Essays in the Metaphysics of Mind

Essays in the Metaphysics of Mind This page intentionally left blank Essays in the Metaphysics of Mind Jaegwon Kim ...

Author: Jaegwon Kim

166 downloads 1542 Views 1MB 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

Essays in the Metaphysics of Mind

This page intentionally left blank

Essays in the Metaphysics of Mind Jaegwon Kim

1

3

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 CapeTown Dar es Salaam HongKong 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 # in this volume Jaegwon Kim 2010 The moral rights of the author Jaegwon Kim 2010 have been asserted Database right Oxford University Press (maker) First published 2010 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 this same condition on any acquirer British Library Cataloguing in Publication Data Data available Library of Congress Cataloging in Publication Data Library of Congress Control Number: 2010933159 Typeset by Laserwords Private Limited, Chennai, India Printed in Great Britain on acid-free paper by MPG Books Group, Badmin and King’s Lynn ISBN 978-0-19-958587-8 (hbk.) 978-0-19-958588-5 (pbk.) 1 3 5 7 9 10 8 6 4 2

Contents Sources and Acknowledgments

vii

Introduction 1. Making Sense of Emergence 2. The Layered World: Metaphysical Considerations 3. Emergence: Core Ideas and Issues 4. “Supervenient and Yet Not Deducible”: Is There a Coherent Concept of Ontological Emergence? 5. Reasons and the First Person 6. Taking the Agent’s Point of View Seriously in Action Explanation 7. Explanatory Realism, Causal Realism, and Explanatory Exclusion 8. Explanatory Knowledge and Metaphysical Dependence 9. Hempel, Explanation, Metaphysics 10. Reduction and Reductive Explanation: Is One Possible Without the Other? 11. Can Supervenience and “Non-Strict” Laws Save Anomalous Monism? 12. Causation and Mental Causation 13. Two Concepts of Realization, Mental Causation, and Physicalism 14. Why There Are No Laws in the Special Sciences: Three Arguments

1 8 41 66

Index

85 105 125 148 167 187 207 234 243 263 282 311

This page intentionally left blank

Sources and Acknowledgments “Making Sense of Emergence,” Philosophical Studies 95 (1999): 3–36. # Jaegwon Kim “The Layered World: Metaphysical Considerations,” published under the title “The Layered Model: Metaphysical Considerations,” in Philosophical Explorations 5 (2002): 2–20. Reprinted with permission of Taylor and Francis Group. “Emergence: Core Ideas and Issues,” Synthese 151 (2006): 547–559. # Jaegwon Kim “‘Supervenient and Yet Not Deducible’: Is There a Coherent Concept of Ontological Emergence?” In Reduction: Between the Mind and the Brain, ed. Alexander Hieke and Hannes Leitgeb, Ontos Verlag, 2009. # Jaegwon Kim “Reasons and the First Person.” In Human Action, Deliberation and Causation, ed. Jan Bransen and Stefaan E. Cuypers, Kluwer, 1998, 67–87. # Jaegwon Kim “Taking the Agent’s Point of View Seriously in Action Explanation.” New. # Jaegwon Kim “Explanatory Realism, Causal Realism, and Explanatory Exclusion,” Midwest Studies in Philosophy 12 (1988): 225–239. Used with permission of Wiley-Blackwell. “Explanatory Knowledge and Metaphysical Dependence,” Philosophical Issues 5 (1994): 51–69. Used with permission of Wiley-Blackwell. “Hempel, Explanation, Metaphysics,” Philosophical Studies 94 (1999): 1–20. Used with permission of Springer. “Reduction and Reductive Explanation: Is One Possible Without the Other?” In Being Reduced: New Essays on Causation and Explanation in the Special Sciences, ed. Jakob Hohwy and Jesper Kallestrup, Oxford University Press, 2007. # Jaegwon Kim

viii

SOURCES AND ACKNOWLEDGMENTS

“Can Supervenience and ‘Non-Strict’ Laws Save Anomalous Monism?” From Mental Causation, ed. John Heil and Alfred Mele, Oxford University Press, 1993. # Jaegwon Kim “Causation and Mental Causation.” In Contemporary Debates in Philosophy of Mind, ed. Brian McLaughlin and Jonathan Cohen, Blackwell, 2007. Used with permission of Wiley-Blackwell. “Two Concepts of Realization, Mental Causation, and Physicalism.” New. # Jaegwon Kim. “Why There Are No Laws in the Special Sciences: Three Arguments.” New. # Jaegwon Kim

