and utilities will be the basis of his action. This view fits well with our analysis of group preferences and the related practical inferences. The other way to see the situation is that the group member is trying to make his decisions on the basis of both his I-mode utilities and his group utilities. He may use group-beneficial transformations to arrive at either I-mode group utilities or to we-mode utilities. In this second case, we are, generally speaking, dealing with a kind of switch problem, namely, a switch from thinking (including preferring) and acting as a private person to thinking and acting as a member of g. (The switch need not be based on direct intentional action but may often be an action that can indirectly be brought about by the agent or by his social environment, including educators; in other, more dramatic cases one can even speak of the ‘‘group taking over’’ the agent.) Starting with the I-mode case, we will now deal with the ethos, E, of the group, g, by simply treating it as a goal of g. The production of E is assumed to give a certain amount of utility, which we denote by u, to all the group members. The goal state E is in the first case to be considered below assumed to be wanted and intended in the I-mode by the agents. Again, we assume that there are only two participants with

244

raimo tuomela

two possible choices (C and D) and that E can be produced even alone. I will here concentrate on the second of the discussed views where our group members are not fully socialized and thus face a switch problem. Thus the main question here is to compare group-related, namely, ethospromoting behavior with I-mode behavior (and especially I-mode behavior which is ethos-promoting only if it suits the individual’s preferences or utilities). The harder problem is to show that we-mode ethos-promoting behavior wins over ethos-promoting I-mode (namely, pro-group I-mode) behavior. Let us now consider under what circumstances it is rational to act so as to promote the ethos of the group. I will discuss the situation mainly from a single participant’s point of view (and thus individual rationality), which makes it possible to speak of one-sided ethos-promotion. I will assume first that the agents (group members) have certain I-mode preferences about the situation, namely, preferences that they have on private grounds. These preferences are contrasted with we-mode preferences related to their group, g. As noted, the we-mode preferences must respect the group’s ethos. Under what conditions is it rational for a person to adopt the group stance and act for a group reason? In the above setup, we make the switch in two steps. First we consider pro-group reasoning and acting. Earlier I dealt with the I-mode case with e = 0 to reflect the fact that the agent may not have acted for the group at all. When an agent finds it rational to cooperate in part for the reason (reflected in parameter e) that the group requires it or that it is to the benefit of the group—and accordingly cooperates—we can say that he is thinking and acting in the pro-group I-mode. Technically we may operate as follows. Above the joint outcome CC was valued by our agent as u + j. Let us suppose that we then have a case in which a group reason for C is operative and assumed to entail that e > 0. Then, if giving e a suitable positive value results in EU(C) > EU(D), we have a case of the pro-group I-mode. Indeed we have a strong case of pro-group I-mode sufficing for solving the collective action problem (here PD) at hand. What value e must have for rational switch to occur of course depends on the particular situation at hand. In our present simple example, we can go about as follows. Letting e = 1 instead of e = 0 and using the earlier values for the other parameters in the first case (thus m = 0, m∗ = 0), we arrive at an Assurance Game with the following expected utilities:

cooperation and the we-perspective 245 EU(C) = 1/2 (2 + 1.50) + 1/2 (2 – 3) = 1.25 EU(D) = 1/2 (2 + 1 − 0) + 1/2 (0) = 1.50, and thus EU(C) < EU(D). It suffices now to assume that mutual defection is sanctioned with the value m∗ = −1 to make cooperation more rational than defection, for then EU(D) = 1 and EU(C) > EU(D). Thus in the pro-group I-mode case less sanctioning is needed than in the earlier pure, possibly fully selfish I-mode case. Next we consider if we can model the we-mode in the present context. In a collective dilemma situation such as the PD above, the group prefers the joint outcome CC to DD and that its members choose C over D. If the members here think collectively rationally they will think in the we-mode. The we-mode depends on the collective acceptance by the group of E (hence CC) as the group’s goal, collective commitment to E and CC as well as forgroupness. In this particular case forgroupness is automatically satisfied due to the group’s acceptance of E as the group’s goal. The important thing to emphasize here is that we basically are dealing with the change of agency from individual group members to the group. If we view the matter of achieving E from the group’s point of view we can say simply that here the group prefers to achieve E to not achieving it: E is better for the group than −E. Thus if the costs for achieving E are such as to make the expected utility of E higher than that of −E, a rational group will try to achieve E. There is in general no Prisoner’s Dilemma or any other collective action dilemma here.29 Our I-mode account concerned an individual’s decision-making, whereas here we are dealing with the group’s decision-making. Thus there would seem not to be an interesting connection to the (pro-group) I-mode treatment. However, while the above situation is what we in principle have, we may still make the educated guess that in actual social life things are not so smooth. Group members must be supposed to more or less blindly follow the group’s directives and recommendations for the theory to work, but that is a strong requirement especially in the case of large groups where the members may not fully identify with the group. Thus, through the backdoor as it were, we get back to a dilemma situation (be it a Prisoner’s Dilemma or some other collective action dilemma). Not only this is the case, but we must face situations where our group is in conflict with some other group or groups. For instance, there may be another group

246

raimo tuomela

competing for the same good (the other group may want to graze their cattle in the same area that we want to use). In such a situation, the chance of ingroup cooperation often increases, based on the old idea that a common enemy unites. Still, individual group members will have an incentive to freeride if their we-mode commitment to the group is not firm enough, so that their group-goal-based we-mode preferences do not preempt I-mode preferences from having motivating force.30 To say something about this kind of real life situation, we must again consider the earlier parameters. What happens to them in the we-mode case? Here the agents are supposed to act together in the we-mode sense on the basis of their shared intention (or at least the operative members in the group are supposed to share the intention). Such a shared we-mode intention provides the participants at least with a minimal social reason to participate in the production of E.31 This is a conceptual element that the I-mode case does not contain. Related to this factor, one may make educated empirical guesses. In general, forming such a we-mode intention and social reason will increase the value of the jointness parameter j, e.g. by increasing the amount of trust and social capital in the group, as compared with the (pro-group) I-mode situation. Furthermore, this shared intention rationally entails collective commitment (which can be taken to entail ‘‘forgroupness’’) to E. This collective commitment will do much work in the we-mode case, although in real life, in which a group seldom is fully homogeneous concerning the collective commitment and its strength, the group will have to monitor the situation and (typically) use positive and negative sanctions. What does collective commitment do in particular? Suppose our agent A is switching from pro-group I-mode to we-mode. In our earlier setup, m and m∗ will then have bigger (negative) values in the we-mode case.32 Here m derives from social sanctions and the possible accompaniment of the agent’s feeling of guilt (‘‘I should really have done C’’) or shame (‘‘others think I should have done C and I would indeed like to have conformed’’). Shame may come about due to (overt) social disapproval and perhaps threats of dismissal from the group. Furthermore, there may be objective sanctions (e.g. fines). If all participants defect there will be no overt social sanction, and m∗ will have a smaller value than m has. In the we-mode case the sucker effect s will either be zero or it will have a positive value (‘‘I am proud to be the sole person to act for the group’’), and in any case here s will have a bigger value than in the I-mode

cooperation and the we-perspective 247 case. Furthermore, the jointness effect j will, or at least may, be larger in the we-mode case due to the social commitment involved in collective commitment. For example, helping and encouraging may occur and bring about a net increase in j, and there will be an increase in social capital. Furthermore, the group members may act in a more disciplined and orderly way now (due to social commitment), and this may bring about a larger jointness effect than in the I-mode case. Let me note that on the group level there will in principle be a second-order collective action dilemma arising from the costs of punishing defectors. I cannot here tackle this complex problem except for making one remark. At least in groups with voluntary membership it is plausible to assume that there exist committed members for whom the cost of punishing defectors (e.g. socially criticizing) is not irrationally high. Perhaps they will be able to construct an impersonal system (‘‘the police’’) for this and persuade others to participate in the costs. This would suffice to block the emergence of higher-order collective dilemmas of punishment.33 In this section it has been shown that we-mode action will in many dilemma cases win over pro-group I-mode action and the latter will win over the pure I-mode case, which fits selfishness and ‘‘for-me-ness’’ better. These effects are likelier to happen when the agents are sensitive to what others think of them and their action. Mainly because of the shared social reason for joint action and the ensuing collective commitment and the social sanctions involved, it is harder to defect in the we-mode case than in the I-mode case and even in the pro-group I-mode case. Broader conclusions related to the matter will be briefly commented on in the concluding section.

5. Conclusion The distinction between the I-mode and the we-mode is a central tool in the theory of social action marking the difference between an attitude or action as a private person and that as a member of a group. We-mode reasoning leads more easily to cooperation simply because the we-mode constitutively is cooperative, and the cooperation also will be more persistent, ceteris paribus. We-mode joint action is constitutively cooperative, as seen from (CWM), basically because the participants’ private motivations do not matter. The participants view the situation as a case of the group acting,

248

raimo tuomela

and they function as ‘‘cogs in a machine’’. In contrast, in the I-mode case the private motivation may more easily change (even if they share a joint plan to act together). This is because the group is not backing it and participants view the situation merely as their acting in a social context of interacting persons rather than as being a part of an acting group. Both we-mode and I-mode functioning are needed in social life. It was shown that we-mode acting will often be more rational for an individual than is I-mode acting (including pro-group I-mode action). Let me finally sum up in broader terms what the debate between the we-mode and the I-mode approaches amounts to, especially in the case of cooperation. It was argued in section 4 that there are cases (that rather well simulate real life) in which the we-mode approach is rationally better than the I-mode in a functional sense. Whether there are cases where the I-mode wins over the we-mode has not been considered above. Indeed, we may also ask whether in those situations where the participants rationally do better when thinking and acting in the we-mode rather than in the I-mode could somehow have taken other I-mode features into their decisionmaking and rationally fared better in that way. This is a question which has two aspects. First there is the matter of the psychology of the agents. Are they somehow better off when acting in the I-mode way than in the we-mode way? This is not a clear question as such. It is central to note that the agents’ basic valuations of the situation, and which factors they thus regard as important, may differ in the case of we-mode and I-mode agents. As we know, the matter is partly culture-relative: in collectivist cultures we-mode valuations are prevalent while in individualist cultures I-mode valuations are more central. Education and environmental influences do seem to matter here, even if people might be genetically disposed to group thinking. (Also recall that group thinking can be shallow, or I-mode, or thick, namely, we-mode.) People thus might in fact act, and indeed act subjectively rationally, either prevalently in the we-mode or prevalently in the I-mode in accordance with their cultural valuations and standards. Yet there are objective elements involved here. For one thing, all cultures that we know have groups and institutions, for instance, and those group notions are conceptually we-mode notions and thus we-mode is the presupposed mode of acting in groups, even if people may act as group members and follow rules also in the I-mode way. These group notions represent something

cooperation and the we-perspective 249 in the social world that is objective, at least epistemically objective.34 Understanding the social world fully requires the use of full-blown group notions, hence the we-mode. The we-mode is not reducible to the I-mode primarily because of its necessary dependence on the group’s view of the situation and of the members regarding a group as an authority for their thinking and acting. In contrast, the I-mode is not dependent on group reasons, except perhaps in a contingent sense. In all, it is plausible to think that the we-mode is conceptually and, at least to an extent, ontologically indispensable.35 As to matters of explanation, it is therefore plausible to think that the ideal best-explaining social theory cannot dispense with we-mode concepts. Explanation depends on understanding, and understanding, as claimed above, here requires the we-mode we-perspective. This means also that in the study of social action, e.g. cooperative action, we-mode accounts will in many cases supersede Imode accounts, which of course tallies with what was said in the previous section of this paper.36 Notes

1. Cf. Richerson and Boyd (2001, 2005). 2. See Tuomela (2000). 3. In the example case one can participate in a group’s doing X without justifying one’s intention by the group’s intention, e.g. a person may as a private person take part in some agents’ joint action of pushing a broken bus group uphill. 4. See Tuomela (2003, 2007) for discussion. 5. See the various conceptual, rational, and empirical arguments in Tuomela (2002, 2003, 2007). For empirical evidence see e.g. van Vugt et al. (2000) and Brown (2000). The best survey in my work is in (2007), chapter 2. Limitations of space do not allow describing the empirical evidence here, but conceptual and rational points will be made in this paper. I will comment on these matters especially in the concluding section of this paper. 6. Here is a more precise characterization of I-mode and we-mode reasons in a simple case relating to the context of a specific group, assuming that a reason is a fact or a fact-like entity expressible by a that-clause (Tuomela, 2007, chapter 1):

250

raimo tuomela

(IMR) Reason R is a group member’s motivating I-mode reason for performing an action X if and only if R is the agent’s main motivating private (‘‘nongroup’’) reason for his performing X. Typically, R is a state that the agent wants or intends to be the case or a state that, according to his belief, obtains; and X is an action that is a means to R or an action that R requires for its obtaining such that the agent is privately committed to performing X for reason R. (WMR) Reason R is a group member’s motivating we-mode reason for performing an action X if and only if R is the agent’s main motivating group reason for performing X. Typically, R is a state that the group in question wants, intends, or requires to be the case or a state that, according to the group’s belief, obtains; and X is an action that is the individual’s part of a collective action that is a means to R or a collective action that R requires for its obtaining, where the group members are collectively committed to performing the collective action for reason R and mutually believing so.

In (WMR) X can be e.g. a collective (or group) action with multiple tokens (e.g. going to church on Sundays) or a joint action like cleaning up a park as a many-person action. As the group members are collectively committed to performing the collective action in question for reason R (a state expressible by a that-clause), they are also socially committed to the group members performing their parts of the collective action for reason R. Having a private commitment means in (IMR) that the person privately (rather than as a group member) has psychologically bound himself to a ‘‘content,’’ e.g. to performing an action for a reason. In general, private commitment is dependent on an intention, here the intention to reach a goal. Notice that functioning in the we-mode is necessarily connected to a group reason, to what one’s group has committed itself in the situation at hand, where the group’s commitment serves as an authoritative reason for the participants. In contrast, functioning in the I-mode is at most only contingently connected to the group reason (when there happens to be one involved). ((WMR) and (IMR) can be generalized to cover any voluntary attitude and action.) 7. See Tuomela (2002, 2003) for more details. 8. As for a more general conceptual discussion of the two perspectives, the reader can be referred to Tuomela (1984, 1995, 2000, 2002, 2003), and Tuomela and Tuomela (2003). The paper will draw on these works and also, especially in section 3, on Hakli and Tuomela (2003). Of other

cooperation and the we-perspective 251 authors that have recently discussed and emphasized the we-perspective I would like to mention especially Gilbert (1989) and Sugden (2000). 9. Preferences are all-relevant-things-considered preferences and can be thought of as comparative rankings of the value or ‘‘betterness’’ of the alternatives by an agent (cf. Tuomela, 2000, chapter 3). They are reflected in his dispositions to make choices. I will distinguish between ‘‘given’’ preferences, namely those with which an agent enters a situation of interaction, and ‘‘final’’ preferences, namely those on which the agent actually acts after considering his present situation. Both given and final preferences can accordingly be all-relevant-thingsconsidered preferences. Obviously, preferences can be institutional and social or they can be private and subjective. 10. Cf. Tuomela and Tuomela (2005). The present account is modeled but improves on the account given in Tuomela (2000) where also the term ‘‘reciprocity cooperation’’ is used for I-mode cooperation and where the adjustment of actions is justified in terms of shifting one’s preferences toward the other’s preferences. 11. The kinds of trust that cooperation in the I-mode case and the we-mode case require are discussed in Tuomela and Tuomela (2005). 12. See Tuomela (1995: chapter 1) for this kind of analysis and justification for clause (3). 13. A shared collective goal is a goal satisfying the so-called Collectivity Condition according to which, it is necessarily (on non-contingent, quasi-conceptual grounds) true that if one or more agents satisfy the goal, then it is satisfied for all participants. The ‘‘quasi-conceptual’’ grounds are taken to entail that the collective goal content is collective due to the collective acceptance of it by the participants as their collective goal (note the intended partial circularity!). The notion of satisfaction for a goal holder means here that the goal state or event comes about due to the collective effort by, or at least under the collective guidance of, the group members. Collective acceptance entails that the goal is shared in a we-mode sense and the participants are collectively committed to the goal and aim at contributing to satisfy its content. 14. See section 3 and Tuomela (1995: chapter 6). 15. Groups can be regarded as conglomerations of (or relational systems consisting of) their members, their interrelations, and, possibly, of relevant artifacts all extended through time. So in this sense groups can

252

raimo tuomela

be taken to exist as entities. However, groups are not strictly speaking agents or persons. I take the conceptual framework of agents and persons to presuppose that persons have bodies and bodily sensations, as well perceptions and feelings, and can refer to themselves by the first person pronoun ‘‘I’’. Groups are not this kind of entity. 16. See Tuomela (2000: chapter 3) and Tuomela and Tuomela (2005). 17. I have discussed and defended a certain account of joint intentions as shared we-intentions e.g. in Tuomela (1984, 1995, 2005a, 2005b, 2007). As to an intention attributable to a group, see Tuomela (1995: chapter 6) and the account (GP) of group preference in a later note below. 18. Bratman has formulated a theory of cooperation that is best regarded as an I-mode one (see Bratman 1999). In it the notion of joint action need not yet be cooperative. Only the addition of a condition of ‘‘minimal cooperative stability’’ makes it so. Bratman’s case of ‘‘unhelpful singers’’ indicates that his account is an I-mode account (see Bratman 1999: 103): The singers are assumed to jointly intend to sing a duet without helping the other one if he, contrary to expectations, does something wrong. This example indicates that I-mode joint action need not be cooperative. In contrast, a we-mode joint action is necessarily (minimally) cooperative, being based on joint intention as a group. Joint intention as group involves the kind of collective commitment to the joint action that excludes the case of the unhelpful singers (the inference schema (W2) assumed to apply to we-intenders also shows this, see e.g. Tuomela (1984, 2005a, 2007). 19. See Tuomela (2000: chapters 11–12) for discussion. I argue in Tuomela (2007: chapter 7), that the we-mode wins over the I-mode also in the iterated case. 20. See Tuomela (2000: chapter 9). 21. See the detailed analyses in Tuomela (2002, 2007), Tuomela and Tuomela (2003). 22. See Tuomela and Tuomela (2005), and Tuomela (2007: chapter 7) for arguments. 23. Here is my detailed account of group preferences, which does not assume that the members have the same preference in the I-mode as the group has. It is assumed that there are ‘‘operative’’ members for the group (possibly consisting of all its members) who have the authority

cooperation and the we-perspective 253 from all the group members to accept preferences and possibly other contents for the group and who perhaps are authorized to give orders to other group members. The preferences here may concern the kinds of action-goal sequences discussed earlier in this paper: (GP) Group g prefers X to Y as a group if and only if there are operative members of g such that: (1) these agents acting as group members in the we-mode intentionally collectively accept with collective commitment for g that X is preferable to Y (namely they jointly form the preference ordering in the we-mode); (2) there is mutual knowledge among the operative members to the effect that (1); (3) because of (1), the (full-fledged and adequately informed) nonoperative members qua members of g tend to tacitly accept the collective commitment—or at least ought so to accept—that their group g prefers X over Y; (4) there is mutual belief or knowledge in g to the effect that (3). Considering the case in which all the group members are operative ones (think of informal egalitarian groups, for instance), clauses (3) and (4) become emptily satisfied. Thus, we can say that, basically, in conditions of mutual knowledge, such a group prefers X over Y as a group if and only if its members collectively (or jointly) accept that X is preferred to Y in group contexts—and if and only if the members of the group as a group prefer X to Y. Thus we can here speak about preference attributions of X over Y to a group and equivalently speak about the members jointly or as a group preferring X over Y. In groups where the decision-making is based on argumentation or informal discussion, this may happen on a case by case basis, but in groups with formal rules and procedures, the initial acceptance of a group decision procedure binds the group members to accept also the results, as in majority voting, for example. Such an institutionalized and codified method may replace the specially authorized operative members, as the collectively accepted and authorized procedure will yield group decisions as its output (cf. the detailed discussion of this kind of ‘‘authority systems’’ in chapter 4 of Tuomela 1995; also see Tuomela,

254

raimo tuomela

2007: chapter 6). In general, the results of the procedure, be it an institutionalized one or simply spontaneous group discussion, may significantly differ from the agents’ original opinions. Thus group preferences need not directly reflect the individual members’ private preferences. 24. See the discussion in Tuomela (1984: chapter 2, and 2000: chapter 5), for we-intentions see Tuomela (2005a, 2007). 25. How commitments affect social action is an important issue in modeling social life. This has been discussed by e.g. Gilbert (1989, 2002), Castelfranchi (1995) and Tuomela and Tuomela (2003). 26. See Tuomela (2000: chapter 11). 27. I will below consider freeriders only in the context of a simple Prisoner’s Dilemma case without attempt to analyze the notion of freeriding. As Professor Amartya Sen pointed out to me, the present kind of context does not allow a freerider to publicly threaten the others, which, in contrast, is the case in the ‘‘threat games’’ studied by Nash in the context of bargaining (Nash 1950). However, even in the simple public good case I will concentrate on below it may well be the case that a player believes that there will be sufficiently many contributors that the good will be produced (assuming it is a step good) and there thus is something to freeride on. 28. See Tuomela (2000: chapter 10). 29. When E amounts to the joint outcome CC, we have CC > CD ∨ DC ∨ DD. Thus there is no PD any more. However, if we allow that E can be achieved also alone, E would amount to CC ∨ CD ∨ DC and we have a dilemma for the members, for while they are collectively committed to avoiding DD each will prefer in the I-mode to choose D over C. (I thank Kaarlo Miller for emphasizing this to me.) In the general case, however, it is very rare that a public good can be produced by a single member. 30. Some social psychologists have investigated situations in which an nperson Prisoner’s Dilemma is embedded in a social context where the ingroup formed of those n persons is competing with another group (outgroup) for the same continuous or binary public good (for recent discussion and experimental results see Bornstein 2004). This is called an Intergroup Prisoner’s Dilemma (IPD) game. In it the more the players contribute to the public good the more the group gains (the ‘‘cake’’ the group gets is bigger), but on the other hand freeriders will

cooperation and the we-perspective 255 always do better than cooperators. This game has the following central strategic properties: (1) Withholding contribution is the dominant individual strategy. In each group a player’s payoff for defecting is higher than her payoff for contributing, regardless of what all other (ingroup and outgroup) players do. (2) The dominant group strategy is for all group members to contribute. The payoff for a player is highest when all ingroup members contribute, regardless of the number of outgroup contributors. (3) The collectively (i.e. Pareto) optimal outcome, the one that maximizes the collective (or summative) payoff to all players in both groups, is for all of them to withhold contribution. The first and second properties taken together define the intragroup payoff structure of the IPD game (for any number of outgroup contributors) as an n-person Prisoner’s Dilemma (PD) game or a social dilemma. The second and third properties define the intergroup payoff structure as a two-party PD game between the two groups. I cannot in this context discuss intergroup dilemmas, but I wish to point out that here we have an invitation to both we-mode and I-mode thinking. Properties (1) and (3) encourage subjects to think and act in in the I-mode while (2) invites we-mode thinking and acting. 31. See Tuomela (2006) for acting for a reason in the case of joint action. 32. Collective commitment is bound to change the probability assignments in the expected utility formula. Thus with full collective commitment a member acting as a group member will give zero probability to another member’s choice of D. In real life the probabilities hardly change so dramatically. 33. See Boyd, Gintis, Bowles, and Richerson (2003) for a recent treatment of the higher-order dilemma of punishment; see also the comments in Tuomela (2007: chapter 7). 34. For epistemic objectivity, see Searle (1995: chapter 1). 35. See Tuomela (2005b, 2007) for discussion and central arguments. In the 2007 book especially chapters 2, 6, and 8 are central. For the case of social institutions, see also Tuomela (2002: chapter 6). 36. I wish to thank Raul Hakli for comments.

256 raimo tuomela References Bornstein, G. 2004. ‘‘Cooperation in Intergroup Social Dilemmas.’’ In R. Suleiman, D. Budescu, I. Fischer, and D. Messick (eds.), Contemporary Psychological Research on Social Dilemmas. Cambridge: Cambridge University Press, pp. 227–47. Boyd, R., H. Gintis, S. Bowles, and P. Richerson. 2003. ‘‘The Evolution of Altruistic Punishment.’’ Proceedings of the National Academy of Sciences (USA) 100: 3531–5. Bratman, M. 1999. Faces of Intention. Cambridge: Cambridge University Press. Brown, R. 2000. Group Processes. 2nd edn. Oxford: Blackwell. Castelfranchi, C. 1995. ‘‘Commitment: From Intentions to Groups and Organizations.’’ In Proceedings of ICMAS’95. Cambridge, Mass.: MIT Press, pp. 41–8. Gilbert, M. 1989. On Social Facts. London: Routledge. Gilbert, M. 2002. ‘‘Considerations on Joint Commitment: Responses to Various Comments.’’ In G. Meggle (ed.), Social Facts & Collective Intentionality. German Library of Sciences, Philosophical Research, vol. 1. Frankfurt a. M.: HänselHohenhausen, pp. 73–101. Hakli, R. and R. Tuomela. 2003. ‘‘Cooperation and We-mode Preferences in Multi-agent Systems.’’ In First International Workshop on Social Life. Workshops and Tutorials, European Conference for Artificial Life, September 14–17, 2003, Dortmund, pp. 90–110. Available at . Nash, J. 1950. ‘‘The Bargaining Problem.’’ Econometrica 21: 155–62. Richerson, P. and R. Boyd. 2001. ‘‘The Evolution of Subjective Commitment to Groups: A Tribal Instincts Hypothesis.’’ In R. Nesse (ed.), Evolution and the Capacity for Commitment. New York: Russell Sage Foundation, pp. 186–220. Richerson, P. and R. Boyd. 2005. Not by Genes Alone: How Culture Transformed Human Evolution. Chicago: The University of Chicago Press. Searle, J. 1995. The Construction of Social Reality. London: Allen Lane/Penguin. Sugden, R. 1993. ‘‘Thinking as a Team: Towards an Explanation of Nonselfish Behavior.’’ Social Philosophy and Policy 10: 69–89. Sugden, R. 2000. ‘‘Team Preferences.’’ Economics and Philosophy 16: 175–204. Tuomela, R. 1984. A Theory of Social Action. Synthese Library. Dordrecht and Boston: Reidel Publishing Company. Tuomela, R. 1995. The Importance of Us: A Philosophical Study of Basic Social Notions. Stanford: Stanford University Press. Tuomela, R. 2000. Cooperation: A Philosophical Study. Philosophical Studies Series 82. Dordrecht: Kluwer Academic Publishers.

cooperation and the we-perspective 257 Tuomela, R. 2002. The Philosophy of Social Practices: A Collective Acceptance View. Cambridge: Cambridge University Press. Tuomela, R. 2003. ‘‘The We-mode and the I-mode.’’ In F. Schmitt (ed.), Socializing Metaphysics: The Nature of Social Reality. Lanham: Rowman and Littlefield, pp. 93–128. Tuomela, R. 2005a. ‘‘We-Intentions Revisited,’’ Philosophical Studies 125: 327–69. Tuomela, R. 2005b. ‘‘Motivational Reasons for Action.’’ Forthcoming in M. Timmons, J. Greco, and A. Mele (eds.), Rationality and the Good: Themes from the Epistemology and Ethics of Robert Audi. New York: Oxford University Press. Tuomela, R. 2006. ‘‘Joint Intention, We-Mode and I-Mode.’’ Midwest Studies in Philosophy 30: 35–58. Tuomela, R. 2007. The Philosophy of Sociality:The Shared Point of View. New York: Oxford University Press Tuomela, R. and M. Tuomela. 2003. ‘‘Acting as a Group Member and Collective Commitment.’’ Protosociology, Understanding the Social II: Philosophy of Sociality 18/19: 7–65. Tuomela, R. and M. Tuomela. 2005. ‘‘Cooperation and Trust in Group Context.’’ Mind and Society 4: 49–84, ms available at . Van Vugt, M., M. Snyder, T. Tyler, and A. Biel. 2000. Cooperation in Modern Society. London: Routledge.

11 Collective Intentions, Commitment, and Collective Action Problems M A RG A RE T G I L B E RT

Introduction Towards the end of his famous 1977 article ‘‘Rational Fools,’’ the distinguished economist Amartya Sen wrote: ‘‘The main thesis has been the need to accommodate commitment as part of behavior’’ (Sen 1977: 344). He means, of course, human behavior. Since Sen published these words, my own work in philosophical social theory has brought me to a conclusion that could in broad terms be similarly phrased. Naturally I do not mean to ascribe to Sen the considerations and arguments that led me to this position. In this paper I explain the particular form of my own conclusion, and relate it to collective action problems as these are conceived of in what has come to be known as the theory of rational choice. One of the questions from which I started was this. What do people refer to when they talk about what ‘‘we’’ intend, want, believe, and so on? Starting with the article ‘‘Modeling Collective Belief,’’ in 1987, and a long book, On Social Facts, in 1989, I have subsequently argued as follows. The phenomena in question—or at least an important range of them—have a special kind of commitment at their core.1 This I refer to as joint commitment. This also lies at the core of, among other things, social groups, social rules and conventions, and everyday agreements, according to central everyday understandings of what these are.2

collective intentions and commitment 259 In the present paper I first sketch my joint commitment account of what it is for people to do something together, as this is understood in everyday life. I discuss the nature of joint commitment at some length. I then discuss the relevance of the preceding material to collective action problems in the sense of rational choice theory. Pertinent supplementary material is from time to time referred to in the notes. Hopefully, the brevity of my exposition here will enable the gist of my main contentions clearly to emerge. I start with some preliminary remarks on the phrase ‘‘collective action’’ and two distinct phenomena to which it may be applied.

1. Two kinds of ‘‘collective action’’ In many contexts, the phrase ‘‘collective action’’ is used to refer to a situation involving a number of agents or actors, each with his personal preferences over the possible outcomes constituted by their individual actions in combination.3 An example is this case made famous by David Lewis (1969). Their telephone call interrupted, the two people involved, each preferring to resume their conversation, have these options: call the other, or wait for the other to call. If both call back at the same time, each will hear a busy signal, and they will not be able to talk. If neither calls back, their call will of course not be reconnected. The situation just described is a paradigm case of what has come to be called a ‘‘coordination problem,’’ following Lewis, or a ‘‘coordination game,’’ following Schelling (1960). In this situation, whatever each chooses, each one’s acting on his choice will constitute a collective action in the sense at issue. That the agents rank the possible collective actions in the same way is a coincidence as far as the relevant conception of collective action is concerned. This, or something close to it, is the construal of ‘‘collective action’’ that predominates in the theory of rational choice. There is nothing wrong with this construal. It is not, however, the only prevailing one. The other construal I have in mind associates the phrase ‘‘collective action’’ with what may also be referred to as collective agency. As it happens, Lewis’s telephone case provides an example. His two agents were engaged in a telephone conversation. That is what they were collectively doing or (in more vernacular terms) doing together.4

260

margaret gilbert

In their 1993 paper ‘‘Rationality in Action,’’ philosopher of social science Martin Hollis and economist Robert Sudgenremark: ‘‘the idea of supra-individual units of agency implies deep revisions to the conventional theory.’’5 By ‘‘conventional theory,’’ here, they mean the theory of rational choice. They are not arguing against such revisions; they simply propose that they will be necessary if sense is to be made of what they refer to as ‘‘supra-individual units of agency.’’ They go on to mention the work in which I set myself ‘‘to make sense of agency in collective terms.’’ What is the relationship of collective action in the sense of collective agency to rational choice theory? Is it something with which the theory should be concerned? Will the theory need deeply to be revised in light of such concern, as Hollis and Sugden suggest? These are good questions. They will best be considered, however, after one has indeed ‘‘made sense of agency in collective terms.’’ I now explain what I take collective agency to be. The account I sketch here is doubtless open to some fine-tuning. As it stands, however, it provides a basis for answering the question: what is the relationship of collective action in the sense of collective agency to collective action problems in the sense of the theory of rational choice?

2. Collective action as involving collective intention My account of collective agency is developed in two stages corresponding to a certain rough understanding of action in general. Consider a human being, Sally, who raises her arm. Raising one’s arm is a paradigmatic human action. There is a clear-cut behavioral component: Sally’s arm rises. But it would not necessarily be true of her that she raised her arm if her arm had simply risen. Perhaps she was completely passive in the matter: her arm was raised by a pulley controlled by someone else. The general philosophical understanding of what it is to raise one’s arm, as opposed to simply having it rise, appeals to intention. Details vary, but the nub of this way of understanding the matter is this. One’s arm must rise by virtue of one’s intention to raise it.6 The type of action exemplified by a person’s raising his arm is sometimes referred to as the category of intentional action. I take the concept of collective action at issue here to be the concept of collective intentional action. As noted by Elizabeth Anscombe (1957), one can distinguish between an intention to do something in the future, and an intention in acting.7 As to

collective intentions and commitment 261 the first, one can intend to perform an action, in the future, without ever actually doing it. As to the second, one who raises her arm, for instance, has an intention by virtue of which the action is occurring. The intention lies within the action, one might say. That said, I start by presenting a generic account of collective intention. This is intended to capture what people refer to when they speak about what ‘‘we’’ intend, where ‘‘we,’’ here, is not understood to be elliptical for ‘‘we both’’ or ‘‘we all.’’ They may be said in this case to be referring to what they collectively intend.

3. Collective intention My proposal about collective intention involves a number of technical terms, which will be explained shortly. I start by simply presenting and explaining the meaning of the proposed account. I say something in defense of it later. The account runs as follows. Persons X and Y collectively intend to perform action A (for short, to do A) if and only if they are jointly committed to intend as a body to do A. Here I say that a joint commitment to intend as a body to do A is both necessary and sufficient for the existence of a collective intention to do A. Possibly some amplification is needed for a sufficient condition. I doubt this would alter the main drift of the discussion. The technical terms in this account are, evidently, ‘‘joint commitment’’ and ‘‘intend as a body.’’ Before explaining what I have in mind here, I introduce one further technical term it will be useful to have to hand. I say that, by definition, those who are jointly committed to do X as a body, for any X, constitute the plural subject of X-ing. The label ‘‘plural subject’’ is not intended to connote any more or less than its definition entails. In the next two sections I focus on the key notion of joint commitment.8 I then return to collective intention.

4. Commitments of the will I start by discussing the relevant general notion of commitment. The term ‘‘commitment’’ appears in many contexts nowadays, both theoretical and

262

margaret gilbert

practical, and it is clearly used in different senses. Therefore in any given discussion it stands in need of some degree of explanation. Consider in this connection some statements by economist Robert Frank. In the first quotation the word ‘‘commitment’’ is in quotation marks; in the second it is not. Thus Frank (1988: 6): ‘‘A person who has not eaten for several days is ‘committed’ to eat.’’ And: ‘‘commitments of this sort ... are merely incentives to behave in a particular way’’ (ibid.). Perhaps all so-called commitments are incentives of a kind, depending, of course, on how one construes ‘‘incentive.’’ In my discussion here, I am concerned with what I shall refer to as ‘‘commitments of the will.’’ One who has not eaten for several days need not be at all committed to eat—according to the conception at issue. He may, indeed, be committed to never eating again. Perhaps he desires to die and is deliberately refraining from eating, with this end in view. This is not to say, of course, that he cannot feel or be hungry. I take the conception I have in mind to be an intuitive or vernacular one, though there are other such conceptions as well. I shall not attempt fully to explicate it, but hope to mark at least some of its key features and instances, and thus to mark it off from some of the conceptions at issue in the writings of other authors. I take the personal decision of an individual human being to involve a commitment of the relevant kind. It is not the only source of such a commitment, but it is a familiar one, and therefore a useful focus for discussion. Suppose, for instance, that Lily decides to go skiing tomorrow. She is thereby committed to skiing then. To say this much is to make it clear that a commitment, in the sense in question, does not necessarily involve more than one person. It also makes it clear that an action to which one is committed need not be one that, in and of itself, one is morally required to perform. Indeed, one may be committed to a certain course of action without there being good reasons of any kind for one’s preferring that action as such to all alternatives. Nor need one think there are such reasons. One may rightly regard one’s decision as essentially arbitrary. I assume that in the case of Lily’s decision to go skiing tomorrow, the associated commitment is solely a function of her having decided. I take it that if anything is an exercise of a person’s will, his making a decision is. That is why I shall call commitments of the kind a decision produces a commitment ‘‘of the will.’’

collective intentions and commitment 263 In fact this label suggests two things, both of which are important features of the case involving a personal decision. First, the commitment is in a clear sense a ‘‘creature of the will.’’ Second, it may be said to ‘‘bind’’ the will. In short, the commitment is of the will, and by the will. There are various senses in which one can said to be ‘‘bound.’’ The binding nature of personal decisions has two salient aspects. First, in an intuitive sense, the one who makes the decision has sufficient reason to act in accordance with it. As I am using the phrase, if by virtue of some fact I have sufficient reason to perform act A, then, in the absence of countervailing factors, this is what reason requires me to do. Thus one might be said to be ‘‘bound,’’ in a sense, to do what one has decided to do. All else being equal, reason requires one to do it.9 I should emphasize that I am not saying that, once I have decided to perform it, an action has something to be said in its favor, in and of itself. Deciding to do something cannot make the action decided upon ‘‘a good thing to do’’ in and of itself. Nor am I saying that one is morally bound to do what one decided to do. That is, I do not say that one’s having reason or the associated ‘‘ought’’ is a moral matter—whatever precisely that amounts to. Second, the form of binding engendered by a decision has a specifiable degree of persistence. Given one’s decision, one will continue to be bound in the first sense unless and until one has carried out the decision—unless or until one rescinds it. I take rescission to be a matter of deliberate repudiation, as when one ‘‘changes one’s mind’’—deciding not to do what one had previously decided upon.10 A decision that has been rescinded ceases to bind in the first sense. I have elsewhere argued for a distinction between decisions and intentions in this regard. In brief, personal intentions, as opposed to decisions, bind in the first sense but not in the second (Gilbert 2005b). A related point is this: intentions do not require rescission or deliberate repudiation. One can decide not to do what one finds oneself with an intention to do and hence deliberately repudiate it. It will then cease to bind in the first sense at issue. It may cease so to bind, however, without any such repudiation. It can simply go out of existence or be replaced by a contrary intention. I return to this distinction later. The personal decision of an individual human being is such that he can rescind it on his own and he is the only one who can rescind it. Thus

264

margaret gilbert

it creates what I refer to as a personal commitment: it is created by a single human being who alone has the capacity to personally rescind it.

5. Joint commitment I now turn to joint commitment. As I have already indicated, I take the concept of a joint commitment to be a fundamental everyday concept in that it is a central element in many other such concepts. I call these the plural subject concepts. In addition, and relatedly, we need to appeal to joint commitment, I would argue, in order fully to understand an important class of obligations and rights.11 In short, I commend this notion to everyone’s attention. For present purposes a full treatment will not be necessary. Perhaps the most striking thing about it is that it is not what I have elsewhere called a singularist notion.12 That is, it goes beyond the conceptual scheme of individual human beings with their personal intentions, beliefs, goals, and so on. (I prefer to talk of ‘‘singularism’’ rather than the more familiar ‘‘individualism’’ here because the latter term is by now quite ambiguous.) In the text below I shall be describing what one can call the basic or simple case of joint commitment. That is, in brief, joint commitment in the absence of any special background understandings. The reader should bear in mind this qualification when reading what follows. For present purposes it is not necessary to go beyond this.13 The joint commitment of Anne and Ben, say, is the commitment of Anne and Ben. It is not the conjunction of a personal commitment of Anne’s and a personal commitment of Ben’s. It is precisely the joint commitment of the two of them. As to the general form of a joint commitment, people are jointly committed to do something as a body—in a broad sense of ‘‘do something.’’ In the case of collective intention, on my account of it, the parties are jointly committed to intend as a body to perform some action. I understand this roughly as follows: the commitment is together to emulate as far as is possible a single body (perhaps better, person) that intends to perform that action. Each party, then, has reason to do what he can to emulate, in conjunction with the others, a single body of the relevant kind. Depending

collective intentions and commitment 265 on the case, discussion among the parties may be called for in order that individual efforts are effective. The concept of a ‘‘single body’’ that I argue is part of the content of a joint commitment is just that: the concept of a single body. It is not supposed to incorporate the idea that this body has a plurality of members. One who is party to a given joint commitment is thereby committed to act as best he can, in conjunction with the others, to constitute a single body of the relevant kind.14 The participation of each of the would-be parties is required in order to create a given joint commitment. One way of putting what is to happen is this: each party must make clear to the others that he is ready to be jointly committed in the relevant way. Thus each must express to the other a certain condition of his will.15 Once these matching states of the will are mutually expressed in conditions of common knowledge, the joint commitment has been created.16 To make this more concrete, here are some situations of the kind I have in mind. First example: Jessica says ‘‘Shall we meet at six?’’ and Joe says ‘‘Sure.’’ I have argued elsewhere that this interchange is best parsed somewhat as follows. Through these mutual expressions made in conditions of common knowledge, each party expresses his or her readiness to be jointly committed to endorse as a body the decision that they meet at six. This interchange, of course, constitutes what would ordinarily be termed ‘‘an agreement’’ in English—an informal agreement, not a legal contract. I say more about agreements in due course. Second example: coming out of the factory one day, two factory workers, Polly and Pam, start a conversation about their day. Each lights a cigarette and they talk till a while after the end of their smoke. This happens again and again. The sequence is broken when one day Pam waits for Polly but she doesn’t turn up. The day after this Polly comes up to her and, with evident sincerity, apologizes for her absence: ‘‘Sorry I didn’t turn up yesterday, I was off sick.’’ ‘‘Not to worry,’’ says, Pam, accepting her apology ‘‘I’m glad we are back on track, though.’’ At no point did Pam and Polly agree jointly to uphold the rule that they are to meet daily outside the factory for a smoke and a chat. Nonetheless, enough may have passed between them by now jointly to commit them to uphold it (as a body). My thought here is that the signs, verbal and non-verbal, in the scene I’ve described suffice to show (if

266

margaret gilbert

this has not been evident before) that the parties are both ready jointly to commit to uphold this informal rule for their behavior. Some may be inclined to see this story—up to the dialogue—in terms of expectations of conformity to the regularity of meeting for a smoke and chat after work, and then to see Pam’s apology as an apology for having risked disappointing or otherwise inconveniencing Pam. There is no need to see it this way, however, and there is reason to see it in the way suggested. Certainly expectations of conformity are a likely outcome of the story up to the dialogue. If this were all that the parties had in mind, however, Polly might well have come out with something like this: ‘‘I hope you weren’t disappointed when I didn’t show up the other day’’ or ‘‘I hope you weren’t put out.’’ And Pam might have alluded to her own disappointment or the lack of it. Certainly, again, in a world—unlike that depicted by many economists and lawyers—where people see their interactions as much in terms of the establishment of joint commitments as in terms of the production of expectations and possibly detrimental reliance, there is no reason to see the expectations story as having any greater initial probability. All the signs in the story point to an interpretation in terms of joint commitment. They are, indeed, relatively subtle, and a full discussion of how they suggest a joint commitment interpretation would take some time. Here, in brief, are some pertinent points. First, it is not clear that an apology would be in order if all that is in question is disappointed expectations created by one’s regular behavior. Yet Polly offers an apology for non-conformity to the regularity of meeting after work for a chat, and Pam accepts her apology—though not in so many words. Such an apology would definitely be in place were they jointly committed to uphold a rule mandating their regular meetings.17 Second, Pam’s saying ‘‘we are back on track’’ suggests that, for the two of them, there is indeed a proper way of going about things—there is, in other words, a rule of the type described that they are jointly committed to uphold. References to ‘‘our rule’’, and so on, can be tendentious. The speaker may not be sure that he and his hearer(s) are party to a joint commitment to uphold a given rule, or may be sure they are not but want to bring about such a joint commitment. If a tendentious utterance like this is accepted without demur by the other party or parties, this can make it the case that

collective intentions and commitment 267 the joint commitment in question exists—that there is a rule that is ‘‘ours’’ in the joint commitment sense.18 If the signs indicated are not enough in the eyes of Pam and Polly to clinch the point that each is ready jointly to commit to uphold this rule, then more data of a similarly subtle kind can be expected to firm it up. The main point to be made is that it is possible for people to establish a joint commitment between them without an agreement. Agreements are useful, however, precisely because by virtue of their explicitness things are made (relatively) clear within a brief span of time—the time between the proposal of the agreement and its acceptance. It seems fair to characterize a joint commitment as a commitment of the will. In this case, it is a commitment of two or more wills. It is true that mutual expressions in conditions of common knowledge are among the necessary conditions for the creation of a joint commitment, failing a special background. What is expressed, though, and what thereby becomes common knowledge, is the state of the will of each party (his or her readiness to be jointly committed with the other party or parties). I take it that a joint commitment can only be rescinded with the concurrence of all the parties, just as it requires their concurrence for its formation. I take such rescinding to involve a deliberate repudiation of the joint commitment. (Remember that I am discussing the basic case here.) In what way or ways do joint commitments bind? First, a standing joint commitment binds the wills of the individual parties in the following way: each has sufficient reason to act in accordance with the commitment. Thus the commitment has the normative force of a personal intention or decision for each of the parties. What of the persistence of this normative force? Is it the case that once a joint commitment comes into being its normative force persists until it has been either satisfied or rescinded, as in the case of a decision? Certainly some joint commitments require rescission if their normative force is to be cancelled prior to their fulfillment. Those created by an agreement are in this category. Agreements are, indeed, the closest cousins to decisions in the realm of joint commitment. As indicated earlier, I take those who enter an everyday agreement thereby to create a joint commitment. I say more on this shortly. Other joint commitments are open to deliberate repudiation but do not appear to require it in order that their normative force with respect to

268

margaret gilbert

their participants cease to exist. They may go out of existence in much the way that they came in, by means that do not rise to the status of an agreement. The joint commitment of the factory workers Pam and Polly would appear to be of this kind. Just as it came into existence by a subtle, gradual process, it may cease to exist in such a way.19 Then its normative force will be at an end, though it has not been deliberately repudiated. Such joint commitments are in this way closer to the personal commitments involved in personal intentions than those involved in personal decisions. At the same time, I take it that all parties must contribute to the demise of the joint commitment in these more intention-like cases also. In sum, a joint commitment binds in the following way: it gives the parties to it sufficient reason to act in conformity to it unless and until it has been satisfied, rescinded or otherwise put to rest with the concurrence of all. Once jointly committed, then, each party has sufficient reason to conform to the commitment subject to the appropriate exercise of both his own will and those of the other parties. Not only does a joint commitment give each party sufficient reason to act in a certain way, reason no one party can unilaterally remove. In addition, it constitutes an important form of relationship between the parties. Here it must suffice to say but a few words about this with only a brief explanation. First: each party is answerable to every other party should he default on the commitment. (Each has the standing to demand an explanation, a standing that is grounded on the jointness of the commitment.) Second: each party can be said to owe every other party conformity to the commitment and to have a right to each other’s conformity. (Each has the standing to demand conformity, a standing that is grounded on the jointness of the commitment.) These things will be understood, if not in so many words, by the parties.20

6. Collective agency I return now to collective agency. Before presenting an account of it, it will be good to repeat my account of collective intention—and to say something in defense of it. The account runs as follows: Persons X and Y collectively intend to perform action A (for short, to do A) if and only if they are jointly committed to intend as a body to do A.

collective intentions and commitment 269 One thing that strongly recommends a joint commitment account of collective intention is this. When people regard themselves as collectively intending to do something, they appear to understand that, by virtue of the collective intention and that alone, each party has the standing to demand explanations of action contrary to the collective intention, and, indeed, to demand of each party that he act in accordance with that intention. A joint commitment account of collective intention respects this fact. Though it would take too long to argue this here, accounts that do not appeal to joint commitment are hard pressed to do so.21 A joint commitment account of everyday agreements is also attractive. I have proposed that the parties to such an agreement can be seen as jointly committing to uphold a certain decision as a body, for instance, the decision that one will do the laundry and the other the dishes, or that they will meet at six o’clock. One can argue, though I shall not do so here, that this account is preferable to the standard approach in moral philosophy.22 This point about agreements is pertinent to my account of collective intention because pre-theoretically a standard way of creating a collective intention and, indeed, of initiating collective agency is by making an agreement. Those who jointly commit to uphold as a body the decision to paint the house, say, are thereby jointly committed to intend as a body to paint the house. I turn now to collective agency—collective action in the second sense mentioned earlier. One can see how to go about characterizing that given the present account of collective intention, along with the idea, mentioned earlier, that action as opposed to mere bodily movement is in some sense driven by an intention. One can, that is, say something along the following lines. Persons A and B are collectively doing A if and only if they collectively intend to do A (according to the previous definition) and each is effectively acting in light of the associated joint commitment.23 In other words, if the stated conditions are fulfilled, A and B are in a position to say ‘‘We are doing A’’ where this is not elliptical for ‘‘We are both doing A’’ or ‘‘We (collectively) are doing A.’’ In addition to its correspondence with the general idea of intentional action, this account of collective action squares with the way people think, talk, and act in the context of what they see as their acting together or, in

270

margaret gilbert

my terms, their collective action. For instance, they may see fit to rebuke one another for actions that are liable to bring the collective action to a premature halt, or for acting as if they are free to stop participating in it as and when they choose, without seeking the others’ concurrence. Once again, I find a plural subject or joint commitment account such as that proposed here is best suited to explain such data. I have argued this at length in a number of publications.24 It may be helpful to address one particular query here.25 Are there not cases of what would be described as doing something together in common parlance that do not answer to the account of collective action proposed here? In this connection the following example might be put forward: a committee passes a resolution by a majority vote. Suppose that the resolution in question is that Bill Jones should be offered a certain job. The committee’s passing it may seem simply to be a matter of the particular profile of the actions of the members of the committee: each member of the committee votes and, moreover, most of them vote to offer Bill Jones the job. Though one might think this, it does not seem to be right. In order for it to be true that a committee passes a resolution as the result of a vote, certain background understandings must be in place. More precisely, something like this must be understood by the committee members before the vote takes place: we intend to reach a conclusion on this matter by voting, where the majority of votes will determine the conclusion we have reached. In, say, raising his hand to exercise his vote, each member acts in light of this collective intention. This understanding of the situation is consonant with to the account of collective action that I have just proposed. According to this account, the background collective intention is a joint commitment with the content indicated, and the collective action is performed as a result of the parties to the collective intention acting in light of this joint commitment. Thus a committee’s passing a resolution is not just a matter of the particular profile of the actions of the committee’s members. It is a matter of the particular profile of the actions of the members of the committee acting in light of a particular joint commitment, a joint commitment to intend as a body to reach a decision in a certain way.

collective intentions and commitment 271

7. Participation in collective agency As the example just discussed makes clear, collective agents as I understand them act through their members. In other terms, in order for a collective to act its members must correctly understand their situation in a certain way and their behavior must in part be explicable in terms of this understanding. What may be referred to as the ‘‘driving force’’ in the situation is the understanding of the parties that they are jointly committed in a certain way. Or, perhaps better, the joint commitment itself is the driving force. The individual parties act in light of it. Now the members of collective agents constituted by human beings are generally possessed of the following important capacities. First, they can make their own decisions, form their own plans, settle on their own goals, and act in light of these. In short, they have the capacity to act as ‘‘singular agents.’’ Second, they are capable of considering the reasons that apply to them for acting in one way or another, considering which reasons outweigh or otherwise ‘‘trump’’ which other reasons, and acting in accordance with what, if anything, reason dictates. Insofar as they act within the bounds of reason they will be acting rationally in an important, intuitive sense. As empirical evidence suggests, the capacity to reason and deliberate may sometimes be put ‘‘on hold’’ as people act on salient factors without consideration of the big picture. Clearly, a salient joint commitment might in this way lead to unthinking participation in collective agency—which I have referred to elsewhere as ‘‘participant agency.’’ Given that the joint commitment was considered the only applicable factor, and there was no question in this case of either unilaterally or jointly rescinding it, conformity would, indeed, be the rational outcome. Recall that I am assuming that the parties to a joint commitment have sufficient reason to conform to it as long as it is not rescinded. So, all else being equal, rationality would require conformity. What I want to touch on now is the relationship between commitments of the will and two other types of consideration: personal inclinations and moral considerations. What would it be rational to do when these factors are opposed to a given commitment of this kind?

272

margaret gilbert

This is a large topic and open to disagreement. In particular, it may be unclear whether the claims made are ultimately matters of opinion or judgment, on the one hand, or of logic, broadly speaking, on the other. I turn first to inclinations—the label is intended to cover urges, promptings, leanings toward, and so on. My suggestion here is that commitments of the will ‘‘trump’’ inclinations with respect to what reason dictates. Why else, one might ask, would people make New Year resolutions? A regular smoker, for instance, resolves to stop smoking. If commitments do not ‘‘trump’’ inclinations at the level of reasoning it is hard to see how they are supposed to operate in order to produce conformity to them. The smoker can be sure that strong contrary urges will arise and somehow the resolution is supposed to be proof against this. It may be objected that most people break their New Year resolutions. This does not prove anything, however. We need not expect people always to act as reason requires. The salience of a strong urge could lead them to ignore their resolutions. Or they might reasonably describe things thus: they ‘‘found themselves’’ reaching out for a proffered cigarette, or they ‘‘couldn’t control’’ themselves though they made some effort to do so. A resolution, New Year’s or otherwise, may be somewhat special. Rather than pursuing that question, we can consider humdrum personal decisions. I think it was Christine Korsgaard who used the nice metaphor of inclination ‘‘proposing’’ that a certain act—such as buying an ice cream—be performed. One way of reacting to such a proposal is to decide to go along with it. ‘‘OK,’’ one says, ‘‘I’ll buy the ice cream.’’ Another way to react is to decide not to go along with it. ‘‘No, I won’t buy that ice cream, I’m on a diet.’’ Either of these things is supposed to close the case. In other words, the decision is assumed to be the last word on the subject, not the inclination. Of course in the case of a personal decision one can change one’s mind, and a nagging urge may well lead one to do so. This could be the story of many New Year’s resolutions: they are rescinded rather than broken. The case of joint commitment is different. Absent special background understandings, no one party is in a position to rescind them or otherwise bring them to a close unilaterally. I now turn briefly to the matter of morality. Precisely what morality amounts to is a hard question. That said, it is not clear that the foregoing discussion introduces morality in any way. The ‘‘trumping’’ of inclinations

collective intentions and commitment 273 by decisions, for instance, is not clearly a moral matter. It is a matter of the nature of decisions, on the one hand, and inclinations, on the other—and it is not clear that morality need intrude into the case. Be that as it may, at least some moral considerations can surely ‘‘override’’ at least some commitments of the will in terms of what rationality requires one to do. An example will make the point more vivid. Suppose that I have decided to do something evil. In addition, suppose that somehow I am incapable of changing my mind, though I am still capable of not doing the thing in question. Then one might argue that insofar as my decision stands I still have reason to do it. Nonetheless rationality requires that, all things considered, I do not do it.

8. Collective agency and collective action problems I now return to collective action problems in the ‘‘other’’ sense of collective action. I start this final section with another look at Hollis and Sugden’s paper ‘‘Rationality in Action.’’ To give some context, Hollis and Sugden are considering, roughly, how one might develop a more nuanced account of human motivation than that involved in rational choice theory. (One might observe that being more nuanced, in this case, would not take much.) The following (quite lengthy) quotation includes a number of quite lengthy gaps but represents, I think, a complete thought that is present in the text (Hollis and Sugden 1993: 30–1): An attractive idea here is to complicate the theory of action by introducing two tiers into deliberation ...A Kantian could try saying that the crucial distinction is between inclinations (lower tier) and duties (upper tier), with duty always serving as trumps ... That ... might ... help with the coordination problem, if a case could be made for treating all encounters surrounded with normative expectations as, somehow, morally charged, so that each player was obliged to consider the interests of both. But none of this is clearly compelling. Kant’s distinction between inclination and duty sets problems of its own; and the proposed treatment of all social norms as, somehow, moral obligations is unlikely to carry general conviction. So we shall merely leave the door open for Kantians who think the line worth exploring further.

In the preceding discussion I have invoked something other than inclinations, on the one hand, and moral obligations, on the other. I

274

margaret gilbert

have appealed to commitments of the will, whose normative force is distinguishable from the force of moral obligations, whatever precisely those are. I am not certain what Hollis and Sugden mean by ‘‘inclinations’’ when they use the term. Sometimes they use ‘‘preferences’’ in the same context. This might raise a doubt as to whether they are using the term ‘‘inclination’’ as I am. Be that as it may, I’ve said what I mean by inclinations and will proceed in terms of that idea. Now, the structure of the collective action problems that have been the focus of study in rational choice theory can exist among inclinations if we allow—as seems reasonable—that inclinations can have different strengths. We can, then, consider the ‘‘inclination’’ versions of the various problems. This will have a degree of realism insofar as personal inclinations often are patterned in the way of these problems. Suppose, then, that two parties face a pure coordination problem version of an interrupted telephone call. Both would like to continue the conversation and there is nothing to choose between the two relevant ways of doing so—as far as their inclinations are concerned. That is, it neither cares which one calls back and which waits for a call. This structure of inclinations is common knowledge. The existence of an appropriate joint commitment would decide the matter for each agent, rationally speaking. That is surely uncontroversial. For, apart from the reason each agent has to conform to the joint commitment, all else is equal. An appropriate joint commitment would, of course, have to be established. In that way joint commitment is not as handy as morality insofar as that is the same at all times and places. It is handier than morality, however, insofar as those who are parties to a joint commitment will at least at some level know this, while it seems possible that not everyone has a fully developed moral sense. In any case, there is really no contest in a pure coordination problem, since morality gives no clue as to how to proceed in such a problem where there are two possible collective actions that tie for best in each player’s eyes. Note that a joint commitment would appear to do better in this situation than two concordant personal intentions or decisions would, even where it is common knowledge that these have been made, for a personal intention or decision is rescindable at will by the one who made it. As I wonder whether or not to push off from the shores of reflection, it may occur to

collective intentions and commitment 275 me that you just might change your mind before you act, knowing that, after all, I might change mine. What of other types of collective action problem? I am not sure what reason has to say about attempting to do as well as possible according to one’s inclinations over the possible outcomes. Perhaps it has nothing to say about this. In other words, whatever you choose to do is OK. It’s up to you. Suppose, though, that reason says one maximize inclination satisfaction, all else being equal. If a commitment always ‘‘trumps’’ an inclination, as may be the case, then, once again, an appropriate joint commitment will settle the issue. If this is indeed so, then we can see how agreements, or other joint commitment phenomena—such as specific collective intentions or goals or rules with appropriate content—can lead to relatively good outcomes for all in collective action problems of all kinds, including the notorious ‘‘Prisoner’s Dilemma.’’ A significant advantage of this relatively nuanced perspective is this: it helps to explain how there can remain a pull in the direction of acting contrary to reason’s dictates. A joint commitment may trump one’s inclinations in the balance of reasons, but it does not obliterate them.26 Notes

1. Gilbert (1987, 1989). I allow that sentences of the form ‘‘We intend ... ’’ may on occasion be used to refer simply to what each of us intends. I am concerned with a different use according to which this is not the case (see Gilbert 1997). A selection of essays in which I have developed these views is to be found in Gilbert (1996, 2000, 2003a). See also Gilbert (2006). 2. On social groups and social conventions see Gilbert (1989: ch. 4) and Gilbert (2006: ch. 8); on social convention, with a critical focus on David Lewis’s account, see Gilbert (1989: ch. 6); on social rules, with a critical focus on H. L. amend A. Hart’s account, see Gilbert (2000: ch. 5); on agreements see Gilbert (1996: ch. 13), and Gilbert, (2006: ch. 10), also the text below. On collective emotions and values, respectively, see, e.g. Gilbert (2000: ch. 7 on collective remorse), and Gilbert (2005a).

276

margaret gilbert

3. For simplicity’s sake I shall in what follows use the generic ‘‘he’’ as substitute for ‘‘he, she, or it’’. 4. As this sentence indicates, I take collective action in the sense in question in this paragraph to be equivalent to the everyday, vernacular concept of acting together or doing something together—where this is more than a matter of an combination of actions of the individual parties. Examples of acting together include: conversing (or talking to one another), walking together, or painting the house together. 5. The passage quoted and the other material quoted in this paragraph is from p. 12. 6. The appearance of circularity can, I take it, be dispelled given a more fine-grained description. I intend only to focus on the fact that, generally speaking, a pertinent intention is held to be necessary to make the rising of a person P’s arm a case of P’s raising it. 7. What follows is not intended to reproduce Anscombe’s own position. 8. This discussion approximates others I have published. Some references to prior discussions that amplify various points are given in the notes below. 9. I mean to assume no particular theory of reason, having reason, and so on here. My thought is that an intuitively plausible theory will explain and, indeed, justify the judgments made in this paragraph. It would seem that there are senses of the terms ‘‘reason’’, ‘‘having reason’’, etc., such that these judgments are justified. For further discussion, and some references to the burgeoning literature on decisions, intentions, and reasons see Gilbert (2006: ch. 2). See also Bruno Verbeek, this volume (ch. 7). 10. The phrase ‘‘deliberate repudiation’’ may be pleonastic. I want, in any case, to emphasize the deliberateness of the repudiation here. 11. See, for instance, Gilbert (1999; reprinted in 2000), and Gilbert (2006: ch. 7). See also Gilbert (forthcoming). 12. Gilbert (1989: 12) and elsewhere. 13. An example of a complex case is discussed in Gilbert (2000: 23–4). See also Gilbert (2006: ch. 9), and elsewhere. 14. Are plural subjects themselves in some sense bodies or persons? That question goes beyond what I am saying here—which is not to say that it lacks an answer. In characterizing the generalized content of a joint commitment, I am drawing on an unanalyzed, intuitive concept of a

collective intentions and commitment 277 body (and, in the alternative formulation, on a general intuitive concept of a person). That is, I am not operating with any fine-grained theory of bodies or persons. A plural subject, as said, is constituted by a set of jointly committed people, the content of their joint commitment being of the form stated in the text above: they are together to emulate as far as is possible a single body (or person) that does the thing in question. Whether or not a plural subject constitutes a body or person in some intuitive sense according to one or another fine-grained theory of either is something I leave open. Different theorists may, evidently, give different answers. (Note added in response to a query from the editors.) 15. For more fine-grained detail see Gilbert (2003b). 16. ‘‘Common knowledge’’ is a term of art deriving originally from the philosopher David Lewis, and (later but independently), the economist Robert Aumann. For Lewis’s discussion, see Lewis (1969). Informally, something is common knowledge between two people if it is ‘‘out in the open’’ between them. For some discussion of the topic in connection with the creation of a plural subject, including the idea that matters go ‘‘beyond infinity’’ see Gilbert (1989). 17. Cf. Gilbert (2006: ch. 11), agreeing with Simmons (1996): there is an important distinction between the generation of expectations in others through regularities in behavior and the generation of rights to conforming behavior. 18. See Gilbert (1989: ch. 4) on tendentious and initiatory uses of ‘‘we’’. 19. See Gilbert (2000: 367-8) for an example of a similar practice that grows and then wanes. 20. For discussion see Gilbert (2000: ch. 4), and Gilbert (2006: ch. 7). 21. See Gilbert (2000: ch. 9) for a critical discussion of alternative accounts from Michael Bratman, John Searle, Raimo Tuomela, and David Velleman, which emphasizes this point. Gilbert (1997) includes some discussion of Michael Bratman’s ‘interlocking personal intentions’ approach. 22. On the standard approaches in moral philosophy see Gilbert (2004). See also Gilbert (2006: ch. 10). 23. By ‘‘effectively,’’ here, I mean ‘‘in such a way as to ensure that the joint commitment is fulfilled.’’

278 margaret gilbert 24. For further discussion see esp. Gilbert (1989: ch. 4, 2000: ch. 4, 2003b, 2006: chs. 6 and 7). For critiques of other approaches see Gilbert (2000: ch. 9). See also Gilbert (2002), which contains a sustained critical discussion of the approach of Christopher Kutz (which is close to that of Seumas Miller), and the general discussion in Gilbert (2001). 25. This paragraph responds to a query from an anonymous referee for the Press, who offered the example I discuss. Space constraints forbid a fuller discussion. 26. Related material has been presented in several places including an Oxford seminar run by the late Michael Bacharach and Gerry Mackie, in 1997, the expert seminar on rationality and intentions, University of Amsterdam, 1999, the conference on values at the Swedish Collegium for Advanced Study in the Social Sciences, Uppsala, and at St Gallen University, 2004. The bulk of the present paper was written when I was a fellow at SCASSS in the spring semester 2004. Warm thanks to that institution for its hospitality, to the late Michael Robins and to David Gauthier for their ‘‘official’’ comments on the presentation in Amsterdam, to John Broome, Sten Nyberg, and Wlodek Rabinowicz, fellow Fellows at SCASSS, for relevant conversation, and to all discussants on the occasions mentioned. Thanks also to the editors of this volume and the two referees for Oxford University Press, for comments, queries, and suggestions, and to Gabriele Taylor for letting me try out on her, long ago, the ideas presented in section 8 of this essay. That said, I should emphasize that responsibility for any claims made here is my own. References Anscombe, Elizabeth. 1957. Intention. Oxford: Blackwell. Frank, Robert. 1988. Passions within Reason: The Strategic Role of the Emotions, New York: W.W. Norton. Gilbert, Margaret. 1987. ‘‘Modeling Collective Belief.’’ Synthese 73: 185–204. Gilbert, Margaret. 1989. On Social Facts. Princeton: Princeton University Press. Gilbert, Margaret. 1996. Living Together: Rationality, Sociality, and Obligation. Lanham: Rowman and Littlefield. Gilbert, Margaret. 1997. ‘‘What Is It for Us to Intend?’’ In G. Holmstrom-Hintikka and R. Tuomela (eds.) Contemporary Action Theory, vol. 2, Dordrecht: D. Reidel, pp. 65–85.

collective intentions and commitment 279 Gilbert, Margaret. 1999. ‘‘Obligation and Joint Commitment.’’ Utilitas 11: 143–63. Gilbert, Margaret. 2000. Sociality and Responsibility: New Essays in Plural Subject Theory. Lanham: Rowman and Littlefield. Gilbert, Margaret. 2001. ‘‘Joint Action.’’ In N. J. Smelser and P. B. Baltes (eds.), International Encyclopedia of the Social and Behavioral Sciences, vol. 12, Oxford: Elsevier. Gilbert, Margaret. 2002. ‘‘Collective Wrongdoing: Moral and Legal Responses.’’ Social Theory and Practice 28/1: 167–87. Gilbert, Margaret. 2003a. Marcher Ensemble: Essais sur les Fondements de la Vie Collective. Paris: Presses Universitaires de France. Gilbert, Margaret. 2003b. ‘‘The Structure of the Social Atom: Joint Commitment as the Foundation of Human Social Behavior.’’ In Frederick Schmitt (ed.), Socializing Metaphysics. Lanham: Rowman and Littlefield, pp. 39–64. Gilbert, Margaret. 2004. ‘‘Scanlon on Promissory Obligation: The Problem of Promisees’ Rights.’’ Journal of Philosophy 101/2: 83–109. Gilbert, Margaret. 2005a. ‘‘Shared Values, Social Unity, and Liberty.’’ Public Affairs Quarterly 19: 25–50. Gilbert, Margaret. 2005b. ‘‘Towards a Theory of Commitments of the Will: On the Nature and Normativity of Intentions and Decisions.’’ In Toni RonnowRasmussen and Wlodek Rabinowicz (eds.), Patterns of Value II. Lund: Lund University. Gilbert, Margaret. 2006. Theory of Political Obligation: Membership, Commitment, and the Bonds of Society. Oxford: Oxford University Press. Gilbert, Margaret. Forthcoming. Rights Reconsidered. Oxford: Oxford University Press. Hollis, Martin and Robert Sugden. 1993. ‘‘Rationality in Action.’’ Mind 102: 1–35. Lewis, David. 1969. Convention: A Philosophical Study. Cambridge, Mass.: Harvard University Press. Schelling, Thomas. 1960. The Strategy of Conflict. Oxford: Oxford University Press. Sen, Amartya.1977. ‘‘Rational Fools: A Critique of the Behavioral Foundations of Economic Theory.’’ Philosophy and Public Affairs 6/4: 317–44. Simmons, A. John. 1996. ‘‘Associative Political Obligations.’’ Ethics 106/2: 247–73.