Introduction All but one of the 14 essays gathered here have been published or written since 1993 when my earlier collection of essays, Supervenience and Mind, appeared. I have not included papers that have been used in my two monographs, Mind in a Physical World (1998) and Physicalism, Or Something Near Enough (2005). Most of the essays have appeared in print; these have received minor corrections and revisions, and some cosmetic touchups. Three are new and published here for the first time (Essays 6, 13, and 14). The essays have been written over many years, and in consequence they do not always speak in a unified voice on some matters. I have not done anything to retrospectively impose consistency and harmony; that would have been difficult and, in any case, pointless. Nor have I tried to eliminate overlapping material between some essays. The essays collectively deal with a variety of issues, from emergence to action explanation to mental causation and special-science laws. Most of them have something to do with the metaphysical issues about the status of minds; a few are not comfortably grouped under the metaphysics of mind, as advertized in the title of the book. I can assure the reader, though, that even these have something to do with minds or metaphysics. The book begins with four papers on emergence and emergentism. The idea of emergence, which goes back to the 19th century, saw its first flourishing in the early 20th century in the works of British emergentists, most notably Samuel Alexander and C.D. Broad. However, emergentism didn’t fare well during the mid-century when analytic philosophy was dominated by logical positivists and hyper-empiricists, and it was often held up for easy refutation, or outright ridicule. But the fundamental idea that animated the movement proved to have a remarkable staying power, finding friends among accomplished scientists as well as philosophers. The

2

INTRODUCTION

idea of emergence seems to have a magical ability for attracting devotees, enthusiasts, and loyalists from every area of intellectual endeavor. Since around the 1980s, emergence has returned with a vengeance, with conferences and workshops seemingly everywhere and numerous books and special journal issues dedicated to the topic. Philosophers, theologians, scientists, and science writers seem to warm up to the idea of emergence and emergentism with enthusiasm and excitement, just as they recoil from reduction and reductionism with disdain and hostility. Most of them seem to look to emergentism as a promising middle ground that avoids the extremes of reductionist physicalism and serious dualism, as some of the early British emergentists did. It seems fair to say that emergentism has yet to gain full philosophical respectability, but it has long shed its cult-like status as a niche ideology. Whatever we think of it, it is a growing movement that cuts across many scientific fields, with many reputable thinkers on its side, and it is something to be reckoned with. I have had an interest in emergentism for some time, alternately intrigued by its promises and disheartened by its difficulties. It is the kind of philosophical view, or approach, that one wishes were correct, or at least coherent and workable. In spite of my predisposition to see the good side of emergentism, my work on emergence has been generally deflationary. “Making Sense of Emergence” is about the best I have been able to do for emergence and emergentism, but its overall message is pretty clearly on the negative side. Classical emergentism, of Broad and others, makes best sense in a layered ontology of the world, according to which reality is a hierarchically organized structure with distinct “levels,” from the micro to the macro, from atoms and molecules to cells and tissues and to organisms and minds. I don’t know when this kind of hierarchical model was first introduced, or by whom, but it is now a familiar picture. In “The Layered World: Metaphysical Considerations,” I explore various issues and problems that arise from this picture taken as a general ontology of the world. “Emergence: Core Ideas and Issues” raises further issues about the concept of emergence and argues that a robust positive characterization of the concept is yet to be formulated, and that “downward” causation, which arguably is emergentism’s very reason for existence, remains an unresolved metaphysical difficulty. In the final paper on emergence, “‘Supervenient and Yet Not Deducible’: Is There a Coherent Concept of Ontological Emergence?” I try to make sense of the idea, found in C.D. Broad and many other writers, that