12 Theories of Team Agency NATA L I E G O L D A ND RO B E RT S UG D E N ∗

In decision theory, it is almost universally presupposed that agency is invested in individuals: each person acts on her own preferences and beliefs. A person’s preferences may take account of the effects of her actions on other people; she may, for example, be altruistic or have an aversion to inequality. Still, these are her preferences, and she chooses what she most prefers. Opposing this orthodoxy is a small body of literature which allows teams of individuals to count as agents, and which seeks to identify distinctive modes of team reasoning that are used by individuals as members of teams. This idea has been around for some time, having been proposed in different forms by David Hodgson (1967), Donald Regan (1980), Margaret Gilbert (1989), Susan Hurley (1989), Robert Sugden (1993, 2003), Martin Hollis (1998), and Michael Bacharach (1999, 2006). Closely related, but less directly concerned with decision theory, is the literature of collective intentions, exemplified by the work of Raimo Tuomela and Kaarlo Miller (1988), John Searle (1990), and Michael Bratman (1993). These ideas have yet to capture the attention of mainstream decision theory. There seems to be a suspicion either that team reasoning is a particular case of individual reasoning, distinguished only by the particular assumptions it makes about preferences, or that it is not reasoning in the true sense of the word. The main contribution of the present paper is to represent team reasoning explicitly, as a mode of reasoning in which propositions are manipulated according to well-defined rules—an approach that has previously been used by Natalie Gold and Christian List (2004). Our basic building block is the concept of a schema of practical reasoning, in which conclusions about what actions should be taken are inferred from explicit

theories of team agency 281 premises about the decision environment and about what agents are seeking to achieve. We use this theoretical framework to compare team reasoning with the individual reasoning of standard decision theory, and to compare various theories of team agency and collective intentionality.

1. Two puzzles of game theory One motivation for theories of team reasoning is that there are games that are puzzles for orthodox decision theory, in the sense that there exists some strategy that is at least arguably rational and that a substantial number of people play in real life, but whose rationality decision theory cannot explain and whose play it cannot predict. In this paper, we focus on two such puzzles, and show how the theory of team reasoning can resolve them. The first puzzle is the Prisoner’s Dilemma, shown in figure 12.1. In specifying the payoffs of this game, we require only that they are symmetrical between the players and that they satisfy two inequalities. The inequality a > b > c > d encapsulates the central features of the Prisoner’s Dilemma: that, for each player, the best outcome is that in which he chooses defect and his opponent chooses cooperate; the outcome in which both choose cooperate is ranked second; the outcome in which both choose defect is ranked third; and the outcome in which he chooses cooperate and his opponent chooses defect is the worst of all. The inequality b > (a + d)/2 stipulates that each player prefers a situation in which both players choose cooperate to one in which one player chooses cooperate and the other chooses defect, each player being equally likely to be the freerider. This condition is usually treated as a defining feature of the Prisoner’s Dilemma.

cooperate defect a > b > c > d; b > (a + d )/2 Player 1

Player 2 cooperate defect b,b d,a a,d c,c

Figure 12.1. The Prisoner’s Dilemma

For each player, defect strictly dominates cooperate. Thus, in its explanatory form, conventional game theory predicts that both players will choose

282

natalie gold and robert sugden

defect. In its normative form, it recommends defect to both players. Yet both would be better off if each chose cooperate instead of defect. Is that a puzzle? If, in fact, almost all human players of Prisoner’s Dilemma games chose defect, and if they construed this choice as rational, it might reasonably be argued that there was nothing to be puzzled about. It would just be an unfortunate fact about rationality that the actions of rational individuals can combine to produce outcomes that, from every individual’s point of view, are sub-optimal. But the truth is that, in experiments in which people play the Prisoner’s Dilemma for money, anonymously and without repetition, the proportion of participants choosing cooperate is typically between 40 and 50 per cent (Sally 1995). If one describes the game to ordinary people (or, indeed, to philosophers or to social scientists who have not been trained in economics), one finds a similar division of opinion about what a rational player ought to do. While some people find it completely obvious that the rational choice is defect, others are equally convinced that rationality requires each player to choose cooperate. The Prisoner’s Dilemma poses practical problems for us collectively, as citizens. Economic and social life constantly throws up real games of the Prisoner’s Dilemma type. (Think of individuals’ decisions about whether to vote in elections, whether to contribute to fund-raising appeals for public goods, whether to reduce consumption of carbon fuels, and so on.) It would be better for all of us if each of us were disposed to be cooperative in such games. The evidence shows that some people do act on this disposition in some circumstances. If we understood better what factors induced cooperation, we might find ways of structuring the social environment so as to make cooperation more common. The Prisoner’s Dilemma also poses a problem for explanatory game theory. Conventional game theory predicts that players will always choose defect, while in fact many players choose cooperate: the theory is failing to explain observed behaviour in games. There is a parallel problem for normative game theory. The theory prescribes defect, but many people have the strong intuition that cooperate is the rational choice. Of course, it is open to the game theorist to argue that that intuition is mistaken, and to insist on the normative validity of the standard analysis. In doing so, the game theorist can point out that any individual player of the Prisoner’s Dilemma does better by choosing defect than by choosing cooperate, irrespective of the behaviour of her opponent. In other words, each individual player can

theories of team agency 283 reason to the conclusion: ‘The action that gives the best result for me is defect.’ But, against that, it can be said with equal truth that the two players of the game both do better by their both choosing cooperate than by their both choosing defect. Thus, each player can also reason to the conclusion: ‘The pair of actions that gives the best result for us is not (defect, defect).’1 It seems that normative argument between these two positions leads to a stand-off. The second puzzle is the game of Hi-Lo. A Hi-Lo game is a game in which each of two players chooses one element from the same set of labels, the pair of payoffs is (ai , ai ) if both choose the same label i (with ai > 0 and (0, 0) otherwise), and there is one label j such that aj is strictly greater than every other ai . Figure 12.2 shows a simple version of Hi-Lo, in which there are just two labels, high and low. Player 2 Player 1

high low

high a, a 0, 0

low 0, 0 b, b

a>b>0

Figure 12.2. Hi-Lo

Hi-Lo combines features of pure coordination games2 and the Prisoner’s Dilemma. Like a pure coordination game, this is a common interest game—that is, a game in which the interests of the players are perfectly aligned, signalled by the fact that, in each cell of the payoff matrix, the two players’ payoffs are equal to one another. There are two pure-strategy Nash equilibria, each associated with a different label and coming about if both players choose that label. In this sense, Hi-Lo poses a coordination problem: each player wants it to be the case that they both choose the same label. The crucial difference from a pure coordination game is that, in Hi-Lo, one of the equilibria is strictly better than the other for both players. At first sight, this makes the coordination problem in Hi-Lo trivial: it seems obvious that the players should coordinate on the equilibrium they both prefer, namely (high, high). Hi-Lo shares with the Prisoner’s Dilemma the feature that, of the outcomes that occur if both players choose the same label, one is better than the other for both players. In this sense, Hi-Lo poses a cooperation problem: both players benefit by their both choosing high rather than

284


low just as, in the Prisoner’s Dilemma, both players benefit by their both choosing cooperate rather than defect. The difference is that in the Prisoner’s Dilemma, (cooperate, cooperate) is not a Nash equilbrium while in Hi-Lo (high, high) is. It might seem that, because of this difference, the cooperation problem in Hi-Lo is trivial too. Certainly, Hi-Lo does not pose practical problems for ordinary people, either individually or collectively. In experiments in which participants play Hi-Lo games, and in which the high and low strategies are given neutral labels, the overwhelming majority choose high.3 But Hi-Lo presents a fundamental problem for game theory. From the assumptions that the players are perfectly rational (in the normal sense of maximizing expected payoff) and that they have common knowledge of their rationality, we cannot deduce that each will choose high. Or, expressing the same idea in normative terms, there is no sequence of steps of valid reasoning by which perfectly rational players can arrive at the conclusion that they ought to choose high. Many people find this claim incredible, but it is true. It is true because, from the assumption of rationality, all we can infer is that each player chooses the strategy that maximizes her expected payoff, given her beliefs about what the other player will do. All we can say in favour of high is that, if either player expects the other to choose high, then it is rational for the first player to choose high too; thus, a shared expectation of high-choosing is self-fulfilling among rational players. But exactly the same can be said about low. Intuitively, it seems obvious that each player should choose high because both prefer the outcome of (high, high) to that of (low, low); but that ‘because’ has no standing in the formal theory.4 If we are prepared to relax the classical assumption of perfect rationality, it is not particularly difficult to construct theories which purport to explain the choice of high. After we have stripped out any information contained in their labels, the only difference between the high and low strategies is that high is associated with higher payoffs; because of this, most plausible theories of imperfect rationality predict that high is more likely to be chosen than low.5 But it seems unsatisfactory to have to invoke assumptions about imperfections of rationality in order to explain behaviour in such a transparently simple game as Hi-Lo. If we find that standard game-theoretic reasoning cannot tell players how to solve the apparently trivial problem of coordination and cooperation posed by Hi-Lo, we may begin to suspect that something is fundamentally wrong with the whole analysis of coordination

theories of team agency 285 and cooperation provided by the standard theory. Conversely, if we could find a form of reasoning which recommends high in Hi-Lo, that might provide the key to solving the problem posed by the Prisoner’s Dilemma. The source of both puzzles seems to be located in the mode of reasoning by which, in the standard theory, individuals move from preferences to decisions. In the syntax of game theory, each individual must ask separately ‘What should I do?’ In Hi-Lo, the game-theoretic answer to this question is indeterminate. In the Prisoner’s Dilemma, the answer is that defect should be chosen. Intuitively, however, it seems possible for the players to ask a different question: ‘What should we do?’ In Hi-Lo, the answer to this question is surely: ‘Choose (high, high)’. In the Prisoner’s Dilemma, ‘Choose (cooperate, cooperate)’ seems to be at least credible as an answer. Theories of team agency try to reformulate game theory in such a way that ‘What should we do?’ is a meaningful question. The basic idea is that, when an individual reasons as a member of a team, she considers which combination of actions by members of the team would best promote the team’s objective, and then performs her part of that combination. The rationality of each individual’s action derives from the rationality of the joint action of the team.

2. Simple team reasoning In propositional logic, a rule of inference—a rule that allows us to derive conclusions from premises—is valid if, whenever the premises are true, so are the conclusions that are derived from them. Here is a simple example of valid reasoning (the propositions above the line are premises, while the proposition below the line is the conclusion): (1) There are no English mountains over 1000 metres high. (2) Snowdon is a mountain which is 1085 metres high. Snowdon is not in England. One can formulate principles of practical reasoning—that is, reasoning that leads to conclusions about what an agent should do—which satisfy analogous criteria of validity. Bacharach (2000) defines a mode of reasoning as valid in games if it is success-promoting: given any game of some very broad class, it yields only choices which tend to produce success, as measured

286


by game payoffs. The fundamental idea is that practical reasoning infers conclusions about what an agent ought to do from premises which include propositions about what the agent is seeking to achieve. Such reasoning is instrumental in that it takes the standard of success as given; its conclusions are propositions about what the agent should do in order to be as successful as possible according to that standard. If the agent is an individual person, the reasoning is individually instrumental. Schema 12.1 shows a simple example of individually instrumental reasoning. Schema 12.1: Individual rationality (1) I must choose either left or right. (2) If I choose left, the outcome will be O1 . (3) If I choose right, the outcome will be O2 . (4) I want to achieve O1 more than I want to achieve O2 . I should choose left. In our analysis, we will interpret the payoffs of a game as specifying what the players want to achieve as individuals or, equivalently, what counts as success for them. Following the conventions of game theory, we will treat payoffs as utility indices in the sense of expected utility theory so that, in situations of uncertainty, a player’s success is measured by the expected value of her payoff. Thus, the following individually instrumental reasoning for Player 1 in the Hi-Lo game of figure 12.2 is valid: (1) I am Player 1 in Hi-Lo. (2) The probability that Player 2 will choose high is 0.5. I should choose high. This is an example of the kind of best-reply reasoning that is analysed in classical game theory. Of course, if we assume only that the players of Hi-Lo have common knowledge of the payoffs of the game and of their rationality, premise (2) is not available to Player 1 and so, although the schema we have described is valid, Player 1 cannot use it to get to the conclusion ‘I should choose high’. But now consider schema 12.2, in which (left, right) denotes the pair of actions ‘I choose left, you choose right’:

theories of team agency 287 Schema 12.2: Collective rationality (1) We must choose one of (left, left), (left, right), (right, left), or (right, right). (2) If we choose (left, left) the outcome will be O1 . (3) If we choose (left, right) the outcome will be O2 . (4) If we choose (right, left) the outcome will be O3 . (5) If we choose (right, right) the outcome will be O4 . (6) We want to achieve O1 more than we want to achieve O2 , O3 , or O4 . We should choose (left, left). Is this schema valid? Given the symmetries between schemata 1 and 2, it seems that, if one is valid, so too is the other. Yet the two schemata seem to be potentially contradictory. For example, consider the Prisoner’s Dilemma. Using a variant of schema 12.1, each player can reason to ‘I should choose defect’; but using a variant of schema 12.2, each can reason to ‘We should not choose (defect, defect)’. (In the second case, the reason for not choosing (defect, defect) is that the outcome resulting from (defect, defect) is one that we want less than we want the outcome of (cooperate, cooperate).) We suggest that the resolution of this problem is that, properly understood, the two sets of premises are mutually inconsistent.6 The premises of schema 12.1 presuppose that I am an agent, pursuing my objectives. Those of schema 12.2 presuppose that we make up a single unit of agency, pursuing our objectives. But instrumental practical reasoning presupposes a unit of agency. If I am to reason instrumentally, I cannot simultaneously think of myself both as a unit of agency in my own right and as part of a unit of agency which includes you.7 We can make this feature of practical reasoning more transparent by writing schemata in forms which include premises about agency. Consider any situation in which each of a set S of individuals has a set of alternative actions, from which he must choose one.8 A profile of actions assigns to each member of S one element of his set of alternative actions. For each profile, there is an outcome, understood simply as the state of affairs that comes about (for everyone) if those actions are chosen. We define a payoff function as a function which assigns a numerical value to every outcome. A payoff function is to be interpreted as representing what some specific agent wants to achieve: if one outcome has a higher numerical value than another, then the relevant agent wants to achieve the first more than he (or she, or it)

288


wants to achieve the second. Now consider any individual i, and any set of individuals G, such that i is a member of G and G is a weak subset of S. We will say that i identifies with G if i conceives of G as a unit of agency, acting as a single entity in pursuit of some single objective.9 Finally, we define common knowledge in the usual way: a proposition x is common knowledge in a set of individuals G if: (i) x is true; (ii) for all individuals i in G, i knows x; (iii) for all individuals i and j in G, i knows that j knows x; (iv) for all individuals i, j, and k in G, i knows that j knows that k knows that x; and so on. Letting A stand for any profile and U for any payoff function, consider the schema 12.3. Schema 12.3: Simple team reasoning (from a group viewpoint) (1) We are the members of S. (2) Each of us identifies with S. (3) Each of us wants the value of U to be maximized. (4) A uniquely maximizes U.

Each of us should choose her component of A. This schema captures the most basic features of team reasoning. Notice that, because of (2), the schema does not yield any conclusions unless all the members of S identify with this group. Because of (4), the schema yields conclusions only when a profile that is the unique maximizer of the team payoff function exists. We will not address the question of what a team reasoner should do when this is not the case but, for our purposes, the answer is not essential. Notice also that we can apply schema 12.3 in cases in which S contains only one individual. In this case, S can be written as {myself}. (1) then becomes ‘I am the only member of the set {myself}’. (2) reduces to ‘I identify with {myself}’, which amounts to saying that the reasoning individual views herself as an agent. And then the schema represents straightforward practical reasoning by an individual agent. Thus, schema 12.3 encompasses both individual and team reasoning. Schema 12.3 represents a mode of reasoning that can be used by people as a group. What does it mean for a number of people to reason as a group? One way to make sense of this is to imagine those people in an open meeting, at which each of a set of premises is announced, and acknowledged as true by each person. Then, the inference to be drawn from those premises is

theories of team agency 289 announced, and acknowledged as valid by each person. In such a setting, it is common knowledge among the members of the group that each of them accepts the relevant premises. That this is common knowledge does not need to be stated explicitly in the schema; it is not an additional premise, but a presupposition of the whole idea of reasoning as a group. For many purposes, however, it is more convenient to represent team reasoning from the viewpoint of an individual team member. If we adopt this approach, schema 12.3 can be rewritten as in schema 12.4. Schema 12.4: Simple team reasoning (from an individual viewpoint) (1) I am a member of S. (2) It is common knowledge in S that each member of S identifies with S. (3) It is common knowledge in S that each member of S wants the value of U to be maximized. (4) It is common knowledge in S that A uniquely maximizes U.

I should choose my component of A. We now consider the implications of schema 12.4 for Hi-Lo and the Prisoner’s Dilemma, on the assumption that, in each game, it is common knowledge that each player identifies with the two-player group {Player 1, Player 2}. This assumption is used merely as a convenient starting point; later, we will relax it. First, consider Hi-Lo. In order to apply schema 12.4 we need to define a payoff function U to represent what each individual wants to achieve, given that she identifies with {Player 1, Player 2}. We shall assume that, when a player identifies with a group, she wants to promote the combined interests of its two members, at least insofar as those interests are affected by the game that is being played. Thus, the values of U can be interpreted as measures of the welfare of the group {Player 1, Player 2}. Since the two players’ payoffs are equal, irrespective of which actions are chosen, it is natural to make the values of U equal to the players’ common payoffs. Then (high, high) is the profile that uniquely maximizes U, and so (provided there is common knowledge of the rules of the game), each player can use schema 12.4 to reach the conclusion that she should choose high.

290


Now, consider the Prisoner’s Dilemma. Again, we need to define a payoff function U for the group {Player 1, Player 2}. If we assume that U treats the players symmetrically, we need to specify only three values of this function: the payoff when both players choose cooperate, which we denote uC , the payoff when both choose defect, which we denote uD , and the payoff when one chooses cooperate and one chooses defect, which we denote uF (for ‘freeriding’). It seems unexceptionable to assume that U is increasing in individual payoffs, which implies uC > uD . Given the condition b > (a + d)/2, it is natural also to assume uC > uF . Then the profile of actions by Player 1 and Player 2 that uniquely maximizes U is (cooperate, cooperate). If there is common knowledge of the rules of the game, each player can use schema 12.4 to reach the conclusion that she should choose cooperate.

3. Is team reasoning necessary to solve the Prisoner’s Dilemma? In the analysis we have just outlined, a rational player of the one-shot Prisoner’s Dilemma can choose cooperate. For many game theorists, this conclusion is close to heresy. For example, Ken Binmore (1994: 102–17, quotation from p. 114) argues that it can be reached only by ‘a wrong analysis of the wrong game’: if two players truly face the game shown in figure 12.1, then it follows from the meaning of ‘payoff’ and from an unexceptionable concept of rationality that a rational player must choose defect. His argument works as follows. Consider Player 1. She knows that her opponent must choose either cooperate or defect. The inequality a > b tells us that, if Player 1 knew that Player 2 would choose cooperate, Player 1 would want to choose, and would choose, defect. The inequality c > d tells us that, if Player 1 knew that Player 2 would choose defect, Player 1 would want to choose, and would choose, defect. So (Binmore concludes) we need only a principle of dominance to conclude that, whatever Player 1 believes about what Player 2 will do, Player 1 should choose defect. Binmore recognizes that rational individuals may sometimes choose cooperate in games in which material payoffs—that is, outcomes described in terms of units of commodities which people normally prefer to have more of rather than less, such as money, or years of not being in prison—are as in figure 12.1.

theories of team agency 291 But that just shows that the payoffs that are relevant for game theory—the payoffs that govern behaviour—differ from the material ones. The first stage in a game-theoretic analysis of a real-life situation should be to find a formal game that correctly represents that situation. Thus, in response to the problem of explaining why cooperate is sometimes chosen in games whose material payoffs have the Prisoner’s Dilemma structure, the methodological strategy advocated by Binmore is that of payoff transformation: we should look for some way of transforming material payoffs into game-theoretic ones which makes observed behaviour consistent with conventional game-theoretic analysis. It has been followed by various theorists who have proposed transformations of material payoffs to take account of psychological or moral motivations that go beyond simple self-interest. One of the earliest proposals of this kind was made by Amartya Sen (1974, 1977). Sen distinguishes between ‘rationality’ (as this is usually understood in economics) and ‘morality’. He points out that many different codes of morality would prescribe cooperation in the Prisoner’s Dilemma. More generally, many moral codes value actions which ‘sacrific[e] some individual gain—given the action of others—for the sake of a rule of good behaviour by all which ultimately makes everyone better off’. Sen proposes an approach in which this core moral principle is ‘expressed in the form of choice between preference patterns rather than between actions’ (1974: 77–8). His idea is to represent different attitudes towards behaviour in a given game as different orderings over the outcomes of the game (or over the strategy profiles that generate those outcomes). Some of these orderings are egoistic, but others are not. Although a person’s moral principles are described by a meta-ranking of the set of alternative orderings, her actual behaviour is explained by whichever of those orderings she chooses to act on. A person who chooses to act on non-egoistic preferences is said to act on commitment, and her choices are said to be counterpreferential (1977: 91–3). Nevertheless, Sen’s account of what is involved in such action retains the formal structure of conventional decision and game theory: the unit of agency is the individual, and each individual’s actions are governed by some ordering over outcomes. In game-theoretic terms, commitment induces a transformation from egoistic payoffs to ‘counterpreferential’ ones. More recently, and in slightly different ways, Ernst Fehr and Klaus Schmidt (1999) and Gary Bolton and Axel Ockenfels (2000) have proposed

292


that, for any given level of material payoff for any individual, that individual dislikes being either better off or worse off than other people. Matthew Rabin (1993) proposes that each individual likes to benefit people who act with the intention of benefiting him, and likes to harm people who act with the intention of harming him.10 The theory of team reasoning can accept Binmore’s instrumental conception of rationality, but rejects his implicit assumption that agency is necessarily vested in individuals. We can interpret the payoffs of a game, as represented in a matrix like that of figure 12.1, as showing what each player wants to achieve if she takes herself to be an individual agent. In this sense, the interpretation of the payoffs is similar to that used by Binmore: payoffs are defined, not in material terms, but in terms of what individuals are seeking to achieve. The theory of team reasoning can replicate Binmore’s analysis when it is applied to players who take themselves to be individual agents: if Player 1 frames the game as a problem ‘for me’, the only rational choice is defect. However, the theory also allows the possibility that Player 1 frames the game as a problem ‘for us’. In this case, the payoffs that are relevant in determining what it is rational for Player 1 to do are measures of what she wants to achieve as a member of the group {Player1, Player2}; and these need not be the same as the payoffs in the standard description of the game. Thus, there is a sense in which team reasoning as an explanation of the choice of cooperate in the Prisoner’s Dilemma depends on a transformation of payoffs from those shown in figure 12.1. However, the kind of transformation used by theories of team reasoning is quite different from that used by theorists such as Fehr and Schmidt. In team reasoning, the transformation is not from material payoffs to choice-governing payoffs; it is from payoffs which govern choices for one unit of agency to payoffs which govern choices for another. Thus, payoff transformation takes place as part of a more fundamental agency transformation. One might wonder whether we need to transform both payoffs and agency. If payoffs have been transformed so that they represent the welfare of the two players as a group, doesn’t conventional game theory provide an explanation of why each individual chooses cooperate? Not necessarily. Consider a Prisoner’s Dilemma in which a = 10, b = 8, c = 6 and d = 0, and assume that the value of the payoff function for the group {Player 1, Player 2} is the average of the payoffs for the two individuals. Then we have uC = 8, uD = 6, and uF = 5. If we treat Player 1 and Player 2 as individual

theories of team agency 293 agents, each of whom independently seeks to maximize the value of U, we have the game shown in figure 12.3. The structure of this game will be familiar: it is a variant of Hi-Lo, in which cooperate corresponds with high and defect with low. Conventional game theory does not show that rational players of this game will choose cooperate. To show that, we need a transformation of the unit of agency.

Player 1

cooperate defect

Player 2 cooperate defect 8, 8 5, 5 5, 5 6, 6

Figure 12.3. A Prisoner’s Dilemma with transformed payoffs

By using the concept of agency transformation, team reasoning is able to explain the choice of high in Hi-Lo. Existing theories of payoff transformation cannot do this. Further, it is hard to see how any such theory could credibly make (high, high) the unique solution of Hi-Lo. Let us interpret the Hi-Lo payoffs as material payoffs, and consider possible transformations. Suppose that, following Fehr and Schmidt and Bolton and Ockenfels, we introduce assumptions about players’ attitudes towards the distribution of material payoffs. In every possible outcome, the two players’ material payoffs are equal. Whatever the players’ attitudes to inequality, it seems that their subjective ranking of the outcomes must correspond with the ranking of material payoffs. Thus, a game which is Hi-Lo in material payoffs will remain Hi-Lo after payoff transformation. Alternatively, suppose we follow Rabin and assume that each player wants to reciprocate other players’ ‘kindness’ or ‘unkindness’ towards him. In a situation in which both players choose low, Player 1 is benefiting Player 2 to the maximum degree possible, given Player 2’s action; and vice versa. So each is reciprocating the other’s ‘kindness’. Reciprocity in Rabin’s sense does not affect the equilbrium status of (low, low). One might reasonably expect a theory of rational choice to account for the intuition that high is the rational choice in a Hi-Lo game, and to explain why this strategy is in fact chosen by apparently rational players. The theory of team reasoning meets this requirement. Once the components of this theory are in place, very little more is needed to explain the choice of cooperate in the Prisoner’s Dilemma. All that is needed in addition is the assumption that (cooperate, cooperate) is the best profile of actions for the two players together.

294


That assumption is hardly controversial: it is presupposed in most accounts of the significance of the Prisoner’s Dilemma—whether that is understood as a puzzle for game theory or as a model of real-world problems of cooperation.

4. Comparing theories of team agency The reasoning represented by schema 12.4 is the common core of theories of team agency. However, there are various theories of team agency, which differ in important ways. They differ in their hypotheses about how teams are formed, or how individual agents come to identify with groups. Group formation has been claimed to be: a requirement of rationality/morality, a response to the psychological impetus of framing, the result of explicit mutual commitment, or a consequence of non-rational assurance. Further, in schema 12.4, there is common knowledge within the group S that each member of S identifies with S. In some situations, this is not a realistic assumption. It may be the case that some members of S (that is, the group with which team reasoners identify) do not identify with S. In this case, we can define the team T as those members of S who do identify with S. The various theories differ in how they recommend that members of T should reason in such cases. 4.1. Team agency required by rationality/morality The first theorists to discuss team reasoning did so in the context of moral and rational requirements on action. Hodgson (1967) was the first person to use the Hi-Lo game, as part of an argument that rule utilitarianism does not reduce to act utilitarianism. Regan (1980) proposed a form of team reasoning in his theory of cooperative utilitarianism. Regan’s theory is normative; it is commended to all of us in our capacities as rational and moral agents. The fundamental principle of this theory is that ‘what each agent ought to do is to co-operate, with whoever else is co-operating, in the production of the best consequences possible given the behaviour of non-co-operators’ (1980: 124). For the moment, consider a world where everyone is a rule utilitarian. In that case, Regan’s rational and moral agents reason according to schema 12.4. In a similar vein, Hurley (1989: 136–59) proposes that we (as rational and moral agents) should specify agent-neutral goals—that is, goals of which it can simply be said that they ought to

theories of team agency 295 be pursued, rather than they ought to be pursued by some particular agent. Then we should ‘survey the units of agency that are possible in the circumstances at hand and ask what the unit of agency, among those possible, should be’; and we should ‘ask ourselves how we can contribute to the realization of the best unit possible in the circumstances’. Regan’s theory gives recommendations for cases in which not everyone is a cooperative utilitarian. The logic of these recommendations can be represented by a variant of team reasoning called restricted team reasoning by Bacharach (2006). This applies to cases in which it is known that certain specific members of S do not identify with S. It is formalized in schema 12.5. Let AT be a profile of actions for the members of T. Then: Schema 12.5: Restricted team reasoning (1) I am a member of T. (2) It is common knowledge in T that each member of T identifies with S. (3) It is common knowledge in T that each member of T wants the value of U to be maximized. (4) It is common knowledge in T that AT uniquely maximizes U, given the actions of non-members of T.

I should choose my component of AT . In cooperative utilitarianism, each of us is told to join with as many others as are willing to do the same, and to cooperate with them in trying to achieve the overall good of all people (or perhaps the good of all sentient beings—but not just the common good of the members of T ). In terms of schema 12.5, S is the set of all people, T is the set of cooperative utilitarians, and the value of U is a utilitarian measure of overall goodness. Regan claims that a cooperative utilitarian ought to identify with S, and that she ought to want to maximize U. While Regan’s theory tells us which group we should identify with, and what we should want to maximize, Hurley does not nominate any particular group or any particular goal as being the rational one to pursue. Nevertheless, for Hurley, the idea seems to be that rationality requires each person to choose the unit of agency in which she participates, and that this choice should be governed by goals which are independent of the unit of agency.

296


4.2. Team agency as the result of framing In contrast, Bacharach’s (2006) theory does not allow the unit of agency to be chosen, and does not admit the concept of a goal that is not the goal of some agent. For Bacharach, whether a particular player identifies with a particular group is a matter of ‘framing’. A frame is the set of concepts a player uses when thinking about her situation. In order to team reason, a player must have the concept ‘we’ in her frame. Bacharach proposes that the ‘we’ frame is normally induced or primed by Hi-Lo games, but is primed less reliably by the Prisoner’s Dilemma. Both games have a property that Bacharach calls strong interdependence. Roughly, a game has this property if it has a Nash equilibrium which is Pareto-dominated by the outcome of some feasible strategy profile. (For a more formal definition, see Bacharach 2006.) Although Bacharach proposes that the perception of this property increases the probability of group identification, he does not claim that games with this property invariably prime the ‘we’ frame. More specifically: In a Prisoner’s Dilemma, players might see only, or most powerfully, the feature of common interest and reciprocal dependence which lie in the payoffs on the main diagonal. But they might see the problem in other ways. For example, someone might be struck by the thought that her coplayer is in a position to double-cross her by playing [defect] in the expectation that she will play [cooperate]. This perceived feature might inhibit group identification. (2006: chapter 2, section 4.2)

The implication is that the ‘we’ frame might be primed; but, alternatively, a player may see the game as one to be played by two separate individual agents. That either framing is psychologically possible reflects the sense in which the Prisoner’s Dilemma itself is puzzling. One the one hand, the positions of the two players are completely symmetrical, which prompts one to focus on strategy profiles in which the two players’ actions are symmetrical. Then, comparing the outcomes of (cooperate, cooperate) and (defect, defect), one sees that the two players have a common interest in their both choosing cooperate. This line of thought leads naturally to a conception of the game as a problem ‘for us’. On the other hand, if one looks at the outcomes of (cooperate, defect) and (defect, cooperate), one sees a conflict of interest between the two players: by choosing defect when one’s opponent chooses cooperate, one can gain at her expense. This line of thought leads to a conception of the game as one in which the two players are in opposition,

theories of team agency 297 each facing her own decision problem. As a metaphor or model, Bacharach often refers to the famous drawing (used in Gestalt psychology) which can be seen either as a duck or a rabbit. In the same way, the Prisoner’s Dilemma can be seen by a player either as a problem ‘for me’ or as a problem ‘for us’. Thus, we should not assume it to be common knowledge that the players of the Prisoner’s Dilemma identify with {Player 1, Player 2}. In Bacharach’s theoretical framework, this dualism is best represented in terms of circumspect team reasoning. We now present this mode of reasoning in the form of a reasoning schema. As before, let S be the set of individuals with which team-reasoners identify, and let T be any subset of S, interpreted as the set of individuals who in fact identify with S.11 Suppose there is a random process which, independently for each member of S, determines whether or not that individual is a member of T; for each individual, the probability that he is a member of T is ω, where ω > 0. We define a proposition p to be T-conditional common knowledge if: (i) p is true; (ii) for all individuals i in S, if i is a member of T, then i knows p; (ii) for all individuals i and j in S, if i is a member of T, then i knows that if j is a member of T, then j knows p; and so on. (As an illustration: imagine an underground political organization which uses a cell structure, so that each member knows the identities of only a few of her fellow members. New members are inducted by taking an oath, which they are told is common to the whole organization. Then, if T is the set of members, the content of the oath is T-conditional common knowledge.) We define a protocol as a profile of actions, one for each member of S, with the interpretation that the protocol is to be followed by those individuals who turn out to be members of T. Let P be any protocol. The schema is: Schema 12.6: Circumspect team reasoning (1) I am a member of T. (2) It is T-conditional common knowledge that each member of T identifies with S. (3) It is T-conditional common knowledge that each member of T wants the value of U to be maximized. (4) It is T-conditional common knowledge that P uniquely maximizes U, given the actions of non-members of T.