INTRODUCTION

3

an emergent property is one that is supervenient on, or determined by, its “basal” conditions but not deducible from them. If my reasoning is generally correct, there is a serious doubt as to whether a coherent sense could be attached to the idea of ontological, or metaphysical, emergence (as distinguished from epistemic emergence). I believe this is the most seriously negative message I have delivered to the fans of emergence. Next comes a pair of essays on action and action explanation, a topic I have thought about from time to time for many years. What has motivated me to think about the explanation, or understanding, of actions has been my dissatisfaction with the prevailing third-person theories like Carl Hempel’s nomological account and Donald Davidson’s causal view. The source of my dissatisfaction was simple: these approaches seemed just wrong, or at least irrelevant, when we consider the agent’s understanding of his own actions. Full agency of the sort we are capable of requires that we understand what we are doing, or why we are doing it, and when we lose this understanding, our agency can be seriously compromised, and can even be aborted. “Reasons and the First Person” was my initial attempt to argue for the importance of the first-person perspective in understanding actions. In “Taking the Agent’s Point of View Seriously in Action Explanation,” which is published here for the first time, I approach the issues in a more systematic way, by revisiting an old dispute between Carl Hempel and William Dray on the nature of historical explanations. As I say in the essay, this was really a dispute about explanation of human action and had nothing specifically to do with “historical” explanations. As is well known, Hempel argued that explanations of human actions, like explanations of natural events, come under his covering-law conception of explanation— namely that human actions are to be explained by subsuming them under general laws of human behavior. Dray defended a radically opposed view, arguing that there is a distinctively normative and evaluative aspect to reason-based explanations of actions, and that the understanding of why an agent did something consists in our seeing that the action was the right, or appropriate, thing to do in the circumstances. What I try to do is to construct, and defend, a normative and agent-centered account that does justice to Dray’s important insights. My account remains sketchy in many ways and there are important issues not adequately discussed or even raised; a fullfledged treatment would require a book-length work. I believe explanation of action continues to be a lively philosophical topic because understanding

4

INTRODUCTION

why we do what we do is constitutive of our understanding of ourselves as reflective agents, and our interest in a theory of action explanation reflects our desire to understand how we understand ourselves. This is followed by a group of papers on the general topic of explanation and related issues. “Explanatory Realism, Causal Realism, and Explanatory Exclusion” is an early paper in which I discuss the relationships between taking a realist view of explanation and causation and the proposition that multiple full explanations of a single event “exclude” one another. This was the first paper of mine to appear on the problem of exclusion (I remember that when I first introduced the term “exclusion,” I was thinking about Wolfgang Pauli’s “principle of exclusion” in particle physics, something I had heard from my physicist friends). What made me think about exclusion was a puzzle: If explanations are cognitive achievements of some kind, as they must be, why is it that multiple explanations of a single explanandum, or multiple distinct stories about how something came about, don’t seem like good news? These explanations seem to compete against one another and seem to create their own “epistemic predicament” for anyone considering them. During the past decade or so, discussions of the exclusion problem have turned away from explanatory exclusion and have almost exclusively focused on causal exclusion, and it is the latter that has lately received much intense attention, especially in the mental causation debates. For me, however, all this began with puzzles about explanation, and I wish I had done more work directly on exclusion as it applies to explanations. It is this issue that I find deeply intriguing; I think it has the potential to yield interesting insights into our ideas about explanation and understanding. According to a well-worn platitude, science seeks not only to ascertain the events and facts of the world but also, more importantly, to achieve an understanding of why these facts hold and why the events occur as they do. In “Explanatory Knowledge and Metaphysical Dependence,” I set out some views on why we need a theory of explanation and what such a theory should be expected to deliver. Explanation of something is supposed to give us an “understanding” of the thing and make it “intelligible,” where this is taken to be an epistemic accomplishment of importance. I argue that above all, a theory of explanation should be a theory that tells us what this epistemic accomplishment is—what sort of cognitive gain is achieved when we come to understand something. That is, a theory of explanation must above all be a theory of understanding as a kind of knowledge. This means that