I should choose my component of P.

298


We can apply this schema to the Prisoner’s Dilemma by setting S = {Player 1, Player 2} and by defining U as before. Let ω (where 0 < ω 1) be the probability that, for any individual player of the Prisoner’s Dilemma, the ‘we’ frame comes to mind; if it does, the player identifies with {Player 1, Player 2}. Assume that, if this frame does not come to mind, the player conceives of himself as a unit of agency and thus, using best-reply reasoning, chooses the dominant strategy defect. We can now ask which protocol maximizes U, given the value of ω. Viewed from within the ‘we’ frame, the protocol (defect, defect) gives a payoff of uD with certainty. Each of the protocols (cooperate, defect) and (defect, cooperate) gives an expected payoff of ωuF + (1 − ω)uD . The protocol (cooperate, cooperate) gives an expected payoff of ω2 uC + 2ω(1 − ω)uF + (1 − ω)2 uD . There are two possible cases to consider. If uF ≥ uD , then (cooperate, cooperate) is the Umaximizing protocol for all possible values of ω. Alternatively, if uD > uF , which protocol maximizes U depends on the value of ω. At high values of ω, (cooperate, cooperate) is uniquely optimal; at low values, the uniquely optimal protocol is (defect, defect).12 If we assume either that uF ≥ uD or that the value of ω is high enough to make (cooperate, cooperate) the uniquely optimal protocol, we have a model in which players of the Prisoner’s Dilemma choose cooperate if the ‘we’ frame comes to mind, and defect otherwise. Bacharach offers this result as an explanation of the observation that, in one-shot Prisoner’s Dilemmas played under experimental conditions, each of cooperate and defect is usually chosen by a substantial proportion of players. He also sees it as consistent with the fact that there are many people who think it completely obvious that cooperate is the only rational choice, while there are also many who feel the same about defect. Bacharach can say that both sets of people are right—in the same way as two people can both be right when one says that the drawing they have been shown is a picture of a duck and the other says it is a picture of a rabbit. Bacharach claims that schema 12.6 is valid, with the implication that, for any given individual, if she identifies with S and wants U to be maximized, it is instrumentally rational for her to act as a member of T, the team of like-minded individuals. He does not claim that she ought to identify with any particular S, or that she ought to want any particular U to be maximized. In the theory of circumspect team reasoning, the parameter ω is interpreted as a property of a psychological mechanism—the probability

theories of team agency 299 that a person who confronts the relevant stimulus will respond by framing the situation as a problem ‘for us’. The idea is that, in coming to frame the situation as a problem ‘for us’, an individual also gains some sense of how likely it is that another individual would frame it in the same way; in this way, the value of ω becomes common knowledge among those who use this frame. 4.3. Team agency produced by commitment Another variety of team agency has it that a group is constituted by public acts of promising, or by public expressions of commitment by its members. This latter idea is central to Gilbert’s (1989) analysis of ‘plural subjects’. Although Gilbert is more concerned with collective attitudes than with collective action, her analysis of how a plural subject is formed might be applied to the formation of teams. There are also hints of this approach in the work of Hollis (1998). Hollis suggests that Rousseau’s (1988[1762]) account of the social contract, with its ‘most remarkable change in man’, can be understood as a transition from individual to group agency that takes place through a collective act of commitment. However, the commitment and the framing understandings of group identity may be more similar than Bacharach’s and Gilbert’s formal analyses suggest. In notes for a chapter of his book that remained unwritten at the time of his death, Bacharach (2000) records the following train of thought about the formation of teams: My current ... idea is something like this: Something in the situation prompts the parties to see that they have action possibilities which provide joint agency possibilities which have possible outcomes of common interest. Each finds herself in a frame which features concepts which describe the conceived possible actions, describe the conceived possible outcomes, and present some of these outcomes positively. Each of us sees that we could write a paper together, or have a pleasant walk round the garden together, or bring down the appalling government together. The holism of frames comes in here. One concept belonging to the frame may bring others with it, or only be activated if others are. Some actions only get conceived if one gets the idea of certain possible outcomes, and conversely. One such situation is that created by one of us being so prompted, then making a verbal suggestion to the other(s), as in ‘Would you like to dance?’

This account has many similarities with Gilbert’s, except that what Gilbert describes in the language of agreement and tacit understanding,

300


Bacharach describes in terms of framing. Bacharach’s concept of framing allows one person to choose an action with the intention of affecting someone else’s frame. While Gilbert would treat the saying of the words ‘Would you like to dance?’ as the first stage in a process which may lead to a common understanding that the two people are a plural subject in relation to a dance, Bacharach treats it as part of a process by which two people influence one another’s frames. Bacharach cannot say (as Gilbert might) that each individual chooses (or agrees, or commits himself ) to view the situation in the ‘we’ frame. But he can, and does, say that group identification tends to be primed by an individual’s recognition that the members of the putative group have common interests that can be furthered by joint action. So, although Bacharach’s individuals cannot choose to create teams with the rational intention of solving problems of coordination or cooperation, such problems tend to induce the kind of mutual adjustment of frames that he describes. Although Gilbert does not offer an explicit model of collective choice, we suggest that schema 12.4 is compatible with her general approach, provided that ‘identifying with’ the group S is understood as some kind of conscious and public act of commitment. On this interpretation, however, the schema is not one of instrumental reasoning. Rather, the rationality of acting as a member of a team derives from the rationality of fulfilling one’s commitments or intentions. Focusing on collective attitudes rather than collective actions, Gilbert claims that membership of a plural subject imposes obligations to uphold ‘our’ attitudes. This claim is conceptual rather than moral: roughly, the idea is that a plural subject is formed by an exchange of commitments, and that to make a commitment is to impose on oneself an obligation to act on it. For Gilbert, there is no problem that S (the group with which individuals identify) may be different from T (the set of individuals who in fact identify with S). In commitment-based theories, it is natural to assume that the set of individuals who act as a team is the same as the group with which they identify, provided we can assume that individuals keep their commitments. 4.4. Team agency and assurance Schemata 12.4, 12.5, and 12.6 share an important common feature. In each case, the conclusion is an unconditional proposition of the form ‘I should choose my component of the best profile’. The unconditional form

theories of team agency 301 of this conclusion is crucial in the resolution of the problem posed by Hi-Lo. In contrast, the best-reply reasoning of classical game theory leads to conclusions about what one agent should do conditional on what other agents can be expected to do. Thus, in Hi-Lo, best-reply reasoning leads only to the conclusion ‘If I expect my opponent to choose her component of the best profile, I should choose mine’, and so to an infinite regress. Schema 12.4 has an additional feature, not shared by the other two team reasoning schemata. Even though it yields an unconditional conclusion, it tells an individual member of S to choose his component of the best profile only in situations in which it also tells the other members to choose theirs. Further, these are always situations in which the other players identify with S: they are framing the decision problem as one ‘for us’. Thus, if each of them is rational, each will act on the conclusions of schema 12.4, as applied to her case. And, since it is common knowledge in S that everyone identifies with S, each player can work all this out. So, whenever schema 12.4 tells an individual to choose his component in the best profile, that individual has the assurance that the others (if rational) will choose theirs too. To put this another way: when the individual chooses his component in the best profile, he can construe this as his part of a collective action that is in fact taking place. However, this property of assurance does not carry over to the theories of restricted and circumspect team reasoning. In these theories, each member of the team T identifies with S. Thus, each member of T wants the value of U to be maximized, where U represents what people want as members of S. Each member of T is told to do his part in a joint action by T to maximize U, given the behaviour of non-members of T; he can be assured only that the other members of T will do their parts. For example, consider the Prisoner’s Dilemma under the assumption that uF > uD (that is, it is better for the two players as a group that one of them chooses cooperate and the other chooses defect than that both choose defect). Suppose that Player 1 identifies with the group S = {Player 1, Player 2}; but suppose also (the case of restricted team reasoning) that Player 1 knows that Player 2 does not identify with S, or (the case of circumspect team reasoning) that Player 1 knows that the probability that Player 2 identifies with S is close to zero. Restricted and circumspect team reasoning both lead to the conclusion that Player 1 should choose cooperate, even though he knows (or is almost certain) that Player 2 is taking a free ride.

302


On some interpretations of the concept of team reasoning, it involves an idea of reciprocity: each member of a team does her part on the understanding that others will do theirs. If team reasoning is viewed in this way, assurance may seem an essential concept. This provides the starting point for yet another interpretation of the status of team reasoning. Sugden (2003) presents a ‘logic of team reasoning’ without making any claims for its validity. In his analysis, a ‘logic’ is merely an internally consistent system of axioms and inference rules. An individual actor may endorse a particular logic, thereby accepting as true any conclusions that can be derived within it, but the theorist need not take any position about whether the axioms of that logic are ‘really’ true or whether its inference rules are ‘really’ valid. Team reasoning is then represented as a particular inference rule which, as a matter of empirical fact, many people endorse. Thus, following this approach, one might re-interpret schema 12.4 as specifying the inference rule ‘From (1), (2), (3), and (4), infer ‘‘I should choose my component of A’’ ’. On this interpretation, however, schema 12.4 does not guarantee assurance. Recall that this schema recommends an individual to choose his component of the U-maximizing profile of actions by the members of the group S, only in situations in which it also recommends the other members to choose their components. So, if it can be assumed that schema 12.4 has a validity that is transparent to all rational people, and if it can be assumed that each member of S is confident that the others are rational, then each member of S has the assurance that, when he chooses his component, the others will choose theirs. But Sugden’s approach does not acknowledge agent-neutral concepts of ‘validity’ and ‘rationality’. It maintains assurance in a different way, which we now formulate as a reasoning schema. Following David Lewis (1969) and Robin Cubitt and Sugden (2003), Sugden uses a theoretical framework in which the central concept is reason to believe. To say that a person has reason to believe a proposition p is to say that p can be inferred from propositions that she accepts as true, using rules of inference that she accepts as valid. On the analogue of the definition of common knowledge, there is common reason to believe a proposition p in a set of individuals T if: (i) for all individuals i in T, i has reason to believe p; (ii) for all individuals i and j in T, i has reason to believe that j has reason to believe p; (iii) for all individuals i, j, and k in T, i has reason to believe that j has reason to believe that k has reason to believe p; and so on.

theories of team agency 303 The following definition is also useful. Within a set of individuals T, there is reciprocal reason to believe that some property q holds for members of T if (i) for all individuals i and j in T, where i = j, i has reason to believe that q holds for j; (ii) for all individuals i, j, and k in T, where i = j and j = k, i has reason to believe that j has reason to believe that q holds for k; and so on. To see the point of this latter definition, consider the Prisoner’s Dilemma and let q be the property ‘chooses cooperate’. In a schema of practical reasoning which is intended to be used by Player 1 in deciding how to play the Prisoner’s Dilemma, we cannot allow the premise that, in the group {Player 1, Player 2}, there is common reason to believe that Player 1 chooses cooperate. That would make it a premise that Player 1 has reason to believe that he himself will choose cooperate, when the whole point of using the schema is to determine which action he should choose. However, we can allow the premise that there is reciprocal reason to believe that members of the group {Player 1, Player 2} choose cooperate, and there may be circumstances in which such a premise would be natural. For example, suppose that Player 1 and Player 2 have played the Prisoner’s Dilemma many times before, and on every such occasion, both have chosen cooperate. They are about to play again, and there is no obvious difference between this interaction and all its predecessors. Then, by induction, Player 2 might have reason to believe that Player 1 will choose cooperate. Attributing this reasoning to his opponent, Player 1 might have reason to believe that Player 2 has reason to believe that Player 1 will choose cooperate; and so on. Sugden’s formulation of team reasoning can be represented as shown in schema 12.7. Schema 12.7: Mutually assured team reasoning (1) I am a member of S. (2) I identify with S and acknowledge U as its objective. (3) In S, there is reciprocal reason to believe that each member of S identifies with S and acknowledges U as the objective of S. (4) In S, there is reciprocal reason to believe that each member of S endorses and acts on mutually assured team reasoning. (5) In S, there is common reason to believe that A uniquely maximizes U.

I should choose my component of A.

304


This schema is not presented as a mode of valid reasoning. It is merely a mode of reasoning that any person might (or might not) endorse; a person commits herself to team reasoning by endorsing the schema. Notice that premises (2) and (3) refer to ‘acknowledging U as the objective of S’ rather than ‘wanting U to be maximized’. On Sugden’s account, a team reasoner who identifies with a group stands ready to do her part in joint actions in pursuit of the group’s objective; but she does not necessarily take this objective as hers in the stronger sense of wanting to pursue it even if other members of the group do not reciprocate. Schema 12.7 is recursive: premise (4) refers to the endorsement of the schema itself. That this is not circular can be seen from an analogy. Consider an international treaty which includes among its conditions that it will come into effect only if and when it has been ratified by a certain number of nations; once this condition is met, it is binding on every nation that has ratified it. To ratify such a treaty is to make a commitment which is binding from that moment, but which is activated only if enough others make the same commitment. Analogously, to endorse mutually assured team reasoning is to make a unilateral commitment to a certain form of practical reasoning, but this reasoning does not generate any implications for action unless one has assurance that others have made the same commitment. Such assurance could be created by public acts of commitment of the kind considered by Gilbert. But it could also be induced by repeated experience of regularities of behaviour in a population. For example, suppose that in some population, some practice of mutual assistance (say, giving directions to strangers when asked) is generally followed in anonymous encounters. Each individual might interpret the existence of the practice as evidence that premises (3), (4), and (5) are true. If so, each individual would be assured that others would choose their components of the U-maximizing profile. But he would still have to decide whether team reasoning was a mode of reasoning that he wanted to endorse.

5. Collective intentions While the problems that motivated the literature on team agency are about why agents would take certain actions, the literature on collective intentions analyses agents’ mental states. When an agent deliberates about what she

theories of team agency 305 ought to do, the result of her reasoning is an intention. An intention is interposed between reasoning and an action, so it is natural to treat the intentions that result from team reasoning as collective intentions. An early analysis of collective intentionality is the work of Tuomela and Miller (1988). The essential features of this analysis can be presented as follows for the case of a two-member group, whose members are Player 1 and Player 2. Consider some ‘joint social action’ A which comprises actions A1 and A2 for the respective individuals. According to Tuomela and Miller, Player 1 has a we-intention with respect to A if: (i) Player 1 intends to do A1 , (ii) Player 1 believes that Player 2 will do A2 , (iii) Player 1 believes that Player 2 believes that Player 1 will do A1 , and so on (p. 375). This analysis reduces we-intentions to individual intentions and a network of mutual beliefs. An apparently unsatisfactory feature of this analysis is that it seems to treat every Nash equilibrium as a case of collective intentionality. For example, consider the version of the Hawk–Dove game shown in figure 12.4.

Player 1

dove hawk

dove 2, 2 3, 0

Player 2

hawk 0, 3 –5, –5

Figure 12.4. Hawk–Dove

As an example of such a game, think of two individuals in a state of nature who come into conflict over some valuable resource. To play dove is to offer to share the resource but to back down if the other attempts to take it all; to play hawk is to demand the whole resource, backed by a readiness to fight for it. We assume that fighting is costly for both parties, and that the utility value of a half share of the resource is greater than half of the utility value of the whole. This game has two pure strategy Nash equilibria: (hawk, dove) and (dove, hawk). Consider the first of these. Suppose it is common knowledge between Player 1 and Player 2 that, in interactions like this, the player in the position of Player 1 almost always chooses hawk and the one in the position of Player 2 almost always chooses dove. Expecting Player 2 to play dove, Player 1 forms the intention to play hawk. Expecting Player 1 to play hawk, Player 2 forms the intention to play dove. Given all this, does each

306


player have a we-intention with respect to the pair of strategies (hawk, dove)? On Tuomela and Miller’s analysis, it seems that they do. We say ‘seems that’ because Tuomela and Miller’s core analysis comes with various qualifications. In particular, it applies only to ‘joint social actions’, defined as ‘situations in which some agents act together, usually or often with the purpose of achieving some joint goal’ (p. 367); this goal is ‘normally (but not necessarily) the goal to perform the total action [in our notation, A]’ (p. 370). Tuomela and Miller also add a condition to the effect that when Player 1 performs A1 , ‘he does it in order for the participating agents to succeed in doing [A]’ (p. 376). Possibly, these conditions are intended to exclude cases like the Hawk–Dove example; but if so, how these cases are excluded remains obscure.13 Intuitively, the Hawk–Dove case does not seem to be an instance of we-intentions. At any rate, it seems unlike the examples that are treated as paradigm cases in the literature of collective intentions: two people singing a duet, two people pushing a car, two players on the same football team executing a pass. But what makes the Hawk–Dove case different? Searle (1990) tries to answer this question. He undertakes to show that we-intentions cannot be reduced to combinations of I-intentions—that we-intentions are ‘primitive’ (p. 404). He presents a critique of Tuomela and Miller’s analysis and then proposes his own. The critique is persuasive at the intuitive level but, on closer inspection, turns out to be questionbegging. Searle asserts that ‘the notion of a we-intention ... implies the notion of cooperation’ (p. 406), and construes cooperation in terms of ‘collective goals’ (pp. 405, 411). He says that, in cases of collective intentionality, individual I-intentions are ‘derivative from’ we-intentions ‘in a way we will need to explain’ (p. 403). But he offers no analysis of the concepts of ‘cooperation’ or ‘collective goal’, and the explanation of the sense in which I-intentions derive from we-intentions never materializes. Searle analyses collective intentions with reference to a case in which Jones and Smith are preparing a hollandaise sauce together, Jones stirring while Smith pours. On Searle’s analysis, Jones’s description of what is going on is ‘We make the sauce by means of Me stirring and You pouring’. The intention in Jones’s mind is: ‘We intend to make the sauce by means of Me stirring’ (p. 412). Searle suggests that the we-intention to make the sauce by means of Jones’s stirring is like an intention to fire a gun by means of pulling the trigger. The idea seems to be that the I-intention to stir is part

theories of team agency 307 of the we-intention to make the sauce. This is not quite the derivation of I-intentions from we-intentions that Searle said we needed. Whatever one makes of this analysis,14 it does not resolve the problem with which Searle began. We can still ask why, in the Hawk–Dove example, Player 1 and Player 2 don’t have a collective intention with respect to (hawk, dove). What is wrong with saying that, in Player 1’s mind, there is a we-intention to play the combination (hawk, dove) by means of Player 1 playing hawk and Player 2 playing dove? The only answer Searle’s analysis can give is that, in playing that combination of strategies, Players 1 and 2 are not ‘cooperating’ in pursuit of a ‘collective goal’. But those concepts are left unanalysed. What is missing, we suggest, is an analysis of practical reasoning in which cooperative reasoning—or reasoning about how to achieve a collective goal—can be distinguished from straightforward individual reasoning. This is what theories of team reasoning provide. Bratman (1993) offers a rather different account of collection intentionality in his analysis of ‘shared cooperative activity’. The activities that Bratman has in mind are ones in which individuals coordinate their actions over a period of time: each has to adjust his behaviour continuously so as to keep it aligned with the behaviour of the other, as in the case of two people singing a duet. Bratman explicitly rules out the kind of ‘pre-packaged cooperation’ that can be represented by the choice of strategies in a normal-form game, such as the Prisoner’s Dilemma or Hi-Lo (p. 339). He uses what he calls a ‘planning conception of intention’ (p. 330), whereby an intention is an action-guiding mental state that is maintained over an interval of time. Thus, for example, someone might have an individual intention to paint her house; this intention would then guide the formation of ‘sub-plans’ for buying paint, cleaning walls, and so on. Bratman argues that shared cooperative activities are governed in a corresponding way by continuing intentions, but in this case the intentions are collective. Collective intentions, as analysed by Bratman, reflect what he sees as the characteristic features of shared collective activity. In such an activity, a set of agents (for ease of exposition, say a pair) coordinate their actions in some joint enterprise. The successful pursuit of this enterprise requires the continuous ‘meshing’ of the separate sub-plans of the two agents. In Bratman’s analysis, each agent has the intention that ‘we’ perform the

308


joint activity through the meshing of ‘my’ sub-plans with ‘yours’. This intention is not linked to any particular combination of sub-plans; rather, it expresses a commitment to engage with the other in a process of ‘mutual responsiveness’ and ‘mutual support’ which is directed towards the meshing of sub-plans in general. Thus, Bratman’s conception of collective intentionality can be thought of as the counterpart in the domain of intentions of group identification in the domain of practical reasons: it expresses a disposition to reason and to act as a member of a group. It leaves open one of the main questions that the theory of team reasoning tries to answer: how the members of a group coordinate their actions.

6. Conclusion We have presented a number of alternative theories of team reasoning, which differ on several dimensions—in particular: how to deal with cases in which not every member of the relevant group can be relied on to identify with the group; whether group identification is a product of psychological framing or conscious commitment; whether each individual’s engaging in team reasoning is conditional on assurance that others engage in it too; and, if so, whether assurance is generated by common knowledge of the psychology of framing, by joint commitment, or by experience. But despite these unresolved issues, we believe that our analysis has shown that team reasoning is just as coherent and valid as the best-reply reasoning of conventional game theory. Notes ∗

Previous versions of this paper were presented at a workshop on Rationality and Commitment at the University of St Gallen, a conference on Logic, Games and Philosophy: Foundational Perspectives in Prague, the Collective Intentionality IV conference in Siena, and the Kline Workshop on Collective Rationality at the University of Missouri at Columbia. We thank the participants at these meetings, and Nicholas Bardsley, for comments. The paper uses material from our contributions as editors to an unfinished book by the late Michael Bacharach, now published as Bacharach (2006).

theories of team agency 309 1. In order to conclude that (cooperate, cooperate) is the best pair of strategies for them, the players have to judge the payoff combinations (a, d) and (d, a) to be worse ‘for them’ than (b, b). 2. A pure coordination game is identical with Hi-Lo except that, for all labels i, ai takes the same value a > 0. 3. In an as yet unpublished experiment, Nicholas Bardsley presented fiftysix Dutch students with two Hi-Lo games. In one game, the ratio of the money payoffs to high and low was 10:1; in the other it was 10:9. In each case, fifty-four subjects (96 per cent) chose high. 4. For fuller statements of this argument, see Hodgson (1967), Sugden (1993), or Bacharach (2006). 5. For example, suppose that each player believes that his opponent is just as likely to choose one strategy as the other. Then both will choose high. Or suppose that each player believes that his opponent believes that he is just as likely to choose one strategy as the other. Then each player will expect his opponent to choose high, and so choose high as a best reply. Or ... 6. Gold (2005) shows a technical sense in which this is so, within a model of reasoning involving the manipulation of propositions. 7. This is not to deny the psychological possibility that a person might simultaneously experience motivational or affective pulls towards both individual and group identity. Our claim is merely that such conflicting pulls cannot be resolved by instrumental reasoning. Consider an analogous case in conventional choice theory. What if an individual faces a choice between two options, feels motivational pulls towards each of them, but cannot settle on any firm preference (or on a firm attitude of indifference)? Clearly, this case is psychologically possible; but if a person is unsure of her own objectives, instrumental rationality cannot tell her what they should be. 8. In game-theoretic language, this is a game form. A game form consists of a set of players, a set of alternative strategies for each player, and, for each profile of strategies that the players might choose, an outcome. In contrast, a game is normally defined so that, for each profile of strategies, there is a vector of numerical payoffs, one payoff for each player. 9. We will say more about how agents might come to conceive of S as a unit of agency in later sections of the paper.

310 natalie gold and robert sugden 10. Rabin’s formulation of this hypothesis is not fully compatible with conventional game theory, since it allows each player’s utility to depend on his beliefs about other players’ beliefs about the first player’s choices. (These second-order beliefs are used in defining the first player’s beliefs about the second player’s intentions.) Rabin’s theory uses a non-standard form of game theory, psychological game theory (Geanakoplos, Pearce, and Stachetti 1989). However, as David Levine (1998) shows, the main features of Rabin’s theory can be reconstructed within conventional game theory. 11. One might wonder why we can’t simply substitute ‘identifies with T’ for ‘identifies with S’ in premise (2) of schema 12.6, and interpret U as a measure of what is good for T. But Bacharach’s theory of framing commits him to these premises as we have written them. His hypothesis is that group identification is an individual’s psychological response to the stimulus of a particular decision situation. It is not itself a group action. (To treat it as a group action would, in Bacharach’s framework, lead to an infinite regress.) Thus, group identification is conceptually prior to the formation of the ‘team’ of people who identify with the group. 12. We can normalize the payoff function by setting uC = 1 and uD = 0. Then, given that uF < 0, the critical value of ω is ω∗ = 2uF /(2uF − 1). The protocol (cooperate, cooperate) is optimal if and only if ω ≥ ω∗ , (defect, defect) is optimal if and only if ω ≤ ω∗ . There is no non-zero value of ω at which (cooperate, defect) or (defect, cooperate) is optimal. 13. These conditions may rule out some dominant-strategy Nash equilibria as cases of collective intention. For example, in the case of the Prisoner’s Dilemma, one might deny that Player 1 plays defect in order for Players 1 and 2 to succeed in playing (defect, defect); rather, Player 1 plays defect because that is best for him, irrespective of what Player 2 does. 14. Nicholas Bardsley (2007) criticizes Searle’s analysis and offers an alternative, intended to be compatible with team reasoning. Bardsley’s alternative to ‘We intend to make the sauce by means of me stirring’ as Jones’s intention would be: ‘I intend my part of the combination (Jones stirs, Smith pours) in circumstances that you and I have this very intention, all of which is to make the sauce’. In Bardsley’s analysis, Smith’s intention has exactly the same sense as Jones’s (although ‘my’, ‘you’, and ‘I’ have different references in the two cases). The self-reference in

theories of team agency 311 these intentions is analogous with the recursiveness of mutually assured team reasoning.

References Bacharach, Michael. 1999. ‘Interactive Team Reasoning: A Contribution to the Theory of Cooperation’. Research in Economics 53: 117–47. Bacharach, Michael. 2000. ‘Scientific Synopsis’. Unpublished manuscript (describing initial plans for Beyond Individual Choice). Bacharach, Michael. 2006. Beyond Individual Choice: Teams and Frames in Game Theory. Natalie Gold and Robert Sugden (eds.). Princeton: Princeton University Press. Bardsley, Nicholas. 2007. ‘On Collective Intentions: Collective Action in Economics and Philosophy’. Synthese 157: 141–59. Binmore, Ken. 1994. Playing Fair. Cambridge, Mass.: MIT Press. Bolton, Gary and Axel Ockenfels. 2000. ‘ERC—A Theory of Equity, Reciprocity and Competition’. American Economic Review 90: 166–93. Bratman, Michael. 1993. ‘Shared Intention’. Ethics 104: 97–113. Cubitt, Robin and Robert Sugden. 2003. ‘Common Knowledge, Salience and Convention’. Economics and Philosophy 19: 175–210. Fehr, Ernst and Klaus Schmidt. 1999. ‘A Theory of Fairness, Competition and Cooperation’. Quarterly Journal of Economics 114: 817–68. Geanakoplos, John, David Pearce, and Ennio Stacchetti. 1989. ‘Psychological Games and Sequential Rationality’. Games and Economic Behavior 1: 60–79. Gilbert, Margaret. 1989. On Social Facts. London: Routledge. Gold, Natalie. 2005. ‘Framing and Decision Making: A Reason-Based Approach’. Unpublished D.Phil. thesis, University of Oxford. Gold, Natalie and Christian List. 2004. ‘Framing as Path-Dependence’. Economics and Philosophy 20: 253–77. Hodgson, David. 1967. Consequences of Utilitarianism. Oxford: Clarendon Press. Hollis, Martin. 1998. Trust within Reason. Cambridge: Cambridge University Press. Hurley, Susan. 1989. Natural Reasons. Oxford: Oxford University Press. Levine, David. 1998. ‘Modelling Altruism and Spitefulness in Experiments’. Review of Economic Dynamics 1: 593–622. Lewis, David. 1969. Convention: A Philosophical Study. Cambridge, Mass.: Harvard University Press. Rabin, Matthew. 1993. ‘Incorporating Fairness into Game Theory and Economics’. American Economic Review 83: 1281–1302. Regan, Donald. 1980. Utilitarianism and Cooperation. Oxford: Clarendon Press.

312


Rousseau, Jean-Jacques. 1988 [1762]. ‘On Social Contract’. In Alan Ritter and Julia Conaway Bondanella (eds), Rousseau’s Political Writings. New York: Norton. Sally, David. 1995. ‘Conversation and Cooperation in Social Dilemmas: A MetaAnalysis of Experiments from 1958 to 1992’. Rationality and Society 7: 58–92. Searle, John. 1990. ‘Collective Intentions and Actions’. In P. Cohen, J. Morgan, and M. E. Pollack (eds.), Intentions in Communication. Cambridge, Mass.: MIT Press, pp. 401–15. Sen, Amartya. 1974. ‘Choice, Orderings and Morality’. In R. Ko¨ rner (ed.), Practical Reason. Oxford: Blackwell [page references are to the paper as reprinted in Amartya Sen (1982), Choice, Welfare and Measurement (Oxford: Blackwell)]. Sen, Amartya. 1977. ‘Rational Fools: a Critique of the Behavioral Foundations of Economic Theory’. Philosophy and Public Affairs 6: 317–44 [page references are to the paper as reprinted in Amartya Sen (1982), Choice, Welfare and Measurement (Oxford: Blackwell)]. Sugden, Robert. 1993. ‘Thinking as a Team: Toward an Explanation of Nonselfish Behavior’. Social Philosophy and Policy 10: 69–89. Sugden, Robert. 2003. ‘The Logic of Team Reasoning’. Philosophical Explorations 6: 165–81. Tuomela, Raimo and Kaarlo Miller. 1988. ‘We-Intentions’. Philosophical Studies 53: 367–89.

13 Identity and Commitment: Sen’s Fourth Aspect of the Self J O H N B . DAV I S ∗

Amartya Sen has made examination of the behavioral foundations of economics and their connection to three different interpretations of the ‘‘privateness’’ of the individual a central theme of his work. He has also associated this examination with the issue of how we might talk about the self or the identity of individual. This paper argues that while Sen frames his views about identity primarily in terms of social identity or in connection with how individuals identify with others, his thinking about social identity derives from his understanding of commitment and its association with a fourth aspect of the self that provides the basis for an account of individual or personal identity. The key to this argument lies in how Sen extends his ‘‘privateness’’ framework to distinguish this fourth aspect of the self as different in kind from the three standard, self-interest-based aspects of the self employed in neoclassical behavioral models. This fourth aspect of the self, or ‘‘commitment self,’’ is linked to Sen’s emphasis on individuals being able to engage in reasoning and self-scrutiny. Individuals understood in terms of this capacity, however, can be seen to be distinct and re-identifiable beings, and thus be said to have personal identities. Seeing them in this way, moreover, links to Sen’s later thinking about functionings and capabilities, and makes it possible to argue that the capability framework either employs or can make use of an understanding of personal identity. This paper offers an interpretation of this possible identity framework for Sen, and also briefly discusses how it might be used to address one of the leading criticisms of Sen’s approach to thinking about capabilities, namely, that it lacks a short-list of essential capabilities.

314

john b. davis

Section 1 first reviews Sen’s ‘‘privateness’’ framework and its relation to commitment, considers recent criticisms of the idea that commitment stands outside of this framework, and then considers how the different forms of ‘‘privateness’’ and commitment differ in terms of individual selfregard and reflexivity. Section 2 discusses Sen’s views on identity as social identification and suggests ways in which these views are connected to his thinking about individual identity and rationality. Section 3 sets forth my earlier application of collective intentionality analysis to account for Sen’s thinking about personal and social identity. Section 4 provides a related but different interpretation of Sen’s thinking about personal and social identity that rather emphasizes the concept of commitment, and draws on Bernard Williams’ understanding of integrity as based on identityconferring commitments. Section 5 briefly addresses Martha Nussbaum’s critique of Sen’s framework that it fails to identify a short-list of essential capabilities, and argues that understanding personal identity as a basic capability provides one way of answering this critique. Section 6 offers four concluding comments on the paper’s motivations.

1. The ‘‘privateness’’ framework and the fourth aspect of the self Sen distinguishes three aspects of the self or concepts of the person that operate in one form or another in standard economics’ characterizations of self-interest, and then contrasts all three with a fourth aspect of the self and concept of the person associated with commitment (Sen 1985; 2002: 33–7, 206–24). The three standard types of ‘‘privateness’’ and aspects of the self are: Self-centered welfare: A person’s welfare depends only on her own consumption and other features of the richness of her life (without any sympathy or antipathy toward others, and without any procedural concern). Self-welfare goal: A person’s only goal is to maximize her own welfare. Self-goal choice: A person’s choices must be based entirely on the pursuit of her own goals. (Sen 2002: 33–4)

Though Sen finds that these different aspects of the self are often lumped together and not well distinguished in standard economics, he shows that

identity and commitment 315 they are independent from one another in a number of ways, and can play distinct roles in different behavioral models. Sen’s fourth concept of the self is associated with individuals making commitments. He originally introduced the concept of commitment by contrasting sympathy and commitment (Sen 1977), but with his subsequent distinction between self-welfare goal and self-goal choice—both of which make individual well-being depend upon something other than the individual’s own consumption and self-centered welfare—making choices based on one’s commitments comes to be understood to be a matter of making choices irrespective of any kind of personal gain or loss. As he puts it then: Commitment ... is concerned with breaking the tight link between individual welfare (with or without sympathy) and the choice of action (e.g., acting to help remove some misery even though one personally does not suffer from it). Sympathy alone does not require any departure from individual-welfare maximization: but commitment does involve rejection of that assumption. (Sen 1982: 8)

Taking all three of these forms of ‘privateness’ to generally be a matter of individual self-regard, we might accordingly say that introducing commitment breaks the ‘‘tight link’’ between individual self-regard in any form and individual choice of action. Recently, however, the idea that commitment somehow escapes selfregard has been challenged. Philip Pettit (2005) focuses on Sen’s distinction between self-goal choice and commitment, and asserts it to be highly implausible. He treats commitment as being either own goal-modifying or own goal-displacing, sees the latter as closer to Sen’s meaning, and interprets commitment as putting aside one’s own goals to act on those of another. But he doubts it ever makes sense to say one could fail to act on one’s own goals, since one’s acting on the goals of another must still be a matter of acting on one’s own goals. This charge, however, mixes together something being one’s own act and that act being one’s own goal by characterizing the former as one’s own goal. Sen of course knows that acting on the goals of another is one’s own act and that intentional action is goal-directed. But that an action is one’s own goal in this intentional action sense still allows for a distinction between acting on one’s own particular goals and ignoring those particular goals to act on goals of others. More

316

john b. davis

generally, action need not always be self-regarding though it is nonetheless one’s own action. A somewhat different criticism of commitment comes from Dan Hausman (2005) who argues that the concept of preference in economics is ambiguous, but that if understood broadly as an all-things-considered type ranking, it could accommodate commitment as one type of preference. Hausman’s proposal to sharpen the concept of preference is entirely reasonable, but it does not imply, as he allows, that commitment is necessarily a kind of preference. In fact, Hausman understands his recommendation to refine the preference concept as a strategic prescription designed to make it easier to deal with the complexity of human motivation, and recognizes that Sen has adopted a different, also reasonable, strategy toward this same goal by distinguishing preference and commitment. That is, he does not believe self-goal choice and commitment cannot be distinguished, and primarily disagrees with Sen over strategic objectives. We might note, then, that Sen’s objective as he has recently stated it is not so much to improve the preference framework but rather to explain different formulations of rationality as being due to differences over how one understands the domain of reason (Sen 2005). In what follows, then, I attempt to get at what this may involve by reexamining the structure of Sen’s ‘‘privateness’’ framework and its relation to commitment along two lines. First, note that his three types of ‘‘privateness’’ can be ranked in the order he provides according to the degree to which they emphasize an individual’s own self-regard. Self-centered welfare concerns only an individual’s own satisfaction (or desire fulfillment), but self-welfare goal allows other individuals’ satisfactions to enter into an individual’s satisfaction through sympathy (or antipathy), and self-goal choice allows for non-welfarist goals that are altogether removed from an individual satisfaction (the pursuit of social justice). Second, if we emphasize the reflexivity of ‘‘privateness’’ in terms of three ways that individuals affect themselves by the choices they make, Sen’s three aspects of ‘‘privateness’’ can be ranked according to the degree to which this reflexive relation incorporates considerations external to the individual. Only self-centered welfare is independent of such considerations. Individuals affect themselves by their choices without any external mediation. But with self-welfare goal individuals affect themselves through concern for others’ welfare, and with self-goal choice individuals affect themselves through concern

identity and commitment 317 for matters that transcend individuals’ welfare altogether. Thus, across the three forms of ‘‘privateness,’’ choice is increasingly non-self-regarding and reflexively affects individuals through considerations increasingly external to the individual. Extending this double characterization of the three types of ‘‘privateness’’ to commitment, then, we may add that commitment eliminates self-regard understood as the individual’s pursuit of own goals altogether as a dimension of choice. But since the reflexivity of ‘‘privateness’’ is a matter of how individuals affect themselves in the choices they make, is commitment, in ‘‘breaking the tight link between individual welfare ... and the choice of action,’’ still a reflexive relation? That is, does commitment’s elimination of self-regard also involve an elimination of any sort of self-referencing? In fact, Sen’s fourth aspect of the self and concept of the person associated with commitment is quite explicitly reflexive. The fourth aspect of the self considers individuals as being able to engage in reasoning and self-scrutiny. ‘‘A person is not only an entity that can enjoy one’s own consumption, experience, and appreciate one’s welfare, and have one’s goals, but also an entity that can examine one’s values and objectives and choose in the light of those values and objectives’’ (Sen 2002: 36). Sen’s clear coupling, as he also puts it, of ‘‘one’s own reasoning and self-scrutiny’’ (Sen 2002: 36) demonstrates that he sees his concept of commitment as pre-eminently possessing a reflexive dimension. At the same time, Sen associates commitment and self-scrutiny with the ‘‘problem’’ of ‘‘the ‘identity’ of a person, that is, how the person sees himself or herself’’ (Sen 2002: 215). Thus, to better understand this further connection, we turn to Sen’s stated views on identity.

2. Sen on identity, commitment, and agency Sen’s primary way of talking about the concept of the identity of a person is that of social identity, or the idea of identifying with others. Once one goes beyond the trivial or logical concept of identity—the idea that an object is necessarily identical with itself—‘‘we shift our attention from the notion of being identical to that of sharing an identity, and to the idea of identifying oneself with others of a particular group’’ (Sen 1999b: 2). One source of this notion of identifying with others in Sen’s thinking is his ordering of aspects of the

318

john b. davis

self and types of ‘‘privateness’’ in standard economics’ characterizations of self-interest. Thus, the long-standing emphasis in standard economic theory on self-interest, he notes, excludes from the outset the idea that individuals might identify with others in deciding on their objectives and making their choices (ibid.). Yet if one intends to broaden the behavioral foundations of economics, one may move progressively through a set of enlargements of the individual, first, by opening the door to the idea that one’s self-interest may be influenced by sympathy (or antipathy) toward others (self-welfare goal), next by including the idea that one’s goals may include considerations other than one’s own welfare (self-goal choice), and finally by supposing that one may act on principles altogether removed from one’s own personal welfare as when one identifies in some way with others (commitment). Thus, identification with others through commitment merely carries Sen’s ordering of types of ‘‘privateness’’ that underlies his classification of the different features of the self but one step further.1 At the same time, this additional step is a significant one, not only in ‘‘breaking the tight link between individual welfare ... and the choice of action,’’ but indeed in raising the issue of whether or how the individual even remains an individual when identifying with others. A second source of the idea of identifying with others is Sen’s early interest in multiple preferences and meta-rankings. One good reason to suppose preference rankings are incomplete and at best offer incomplete quasi-orderings is that individuals might have multiple preferences that they cannot immediately reconcile. Sen’s original interest in multiple preference rankings stems from his desire to improve the behavioral foundations of economics, analyze various moral dilemmas, and investigate social cooperation, but his explanation of why we might have multiple rankings is that we have multiple social identifications with others. Community, nationality, race, sex, union membership, the fellowship of oligopolists, revolutionary solidarity, and so on, all provide identities that can be, depending on the context, crucial to our view of ourselves, and thus to the way we view our welfare, goals, or behavioral obligations. (Sen 2002: 215)

We might then say that one sees oneself as a certain type of person, and having different meta-rankings means one can also see oneself simultaneously as different types of persons. Note that seeing oneself as a certain type of person suggests one is still a distinct individual, albeit under some

identity and commitment 319 description. Yet at times Sen’s association of social identification and commitment threatens to undermine this interpretation, as when he asserts that seeing oneself as a member of a social group goes beyond sympathizing with members of that group, and involves ‘‘actually identifying with them’’ (ibid.).2 Indeed, this strong sense of social identification is important to his view of the seriousness with which we need to look at the issue of our having multiple social identifications, since having multiple meta-rankings produces ‘‘conflicting demands arising from different identities and affiliations’’ (Sen 1999b: 30). From this perspective, we might thus say that for Sen the issue of whether or how an individual identified with and having commitments to others remains a distinct individual is a matter of whether the individual is able to sustain a personal unity and integrity across these conflicting multiple associations. Interestingly, Sen suggests we might understand social identification in terms of our use of first-person plural speech. The nature of our language often underlines the forces of our wider identity. ‘‘We’’ demand things; ‘‘our’’ actions reflect ‘‘our’’ concerns; ‘‘we’’ protest at injustice done to ‘‘us.’’ This is, of course, the language of social intercourse and politics, but it is difficult to believe that it represents nothing other than a verbal form, and in particular no sense of identity. (Sen 2002: 215; also cf. p. 41)

Since by ‘‘sense of identity’’ Sen means social identity, we have a suggestion here that social identity as reflected in first-person plural speech might be understood in collective or shared intentionality terms.3 Though he does not develop this idea systematically, he nonetheless alludes to the convention-based, reciprocal feedback framework used by some contributors to the collective intentionality literature (e.g., Tuomela 1995; Davis 2003a) when he comments on an apparent tendency in experimental game theory for players to want to know the identity of fellow players (Sen op. cit.). Thus he associates ‘‘we’’ language and social identification with the notion that not only players’ choices but also identities are somehow mutually constitutive of one another in games. Sen’s focus is on how one-shot Prisoner’s Dilemma games may become Assurance games, but if we broaden the framework to, say, repeated game formats where the Folk Theorem applies, or allow for pre-play communication considerations, all sorts of results are possible, some of which it seems fair to say might make

320

john b. davis

use of explanations of social identity in some form of first-person plural speech, collective intentionality analysis. Returning to the question of personal identity, then, how does Sen’s thinking about social identification fit with his emphasis on individual choice as central to the behavioral foundations of economics? Sen maps out his answer to this question in relation to communitarian critiques of liberal theories of justice. Such critiques, he notes, have at times been employed to argue that individuals’ reasoning abilities are effectively captive to their (pre-existing) social identities, which are ‘‘not a relationship they choose (as in a voluntary association) but an attachment they discover, not merely an attribute but a constituent of their identity’’ (Sandel 1998: 150; quoted in Sen 1999b: 16). On this view, one’s identity is something one discovers—communitarians sometimes say ‘perceive’—and the way that one reasons then depends upon this identity. Sen’s response to this is to reject the false dichotomy it involves between perfectly autonomous choice and perfectly determined choice. The alternative to the ‘‘discovery’’ view is not choice from positions ‘‘unencumbered’’ with any identity (as communitarian expositions often seem to imply), but choices that continue to exist in any encumbered position one happens to occupy. Choice does not require jumping out of nowhere into somewhere. (Sen 1999b: 23)

That is, choice may be influenced but is not determined by social identification. Social identities are indeed important. So is choice. But while we may agree with him about this in principle, one would also like to know more about how this might be the case. Hints come in Sen’s characterization of the fourth aspect of the self associated with commitment where he links ‘‘one’s own reasoning and self-scrutiny’’ (Sen 2002: 36). We saw above that on this view a person is someone who is able to ‘‘examine one’s values and objectives and choose in the light of those values and objectives’’ (Sen 2002: 36). As if answering the communitarians again, this passage continues, Our choices need not relentlessly follow our experiences of consumption or welfare, or simply translate perceived goals into action. We can ask what we want to do and how, and in that context also examine what we should want and how. (Sen ibid.; emphasis added)

identity and commitment 321 Note, then, that this ability individuals are said to have to stand apart, as it were, from their goals and objectives in order to scrutinize and evaluate them in a reasoned way gives them a status as agents that is absent in Sen’s first three aspects of the self. First, from the vantage point of the individual, there is a fundamental difference between this sort of reflexive relationship and those associated with the other forms of the self Sen distinguishes. For self-centered welfare, self-welfare goal, and self-goal choice, when individuals make choices they are affected by own consumption, sympathy or antipathy, and non-welfare goals respectively. But individuals engaged in reasoned self-scrutiny are not only affected but also affect themselves in virtue of adding their evaluation of the effects of their choices upon themselves to those effects themselves. Second, from the vantage point of others, whereas reflexivity operates in Sen’s three original aspects of the self in terms of increasingly non-private, other-referencing sorts of considerations, in the case of his fourth aspect of the self, reflexivity involves a relation between an—at least loosely—identifiable social group and the individual with a now elevated status as an agent able to engage in some form of self-evaluation. I thus suggest that Sen’s response to (strong) communitarian thinking about individuals and their social identities is to say that individuals have identities that are in some sense independent of their social identities. That is, they have both social identities and personal identities.

3. Self-scrutiny, shared intentions, and personal identity Here I review the argument I previously used to develop a conception of personal identity for Sen’s capability framework (Davis 2002, 2003b: ch. 8), in order to compare it to one that makes use of Sen’s emphasis on individuals as engaged in reasoning and self-scrutiny. That argument proceeded by applying collective intentionality analysis to Sen’s understanding of the individual agent, and then asking how individuals thus understood might be thought distinct and re-identifiable across change. The point of entry for the argument was Nancy Folbre’s ‘‘structures of constraint’’ interpretation of individuals having multiple, conflicting social identifications. As she expresses it:

322

john b. davis

Individuals cannot be located by a single set of coordinates, because they operate in many different collective dimensions, within many different chosen and given groups. Nor can they be located by a list of all the given groups to which they belong, by a simple ‘‘adding up’’ of separate positions. The interaction between different dimensions of collective identity affects the choices individuals make about which collective interests to pursue. (Folbre 1994: 52–3)

How, then, do individuals manage the ‘‘interaction between different dimensions of [their] collective identity?’’ In her attention to women in particular, Folbre stressed the need for women to be able to move back and forth between their different roles, and not be confined to any single set of roles. This suggests that women need to possess a special capacity not only to be able to move between roles, but also to do so with versatility, that is, to be able to move successfully in and out of often very different roles without high adjustment costs to themselves and others.4 Having a special capacity of this sort would provide women (or individuals in general) a measure of detachment and independence from their competing social roles, but at the same time it would enable them to actively embrace these different, often conflicting roles. One way, then, to understand this special capacity as a combination of independence and affiliation with groups is to apply collective intentionality analysis to how women affiliate with social groups. Broadly speaking, collective intentionality analysis examines how distinct individuals may form shared intentions in their interaction with one another (cf. Gilbert 1989; Tuomela 1995). My approach to collective intentionality analysis is to explicate the normal success conditions associated with first-person plural speech or the use of ‘‘we’’ language. Particularly important here is that such language is used in a performative manner. To say that a form of speech has a performative character is to say that individuals not only communicate in using it but also accomplish some action in doing so. Thus, when individuals use ‘‘we’’ language to express intentions they believe to be shared, they might be thought to be intent upon establishing some shared understanding with others to whom their use of the term ‘‘we’’ applies in regard to the content of the proposition they have expressed. For example, if I say ‘‘we are happy with our work,’’ I not only express my own view about our work, but by including others to whom the ‘‘we’’ applies I also suppose that others have the same view of it. Emphasizing the normal success conditions associated with using first-person plural speech, then, is

identity and commitment 323 a matter of whether my action is successful, where an important indication of success is that others do not challenge what I have said. Individuals having a special capacity to move comfortably back and forth across their different social affiliations may now be understood in connection with their capacity to use ‘‘we’’ language. The capacity to move comfortably across one’s different social affiliations requires being able to both identify with others and yet still preserve an independence and detachment from them. On the one hand, the identification-with-others side of this is captured in how the normal use of ‘‘we’’ language requires identification with the intentions of others to whom this language applies. On the other hand, an individual’s expression of a collective intention is still an individual’s intentional expression, and since shared intentions can only be formed by those who have them, such language also requires that the individual have an independence and detachment from others to whom their ‘‘we’’ language applies. We might thus say that under normal conditions individuals freely constrain themselves in their expression of collective intentions. Thus individuals might be said to have a special capacity to move comfortably back and forth across their different social affiliations when they are able to successfully exercise their capacity to express collective intentions in social groups. Needless to say, much more needs to be said about how having a capacity to freely move across one’s social affiliations can be related to the capacity to successfully express collective intentions within the groups in which one operates. Here, however, I want to emphasize how anchoring an independence-preserving mobility across groups in how individuals freely constrain themselves within groups—as captured in the logic of normal expression of collective intentions—tells us something about personal identity and Sen’s views. One argument is that the concept of personal identity requires some understanding of individuals apart from their social identification with others, and that Sen’s emphasis on individuals as reflexively detached agents suggests a way of developing this understanding. Let me now try to explain this in connection with individuals’ expression of collective intentions. An individual’s expression of a collective intention, it seems fair to say, involves just the sort of reflexive self-scrutiny and evaluation of own objectives that Sen emphasizes. When an individual asserts, ‘‘we are happy with our work,’’ it is normally the case that the individual

324

john b. davis

considers whether individually expressing this intention aligns her or him with what others believe about their work.5 That is, the individual must scrutinize and evaluate what she or he intends relative to what others might intend. There are two reasons for saying that this distinguishes or individuates the individual. First, the self-scrutiny involved in considering one’s own expression of a collective intention relative to what others intend itself distinguishes the individual from others. In effect, the individual distinguishes herself or himself in an exercise of comparative thinking. Second, actually expressing a collective intention requires that the individual freely bind herself or himself to whatever the content of that collective intention implies. But since only the individual can freely bind herself or himself to something, this act is self-individuating. The interpretation that I thus previously offered of Sen’s thinking in regard to personal identity begins with the problem of social identification, builds on his emphasis on individuals as agents having a capacity for reflexive self-scrutiny, develops this idea in terms of individuals’ expression of collective intentions in groups, and explains this as a way of seeing individuals as distinct and independent. The focus on a self-individuating capacity—individuals having a capacity to not only move across social groups while maintaining a relative independence, but also a capacity within groups to freely tie themselves to the consequences of the collective intentions they express—was meant to be a step in the direction of the capability framework. My view was that individuals have the native capacities stated above, but may fail to succeed in exercising them, and thus fail to develop associated capabilities. If we think of personal identity as requiring that individuals be both distinct and re-identifiable beings, then individuals who regularly exercise this self-individuating capacity in interaction with others are re-identifiable in terms of this capacity. But individuals who fail to regularly exercise this native capacity lack personal identities and lose their status as individuals to their social identifications with others. Thus, this interpretation of Sen’s thinking makes personal identity a particular capability among many that individuals may develop. I argue in section 5 that it is a centrally important capability. What might be thought to be missing from this discussion, however, is Sen’s own emphasis on commitment. I consequently turn in the following section to argue that many of the same themes as appear in the analysis above re-emerge when we use Sen’s thinking about commitment to develop an account of personal identity.

identity and commitment 325

4. Self-scrutiny, rules, and integrity-conferring commitments Sen links commitment and the status of individuals as independent, selfscrutinizing agents, but he also links commitment to social identification. How are these two ways in which he speaks of commitment compatible with one another? Here I first review the development in Sen’s thinking about the concept of commitment from his early ‘‘Rational Fools’’ paper to his later thinking that associates commitment and the self-scrutinizing aspect of the self. Then I review Bernard Williams’ thinking about commitment that dissociates it from desire-endorsement, and interprets it rather as an identity-conferring behavior of the individual. Finally I draw these two sets of views together to suggest an account of how Sen might be said to understand personal identity and social identity in a manner that recalls the collective intentionality-inspired account of these concepts in the previous section. 4.1. Sen on commitment As seen above, if we go back to Sen’s ‘‘Rational Fools’’ discussion of the concept of commitment, we find him primarily emphasizing the idea that commitment ‘‘drives a wedge between personal choice and personal welfare’’ (Sen 1977: 329). Comparing sympathy and commitment, sympathy involves a concern for others that affects one’s welfare directly, while in the case of commitment one’s welfare is only incidentally related to one’s choice and certainly not the reason for it. One way of expressing this is to say that action based on sympathy is in a sense ‘‘egoistic’’ whereas action based on commitment is by this standard ‘‘non-egoistic’’ (ibid.: 326). Alternatively commitment involves ‘‘counterpreferential choice’’ (ibid.: 328), though it is still possible that acting on the basis of a commitment may coincide by chance with the maximization of personal welfare. These characterizations of commitment in terms of what it is not are supplemented by suggestions regarding what might motivate commitment: a sense of duty, one’s morals, or a sense of obligation going beyond the consequences. At the same time, Sen does not require that commitment draw on moral motives, and emphasizes that morality or culture both offer individuals a basis for the commitments they make. Indeed, in response to the suggestion that we may have two kinds of preferences, subjective and ethical, with the

326

john b. davis

latter reflecting a kind of impartiality (Harsanyi 1955), Sen notes that an individual may be quite partial in making commitments to ‘‘some particular group, say to the neighborhood or to the social class to which he belongs’’ (ibid.: 337). But Sen’s explicit introduction of the reasoning and self-scrutinizing aspect of the self after ‘‘Rational Fools’’ brings out a new and somewhat different dimension to his thinking about commitment and identity. Thus in one important later discussion, where he links the reasoning and self-scrutinizing self and social identification, he also emphasizes the noninstrumental, ‘‘intrinsic importance ... attached to following certain rules of behavior’’ that are operative in the groups with which one identifies (Sen 2002: 217n): One of the ways in which the sense of identity can operate is through making members of a community accept certain rules of conduct as part of obligatory behavior toward others in a community. It is not a matter of asking each time, What do I get out of it? How are my own goals furthered in this way?, but of taking for granted the case for certain patterns of behavior toward others. (Sen 2002: 216–17)

We might infer from this that only reasoning and self-scrutinizing individuals, who are themselves detached in this aspect of their selves from the instrumentality of self-goal choice, are able to recognize in a correlative manner that certain social rules of behavior also have non-instrumental, intrinsic value. Social identification with others, Sen is arguing, presupposes rather than eliminates individual detachment and independence from those with whom one identifies, and it does so just because commitment to others takes the form of a reasoned appreciation of the intrinsic value of the rules operating in those groups. Commitment, consequently, is not an unreflective type of attachment to others, but rather a rational recognition of rules associated with social membership that can only be achieved by individuals who have distanced themselves from their own interest as self-scrutinizing individuals. 4.2. Williams on commitment and integrity The origins of recent interest in the concept of commitment are Williams’ use of it in his critique of the theory of action implied by act utilitarianism.

identity and commitment 327 Utilitarianism, he tells us, offers ‘‘a general project of bringing about maximally desirable outcomes,’’ but leaves out certain other projects, interests, and causes—both small-scale and large-scale—to which individuals commonly find they have made commitments (Williams 1973: 110–11). This omission cannot be remedied by attempting to interpret commitment in utilitarian terms. While pursuing one’s commitments may happen to make one happy, ‘‘it does not follow, nor could it possibly be true, that those projects [to which one is committed] are themselves projects of pursuing happiness’’ (ibid.: 113). On one level, one’s commitments ‘‘flow from some more general disposition toward human conduct and character, such as a hatred of injustice, or of cruelty, or of killing’’ (ibid.: 111). But on another level, an individual’s commitments stem more fundamentally from ‘‘what his life is about’’ (ibid.: 116). Thus Williams comments, It is absurd to demand of such a man [one with commitments], when the sums come in from the utility network which the projects of others have in part determined, that he should just step aside from his own project and decision and acknowledge the decision that the utilitarian calculation requires ... [T]his is to neglect the extent to which his actions and his decisions have to be seen as the actions and decisions which flow from the projects and attitudes with which he is most closely identified. It is thus, in the most literal sense, an attack on his integrity. (ibid.: 116–17)

One’s commitments, then, are not only not instrumental to happiness or other good consequences, but because they ‘‘flow from the projects and attitudes with which [the individual] is most closely identified,’’ they additionally enable individuals to provide themselves a sense of their own integrity as individuals. As Williams (1981) later further developed this view, the ‘identity view of integrity’, individuals have and act with integrity when they act on their commitments—those motives, interests, and attitudes that reflect who they are in the most fundamental way. Or, individuals act with integrity when they make what Williams terms ‘‘identity-conferring commitments.’’ The emphasis in this idea suggests that commitment is a relation between individuals and others that arises out of positions individuals reflexively take toward themselves.6 Because—or when—individuals make commitments to others, they are able to confer identity upon themselves. Put the other way around, if individuals fail to make commitments to others, they fail

328

john b. davis

to create a sense of personal identity for themselves. But what are we to say about individuals having different, often competing commitments? Williams’ answer, it seems, must be that while individuals’ competing commitments appear to fragment them, their very capacity for having and making commitments per se invests them with integrity, which itself makes them distinct and unitary beings. Indeed, in his original critique of utilitarianism, Williams saw the absence of any account of the individual as an agent as an important flaw. Utilitarian calculations always exhibit an abstract, impersonal—even ‘‘churchy’’—quality, tending to slide off toward some kind of ‘‘impersonally benevolent happiness-management’’ (Williams 1973: 110, 112). Commitments, however, are generally made by individuals.7 That is, by nature they are inherently personalizing. Let us, then, draw on this understanding of commitment to further develop Sen’s own understanding. 4.3. Sen on commitment and identity While their immediate concerns are different, with Sen interested in the behavioral foundations of economics and Williams interested in moral theory, they are both concerned to develop an adequate theory of action, and both agree that the concept of commitment cannot be accommodated within the framework of utilitarian consequentialist reasoning. What additionally appears to be shared between them is the view that personal identity is dependent upon individuals being able to make commitments to others, or, that social identification is a means to achieving personal identity. Whereas standard preference-based conceptions of the individual may be seen as seeking to explain the personal identity of individual economic agents atomistically, or apart from their interactions with others (Davis 2003b), the Sen–Williams commitment-based conception of the individual may be seen as seeking to explain the personal identity of individual economic agents relationally, or through their interactions with others. For both Sen and Williams this relational view of personal identity relies on commitment having a reflexive dimension. Sen holds that only reasoning and self-scrutinizing individuals, detached in this particular aspect of their selves from the instrumentality of self-goal choice, are able to make commitments to others, because only as individuals thus self-engaged are they able to recognize the correlative, intrinsic, non-instrumental value of social rules. Similarly, for Williams, individuals act upon themselves when

identity and commitment 329 they confer integrity upon themselves by making commitments to others that put aside impersonal, consequentialist utilitarian reasoning. How, then, does this reflexive self-scrutiny or identity self-conferral invest individuals with personal identity? Taking distinctness as the first and primary element in any account of personal identity, the act of taking oneself as a separate object of consideration is equivalent to treating oneself as distinct and independent. When individuals adopt a position of reflexive self-regard, they self-individuate themselves. For Sen and Williams, then, making commitments to others individuates the commitment-maker. This account of the individuation side of personal identity focused on commitment is similar to the collective intentionality-based account of individuation in the previous section. There I argued individuals have a native capacity to freely self-constrain themselves when they express we-intentions in social groups. That only individuals can freely selfconstrain themselves effectively distinguishes them as distinct individuals. Here, commitment functions in essentially the same way. We might infer, then, that commitment is an act in which individuals freely self-constrain themselves to others. Thus it is interesting that Sen employs ‘‘we’’ language (Sen 2002: 215) when he speaks of commitment and social identification, since this suggests that we are entitled to transfer the individuating character of we-intention behavior to commitment behavior. Note also that I suggested that individuals’ being able to freely self-constrain themselves to what their we-intentions require of them in social groups is a native capacity. In terms of the Sen–Williams understanding of commitment, we might accordingly say that being able to make commitments is also a native individual capacity. This raises the question of whether the second element in any account of personal identity, namely, the re-identification requirement, plays a role in the commitment framework. In the collective intentionality account, re-identification is contingent upon individuals being able to regularly exercise their capacity to freely constrain themselves in social groups over time and across different types of social settings in which they participate. That is, re-identification is a matter of individuals developing a special capability. The commitment account of Sen’s thinking, I believe, has a slightly different emphasis, since while we may imagine that individuals lose or fail to develop this special capability due to a variety of factors having to do with the nature of their lives and experience and with the ways

330

john b. davis

in which societies are organized, when we rather speak of commitment it seems odd to say people are not always able to make commitments to others. But if they are always able to form commitments to others, then on the argument here they would always be re-identifiable as distinct, commitment-making individuals, and would thus always have personal identities. That is, having personal identity would not be a capability, but rather a native capacity. In the section that follows, I rather opt for the capability view of re-identification and personal identity as the preferred interpretation of Sen’s thinking about personal identity, and accordingly assume that whether individuals are able to consistently form commitments to others over their lifetimes depends on the same variety of factors having to do with the nature of their lives and the ways in which societies are organized. My grounds for this is that this makes whether individuals have personal identities a contingent matter endogenous to the economic process, and thus an object of social-economic policy. In conclusion, therefore, I define personal identity for Sen as a special capability whereby individuals exercise a reflexive capacity to make commitments in social settings in a sustained way.

5. Personal identity and the capability framework A familiar and oft-repeated criticism of Sen’s capability framework is that it lacks a short-list of essential capabilities that, a` la Aristotle, would provide a single, universal set of policy recommendations regarding capability development (e.g., Nussbaum 2003). Sen has consistently resisted the idea that there ought to be an essential short-list of human capabilities on the grounds of flexibility in application and social diversity, but this reply has not been persuasive to some, perhaps in part due to the appeal of having a single, essential view of the individual and a set of universal policy prescriptions. In this section, then, I briefly attempt to reconcile these two positions by arguing for the central importance of one capability: having a personal identity.8 One way of looking at capabilities takes them to be freedoms (Sen 1999a). We have many freedoms according to the many capabilities we develop. More generally, freedom is a central value and behavioral feature of individuals according to Sen. Can individuals, then, be thought to be

identity and commitment 331 free beings if they are unable to generally sustain personal identities over their lifetimes? The type of freedom at issue here is a positive freedom to carry out one’s plans and goals as an agent. In connection with the concept of freedom, Sen understands being an ‘‘agent’’ not as in standard principal– agent analysis, but rather in its ‘‘older—and ‘grander’—sense as someone who acts and brings about change, and whose achievements can be judged in terms of her own values and objectives’’ (Sen 1999a: 19). But if individuals are unable to regularly exercise a capacity to freely bind themselves by the obligations and requirements of groups with which they associate, or if they are unable to genuinely form commitments to others, it is not clear how they can be regarded as agents able to act and bring about change with achievements judged in terms of own values and objectives. Thus it seems fair to say that the entire capabilities-as-freedoms framework depends on the one central freedom or capability of being able to sustain a personal identity. That is, underlying the development and exercise of all our more particular capabilities is a general capability of being able to freely sustain oneself in an environment that everywhere involves social interaction. The ‘‘Aristotelian’’ capability approach requires either a definite list of essential capabilities or some hierarchical organization of capabilities by importance. If we prescribe social policies aimed at ensuring that individuals have opportunities to develop a personal identity capability, then we indeed make one capability both essential and prior to all other capabilities. But we do not go the full Aristotelian route by filling out a complete list of essential capabilities or by organizing them in some hierarchical order. There seem to be at least two reasons for applying this reading to Sen’s framework. First, it still preserves his intuition that the capability framework works best when it flexibly accommodates social diversity. Second, it reinforces the role of the concept of freedom in that framework by further rooting it in a re-characterization of Sen’s strong sense of agents as beings who also seek to maintain personal identities. As a reconciliation of Sen’s and the more classic Aristotelian approach, this strategy offers perhaps more determinacy in policy determination than Sen’s own open-ended strategy, is still Aristotelian in making one feature of human life essential, but departs from Aristotle in providing a different view of the good life, and in giving pride of place to the more modern value of freedom.