INTRODUCTION

5

theory of explanation should be part of epistemology, with explanatory understanding as its central topic. But, as we know, traditional epistemology is completely silent on such issues. When we look at the active debate on explanation from around the middle of the last century through the ’80s, what we find is a plethora of “models” of explanation, with only scant attention paid to the question just how would-be explanations fitting these schematic models manage to explain. Almost invariably objections to these models took the form of counterexamples, and the purported counterexamples would be tested against unexamined raw intuitions. The correct procedure would have been to evaluate the worth of these models by considering whether and how they help generate explanatory understanding. Our primary need, then, is for a theory, not models, of explanation. I believe that this still is a valid comment on the way discussions about explanation are being carried on in philosophy of science. In “Hempel, Explanation, Metaphysics,” given at an APA symposium in memory of Carl Hempel, my teacher at graduate school, I argue that if Hempel had been willing to use in his work on explanation some of the metaphysical tools that are now considered uncontroversial, he would have been able to sidestep many of the familiar objections and counterexamples against his views. His positivist commitments had severely restricted the range of concepts that he found philosophically acceptable, and he was not able to fend off even some of the simplest objections in a convincing way. This paper also contains a brief defense of a central thesis of Hempel’s theory of explanation that explanations are logical derivations or deductions. The title of “Reduction and Reductive Explanation: Is One Possible Without the Other?” says exactly what the paper is about. It has been quite common for philosophers to distinguish reduction and reductive explanation, with the assumption that even where reduction fails, reductive explanation is often possible and successful. The paper examines this assumption in some detail, with regard to three models of reduction, bridge-law reduction, identity reduction which replaces bridge laws with identities, and functional reduction based on functional analyses of properties to be reduced. I try to show that bridge-law reduction gives us neither genuine reduction nor reductive explanation, that identity reduction gives us reduction but not reductive explanation, and that functional reduction yields reductive explanation and, arguably, reduction as well.

6

INTRODUCTION

During the past couple of decades, I probably have done as much work on mental causation as any other issue, and the next three papers concern mental causation in one way or another. As is well known, Davidson’s “anomalous monism” was taken to task by a number of critics, including myself, on the ground that it leads to epiphenomenalism of mental properties. In his “Thinking Causes,”1 Davidson attempted to rebut this criticism, arguing that he could save mental causation by appealing to mind-body supervenience and psychophysical laws that are not strict and exceptionless. “Can Supervenience and ‘Non-Strict’ Laws Save Anomalous Monism?” argues that Davidson’s replies all fail, and points to other problematic aspects of his views on the mind-body problem. This paper is part of my efforts to show that nonreductive physicalism, of which Davidson’s anomalous monism is a version, cannot account for mental causation. “Causation and Mental Causation,” written for Contemporary Debates in Philosophy of Mind,2 is an overview statement of my current thoughts on mental causation. Sydney Shoemaker is one of the many philosophers who believe that mental causation can be vindicated under nonreductive physicalism. In an important and admirable new book, Physical Realization,3 Shoemaker develops a defense of this position on the basis of what he calls the subset view of realization. On this approach, a property is a realizer of another just in case its causal powers include those of the latter as a subset. It is then argued that a property and its realizers don’t compete for causal status, and therefore that the causal powers of a mental property are not preempted, or excluded, by those of its physical realizers. In “Two Concepts of Realization, Physicalism, and Mental Causation,”4 I compare Shoemaker realization with the standard “second-order” account of realization, and try to show why I find Shoemaker’s treatment of mental causation less than fully satisfying, and why his overall view is best construed as a form of type physicalism, not nonreductive physicalism. I hope that the final essay “Why There Are No Laws in the Special Sciences: Three Arguments,” which is new, delivers what its title promises—at least, 1 In Mental Causation, ed. John Heil and Alfred Mele (Oxford: Oxford University Press, 1993). 2 Ed. Brian P. McLaughlin and Jonathan Cohen (Oxford: Blackwell, 2007). 3 (Oxford: Oxford University Press, 2007). 4 Based on a paper presented at an APA book symposium on Shoemaker’s book in 2009; my symposium paper, “Thoughts on Sydney Shoemaker’s Physical Realization,” is to appear in Philosophical Studies.