332

john b. davis

6. Concluding remarks I close by way of brief comment on four general themes that motivate this paper rather than with a summary of the discussion of the paper itself. First, the paper follows a development in Sen’s own thinking which moves from an examination of standard economics’ characterization of selfinterest to a fourth aspect of the self associated with reasoned self-scrutiny, and which gives increased importance and meaning to the concept of commitment. My view is that the issue of personal identity only emerges once this fourth aspect of the self is clearly introduced, and introduced in connection with the tensions it creates between personal and social identity. Accordingly the paper takes the development of Sen’s thinking along this pathway to have itself created an agenda for examination of personal identity where it formerly was absent. Second, as this previous remark implies, the approach to personal identity here and in thinking about individual behavior in economics in general need not recall the approaches to personal identity taken in philosophy where in recent years different sorts of issues have been investigated. Thus the treatment of personal identity in economics may not, and it seems need not, conform to other concerns about the concept in philosophy and elsewhere. There are obviously many ways of talking about identity, personal and social, and the discussion in this paper is intended to be specific to a particular history of thinking about individuals in economics. Third, what seems to be key to this particular history of thinking about individuals in economics is a failure in most of the literature on the individual to ask whether individuality is endogenous to the social-economic process. That is, this literature fails to establish what makes individuals distinct, and this explains why much of my discussion here and elsewhere makes the individuation requirement a focus. Fourth, what especially distinguishes the treatment of personal identity here is the emphasis on reflexivity. Partly this reflects the emphasis that this concept possesses in thinking about individuals in economics, both in standard economics and in Sen’s thinking. But it also reflects what I perceive likely to be the strongest route to successfully explaining individual distinctness and independence given the various determining influences on individual behavior. The discussion in this paper, then, aims to make this concept central by bringing out its role in the development from standard

identity and commitment 333 economics’ characterizations of the self-interest aspects of the self to Sen’s commitment sense of the self. Notes ∗

The author is grateful to Solange Marin, Ingrid Robeyns, Fabienne Peter, Bernhard Schmid, and the referees for this volume for comments on earlier versions of the paper. An earlier version of this paper appeared as Davis (2004). 1. Note that Sen recognizes that not all commitments are to individuals but may also be to such things as causes and principles. 2. This is reinforced in something of an aside where Sen suggests, following Marx, that it might even be said that private interest itself is socially determined (Sen 2002: 215n). But Sen’s treatment of the different senses of ‘‘privateness’’ as all aspects of the self goes against this. 3. Anderson (2001) draws this connection, and links it to the concept of identity in Sen. 4. For example, one might argue that an important dimension of antidiscrimination law in the United States with respect to women is that it prevents employers from inquiring about family status, and thus restricting or imposing high costs on their mobility between domestic and employment roles. 5. Here I put aside complications associated with what happens when normal success conditions are not fulfilled, such as when others disagree but do not show it, when there is deceptive use of ‘‘we’’ speech, or when one individual imposes a ‘‘we’’ statement on others. These kinds of cases may be argued to be parasitic on the normal use of ‘‘we.’’ 6. It can be argued that, for Williams, commitments need not be made to others, but that individuals may simply have certain commitments. I focus here on the case where commitments are at least implicitly made to others. 7. However, Margaret Gilbert, in her contribution to this volume (ch. 11), argues for a concept of joint commitment. 8. For procedural approach to reconciliation that emphasizes social deliberation, see Robeyns (2003).

334 john b. davis References Anderson, E. 2001. ‘‘Unstrapping the Straitjacket of ‘Preference’: A Comment on Amartya Sen’s Contributions to Philosophy and Economics.’’ Economics and Philosophy 17: 21–38. Davis, J. 2002. ‘‘Using Sen’s Real Opportunities Capabilities Concept to Explain Personal Identity in Folbre’s ‘Structures of Constraint’ Analysis.’’ Review of Political Economy 14/4: 481–96. Davis, J. 2003a. ‘‘Collective Intentionality, Complex Economic Behavior, and Valuation.’’ Protosociology 18: 163–83. Davis J. 2003b. The Theory of the Individual in Economics. London: Routledge. Davis, J. 2004. ‘‘Identity and Commitment: Sen’s Conception of the Individual.’’ Tinbergen Institute Discussion Paper, 04-055/2. . Folbre, N. 1994. Who Pays for the Kids? Gender and the Structures of Constraint. London: Routledge. Gilbert, M. 1989. On Social Facts. London: Routledge. Harsanyi, J. 1955. ‘‘Cardinal Welfare, Individualistic Ethics, and Interpersonal Comparisons of Utility.’’ Journal of Political Economy 63/4: 309–21. Hausman, D. 2005. ‘‘Sympathy, Commitment, and Preference.’’ Economics and Philosophy 21/1: 33–50. Nussbaum, M. 2003. ‘‘Capabilities as Fundamental Entitlements: Sen and Social Justice.’’ Feminist Economics 9/2&3: 33–59. Pettit: 2005. ‘‘Construing Sen on Commitment.’’ Economics and Philosophy, 21/1: 15–32. Robeyns, I. 2003. ‘‘Sen’s Capability Approach and Gender Inequality: Selecting Relevant Capabilities.’’ Feminist Economics 9/2&3: 61–92. Sandel, M. 1998. Liberalism and the Limits of Justice, 2nd edn. Cambridge: Cambridge University Press. Sen, A. 1977. ‘‘Rational Fools: A Critique of the Behavioral Foundations of Economic Theory.’’ Philosophy and Public Affairs 6: 317–44. Sen, A. 1982. Choice, Welfare and Measurement. Oxford: Blackwell. Sen, A. 1985. ‘‘Goals, Commitment, and Identity.’’ Journal of Law, Economics and Organization 1/2. Reprinted in Sen 2002, pp. 206–24. Sen, A. 1999a. Development as Freedom. New York: Knopf. Sen, A. 1999b. Reason before Identity. New Delhi: Oxford University Press. Sen, A. 2002. Rationality and Freedom. Cambridge, Mass.: Belknap Press. Sen, A. 2005. ‘‘Why Exactly is Commitment Important for Rationality?’’ Economics and Philosophy 21/1: 5–14.

identity and commitment 335 Tuomela, R. 1995. The Importance of Us: A Philosophical Study of Basic Social Notions. Stanford: Stanford University Press. Williams, B. 1973. ‘‘Integrity.’’ In J. Smart and B. Williams (eds.), Utilitarianism: For and Against. Cambridge: Cambridge University Press, pp. 108–18. Williams, B. 1981. Moral Luck: Philosophical Papers 1973–1980. Cambridge: Cambridge University Press.


Comment


Rational Choice: Discipline, Brand Name, and Substance A M A RT YA S E N

I It is wonderful to read this collection of engaging essays. I am most grateful to Fabienne Peter and Hans Bernhard Schmid for stewarding this book. They led the project right from the pioneering idea of an interchange of this kind to this happy moment when the collective work is all done and the book is about to be published. They planned the program of research, selected authors and topics, arranged a lively workshop on this subject at St Gallen in May 2004, placed a subset of papers from this collection for publication in the journal Economics and Philosophy, edited the volume, and wrote an elegant and helpful introduction to it. As foot soldiers who cheered them on, we have excellent reasons to be grateful to them. As a point of entry into the debates on ‘‘rational choice’’ in this volume of essays, let me begin right there, with a simple—perhaps na¨ıve-looking—question. It concerns the relation between reasoning and rationality. We may have convincing reason to be grateful to Peter and Schmid, but would it actually be ‘‘rational’’ for us to be grateful to them? I would propose that prima facie there is an excellent case for an affirmative answer. If we do have decisive reasons to do something, then the presumption must be that it should be rational enough for us to do just that. Is this really as simple as that, or have I compounded the frivolity of a silly question with the flippancy of a ridiculous answer? I would argue that the connection is really straightforward enough, even though the implications of straightforward things are often quite demanding

340

amartya sen: comment

and far-reaching. In the understanding of the discipline of rational choice that I have been trying to advance over several decades, rationality of choice is primarily a matter of basing—explicitly or by implication—our volitional choices on sustainable reasoning. The discipline of rational choice, I submit, is foundationally connected with bringing our choices into conformity with the scrutiny of reason. ‘‘Rationality is interpreted here, broadly, as the discipline of subjecting one’s choices—of actions as well as of objectives, values and priorities—to reasoned scrutiny’’ (Sen 2002: 4).1 My answer to the simple question has taken the form of an assertion, but it is, in effect, also a substantive denial. Indeed, this view takes us to disputations that I have tried to present in earlier writings: that, contrary to prevailing mainstream interpretations, rational choice is not constitutively—or necessarily—about (1) maximizing one’s well-being, or (2) maximally advancing one’s self-interest, or (3) single-mindedly promoting self-goals, or (4) fulfilling so-called ‘‘internal consistency of choice.’’2 The contingent role of these more specific demands would turn on special conditionalities (for example, on the nature and exact content of our reasoned objectives), and depending on that, these old-fashioned demands may sometimes be met and at other times not. The prima facie link between what we have reason to do and what would be rational for us to choose to do has to be supplemented by critical examination of the viability of the underlying reasons. Our self-examination cannot end just with an unscrutinized conviction that we have excellent reasons to do something. We must ask further whether the reasons we have are really sustainable, and in particular whether they would survive scrutiny based on closer reasoning (with adequate reflection, and when relevant, dialogue), and on taking note of more information if and when they are available or accessible. We can not only assess our decisions given our objectives and values, we can also scrutinize the critical sustainability of these objectives and values themselves. This approach to rational choice may appear to be so general that there might well be some temptation to think that this does not amount to saying anything much at all. In fact, however, the view in question does make quite strong claims of its own, while rejecting a variety of other claims about the nature of ‘‘rational choice’’ (displacing, for example, the well-known and much-used canons of the brand-named ‘‘Rational Choice Theory’’).

rational choice: discipline, brand name, and substance 341 It is indeed easy to see—and illustrate—the constructive assertions as well as confrontational denials that go with this general approach to the understanding of rationality of choice (on this see Sen 2002: 37–52).

II What is the relevance of all this elucidation to the debates presented in this collection of essays? There are, in fact, several points of relevance. For example it leads us to notice—and thus take note of the fact—that none of the contributions in this collection actually attempts to re-establish the viability or claimed cogency of the mainstream approaches I have been arguing against (even though this may not be invariably clear in every case from the chosen language of the essays).3 They deal, in fact, with quite different—and more interesting—issues, sometimes involving revising the claims of mainstream theory in different ways that the authors see as more sustainable. But is there some affirmative connection as well? I would argue there is. Indeed, many of the contributions are largely—even entirely—constructive. Indeed, a number of the contributions included here helpfully illuminate the extensive reach of reasoning about values and priorities as well as about instruments and procedures. As it happens, many of these contributions are explicitly in conformity with my own position, while going, often enough, very substantially beyond what I have tried to say. I have obvious reason to be grateful for the explorations presented in these contributions. This type of constructive intellectual engagement can be found, for example, in John Davis’ (2007) illuminating analysis of the different ‘‘aspects of the self’’ that identitybased reasoning can bring out. It is also plentifully present in Hans Bernhard Schmid’s definitive (2005, 2007) exploration of ‘‘the role of shared desires.’’ The relation between social psychology and rational choice gets skilfully examined in Margaret Gilbert’s (2007) investigation of the place of ‘‘collective intentions’’ in behavioral commitment. These and a number of other essays have explored different features of decisional reasoning in making sensible choices. There are many constructive achievements there, illustrating the reach of reason, with which I am deeply concerned in my own understanding of rationality of choice.

342


In other cases, the contributions have taken the form of showing ‘‘where Sen goes wrong’’ by presenting rich explorations of reasons for action of a kind that contrasts with, it is presumed, my own thinking. Two excellent examples of this genre of contributions are Philip Pettit’s (2005, 2007) investigation of goal-based reasons for commitment when those reasons are not confined to ‘‘one’s pre-existing, integrated goals’’ (p. 31), and Daniel Hausman’s (2005, 2007) discussion of the plausibility of preference-based explanation of reasoned choices provided preferences are defined in a particular way (it is in this context that Hausman offers his proposal that we see preference as ‘‘all-things-considered rankings,’’ 2005: 50). While I greatly appreciate these contributions too (I have learned a lot from them), I must discuss why I am unable to agree entirely with their proposed take. But that must come later. First I want to talk briefly about a basic interpretational issue, and comment on some other essays, before returning to Pettit and Hausman.

III I must begin by noting that I am impressed to see how often the alleged differences—when there are differences—seem to turn on the ambiguity of the basic terminology involved. The lack of clarity relates not only to the understanding of such central concepts as ‘‘personal goals’’ and ‘‘individual preferences,’’ but also to the foundational idea of ‘‘rational choice’’ itself and to its relation with so-called ‘‘Rational Choice Theory’’ as a specific—brand-named—theory of a particular school of thought. I must confess to some surprise in seeing the belief, directly or indirectly presented in more than one paper, that I must somehow be hostile to the idea of rational choice itself (though, happily, they do not think that I am entirely beyond hope of reform), rather than being engaged—as I see my own position—in protesting against inadequate formulations of rationality of choice precisely because rational choice is so important to us. To a great extent, the misunderstanding arises from identifying my criticism of the standard formulations of rational choice in mainstream economics (and particularly my rejection of what is called ‘‘Rational Choice Theory’’—a theory of a particular and highly influential school of thought) with some kind of disapproval of the idea of rational choice itself.4 The latter is

rational choice: discipline, brand name, and substance 343 not—and has never been—any part of my claims or beliefs (or, for that matter, temptations). On the contrary, it is precisely because I believe that the idea of rational choice is viable, cogent, and critically important that I have to be so insistent that we should not shortchange it.5 Indeed, the domain of practical—and fruitful—application of the idea of rational choice is, I have argued, very large. Among the possible helps we can get from the idea of rational choice are the following: (1) examining the rationality of choices can be an extraordinarily valuable exercise for anyone to make sure that he or she is acting sensibly; (2) the underlying rationale of people’s choices can explain a lot of the decisions we actually take (though, alas, not all the time—that is why scrutiny is so important); (3) presumption of reasoned choice can be very helpful in trying to understand and interpret what others are trying to do; (4) the idea of rational choice can have great moral and political usefulness if we insist on making room for reasons that reflect relevant normativities (for example, as John Rawls (1971) has done through the ideas of ‘‘fairness’’ and ‘‘reasonableness,’’ for which the discipline of rational choice is necessary but not sufficient); and (5) examining the rationality of public decisions can greatly contribute to making—and assessing—public policy.6 While I am very aware that sometimes people do not reason very clearly or cogently, and choose one thing rather than another without adequate reflection or with insufficient engagement in the decisional process, I do not believe that, in general, dissociation of choice from reasoning is a sweeping characteristic of the world in which we live. While more scrutiny of decisions is sometimes, indeed quite often, needed in reviewing decisions, that itself tells us something about the importance of the idea of rational choice as a critically evaluative discipline. If I am supportive of the idea of rational choice, then—it can very well be asked—what am I grumbling about? As was stated earlier, aside from my constructive claims, my concentration has been on identifying why some of the mainstream models of rational choice go hopelessly wrong. My old essay, published thirty years ago, ‘‘Rational Fools: A Critique of Behavioural Foundations of Economic Theory’’ (Sen 1977), began with pointing to

344


the limitations of the model of rationality used by the great economist Francis Edgeworth (1881) as a foundation for his behavioral investigation of ‘‘mathematical psychics’’: ‘‘the first principle of economics is that every agent is actuated only by self-interest.’’ I argued for a broader approach with ‘‘structural extensions in the conception of preference made necessary by behaviour based on reasoned assessment of commitment.’’7 The more recent formulations of the unique dominance of ‘‘self-love,’’ as Adam Smith called it (not in praise—I should add—contrary to a common ‘‘rumour’’), which can be found plentifully in what is now called ‘‘Rational Choice Theory,’’ take somewhat less constricted form. They do rely still on tying up rationality of choice invariably with maximizing one’s own utility. Gary Becker’s recent work on ‘‘accounting for tastes’’ (Becker 1996) illuminatingly explores the implications of the fact that people need not be self-centered and may take note of others’ interests within their own utility. But Becker’s (1996) new analysis, while breaking considerable fresh ground, does not depart from the basic beliefs presented earlier in Becker (1976): ‘‘all human behavior can be viewed as involving participants who (1) maximize their utility (2) from a stable set of preferences and (3) accumulate an optimal amount of information and other inputs in a variety of markets’’ (p. 14). The bone of contention, thus, remains. It should not be hard to see that my criticism of this particular way of ‘‘defining’’ rational choice (including my rejection of the so-called ‘‘Rational Choice Theory’’) implies no skepticism whatsoever of the idea of rational choice in general—nor of its relevance, power, and reach when it is formulated with adequate room for the versatility of human motivations.8 I had assumed that I had left no great opportunity of misunderstanding, since I had pointed repeatedly to the distinction—and to the dissonance—between the so-called Rational Choice Theory, on the one hand, and rational choice as a general discipline, on the other. I had, in fact, gone on to supplement that discussion by expressing my wonder at the general strategy used in so-called Rational Choice Theory of trying to reshape and sharply reduce general and capacious ideas—in this case of rational choice—through the simple device of ‘‘the hidden force of a definition.’’ I noted that I have seen the use of the same dubiously implicit technology in other fields as well: ‘‘I have wondered about this general strategy also while drinking the brand-named ‘‘Best Bitter’’ in England and while staying in the ‘Best Western Grosvenor Hotel’ at the San Francisco

rational choice: discipline, brand name, and substance 345 airport’’ (Sen 2002: 27). There should be no great difficulty in seeing what the target of my criticism has been.

IV Many of the papers in this collection have made substantial contributions to a variety of specific issues and subjects in the discipline of rational choice. I will not, however, attempt to identify each of the departures that seem to me to be important, for there are so many of them. Nevertheless, I would like to point briefly to a few of the more general ideas presented that could be easily acknowledged. Herlinde Pauer-Struder’s (2007) analysis of the respective disciplines of instrumental rationality and broader public reason, taking note of the reasoning behind goals, ends, and commitments, is clearly a very helpful contribution, which clarifies an important distinction. Her pointer to the similarity between some of my claims and the robust discipline of Kantian reasoning in different forms is also perceptive and helpful. Since the assessment of rationality of choice can involve broadly inclusive as well as narrowly instrumental arguments, her analysis helps to show the possibility of distinct kinds of evaluations that lie behind the idea of rational choice. Geoffrey Brennan’s (2007) distinction between ‘‘the rational actor approach’’ as ‘‘an approach rather than as a theory’’ (italics added) seems to me to be a good way of seeing the contrast between a general discipline, on the one hand, and a specific set of theories, on the other. I am particularly grateful that Brennan clarifies that I fall ‘‘firmly within the rational actor school,’’ when that is seen ‘‘as an approach,’’ rather than in terms of specific theories advanced by the so-called ‘‘Rational Choice Theory’’ (henceforth, RCT) school. Given all this, I am a little surprised that he does not attach much importance to the distinction between ‘‘commitment’’ and ‘‘sympathy.’’ Sympathy as an idea—whereby someone else’s well-being affects yours—is widely included in standard RCT, in the ‘‘second-generational formulations,’’ such as Becker (1996). In contrast, the recognition of ‘‘commitment,’’ which takes us (1) beyond one’s own well-being to having more inclusive goals, and (2) sometimes even beyond the single-minded pursuit of one’s own goals (for example, through selfimposed constraints of social behavioral norms) is fairly uniformly absent

346


in these RCT theories, in newer as well as older versions. I will not, however, pursue the distinction further here, particularly since my own contribution in this volume (‘‘Why Exactly is Commitment Important for Rationality?’’) deals with that subject, among others. Güth and Kliemt (2007) clarify the importance of commitment in a very useful way, bringing out a number of relevant contrasts. Their distinction between the specificities of RCT and the generalities of ‘‘rational choice modelling’’ is also illuminating, and, I would argue, it relates closely to the distinction, which has been one of the points on which I have focused, between rational choice as ‘‘a discipline’’ and the particular claims of RCT as a very specific theory. Their analysis, including that of the features of rational choice modelling, has the effect of illuminating the nature of the basic contrasts involved. Bruno Verbeek’s (2007) essay comments on different types of reasoning that can go into what he calls ‘‘rational self-commitment.’’ This kind of investigation, which adds substantially to our understanding of the general discipline of rationality, is very much in line with what I believe is the real content of the subject of rational choice (as opposed to the mechanics of RCT). There is much enrichment also in the experimental perspective carefully explored by Gächter and Thöni (2007). Their investigation of reasons for cooperation in games like the Prisoner’s Dilemma is certainly very helpful, whether or not we agree with the particular explanations they prefer over other lines of explanation of experimentally observed cooperation. There is very great force in their general point that in understanding the reasoning underlying rational choice it is extremely useful to look at the way people actually solve the cooperative problems.9 While the literature on which I have been drawing (based mainly on scrutinies of actual societal behavior, as opposed to observations in specially designed experiments), has the merit of being less artificial and lab-oriented, the experimental games have the alternative merit of being more easily interpretable with clearer identification of distinct influences, without the ambiguities that societal observations typically include. I have already commented on the constructive projects of exploring group objectives and motivations explored by Schmid (2007), Gilbert (2007), and Davis (2007), and I certainly appreciate the clarity they bring to this difficult field.10 The pursuit of social objectives has also been forcefully

rational choice: discipline, brand name, and substance 347 investigated by Gold and Sugden (2007) in terms of ‘‘team agency,’’ with alternative theories of team-based reasoning. These issues relate closely also to what Raimo Tuomela (2007) calls the ‘‘we-perspective,’’ which is extensively explored in his paper in this volume (chapter 10).11 One of the issues to be addressed in the use of the ‘‘we-perspective’’ is the plurality of identities that we have, which makes the ‘‘we-perspective’’ also, frequently enough, irreducibly plural. ‘‘What we do together’’ cannot but be parametric in form, through context-dependent variability of the content and composition of what we take to be the relevant ‘‘we,’’ so that the divergence between ‘‘I’’ and ‘‘we’’ is very far from a unique and straightforward contrast (on this see Sen 2006: ch. 2). These are, in fact, as Davis (2007) insightfully explores, significant aspects of the multiple-identities of the individual.

V In reporting on the confrontations with which this book is inter alia concerned, and on which the subset of the papers that formed the symposium in Economics and Philosophy (2005) particularly concentrated, the editors, Fabienne Peter and Hans Bernard Schmid, put a central problem in the form of a conundrum at the very beginning of their Introduction (Peter and Schmid 2005: 1): In his critique of rational choice theory, Amartya Sen claims that committed agents do not (or not exclusively) pursue their own goals. This claim appears to be nonsensical since even strongly heterogenous or altruistic agents cannot pursue other people’s goals without making them their own.

This is an interesting line of scrutiny which is closely connected, as will be presently discussed, with the extensive arguments presented by Pettit (2005, 2007) and Hausman (2005, 2007). Is there a real dilemma here? I would argue not. A person’s decision not to act only—or even mainly—on the basis of his or her own goals does not require that the person must then be acting according to some ‘‘other people’s goals.’’ One can take note of other people’s goals and priorities and decide to constrain the unifocal pursuit of one’s own goals

348


with behavioral constraints and other restrictions, without that self-restraint being interpreted as the pursuit of the goals of others. Schmid (2005, 2007) himself has presented a way of getting out of this apparent contradiction by bringing in ‘‘shared goals’’ that individuals may have. That is certainly one important direction in which the relation between objectives and choices can be enriched, and having proposed some ideas myself on this (for example, in Sen 1977, 1985), I am very happy that Schmid (2005, 2007) has investigated this kind of enrichment in a definitive way. Self-goal choice as an essential part of rational choice certainly does run into difficulty when goals take this kind of ‘‘joint’’ form. So far so good, but the question that has to be asked is this: is the existence of the shared goals the only reason for removing self-goal choice as a necessary condition for rational choice? Are we forced to accept the diagnosis that the dethroning of self-goal choice would be simply ‘‘nonsensical’’ in the absence of such joint goals? I would argue that the problematic nature of self-goal choice is much more extensive than that. The central issue that the presentational passage of Peter and Schmid misses is that in denying that rationality demands that you must act only according to your own goals (subject only to constraints that are not selfimposed), you do not necessarily endorse the goals of others. The possibility of following voluntary rules of behavior restrains the unique dominance of single-minded pursuit of your own goals, and that is a critically important limitation of seeing rationality exclusively in terms of ‘‘self-goal choice.’’ Let me consider, first, a simple example. You are occupying a window seat in a plane journey, with the window shade up, when you are requested by the occupant of the aisle seat next to you to pull down the shade so that he can see his computer screen better to be able to devote himself fully to playing some computer game, which in your view is a ‘‘plainly silly’’ game (‘‘what a waste of time!’’). You are frustrated that there is so much ignorance around, with so many people playing inane games rather than reading the news—to bone up on what is actually happening in Iraq, or Afghanistan, or for that matter in inner-city America. You decide, nevertheless, to comply with the game-enthusiast’s request to help him see the computer screen, and you oblige him by pulling the shutter down. How do you explain your choice? There is no difficulty in understanding that you are not averse to helping your neighbor—or anyone else—pursue his or her well-being, but it so happens that you do not think that your

rational choice: discipline, brand name, and substance 349 neighbor’s well-being is, in fact, best advanced by his wasting his time—or by your helping him to waste his time—on a silly game (you remain absolutely willing to lend him your copy of The New York Times, reading which would be, you are convinced, much better for his edification and well-being). Your action is not a corollary of any general pursuit of well-being of all. The central issue, rather, is whether to impose—or refuse to dismantle—barriers to the pursuit of other people’s goals, when these goals are not in any sense evil, even if—as in this case—you think that they are not conducive to promoting their own well-being. Perhaps you are reluctant, as a general rule, to be unhelpful to your neighbors (no matter what you think of their goals). Or perhaps you judge that while having a window seat gives you control over the proximate shutter, this incidental advantage should not be used without taking into account what others want to do (even though you yourself were rather enjoying the sun which would now be shut out and even though you do not think much of the goal that the other guy wants to pursue). Is it right to interpret that your socially normed behavior shows that your objective is to help all others to pursue their own goals, no matter what you think of their goals? Thanks to your acceptance of social norms of behavior, you have certainly ended up helping the guy next to you to pursue his own goal. But it is surely too much to say either that your objective is to maximally help everyone to pursue his or her goals, or that his goals have somehow become yours as well. Rather, you are just following a norm of good behavior you happen to approve of (to wit ‘‘let others be’’), which is a self-imposed behavioral restraint you end up accepting in your choice of what to do.

VI What is the connection of all this to the critiques of Philip Pettit (2005, 2007) and Daniel Hausman (2005, 2007)? Some of the connections are fairly straightforward. Peter and Schmid’s summary statement is, in fact, a derivation from Pettit’s arguments, and my response to this part of Pettit’s critique would involve the same line of reasoning that I have already presented in responding to the Peter–Schmid statement.

350


But before I go into that issue, let me comment on a misinterpretation of my arguments that seems to have crept into Pettit’s presentation (this is one of two misinterpretations that I must correct—the second will be taken up later). Pettit (2005) states in the very first sentence of his engaging essay that I claim ‘‘that there is an altruistic attitude toward others that does not make sense within the terms of rational choice theory, however broadly that theory is construed’’ (p. 15). On the contrary, my claim is that such attitudes can be very well accommodated within a theory of rational choice if it is interpreted in an adequately broad way, which the narrow format of RCT (brand-named ‘‘Rational Choice Theory’’) fails to take into account. I have already discussed this misinterpretation earlier on in this essay (in particular in section III), and will not comment on it again here. So I turn now to what clearly is the main bone of contention. Pettit says (2005, pp. 15–16; see also 2007, p. 28): On Sen’s conceptualization, it [commitment] involves putting aside one’s own goals and acting on those of another. But how could one ever fail to act on one’s goals? The picture offends against a picture of human psychology accepted on almost all sides, whether among the professionals or among the folk.

There are two problems in this passage. The first does not, in fact, involve psychology at all. Commitment, which was defined as ‘‘breaking the tight link between individual welfare ... and the choice of action’’ (Sen 1982: 8), need not necessarily involve, unlike what Pettit assumes, ‘‘putting aside one’s own goals.’’ Indeed, the insistence that rationality demands exclusive pursuit of self-interest or promotion of self-welfare can be disputed both by questioning (1) whether one’s own goals must be based on self-interest only, and (2) whether one’s own goals must be the only determinants of reasoned choice. It would be committed behavior all right if one acts entirely according to one’s goals when those goals include consideration of the interests of others—over and above the extent to which one’s own well-being or interest is directly influenced by the well-being or interest of others. Committed behavior, in the sense I outlined, can arise from violating either (1) or (2), or of course both. The role of ‘‘commitment’’ is, thus, much broader than the violation of (2) only, contrary to Pettit’s interpretation.12 The second problem does relate to the violation of (2). This is the context in which Philip Pettit invokes professional as well as folk psychology.

rational choice: discipline, brand name, and substance 351 Committed behavior can—though not necessarily—take the form of violating the exclusive dominance of self-goal choice, so that considering the priorities of others can influence action even without its being a part of one’s self-understood goal. I can understand Pettit’s belief that our ‘‘common sense about action’’ would be offended by our ‘‘putting aside one’s own goals and acting on those of another.’’ This is where the second misinterpretation is quite prominent: one need not be guided entirely by one’s own goals and yet the departures need not take the form of pursuing other people’s goals. As was discussed in the last section, to restrain the single-minded pursuit of one’s own goals in consideration of others does not require one to follow those other people’s goals; indeed behavioral restraint is a very different subject from acting in pursuit of other people’s goals without making them one’s own. I don’t know how upset the professional or folk psychologists would be if the occupant of the window seat (discussed in the last section) were to confess that he is agreeing to pull down the shade without making it his ‘‘own goal’’ that his neighbor plays his silly games. I hope not much, but if their good sense is offended (assuming that Pettit is right), then I would strongly urge the professional and folk psychologists to re-examine their theories. I must, however, add that I am touched by the effort that Pettit has put in, in his characteristically generous and intellectually stimulating way, to try to explain what could have driven me to set myself ‘‘against our basic common sense about action’’ (p. 21). In fact, the explanation is not as complex as the rather sophisticated ‘‘errors’’ that he sympathetically attributes to me and patiently corrects (pp. 22–31). Can we interpret my attempted explanation of the window-seat occupant’s socially normed choice in line with Pettit’s (2005) point that I confuse ‘‘the sort of deliberation that involves the formation of a novel, perhaps occasion-specific goal in which the good of another is prioritized’’ as ‘‘deliberation on the basis of commitment’’ (p. 15). I would argue against this view, but let me begin by distinguishing between (1) Pettit’s insistence that the self-restraining guy is actually pursuing ‘‘the good of another,’’ and (2) his insistence that this pursuit is really that of my own ‘‘novel’’ goal, which I have come to adopt (perhaps as an ‘‘occasion-specific goal’’). To take the idea of ‘‘good’’ first, the invoking of the idea that ‘‘the good of another’’ is being pursued runs into the problem that the shade-puller

352


does not think that it is particularly good for the game-fanatic to play his silly game (rather than, say, reading The New York Times which is available to him thanks to the courtesy of the shade-puller). Rather, as the shade-controller occupying the decisional position of occupying the window seat, you are merely allowing the game-maniac to do what he wants to do (which may or may not be good for him—but that you see as ‘‘his business’’). Now to turn to the second—and principal—problem, we run into the reasoning that Peter and Schmid nicely summarize: people ‘‘cannot pursue other people’s goals without making them their own.’’ Your shade-pulling is, thus, to be seen, in this line of understanding, as the pursuit of your own—allegedly recently adopted and ‘‘novel’’—goal. But is that really what is going on? By agreeing to pull the shade down, are you to be seen as acting according to the game-player’s goal? That, as was discussed earlier, is not at all a good description of what is actually happening. You are not actively ‘‘pursuing’’ another person’s goals, but just letting that person pursue his: you think this is a reasonable restraint on your own behavior. It would be fairly odd to claim that through this restraint you actually show that you share your neighbor’s goal that he should devote himself to that mindless game (that would surely be to add insult to injury!). Indeed, even when we consider the converse case, in which you want others to respect your goals rather than making it impossible for you to pursue them, you need not go so far as to demand that others must forthwith take your goals to be theirs (that would be ‘‘bossiness’’ beyond the call of ‘‘live and let live’’). All you would reasonably want is that they, in letting you be, refrain from those actions that would make it infeasible for you to pursue your goals.

VII Daniel Hausman’s (2005, 2007) main thesis relates to the issue of maximization as well, though this may not be entirely obvious from the way he presents what he sees as his principal argument. He sees his essay as contesting my ‘‘view that economists should recognize multiple notions of preference,’’ presenting a counter-thesis that ‘‘Sen’s concerns are better

rational choice: discipline, brand name, and substance 353 served by embracing a single conception of preference and insisting on the need for analysis of the multiple factors that determine ‘preference’ so conceived’’ (Hausman 2005: 33). I have indeed argued that economists—and philosophers too—use the term ‘‘preference’’ to represent many different things. My main concern in making this clarificatory point is that we should not be confused into thinking that all the uses of the term preference involve the same subject matter. I have given examples to indicate how some of the actual difficulties in the literature have arisen from not taking enough note of the plurality of senses in which the term preference is used. I am far less worried about whether the right way of sorting out the confusion is to recognize explicitly the different senses of the term preference (which does not go to battle with prevailing language but insists on clarity of use in each case), or to demand that the idea of preference be restricted to a ‘‘single conception,’’ in particular as an ‘‘all-things-considered ranking’’ (what Hausman recommends). Indeed, if that is all that is involved, then I am willing to encourage Hausman to give his favorite solution a shot, though I will certainly wait to see how easily he might be able to change the prevailing usage of language. In my attempts at construction, rather then critique, I have tried to argue for the view that, subject to some interpretational clarification, ‘‘the discipline of maximization (without ruling out incomplete preference, menu dependence, or process sensitivity) can give us a good understanding of an important part of the discipline of rational choice’’ (Sen 2002: 41). If Hausman wants to call that maximand by the name ‘‘preference’’ and manages to persuade all others not to use the word ‘‘preference’’ in any other sense, I would certainly not grumble (even though the hidden linguist in me would feel a little neglected). So that is not a central issue that would irreconcilably divide Hausman and me.13 In fact the division between Hausman’s substantive—as opposed to linguistic—position and mine is not that great. Insofar as there is a division, it links with the issue discussed in my response to Philip Pettit. Hausman’s defense of his insistence that it is only an all-things-considered ranking that can correctly predict—or explain—a person’s choice of action is based on having decisional problems explicated in terms of simple maximization when all constraints are externally imposed. But the violation of self-goal choice is arising here from the normative restraint we may voluntarily

354


impose on ourselves on grounds of recognizing other people’s pursuits and goals, without in any substantive sense making them our own goals. That issue has been discussed rather extensively already, and I will not repeat the same arguments here. We do, of course, know that mathematically any constraint can be seen as a surrogate goal—I need not launch a huge discussion of the role of ‘‘Lagrangean multipliers’’ in decisional analysis to make this point. But what we are talking about here is not the mathematical equivalence of goals and constraints in specified circumstances, with the invoking of ‘‘surrogate goals’’ facilitating the ‘‘traverse’’ from one to the other. What is at issue here is the diagnosis of what can be rightly described as your actual goals that you want to pursue, not an ‘‘as if maximand’’ inter alia incorporating the effects of self-imposed constraints that you decide you should not violate.

VIII So how does behavior based on ‘‘commitment’’ diverge from what RCT demands, even in its more sophisticated form, as elucidated in particular by Becker (1996), going beyond Becker (1976)? Commitment involves, in general, surmounting the RCT-postulated ‘‘tight link between individual welfare and the choice of action’’ (Sen 1982: 8). Even in the later sophisticated version of RCT (as in Becker 1996), which avoids the limiting assumption of ‘‘self-centered welfare’’, there is still complete adherence to ‘‘self-welfare goal’’ and ‘‘self-goal choice.’’ These are the points at which possible departures have to be allowed to do justice to the general idea of rational choice, rather than the brand-named doctrine of ‘‘rational choice’’ codified in RCT. Commitment would be involved in both the departures if they were to occur. Committed behavior may work either (1) through having broader goals by taking note of other people’s interests or concerns even beyond the extent to which those interests and concerns directly influence one’s own well-being (this would involve violation of ‘‘self-welfare goal’’), or (2) through restraining one’s single-minded pursuit of one’s own goals without paying any attention to the priorities of others (this would involve transcending ‘‘self-goal choice’’). They can, of course, also work together.

rational choice: discipline, brand name, and substance 355 If one’s goals are modified to take into account other people’s goals (even beyond the extent to which those goals directly influence one’s own wellbeing), and if one single-mindedly pursues one’s own thus-modified goals, subject only to externally imposed—and given—constraints, then there would be no violation of self-goal choice. But it would still be a case of committed behavior since the goals one would pursue fully and single-mindedly would no longer be simply a reflection of one’s own well-being. If, on the other hand, one restrains the pursuit of one’s goals to make room for others to pursue their goals (whether or not one has already taken some note of other people’s concerns in one’s own goals), then self-goal choice would indeed be violated, and that is a second way in which committed behavior may be observed. The two types of departures may, of course, act together, and sometimes they may be hard to disentangle precisely, but analytically there are two distinct issues here that need to be separated out for clarity about the demands—and non-demands—of rational choice as a discipline. However, even the latter scenario of violation of self-goal choice does not imply that one has decided to maximally advance ‘‘the goals of others.’’ Awareness and concern about other people’s goals can lead to voluntary imposition of some constraints on one’s own actions when others are also involved, without that being interpretable as simply devoting oneself to the pursuit of ‘‘other people’s goals.’’ There is, after all, a difference between restraint and conversion. Notes

1. My early attempts at this way of understanding rationality (and the extensive departures that they demand, away from the more conventional ways of seeing rationality) include Sen (1970: ch. 1; 1973, 1974, 1977). My later efforts, with further exploration of the variety and reach of social as well as individual reasoning, are presented in, among other places, Sen (1985, 1993, 1997, 2002). 2. While the first three of the identified denials are easy enough to interpret (though for many mainstream theorists of rationality, perhaps hard to accept), the last disclaimer calls for some clarification. The denial of ‘‘internal consistency of choice’’ does not, of course, imply that rationality does not have any consistency requirements. Indeed any use of reasoned scrutiny must inter alia involve consistent use of

356


reasoning. Rather, the claim is that, applied to choices of actions, the idea of ‘‘internal consistency’’ is an inapplicable notion (involving a ‘‘category mistake’’), since internal consistency is the kind of thing that happens (that is, gets identifiably fulfilled or violated) to statements or assertions or claims, rather than to actions, or choices of actions. To check the congruity of choices made, we have to bring in some interpretations of the motivations underlying the observed choices and this, of course, is ‘‘external’’ to the choices themselves (and therefore, is not a matter of internal consistency of choice). If, on the other hand, motivational interpretations are left free, then they can always be suitably adapted to make any set of choices of actions look entirely congruous. For fuller—and more formal—discussion of the analytical and interpretational issues involved, see my two Econometrica papers on this subject, Sen (1993, 1997), and also an entirely informal—indeed chatty—essay, Sen (1996). 3. The non-pursuit of ‘‘old follies’’—if I may call them that—contrasts not only with the content of the dominant interpretations of rational choice in mainstream economics, but also with a substantial part of contemporary political and legal analyses, in the form of ‘‘rational choice politics’’ and mainstream ‘‘law and economics.’’ 4. Indeed, since the expression ‘‘rational choice theory’’ is used in some of the papers both as the term for a discipline and as the name of a particular school and a specific theory (to wit, RCT), the ambiguity is hard to avoid in these expositions. 5. The nature of my claims and proposals and related issues has been illuminatingly discussed and precisely clarified by Elizabeth Anderson (2005) with her characteristic lucidity. 6. See Sen (1973, 1974, 1982, 1985, 1997), and the summary presentation in my introductory essay (‘‘Rationality and Freedom’’) in Sen (2002), including the discussion on ‘‘What’s the Use of Rationality?’’ (pp. 42–52). 7. Sen (1977); in the reprint of this essay in Sen (1982), pp. 84, 105. See also the critiques presented by Walsh (1996), Jolls, Sunstein, and Thaler (1998), Rabin (1998), Tirole (2002), and Benabou and Tirole (2003). 8. See Anderson (2005) for a clear statement of what I am trying to say and why. See also her own analysis of the demands of rational choice in Anderson (1993).

rational choice: discipline, brand name, and substance 357 9. On this see also Rabin (1993, 1998), Fehr and Schmidt (1999), Fehr and Gächter (2000, 2002), Tirole (2002), and Fehr and Fischbacher (2003). 10. See also the far-reaching analytical and conceptual investigations in Akerlof (1984) and Akerlof and Kranton (2000, 2005). 11. On related issues, see also Davis (2003), Schmid (2003), and Kirman and Teschl (2004). 12. Indeed, beginning with my early writings on this subject (Sen 1973, 1974, 1977), I have been attempting to persuade mainstream economists to break ‘‘the tight link between individual welfare and the choice of action,’’ which so limits the subject (a limitation that would be later embraced also by the—impressively self-immolating—‘‘rational choice political theorists’’ and mainstream exponents of ‘‘law and economics’’). Adapting one’s goals to take note of the interests or priorities of others (even beyond the extent to which those interests and priorities directly influence one’s own well-being) and going beyond one’s own goals—broad or narrow—to restrain the singleminded and exclusive pursuit of one’s own goals are two distinct ways of surpassing ‘‘the tight link between individual welfare and choice of action.’’ 13. There are, of course, also other ‘‘parts’’ of the discipline of rational choice (as I have tried to discuss), including the need to subject our ‘‘objectives, values and priorities to reasoned scrutiny’’ (Sen 2002). I cannot think that Hausman would dissent from that necessity.

References Akerlof, George A. 1984. An Economic Theorist’s Book of Tales. Cambridge: Cambridge University Press. Akerlof, George A., and Rachel E. Kranton. 2000. ‘‘Economics and Identity.’’ Quarterly Journal of Economics 115: 715–53. Akerlof, George A., and Rachel E. Kranton. 2005. ‘‘Identity and the Economics of Organization.’’ Journal of Economic Perspectives 19: 9–32. Anderson, Elizabeth. 1993. Value in Ethics and Economics. Cambridge, Mass.: Harvard University Press. Anderson, Elizabeth. 2005. ‘‘Critical Notice of Amartya Sen, Rationality and Freedom.’’ Philosophical Review 114/2: 253–71.

358 amartya sen: comment Becker, Gary S. 1976. Economic Approaches to Human Behavior. Chicago: Chicago University Press. Becker, Gary S. 1996. Accounting for Tastes. Cambridge, Mass.: Harvard University Press. Benabou, Roland, and Jean Tirole. 2003. ‘‘Intrinsic and Extrinsic Motivation.’’ Review of Economic Studies 70: 489–520. Brennan, Geoffrey. 2007. ‘‘The Grammar of Rationality’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 313–35 (this volume). Davis, John B. 2003. The Theory of the Individual in Economics: Identity and Value. London: Routledge. Davis, John B. 2007. ‘‘Identity and Commitment: Sen’s Conception of the Individual.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 313–35 (this volume). Edgeworth, Francis Y. 1881. Mathematical Phychics: An Essay in the Application of Mathematics to the Moral Sciences. London: C. Kegan Paul and Co. Fehr, E., and U. Fischbacher. 2003. ‘‘Nature of Human Altruism.’’ Nature 425: 785–91. Fehr, E., and S. Gächter. 2000. ‘‘Cooperation and Punishment in Public Goods Experiments.’’ American Economic Review 90: 980–94. Fehr, E., and S. Gächter. 2002. ‘‘Altruistic Punishments in Humans.’’ Nature 415: 137–40. Fehr, E., and K. Schmidt. 1999. ‘‘A Theory of Fairness, Competition, and Cooperation.’’ Quarterly Journal of Economics 114: 817–68. Gächter, Simon, and Christian Thöni. 2007. ‘‘Rationality and Commitment in Voluntary Cooperation: Insights from Experimental Economics.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 175–208 (this volume). Gilbert, Margaret. 2007. ‘‘Collective Intentions, Commitment, and Collective Action Problems.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 258–79 (this volume). Gold, Nathalie, and Robert Sugden 2007. ‘‘Theories of Team Agency.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 280–312 (this volume). Güth, Werner, and Hartmut Kliemt. 2007. ‘‘The Rationality of Rational Fools: The Role of Commitments, Persons, and Agents in Rational Choice

rational choice: discipline, brand name, and substance 359 Modeling.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 124–49 (this volume). Hausman, Daniel M. 2005. ‘‘Sympathy, Commitment, and Preference.’’ Economics and Philosophy 21: 33–50. Hausman, Daniel M. 2007. ‘‘Sympathy, Commitment, and Preference.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 49–69 (this volume). Jolls, Christine, Cass Sunstein, and Richard Thaler. 1998. ‘‘A Behavioral Approach to Law and Economics.’’ Stanford Law Review 50: 1471–550. Kirman, Alan, and Miriam Teschl. 2004. ‘‘On the Emergence of Economic Identity.’’ Revue de Philosopie Economique 9: 59–86. Pauer-Struder, Herlinde. 2007. ‘‘Instrumental Rationality versus Practical Reason: Desires, Ends, and Commitment’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 73–104 (this volume). Peter, Fabienne and Hans Bernhard Schmid. 2005. ‘‘Introduction. Symposium on Rational Commitment.’’ Economics and Philosophy 21: 1–4. Peter, Fabienne and Hans Bernhard Schmid. 2007. ‘‘Introduction. Rational Fools, Rational Commitments.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 3–13 (this volume). Pettit, Phillip. 2005. ‘‘Construing Sen on Commitment.’’ Economics and Philosophy 21: 15–32. Pettit, Phillip. 2007. ‘‘Construing Sen on Commitment.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 28–48 (this volume). Qizilbash, Mozaffar. 2002. ‘‘Rationality, Comparibility and Maximization.’’ Economics and Philosophy 18: 141–56. Rabin, Matthew. 1993. ‘‘Incorporating Fairness into Game Theory and Economics.’’ American Economic Review 83: 1281–302. Rabin, Matthew. 1998. ‘‘Psychology and Economics.’’ Journal of Economic Literature 36: 11–46. Rawls, John. 1971. A Theory of Justice. Cambridge, Mass.: Harvard University Press. Schelling, Thomas C. 1984a. Choice and Consequence. Cambridge, Mass.: Harvard University Press. Schelling, Thomas C. 1984b. ‘‘Self-Command in Practice, in Policy, and in a Theory of Rational Choice.’’ American Economic Review 74: 1–11.

360 amartya sen: comment Schmid, Hans Bernhard. 2003. ‘‘Can Brains in Vats Think as a Team?’’ Philosophical Explorations 6: 201–18. Schmid, Hans Bernhard. 2005. ‘‘Beyond Self-Goal Choice.’’ Economics and Philosophy 21: 51–64. Schmid, Hans Bernhard. 2007. ‘‘Beyond Self-Goal Choice.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 211–26 (this volume). Sen, Amartya K. 1970. Collective Choice and Social Welfare. San Francisco: HoldenDay; reprinted, Amsterdam: North-Holland, 1979. Sen, Amartya K. 1973. ‘‘Behaviour and the Concept of Preference.’’ Economica 40: 241–59; reprinted in Sen (1982). Sen, Amartya K. 1974. ‘‘Choice, Ordering, and Morality.’’ In S. Kroner (ed.), Practical Reason. Oxford: Blackwell, pp. 4–67; reprinted in Sen (1982). Sen, Amartya K. 1977. ‘‘Rational Fools: A Critique of the Behavioural Foundations of Economic Theory.’’ Philosophy and Public Affairs 6: 317–44; reprinted in Sen (1982). Sen, Amartya K. 1982. Choice, Welfare and Measurement. Oxford: Blackwell; reprinted, Cambridge, Mass.: Harvard University Press, 1997. Sen, Amartya K. 1985. ‘‘Goals, Commitment, and Identity.’’ Journal of Law, Economics and Organization 1: 341–55; reprinted in Sen (2002). Sen, Amartya. 1993. ‘‘Internal Consistency of Choice.’’ Econometrica 61: 495–521; reprinted in Sen (2002). Sen, Amartya. 1994. ‘‘Non-Binary Choice and Preference: A Tribute to Stig Kanger.’’ In D. Prawitz et al. (eds.), Logic, Methodology and Philosophy of Science IX. Amsterdam: Elsevier Science, pp. 913–24; reprinted in Sen (2002). Sen, Amartya. 1996. ‘‘Is the Idea of Purely Internal Consistency of Choice Bizarre?’’ In J. E. J. Altham and Ross Harrison (eds.), World, Mind, and Ethics: Essays on the Ethical Philosophy of Bernard Williams. Cambridge: Cambridge University Press, pp. 19–31. Sen, Amartya. 1997. ‘‘Maximization and the Act of Choice.’’ Econometrica 65: 745–80; reprinted in Sen (2002). Sen, Amartya. 2002. Rationality and Freedom. Cambridge, Mass.: Harvard University Press. Tirole, Jean. 2002. ‘‘Rational Irrationality: Some Economics of Self-Management.’’ European Economic Review 46: 633–55. Tuomela, Raimo. 2007. ‘‘Cooperation and the We-Perspective.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 227–57 (this volume).

rational choice: discipline, brand name, and substance 361 Verbeek, Bruno. 2007. ‘‘Rational Self-Commitment.’’ In Peter, Fabienne and Hans Bernhard Schmid (eds.), Rationality and Commitment. Oxford: Oxford University Press, pp. 150–74 (this volume). Walsh, Vivian. 1996. Rationality, Allocation and Reproduction. Oxford: Clarendon Press.


Index Definitions are indicated in bold type. action 22–3, 37–42, 53, 55, 75, 78, 81, 84, 88–90, 91–2, 94–8, 107–8, 121, 124, 127, 136, 145, 151, 154, 157, 164, 197, 211, 215–16, 218–19, 223 n. 2, 228–33, 237, 239, 242–3, 247–9, 259–64, 268, 270, 280, 282, 285, 287, 291, 296, 304–5, 307, 315–16, 320, 322–3, 325, 327, 351–2, 356 n. 2 collective 94, 96, 204 n. 4, 219, 221, 238, 240, 259–60, 269–70, 273–4, 299–301 committed 5, 41, 42, 92–3, 211–14, 216, 218, 222 cooperative 241, 249, 252 n. 18, 307 explanation of 53, 113, 353 intentional 32, 35, 232, 243, 260, 269, 315 joint 220, 234, 247, 249 n. 3, 252 n. 18, 259, 269–70, 285, 300, 304–6 moral 5, 91 non-deliberative 155 action theory 221, 273, 326, 328 actor, see agent advantage, expected 50–1, 54–58, 60–2, 64–5; see also utility, expected agency 89, 212, 215–17, 219, 260, 271, 280, 287, 292, 299, 236 collective 94, 218–21, 259–60, 268-69, 271 non-reductivist account of 220–1 joint 220–2, 236, 299, 301 transformation 292–3, 299 unit of 260, 287–8, 291, 293, 294–6, 298 agent 23, 35, 37, 42, 43, 44, 89–92, 108, 113, 125, 130–4, 142, 144, 212–13, 215–16, 218, 220, 228, 323, 231, 233, 235–6, 238, 240, 242–4, 246, 248, 251 n. 9, 253–4 n. 23, 259, 271, 274, 280–1, 286–7, 304, 307, 321, 328, 234–5, 331 collective 93, 97, 219, 271, 294, 302, 321

corporate 125, 130 rational 3, 19, 22, 35, 37, 38, 107, 124–5, 127, 128 152, 154, 169, 213, 242, 294 agreement 258, 265, 267–9, 275, 299 aims, see ends and goals Ainslee, George 126, 130 Akerlof, George A. 25 n. 9, 357 n. 10 Alchian, Armen 142 altruism 28, 30, 57, 58, 65, 68 n. 11, 108, 122 n. 1, 178, 183–4, 194, 199, 200–1, 212, 213, 215, 218, 280, 229, 280, 347350 Anderson, Elizabeth 25 n. 2, 67 n. 4, 74, 88, 92–7, 224 n. 9, 333 n. 3, 356 n. 5, 356 n. 8 Andreoni, J. 176, 179 animals 23, 37, 39, 74 Anscombe, Elizabeth 260 antipathy, see sympathy Aoki, M. 26 n. 11 Aristotle 330–1 Arrow, Kenneth J. 18, 237 Arrow’s impossibility theorem 237 association, voluntary 320 assurance 294, 301–2, 304, 308 Aumann, Robert 277 n. 16 authority 165–9, 249, 252–4 n. 23 autonomy 84, 86, 91, 96, 220 Axelrod, Robert 178, 199 Bacharach, Michael 179, 203, 280, 285, 295–300, 308, 309 n. 4 backward induction 126, 135, 139–41, 144 Baier, Annette 100 n. 11, 216, 221 Bardsley, Nicholas 186–8, 309 n. 3, 310 n. 14, 308 bargaining 96, 237, 254 n. 27 Becker, Gary 19, 344–5, 354 behaviour 3, 21, 29, 35, 37, 56, 61, 106, 113, 120, 136 145, 177–9, 183, 195,

364 index behaviour (cont.) 200, 258, 260, 271, 284, 291, 294, 307, 326, 332, 344 adaptive 199 as if 133–4 committed 20–1, 22, 47, 122, 350–1, 354–5 explanation of 21, 22, 46 maximizing 29, 32, 40, 76; see also maximization non-rational 12 n. 1 observed 3, 282, 291 patterns of 4, 20, 326 prediction of 20–1, 46, 109, 111–12, 151, 171 n. 12, 176, 178–9 rule following 143, 144 behavioural models 313, 315 belief 35, 37–41, 53–5, 60, 64, 107–8, 113, 166, 183–4, 188–90, 228, 230–1, 235, 238, 264, 280, 284, 310 n. 10 mutual 232–7, 305; see also common knowledge, higher-order 37–40, 232, 310 n. 10 belief-desire schema 35–7, 39–40, 42–5 belief-forming process 38, 40 Benabou, Roland 356 n. 7 Benn, S. I. 223 n. 1 Ben-Ner, A. 25 n. 2 Binmore, Ken 68 n. 10, 125, 179–80, 290–2 biology 202 Blackburn, Simon 100 .n 26 body, intend as a 261, 264, 269–70 Bolton, Gary 202, 291, 293 bootstrapping 161, 163, 165, 167–8, 169 Bornstein, G. 254 n. 30 Bovens, Luc 169 Bowles, S. 200, 255 n. 33 Boyd, R. 181, 200–1, 249 n. 1, 255 n. 33 Brandom, Robert B. 212 Brandstätter, H. 203 Bratman, Michael 154, 171 n. 15, 171 n. 19, 172 n. 22, 252 n. 18, 277 n. 21, 280, 307–8 Brennan, Geoffrey 8, 114, 118, 125, 145, 345 Brentano, Franz 234 Broome, John 41, 67 n. 2, 101 n. 26, 146 n. 6, 161–5, 169, 278 n. 26 Brown, R. 249 n. 5 Brundtland, Gro Harlem 23–4

Buchanan, James 111–12 Buddha, Gautama 24–5 Burdian’s Ass 168 Burlando, R. 187 business ethics 22 Camerer, Colin 175, 202 capabilities 313–14, 321, 324, 329–31 short-list of essential 313–14, 330–1 Castelfranchi, C. 254 n. 25 category mistake 219–20, 356 n. 2 chain store paradox 144 Chammah, A. 176 choice 4, 5, 17–20, 29–30, 31, 50, 52, 54–55, 56, 58, 61, 73–4, 78, 110, 124, 125, 133, 134, 136–7, 139, 143, 144, 152, 161, 170 n. 7, 181, 214–15, 244, 251, 282, 286, 291, 293, 309 n. 7, 315–18, 320–2, 325, 343, 348, 356 n. 2; see also rational choice and welfare, wedge between 4, 5, 19, 30, 75, 214, 217, 315, 317–8, 325, 350, 354 ascription of 74 autonomous 320 collective 300 consequences of 115 content of 74 counter-preferential 57–9, 64–5, 119–22, 151–3, 156, 291, 325 hypothetical 52, 61–2, 118 logic of 111–12, 117 reasoned 350 science of 111 volitional 340 circularity 276 n. 6 Clark, K. 204 n. 8 cognitive dissonance 117 Cohen, P. R. 219 collective action problem 96, 258–60, 273–5; see also cooperation problem; dilemma collective intentionality, see intentionality; action, joint; intentions, shared commitment 4, 5, 19, 20–1, 22, 24–5, 28, 29, 30–1, 32, 40, 45, 47, 56–8, 73, 75–6, 78, 86, 94, 96, 106, 108, 118–22, 125, 126, 128–9, 134, 138, 143, 144, 151–2,156, 159, 211–14, 217, 222, 229–30, 246–7, 258, 261–4, 267–8, 271–2, 275, 291, 299–300

index 365 304, 308, 313–20, 324–5, 327–32, 341–2, 344–6, 350–1, 354 absolute 129–31 causal 152–3, 159–61 collective 228–30, 238, 245–7, 252 n. 18, 253 n. 23 competing 328 devices 151, 153, 197 externalist account of 213–14 goal-displacing 31, 32, 34–5, 36, 40, 42, 45–6, 315 goal-modifying 31, 32, 40, 44, 315 Humean account of 213 identity-conferring 314, 325, 327 internal 125, 130–2, 147 n. 23 internalist account of 213–14 joint 258–9, 261, 264–72, 274–5, 308, 333 n. 7 Kantian account of 213 modelling of 126, 131–2, 138 mutual 294 non-reflective 229 of the will 262–63, 267, 271–4 oppressive 96 option of 129, 136–7 prima facie 76 private 229–30 public expression of 299 rational 159–61, 165, 169 reflexive dimension of 317, 328 relative 129–31 self- 165, 169, 171 n. 13, 346 strategic 139 structure of 211–12, 218, 222 technique 128 to principles 333 n. 1 common good 295 common knowledge 63, 125, 129, 238, 253 n. 23, 265, 267, 274, 284, 286, 288–9, 294–5, 297, 299, 301–2, 305, 308 communication, pre-play 130, 319 communitarians 320 community 318, 326 conflict 58, 65, 94, 127, 200, 213, 219, 227–8, 234, 245, 296, 305, 319, 321–2 conformism 201 consistency 3, 38, 110, 127, 145, 146 n. 7, 153, 155, 202, 340 constraint 18, 32, 38–9, 48 n. 5, 55, 78, 154–6, 167, 321, 352, 354–5

externally imposed 53, 59–60, 154, 157, 159, 353, 355 mathematical equivalence of 354 normative 353 self-imposed 58–9, 323, 329, 345, 347–9, 351–5 conventions, social 78, 258 Cooper, R. 176, 179 cooperation 102 n. 42, 124–5, 176–80, 184–5, 188–9, 193–5, 197–201, 203 n. 2, 227–3, 236, 239, 243, 245, 247, 248, 282, 285, 318, 306–7; see also Prisoner’s Dilemma ability of 227 altruistic 184, 200 behaviour 175, 193–4, 197 conditional 181, 183–92, 194 contribution 182–7, 179, 190, 192–8, 202–3, 240–1, 255 n. 30 games 179, 195, 200, 259, 283, 346 global 94, 96 patterns of 181, 196 rational 182–3, 193, 227–9, 242 unconditional 183 cooperation problem 175–7, 194, 201, 284, 300, 346 cooperativeness 188, 193 coordination 124–5, 157, 227, 233, 285, 307 coordination problem 166, 172 n. 31, 234, 259, 273–4, 283, 300 corruption 22 Croson, R. 183–4 Cubitt, Robin 302 culture 200, 325 collectivist/individualist 248 Dancy, Jonathan 83 Davidson, Donald 169 n. 1 Davis, Erik 145 Davis, John B. 11–12, 319, 321, 328, 341, 346 Dawes, R. M. 176, 204 n. 5 Debreu, G. 18 decision 150–2, 154–61, 163–9, 228, 262–3, 265, 267, 269–74, 282, 285, 327, 340, 343, 347 future-directed 167–8 personal 263, 267–8, 272, 274

366 index decision problem 155, 157, 297, 301 decision-making 35–6, 40–1, 38, 40, 175, 230–1, 245, 253 n. 23 decision theory 52–4, 117, 280–1, 291; see also rational choice theory definition, hidden force of 344 deliberation 36–43, 46–7, 55, 60, 73, 76, 80, 87, 92, 94–5, 98–9, 150, 154–6, 158–9, 162, 165–6, 211, 273, 351 collective 93 integrated 36, 43–5, 47 selfish 36, 40–3 deliberative economy 155, 163 demand curve 111–12 democracy 114, 117 Den Hartogh, Govert 161, 165–9 desire 35, 37–42, 44, 53, 60, 77–8, 94, 107–8, 110, 113, 211–13, 215–16, 222, 231, 262 first-order/second order 78, 213 fulfilment 77 object of 118–19, 122 pathological 41 primitive 44–45 rational 223 n. 2 shared 341 desire-endorsement 325 dialogue 340 Diamond, Peter 172 n. 27 dilemma 347 collective action 227, 233, 236, 240, 244–5 247; see also collective action problem moral 318 second-order collective action 247 dilemma situation 227, 245 social 113, 255 n. 30 disapproval, social 246 dissidence 221 distribution principle 237 diversity, social 330–1 Dore, R. 26 n. 11 Downs, Anthony 114 Dreier, James 86, 100 n. 17 Dufwenberg, M. 203 duty 55, 75, 273, 325 economic practice 62, 202 economics 5, 20, 35, 22, 52–5, 60, 75, 107–8, 111, 133, 141–2, 145, 181, 195, 202, 217, 282, 291, 316, 332, 344

behavioural foundations of 313, 318, 320, 328 experimental 5, 175, 177, 180 standard 53, 75, 195, 314, 318, 332–3, 342, 356 n. 3 economists 5–7, 23, 49–52, 54–6, 59–60, 62, 66, 18, 112, 119, 133–4, 142, 147, 258, 260, 262, 266, 277 n. 16, 344, 352–3, 357 n. 12 Edgeworth, Francis Y. 23, 344 education 248 egalitarianism 121–2 egoism 28, 74, 96, 110 elections, see voting Elster, John 109, 132, 151–2, 156 emotional reaction, patterns of 212 emotions 181, 195, 197–9 collective 275 n. 2 negative 197–9 endorsement, reflective 80, 92, 97, 100 n. 11 ends 74, 76–83, 86–7, 90, 98, 109, 155 , 162–3, 345; see also goals environment 21, 23–5, 94, 204 n. 4 social 37, 179, 199, 228, 234, 243, 249, 281–2, 331 equality 120 equilibrium 66, 130, 134, 180, 189, 200–2, 235, 283 equilibrium theory, general 18 ethics 22, 25, 73–4 evaluation 53–4, 60, 80, 88, 127, 132–3, 211, 321, 323, 345 evidence empirical 21, 50, 116, 142, 176, 180, 202, 249 n. 5, 271, 302, 304 evolution 178–9, 199, 200–2 expectations 166, 267, 284 normative 241, 273 experience 308, 329 experiments 64, 113, 142, 175–6, 178–80, 182, 185–8, 191–4, 197–9, 201–2, 203 n. 2, 227, 236, 254 n. 30, 282, 284, 298, 346 explicitness condition 125–6, 136, 139 externalism 223 n. 3 fairness 343 Falk, A. 203 famine 201 feelings 246, 252 n. 15 of solidarity 31, 229

index 367 we- 204 n. 5, 229 Fehr, Ernst 175, 179, 182, 184–5, 188, 194–7, 199, 202, 291–3, 357 n. 9 Ferejohn, John 105–6, 112 Fessler, D. 197 first person 252 n. 15 first-person plural speech 319–20, 322 Fischbacher, Urs 175, 183, 185, 187, 190–1, 199, 203, 357 n. 9 fitness 200 Fraassen, Bas van 171 n. 17 focal point 166 Folbre, Nancy 321–2 folk psychology 35–7, 40, 44–5, 53 Folk Theorem 319 Foot, Philippa 87, 101 n. 39 forgroupness 238, 245 forward induction 134–5 frames 296, 298–300 we- 296, 298, 300 framing 292, 294, 295–6, 299–301, 310 n. 11 psychological 294, 308 Frank, Robert 181, 197, 262 free market 109 free will 88 freedom 23–4, 31, 109–10, 152, 330–1 free-riding 178–9, 181–93, 195, 197–200, 235, 240–1, 246, 281, 290, 301 Frey, Bruno 25 n. 3 Friedman, D. 203 n. 2 Frisch, Ragnar 30 Fudenberg, D. 178 functionings 313 future generations 23–4 Gächter, Simon 9, 175, 179, 182, 183, 185, 187, 190–2, 194–7, 199, 346 game form 309 n. 8 knowledge of 64 game modelling 124, 126 cooperative 125 non-cooperative 125–6, 131, 133 game theory 33, 54, 62–5, 67, 134, 139, 144, 179, 202, 284–6, 291, 294 conventional 135, 233, 281–2, 285–6, 292–3, 301, 308 cooperative 125 experimental 21, 319, 346 eductive 125–6, 141, 143 explanatory 281–2

non-cooperative 124–5, 142 normative 281–2 orthodox interpretation of 63 psychological 310 n. 10 games 52, 66, 128–30, 134, 137, 176–7, 180, 182, 185–6, 191, 194, 202–3, 204 n. 7, 235, 281, 286, 289, 292, 296, 305, 309 n. 8, 319; see also Prisoner’s Dilemma Assurance 180, 234, 244, 319 battle of the sexes 134–5 centipede 62 Hawk-Dove 305–7 Hi-Lo 283–5, 289, 293–4, 296, 301, 307 public goods (PG) 176–177, 178–80, 182, 184–8, 190–5 ultimatum 129–30, 198 normal form 63, 136, 307 one-shot 178–9, 192–3, 195, 198–9, 202, 227 repeated 178–9, 190, 193, 199, 227, 319 sequential 184, 186 sub-game 128, 131, 135 Gaus, G. F. 223 n. 1 Gauthier, David 278 n. 26 Geanakoplos, John 310 n. 10 gender relations 96 gene-culture co-evolution 200 generosity 23, 30 genetic disposition 248 Gilbert, Margaret 11, 224 n. 11, 251 n. 8, 254 n. 25, 263, 275 n. 1, 275 n. 2, 275 n. 11, 276 n. 12, 276 n. 13, 277 n. 15–22, 278 n. 24, 280, 299–300, 304, 322, 333 n. 7, 341, 346 Gintis, Herbert 199, 200, 255 n. 33, 254 n. 25 goal 18–20, 31–5, 37, 39, 40, 44, 47, 58–9, 65, 73–4, 76, 214–21, 228–33, 234–6, 238–9, 243, 271, 296, 314–15, 317–18, 321, 331, 342, 345, 348, 350–2, 355 agent neutral 294–5 as conditions of satisfaction of intentions 215–16 as if 33–4, 217 collective 221, 233–4, 237–9, 241, 306–7 common 31, 34 contributive 220–1

368 index goal (cont.) internalized 36, 45–7 joint 306, 348 non-welfare 316, 321 of a group 33, 34 of others 18–20, 28, 31, 32, 33, 45, 315, 347, 349–52, 355 own 18–20, 28, 30, 32–3, 45, 58–9, 74, 264, 271, 314–15, 317, 326, 345, 347–52, 354–5 participative 220–1 perceived 320 mathematical equivalence of 354 private 220–1, 233, 236 self-centred 73–5 shared 33, 218–22, 251 n. 13, 348, 352 surrogate 354 transformation of 31 we-derivative 220–1 goal-maximization schema 36 goal-seeking states 37, 38, 39 Gode, Dhamanjay 142 Gold, Natalie 11, 280, 309 n. 6, 347 good life 331 good, concept of 81 government 114, 146 n. 4 group 177, 187, 189, 192–5, 199–201, 218, 227–9, 234–42, 244–9, 290, 299, 303, 217, 322–4, 326 attitudes 228, 237 context 227–8, 230, 239, 240 decision 253 n. 23 dispersal 199 ethos 236, 238, 239–40, 243–4 formation 294 goal 229, 234, 236, 239, 245, 304, 346 homogeneous 246 identification 94, 229, 245, 288–9, 294, 296–8, 300–1, 308 intention 249 n. 3 member 93, 227–9, 231, 233–40, 243–9, 289, 292, 294, 298, 300, 303–4, 308, 319 operative 246, 252–3 n. 23 membership 217, 326 mobility across 323–4 norms 236 perspective 229, 234, 244, 245, 249, 288 preferences 230, 233, 236–7, 239, 243, 252 n. 17, 252–4 n. 23 reasons 228–9, 239, 244, 249

selection 200–1 social 228, 231, 233, 236, 258, 321–3, 329 thinking 248 utility 233, 237, 243 we- 228 welfare 289, 292 Guala, F. 187, 203 n. 2 guilt 241–2, 246 Güth, Werner 8–9, 126, 346 Hahn, Susanne 145 Hakli, Raul 250 n. 8, 255 n. 36 Haley, J. K. 197 Hamilton, W. 178, 199 Hamlin, Alan 118 Hammerstein, P. 175 happiness 84–5, 197, 327–8 Harsanyi, John C. 146 n. 13, 326 Hart, H. L. A. 146 n. 4, 275 n. 2 Hausman, Daniel M. 6–7, 25 n. 2, 60, 68 n. 10, 145, 147 n. 15, 316, 342, 347, 349, 352–3 hedonism, psychological 68 n. 8, 215 Hegel, G. W. F. 93–4, 96 Heidegger, Martin 223 n. 5 Heiner, Ron 145 Henrich, Joe 199 Herrmann, B. 185–7 Hicks, John 127 Hirshleifer, J. 181, 197 Hobbes, Thomas 3–4, 12 n. 1, 147 n. 22 Hodgson, David 280, 294, 309 n. 4 Hollis, Martin 222–3, 260, 273–4, 280, 299 Homo oeconomicus 80, 142, 146 n. 12, 147 n. 22, 211 Houser, D. 187 human flourishing 117 humanity 24, 87, 89–90, 93–4, 96 Hume, David 77–8 80–3, 85–6, 212–13, 222 Humean rationality 77–9, 86–7, 97, 212 Humean theory of motivation 85 Humeanism 100 n. 17 Hurley, Susan 280, 294–5 identification 218 collective 96 social 75, 95, 314, 318–20, 321, 323–6, 328–9, 347

index 369 with others 313, 317–19, 323 identity 33, 73, 75, 89–98, 214, 217–18, 309 n. 7, 313–14, 317, 319–20, 326–8, 332 collective 322 cosmopolitan 96 moral 91, 96, 102 n. 42 personal 132–3, 313–14, 320–1, 323–5, 328–31 practical 89–92, 95–6 sense of 217, 319, 326 shared 317 social 92, 93, 95, 97–8, 204 n. 5, 299, 309 n. 7, 313–14, 317, 319–20, 325, 332 Ikegami, E. 26 n. 11 imitation 22 I-mode 228–31, 233, 229, 236–9, 243–9, 252 n. 28 cooperation (CIM) 230, 231–2, 233, 238, 242 decision-making 228, 231, 245 goals 230, 238 groups 237 perspective 228 preferences 244, 246 rational cooperation 231–2 reasons (IMR) 228–9, 233, 235, 250 n. 6 impartiality 326 imperative categorical (CI) 74, 84–6, 88–92, 95 semantic 212–3 hypothetical 84, 87, 89–90 moral 85–6 impulse 78, 83, 88, 98 incentive 175, 180, 188, 227, 241, 262 inclination 211–12, 271–5 income, private 176, 183 incomparability 168 incompleteness 29, 49 individualism 264 inequality 202, 293 inequality aversion 183, 202, 280 information 63, 132, 137–8, 140, 146 n. 13, 147 n. 18, 154, 155, 167, 232, 284, 340, 344 ingroup 234, 255 n. 30 cooperation 246 institutions 109–11, 228, 248, 255 n. 35 instrumental principle (IP) 82–4

integrity 314, 319, 327–9 intention formation 230–1 intentional behaviour 219 intentionality 133, 215–16, 231 collective 220, 281, 299–300, 305–6, 308, 314, 319–22, 325, 329 shared 319 intentions 128, 133–4, 136, 138–9, 161, 169, 215–16, 220, 230–1, 235, 249 n. 3, 260–1, 263, 269, 305–8, 310 n. 10, 322–4 collective 211, 261, 264, 268–70, 275, 304–7, 323–4, 341 expression of collective 323–4 individual 305 joint 233–6, 238; see also body, intend as a participative 220–1 personal 264, 267–8, 274 planning conception of 307 rational 300 shared 221, 246, 322–3 we- 235, 329, 305–7 we- vs. I- 306–7 we-derivative 220–1 interaction 251, 266, 322, 324, 328 social 142–3, 319, 331 strategic 52, 124, 145 interdependence, strong 296 interests common 24, 33, 118, 201, 236, 273, 283, 289, 296, 299–300, 322 individual 51, 56, 74, 77, 90, 110–11, 114, 201, 327; see also self-interest of others 33, 90, 121, 344, 350, 354 I-perspective 230 irrationality 78, 84, 106–7 Jevons, W. S. 23 Johnson, D. 179 jointness 240, 242, 246–7, 268 Jolls, Christine 20, 356 n. 7 judgement 158, 161, 167 future 158 moral 88, 100 n. 11 of value 156–7 justice 96, 115, 117, 320 social 18, 316 justification 85–6, 92, 98, 152–3 normative 80 transcendental 82, 87

370 index Kagel, J. 203 n. 2 Kahneman, Daniel 26 n. 10 Kant, Immanuel 18, 73–4, 76–7, 81–2, 84–99, 100 n. 17, 213, 222, 273, 345 Kantian Kingdom of Ends 94, 102 n. 49 Kantian rationality 97, 345 Kantianism 74, 88, 98, 101 n. 26 Kaufman, Charlie 217 Keser C. 179 Kirchsteiger, G. 203 Kirman, Alan 357 n. 11 Kliemt, Hartmut 8–9, 125, 346 Knights, S. 179 Kohlberg, Elon 134 Korsgaard, Christine 74, 81–4, 88–91, 93, 95, 97, 100 n. 11, 100 n. 17, 272 Kosfeld, M. 184, 204 n. 10 Kragt, Alphons J. C. van de 204 n. 5 Kranton, 357 n. 10 Kreps, D. 194 Kurzban, R. 187 Kutz, Christopher 278 n. 24 Lagrangean multipliers 354 Lahno, Bernd 145 language 55, 212, 322–3 nature of 319 Law and Economics 356 n. 3, 357 n. 12 laws 88–9 moral 85, 88, 91, 100 n. 17 of anti-discrimination 333 n. 4 learning 190, 194 Ledyard, J. 176, 177 Levesque, Henry J. 219 Levi, Isaac 170 n. 8 Levine, David 310 n. 10 Levy, David 145 Lewis, David 171 n. 17, 172 n. 29, 259, 277 n. 16, 302 like-mindedness 191–4, 298 List, Christian 280 Little, Ian 52, 127 Lomasky, Loren 114, 118 lotteries 127 loyalty 75 majority rule 125 majority voting 253 n. 23, 270; see also voting maladaptation hypothesis 178–9, 195, 199 Malkovich, John 217

market 109, 115, 116, 117, 142, 344 Marshall, Alfred 23 Marx, Karl 22, 333 n. 2 Maskin, E. 178 maximand 19–20, 29, 32, 74–6, 353–4 maximization 3, 40, 74–6, 97, 99, 144, 352–3; see also utility, maximization of expected payoff 284, 289, 295, 298, 301 of inclination satisfaction 275 of self-interest 17, 30, 75, 80 of wealth 108 of welfare 18, 30, 107, 314–15, 325 of well-being 340 maximizing behaviour, see behaviour, maximizing maxims 90–2 McClennen, Edward 140, 170 n. 7 McGeer, V. 38 McNaughton, D. 212 McPherson, M. S. 25 n. 2 means 80, 82, 84, 94, 101 n. 26, 155, 162–3, 230–1 adequate 79, 84 effective 77–8, 83–4 means-end coherence 155 principle 80 reasoning, see reasoning, means-end relation 77 mental states 228, 304, 307 menu-dependence 47 n. 1, 353 Mertens, Jean-Fran¸cois 134 meta-preferences 140 meta-ranking 214, 291, 318–19 methodology 109, 180 Mill, John Stuart 23 Miller, J. 176, 179 Miller, Kaarlo 254 n. 29, 280, 305–6 Miller, Seumas 278 n. 24 mind 143 Moffatt, G. 186–8 money pump 127 morality 1, 22, 31, 74, 77–9, 81, 84–93, 95, 97–8, 100 n. 17, 114, 117, 271–3, 274, 291, 294, 300, 325, 343 demands of 3, 90, 92, 262–3, 294 Morgenstern, Oskar 134 Morishima, M. 26 n. 11 motivation 54, 55, 57–8, 86, 108, 113–14, 133, 145, 156, 180, 181–4, 188–9,

index 371 202, 211–13, 222–3, 246, 273, 309 n. 7, 344, 346, 356 n. 2 complexity of human 316 cooperative 188 ethical 23 heterogeneous 184, 188–9 moral 291, 325 private 247–8 psychological 291 types of 4 work 21, 75 motives 133–4, 141, 327 altruistic 58 self-interested 55 Muller, L. 187 mutual assistance 304 Nash equilibrium 176, 202, 204 n. 7, 283–4, 296, 305, 310 n. 13 Nash, John 254 n. 27 nationalism 117 necessity, practical 84 needs 23–4 negotiations 237 Neumann, John von 134 neuroscience 180, 198 New York Times 349, 352 norm compliance 119 normative force 267–8, 274 normative neutrality 77 normativism 141 normativity 83, 85–7, 89, 221, 343 norms 73, 91 behavioural 22 moral 74 social 194, 241, 273, 349 Nowak, M. 199 Nussbaum, Martha C. 314, 330 Nyberg, Sten 278 n. 26 Oberholzer-Gee, F. 177 objective 287, 317–18, 320–1, 323, 331, 340, 346, 349 objectivity, epistemic 255 n. 34 obligation 75, 241, 299, 300, 318, 325, 331 moral 229, 273–4 Ockenfels, Axel 202, 291, 293 opportunism 141–2, 144, 146 n. 5 optimality, social 175 optimization 29 options 113, 125, 127, 137, 152, 155

Orbell, John M. 204 n. 5 order, legal 146 n. 4 outcomes 52, 63, 143, 179, 259, 275, 282, 286–7, 290–1, 293, 296, 299, 327 comprehensive 64 culmination 64 social 115–16 subjective ranking of 293 outgroup 254 n. 30, 255 n. 30 Pareto efficiency 129, 138, 233, 235, 237, 255 n. 30 participation 265, 271 passions 77 Pauer-Studer, Herlinde 8, 345 payoff 63, 128, 135, 136, 138, 140, 176–7, 179–82, 186, 195, 227, 242, 255 n. 30, 281, 283–4, 286, 290–2, 296, 298 allocation 202 collective 255 n. 30 function 180, 186, 287–9, 310 n. 12 material 143, 202, 290–1, 293 modification 128–9, 132 objective vs. subjective 135, 142–3 transformation 291, 293 Pearce, David 310 n. 10 Peikoff, A. 223 n. 2 person 88–90, 95, 98, 142, 144, 223 n. 3, 234, 249 n. 3, 280, 317, 320 concepts of 314, 317 prosocial 229 personal unity 319 perspective, see group, perspective; I-mode, perspective; I-perspective; team perspective; we-perspective Peter, Fabienne 99, 100 n. 10, 145, 339, 347–9, 352 Pettit, Philip 6–7, 33, 34, 35, 36, 37, 38, 41–2, 47 n. 1, 47 n. 5, 220, 224 n. 12, 315, 342, 347, 349–51, 353 philosophers 6, 49, 60, 85, 99, 140, 175, 282, 353 philosophy 5, 141, 332 continental 216 Kantian 73, 98 moral 269, 328 of mind 8 practical 80 picoeconomics 126 planning 137–8, 154–7, 169, 235, 307 primacy of 157

372 index plans 136–7, 147 n. 19, 154–7, 159–60, 163–4, 169, 271, 307–8, 331 higher-order 155 inertia of 163, 167, 169 joint 248 sub-plans 307–8 players 125–6, 130–4, 136, 139, 142, 144, 176–8, 180, 184, 198, 255 n. 30, 282–4, 286–7, 290, 293, 296–8, 303, 305, 319 computerized 198 zero intelligence 142 plural subject 261, 264, 270, 276 n. 14, 299–300 policy-making 330–1 politeness 78, 95, 118–20 politics 115, 319 Poundstone, W. 176 power 24–5 practices 228 prediction 20–1, 109, 111–12, 148, 151, 171 n. 12, 178–9, 281, 353 preference 17–18, 49–55, 56–7, 59–61, 63, 65–6, 73, 77–8, 96, 107–8, 110–11, 113–14, 119–22, 124–5, 127, 132–3, 136, 139–41, 143, 151, 161, 177, 179–80, 184–5, 188, 201–2, 214, 228, 230, 233, 235–9, 242, 244, 251 n. 9, 251 n. 10, 259, 274, 280, 285, 309 n. 7, 316, 325, 342, 353 all-things-considered 54, 57, 59–62, 64–7, 127, 133, 140–3, 251 n. 9, 316, 342, 353; see also rankings, all-things-considered altruistic 235 as hypothetical choice 52 complexity of 49 concepts of 49–50, 53–5, 59–62, 66–7, 156–7, 316, 344, 352–3 content of 105, 110–12, 115 context-dependence 49 determination of 17, 62 electoral 115, 117–18 elicitation experiments 188 ethical 325 expression of 115–16 framework 316 future 158 given 17, 143, 145, 251 n. 9 heterogeneity 181, 186, 189–90 incompleteness 49, 353

joint 234 market 115, 117–18 multiple 318 non-selfish 175, 181, 201 observed 181 ordering 33 ordinary language notion of 55, 62 patterns 291 ranking 3, 50–1, 318 reciprocal 203 revealed 68 n. 10, 113, 115, 127, 180, 184–5, 223 n. 7 reversal 111, 119 self-interest 114 selfish 175, 235 shifting 159–60 stable set of 344 structure of 111–12, 117 subjective 325 theory of revealed 49–52, 188 true 114, 178 preference-based explanation of choice 342 preference-satisfaction 43, 110, 120 premises 281, 285–9, 303–4, 310 n. 11 principal-agent analysis 331 principles 55, 57–8, 73, 76, 93–6 moral 53, 60, 82, 87–8, 90, 94–5, 97–8, 291 normative 82–5, 92 subjective 90–1 Prisoner’s Dilemma 62–6, 93–4, 138, 144, 176, 177–80, 184–5, 188, 223 n. 7, 235, 237, 240, 243, 254–5 n. 30, 275, 281–5, 287, 289–94, 286–8, 301, 303, 307, 310 n. 13, 346 cooperation 5, 64–6, 147 n. 18, 281–4, 287, 290, 292–4, 296, 298, 303 defection 176, 178–80, 200, 227, 240–2, 245–7, 281–4, 287, 290, 296, 298 experiments 64, 176, 179, 180, 182, 282, 298; see also experiments Intergroup (IPD) 254–5 n. 30 one-shot 235, 241–2, 244–5, 235, 319 sequential 136–7, 184, 204 n. 8 privateness 313–15, 316–18 pro-attitude 221 probabilitiy 125, 255 n. 32 subjective 139 probability distribution 147 n. 19 probability function 142 stable 46

index 373 promises, credible 128 promising, public acts of 299 prudence 23, 55, 78, 82, 84–5 psychologists 53, 60, 66, 204 n. 9, 254 n. 30, 351 psychology 145, 199, 228, 248, 350; see also folk psychology cognitive 133, 141, 142–3, 144, deliberative 157 social 341, 351 Gestalt 297 public choice 115, 117 orthodoxy 117 public goods 113, 177, 181, 183, 189, 190, 195, 199, 240, 254 n. 23, 354 n. 27, 282 game, see games, public goods provision of 204 n. 4, 204 n. 11 public policy 343 public spirit 23 punishment 176, 179, 194–8, 200–2, 247 altruistic 199–21 Putterman, L. 25 n. 2 quasi-orderings, incomplete 318 Quervain, J.-F. de 198 Quinn, Warren 80 Rabin, Matthew 203, 292–3, 356 n. 7, 356 n. 9 Rabinowicz, Wlodek 278 n. 26 race 318 ranking 50–2, 54–6, 60–2, 127–8, 141, 214, 237, 251 single 56, 59 all-things-considered 53–7, 59–62, 63–4, 65, 353 of outcomes 64 Rapoport, A. 176 rational behaviour, standard model of 211, 214 rational beings 84, 86, 88, 95, 143 rational choice 17–9, 20–1, 47, 59, 66, 76, 94, 99, 108, 144, 154, 178, 340–6, 348, 350, 354 demands of 17 forward-looking 135, 139–41 nature of 21 principle of 97–8 rational choice analysis 179

rational choice modelling (RCM) 126–7, 130–32, 134, 138, 140, 143, 145, 343, 346 agent form 132–3, 136, 139, 141–2, 145 rational choice paradigm 76, 94, 99 Rational Choice Politics 356 n. 3, 357 n. 12 Rational Choice Theory (RCT) 3–5, 17, 19, 20, 28–9, 32, 34–6, 40, 43–5, 47, 59, 62, 66, 74, 76, 99, 102 n. 40, 109, 126–7, 134, 140, 142–3, 145, 151, 153, 202, 258–60, 273–4 economic and social scientific application 45–6 critique of 4, 20, 66, 79, 347 explanatory power of 5 minimal 29, 32, 34–5, 37 substantive assumptions of 126–7 vs. discipline 340, 342–6, 350, 353–5 rational fool 56, 59, 60, 62, 126, 127, 142–3, 145 rationality 5, 17, 20–1, 25, 31, 59, 78, 80, 84, 86–7, 93–98, 105, 107, 109, 113, 121, 122, 127, 142–3, 151, 156, 163, 176, 203, 227, 229–30, 233, 235, 244, 271, 273, 281–2, 284, 286, 290–1, 294, 300, 302, 339–40, 346, 348, 350 approach 105–6, 108, 112–13, 115, 117, 118, 121–22, 345 assumptions of 21, 105, 109–11, 118, 121, 195, 284 bounded 143, 145, 145 n. 3, 167, 203, 230, 284 collective 176, 227, 233 conception of 5, 20, 76, 79, 81–2, 93, 99, 106, 110–11, 126 conditions of 76–80, 82, 97, 106, 110–11 demands of 22, 348, 350 economic 73–4, 94, 211 formulation of 17, 316 goal- 230 grammar of 112–13, 121–22 Humean 77–9, 86–7, 97, 212 in action 212–13, 217 individual 144, 227, 233, 244, 286, 314 instrumental 76–7, 79–83, 97–9, 230, 345 instrumental conception of 292 limitations 154–5, 159, 169; see also bounded rationality

374 index rationality (cont.) means-end 77, 84, 98, 100 n. 18; see also reasoning, means-end and Humean rationality model, limitations of 344 nature of 21 of choice 340–1, 343–5 of following rules of politeness of plans 157 of voting 114 opportunistic 124, 129–30, 136, 142–3, 145 n. 3 practical 74, 76–8 80–2, 86, 93–5, 97–8, 102 n. 40, 223 n. 1 Kantian 76–7, 81–2 requirements 21, 35, 273, 294–5 self-interested 77, 80, 99 theory of 4, 22, 76 Rawling, P. 212 Rawls, John 102 n. 49, 115, 343 Raz, Joseph 171 n. 20 real life situation 142, 246, 281, 291, 294 real world 21, 132, 142, 294, 343 reason 17, 20, 31, 37, 76–7, 80, 84, 87, 90, 100 n. 11, 143, 263, 271, 275 conception of practical 4 demands of 35 domain of 17, 316 practical 3, 5, 25, 82, 84–6, 88–9, 212, 308 public 345 reasonableness 343 reasoning 19, 38, 40, 76, 80, 86, 92, 98, 126, 127, 178, 272, 280, 285, 287–8, 294, 303, 305, 313, 317, 320–1, 326, 328, 339–41, 343, 345–6 abilities 320 as a group 289; see also team reasoning best-reply 286, 298, 301, 308 capabilities of 154 cooperative 307 descisional 341 dialectical 93 environmental 25 identity-based 341 impaired 20 individual 280–1, 286, 307 instrumental 79, 84–5, 286–7, 300 means-end 74, 77–80 mistakes 154, 157–8, 163, 167–9 mode of 280, 285, 288, 304

practical 20, 154, 285–6, 288, 304, 307 schema of practical 280, 286–9, 295, 297, 303 practical 76, 81, 85–6, 155 principle of 81–2 sustainable 340 theory of practical 79 utilitarian consequentialist 328–9 valid 284–5, 304 we-mode 247 reasons 36, 73, 75, 82–3, 88, 92, 94–5, 127, 133, 140–1, 143, 157–8, 160, 161, 162, 163–4, 165–9, 212–13, 221–3, 228–9, 232–3, 239, 262–4, 271, 273, 275, 339–40, 342–3 categorical normative 75 content-independent 172 n. 31 decisive 339 desire-independent 212–3 external 94 for participation 233 goal-based 342 instrumental 83 moral 81 normative 81, 87 of fairness 165 pro-group 229 social 246 sufficient 263, 267–8 to believe 302, 303, reciprocation 63 reciprocity 65, 183, 199, 202, 204 n. 11, 251 n. 9, 277, 302 kin 277 strong 199, 201–2 taste for 202 reconsideration 156, 159–06 reflection 340, 343 reflexivity 314, 321, 332 Regan, Donald 280, 294–5 regret 158 re-identification 329–30 relations, inter-individual 125 relationship 268, 320–1 remorse, collective 275 n. 2 representative, role of 33–4 reputation 63 requirements 162–3, 165 respect 90 responsibility 25 mother toward child 25

index 375 toward other species 24 restrictions, see constraints revenge 195, 198 reward 240 centre 198 Richerson, P. J. 200, 249 n. 1, 255 n. 33 Riedl, A. 204 n. 10 right 268 risk 127 Robins, Michael H. 171 n. 18, 278 n. 26 roles, social 89, 322 Rosenberg, J. 222, 224 n. 10 Roth, A. E. 203 n. 2 Rousseau, Jean-Jacques 299 Rovane, Carol 224 n. 9 rule of inference 285, 302 rules 76, 265–7, 291 behavioural 179 heuristic 179 of behaviour 326, 348 of conduct 18, 33, 58, 326 of the game 124–5, 126, 133, 138, 140, 143, 289–90 recognition of 75, 326 social 65, 75, 258, 328 Ryle, Gilbert 219 Sally, David 282 Samuelson, Paul 52 sanctions 245–6; see also punishment social 241–2, 247 Sandel, Michael 320 Sanfey, A. G. 198 satisfaction 251 n. 13, 316 mental 60–1 Schelling, Thomas 172 n. 29, 259 Schmid, Hans Bernhard 10, 145, 220, 341, 346–9, 352 Schmidt, Klaus 202, 291–3, 357 n. 9 Schmitz, David 100 n. 18 Schumpeter, Joseph 114 sciences behavioural 175, 181, 202 scrutiny 35, 39, 40, 76, 98–9, 321, 340, 343, 347, 357 n. 13; see also self-scrutiny Searle, John R. 94, 215, 220, 255 n. 34, 277 n. 21, 280, 306–7 Sefton, M. 204 n. 8 selection kin 199 pressure 31, 200

self 17, 29, 73, 98, 313–15, 318, 321, 325–6, 333 aspects of 313–5, 317–18, 320–1, 332, 341 self-centred welfare 18–19, 30, 58, 98, 314–16, 321, 354 self-goal choice 18–19, 21, 30, 32, 33–4, 45, 46, 58, 98, 214–223, 314–16, 318, 321, 326, 328, 340, 348, 351, 353–5 self-individuation 324, 329, 332 self-interest 4, 19, 20, 28–33, 51, 68 n. 8, 74, 109–10, 117, 122, 291, 314, 318, 332–3, 340, 344, 350 assumption of 105 selfishness 191, 202–3 assumption of 175, 181, 195, 202 self-love 344 self-management 151, 170 n. 6 self-referencing 317 self-regard 314–17, 329 self-scrutiny 31, 75, 98, 271, 313, 317, 320–2, 323–6, 328–9, 340; see also scrutiny self-welfare goal 18–19, 21, 58, 98, 214, 314–16, 318, 321, 350, 354 Sellars, Wilfrid 220 Selten, Reinhard 124 Sen, Amartya 4–12, 17, 19, 25 n. 8, 28–36, 39–40, 43–7, 49–62, 64–7, 73–6, 78–9, 84, 88, 92–4, 96, 97–99, 105–6, 108, 113–14, 118, 121, 126, 156, 175, 211–18, 220, 223, 254 n. 27, 258, 291, 313–21, 323–6, 328–33, 340–3, 345, 347–8, 350, 352–4 ‘Rational Fools’ 4–6, 19, 50, 56–8, 105–6, 113–14, 118, 121, 126, 127, 175, 214–15, 217, 258, 291, 315, 325–6, 343 sensations, bodily 252 n. 15 sentiments 31 separability principle 140 sex 318 shame 241–2, 246 Shapiro, Scott 153–5, 168 shared understanding 322 Sigmund, K. 199 signalling of plans 134, 139; see also plans signalling theory 199 Simmons, A. John, 277 n. 17 singularism 224 n. 11, 264 skills, rules of 84–5

376 index Slovic, P. 26 n. 10 Smit, Houston 81 Smith, Adam 18, 22, 30, 344 Smith, Michael 37, 41–2, 101 n. 34 Smith, Vernon 142 social capital 240, 246–7 social contract 299 social sciences 20, 105, 141, 202 social scientists 275, 282 social theory 145, 249, 258 society 227, 330 solidarity 75 Solow, Robert 23–24 speech, performative 322 Spinoza, Baruch 147 n. 22 Stachetti, Ennio 310 n. 10 standard of living 23–5 state of nature 305 Stopka, P. 179 Stoutland, Frederick 216 strategy 52, 134, 136, 178, 281, 284, 307, 309 n. 1 choice 63–7, 137–8, 307 dominant 63, 66, 144, 176–7, 255 n. 30, 298, 310 n. 13 dominated 135 joint 93 profiles 291, 296 tit-for-tat 178 Sugden, Robert 11, 135, 172 n. 29, 179, 203, 204 n. 11, 251 n. 8, 260, 273–4, 280, 302, 304, 309 n. 4, 347 Sunder, Shyam 142, 203 n. 2 Sunstein, Cass 20, 356 n. 7 sustainable development 23–4 switch problem 243–4 sympathy 4, 5, 18, 19, 20, 22, 24, 28, 30–1, 56–8, 75, 106, 118–121, 122 n. 1, 214, 314–16, 318–19, 321, 325, 345 Tabarrock, Alex 145, 147 n. 23 taste 61, 344 Tawney, R. H. 22 tax 111 team 219, 280, 285, 294, 300 members 179, 280, 285, 289, 295, 297, 302 objective of 285 agency 281, 285, 294, 299, 304, 347 formation 294, 299, 310 n. 11 perspective 179

team reasoning 179–80, 182, 191, 203, 280–1, 289, 292–4, 296–7, 301–2, 303–4, 307–8, 347 circumspect 297–8, 301 logic of 302 restricted 295, 301 temptation 151 Teschl, Miriam 357 n. 11 Thaler, Richard 20, 356 n. 7 Thöma, Dieter 12 Thöni, Christian 9, 186–7, 191–2, 194, 346 threats 246, 254 n. 27 credible 129–30 Tirole, Jean 356 n. 7, 357 n. 9 tit-for-tat strategy 178 togetherness 220 toxin puzzle 171 n. 19 tradition, social 22 transcendental argument 89, 93, 95 transfers 120–21 transitivity 107, 111, 117 Trivers, R. L. 178, 199 trust 213, 241, 246, 251 n. 11 trust games 128–9 trustee 33–4 Tuomela, Maj 251 n. 10–11, 252 n. 16, 252 n. 21–2, 254 n. 25 Tuomela, Raimo 10–11, 221, 224 n. 12, 249 n. 2, 249 n. 4–6, 250 n. 7–8, 251 n. 9–12, 251 n. 14, 252 n. 16–22, 254 n. 23–6, 254 n. 28, 255 n. 31, 255 n. 33, 255 n. 35, 277 n. 21, 280, 305–6, 319, 322 Tversky, Amos 26 n. 10 Ulysses 128–9, 131, 150–4, 156–61, 163, 165, 167–9 uncertainty 286 universalization principle 93–4, 96–7 utilitarianism act 294, 326 cooperative 294–5 critique of 326–8 rule 294 utilitarian calculation 327–8 utility 29, 63–5, 68 n. 6, 74, 102 n. 48, 127, 132, 133, 139–42, 147 n. 21, 180, 204 n. 7, 231, 235, 237, 239–45, 255 n. 32, 305, 344 as index of preferences 52, 68 n. 6

index 377 indices 286 maximization 74, 80, 124, 133–4, 145, 233, 235, 344 maximization of expected 93–4, 96, 103 n. 57, 114, 116, 179, 229–32 payoff 64 von-Neumann-Morgenstern 204 n. 7 utility function 19, 56, 73, 110, 124–5, 127–8, 133–5, 141, 142–3, 147 n. 19, 202, 231 Beckerian 19 content of 112–13 modern concept of 127 stable 46 structure of 112 utility theory expected utility theory 62, 286 pioneers of 23 validity 302 Vallentyne, Peter 223 n. 4, 224 n. 12 value of opportunities 24 value order 132 values 25, 31–2, 35, 56, 60, 61, 76, 92, 317, 320, 331, 340–1 collective 275 n. 2 dimensions of 127 instrumental 65 veil of ignorance 115–16 veil of insignificance 116 veil of utility representation 127, 143, 145 Vellemann, David 277 n. 21 Verbeek, Bruno 145, 147 n. 16, 170 n. 10, 171 n. 12, 172 n. 27, 172 n. 29, 276, 346 Violante, Luciano 26 n. 13 Vries, Willem de 99 voting 102 n. 48, 106, 112, 114–16, 118, 122, 125, 204 n. 4, 282 Vugt, M. van 249 n. 5 Waldfogel, J. 177 Wallace, Jay 99

Walsh, Vivian 356 n. 7 Weber, Max 22 Weibull, J. W. 184, weighing of goods 127 Weirich, P. 211 we-language 319, 322–3, 329 welfare 4, 18–20 29–31, 57–8, 73, 75, 214; see also choice and welfare individual 314–18, 320, 325 non-self-centred 20 welfare economics 58 well-being 29, 84, 96, 110–11, 120–22, 202, 315, 345, 348–50, 354–5, 357 n. 13 levels of 127 material 180 we-mode 228–30, 233–4, 236–9, 243–9, 252 n. 18 cooperation (CWM) 230–233, 234–5, 238, 247 preferences 236, 243–4, 46 reasons (WMR) 228–9, 233, 250 n. 6 we-perspective 230, 347 White, M. 177 will 73, 78, 88, 91, 262, 265, 267–8 binding of 263, 267 necessitation of 85 weakness of 151, 153, 158, 170 n. 6 Williams, Bernard 211–2, 222–3, 223 n. 3, 314, 325–9 Winden F. van 179 wish 215, 216 women 96, 322 World Commission on Environment and Development 23 World Health Organization (WHO) 23 xenophobia 117 Zahavi, Amotz 199 Zahavi, Avishag 199 Zamagni, S 25 n. 2

Rationality and Commitment

Commitment

Commitment

Reason and Commitment

Presentism and Ontological Commitment

Paula's Commitment

Mr. Commitment

Paula's Commitment

Mr. Commitment

Commitment Issues

Commitment Hour

Unbound Commitment

Paula's Commitment

Paula's Commitment

Rationality and Intelligence

Uncertainty, Rationality, and Agency

Reason and Rationality

Rationality and Scientific Discovery

Rationality and Freedom

Rationality and Intelligence

Rationality and Theory Choice

Rationality and Logic

Decision Theory and Rationality

Consistency, Choice, and Rationality

Reality and Rationality

Reason and Rationality

Reason and Rationality

Decision Theory and Rationality

Dialogue on Individualism and Commitment

Avoiding Commitment

Mr. Commitment

Rationality and Commitment

Commitment

Commitment

Reason and Commitment

Presentism and Ontological Commitment

Paula's Commitment

Mr. Commitment

Paula's Commitment

Mr. Commitment

Commitment Issues

Commitment Hour

Unbound Commitment

Paula's Commitment

Paula's Commitment

Rationality and Intelligence

Uncertainty, Rationality, and Agency

Reason and Rationality

Rationality and Scientific Discovery

Rationality and Freedom

Rationality and Intelligence

Rationality and Theory Choice

Rationality and Logic

Decision Theory and Rationality

Consistency, Choice, and Rationality

Reality and Rationality

Reason and Rationality

Reason and Rationality

Decision Theory and Rationality

Dialogue on Individualism and Commitment

Avoiding Commitment

Mr. Commitment

Recommend Documents