INTRODUCTION

7

one argument that works, if not three. The first of the arguments begins with a consideration of Davidson’s argument for his well-known thesis that there are no laws about intentional mental phenomena; however, the argument that I build is not particularly Davidsonian. It is a metaphysical argument, like the other two, based on a certain construal of “causal closure,” not Davidsonian considerations on the holistic and normative character of mentality. The second argument is based on some of J.J.C. Smart’s astute observations concerning biology and its relationship to physics and chemistry. His insightful claim is that, unlike physics, biology does not aim at the discovery of laws, and that engineering, not physics, is the correct model for understanding the scientific status of biology. I construct an argument that takes off from Smart’s remarks about biology, and generalize it to other special sciences. As a bonus, this yields a simple metaphysical argument for Davidson’s anomalism of the mental. The last of the three arguments is based on my earlier work on multiply realizable properties and their projectibility. I am inclined to believe that the second, Smart-derived argument, is the most fundamental one, and that the other two arguments could be shown to be secondary and derivative. Over the years, Brown University has been generous with its support of my work; I am very grateful. I want to thank the participants in my seminars the past two decades, at Brown, Notre Dame, and the Graduate Center of the City University of New York, who made me think, and rethink, with searching questions and unexpected challenges. With his eyes as sharp as his mind, Chiwook Won, my graduate assistant at Brown, has been extremely helpful with many chores involved in the preparation of the book. Finally, I would like to express my gratitude to Peter Momtchiloff, my OUP editor, for his warm encouragement and support.

1 Making Sense of Emergence I It has been about a century and half since the ideas that we now associate with emergentism began taking shape.1 At the core of these ideas was the thought that as systems acquire increasingly higher degrees of organizational complexity they begin to exhibit novel properties that in some sense transcend the properties of their constituent parts, and behave in ways that cannot be predicted on the basis of the laws governing simpler systems. It is now standard to trace the birth of emergentism back to John Stuart Mill and his distinction between “heteropathic” and “homopathic” laws,2 although few of us would be surprised to learn that the same or similar ideas had been entertained by our earlier philosophical forebears.3 Academic philosophers— like Samuel Alexander and C.D. Broad in Britain, A.O. Lovejoy and Roy Wood Sellars in the United States—played an important role in developing the concept of emergence and the attendant doctrines of emergentism, but it is interesting to note that the fundamental idea seems to have had a special appeal to scientists and those outside professional philosophy. These include the British biologist C. Lloyd Morgan, a leading theoretician of the emergentist movement early in this century, and, more recently, the noted neurophysiologist Roger W. Sperry. In spite of its obvious and direct relevance to some of the central issues in the philosophy and methodology of science, however, emergentism failed 1 For helpful historical surveys of emergentism see Brian McLaughlin, “The Rise and Fall of British Emergentism”, and Achim Stephan, “Emergence—A Systematic View on Its Historical Facets”, both in Emergence or Reduction?, ed. A. Beckermann, H. Flohr, and J. Kim (Berlin: De Gruyter, 1993). 2 In A System of Logic (1843), Bk. III, ch. vi. 3 It appears that Galen (ad 129 – c.200) had a clear statement of the distinction between emergent and nonemergent properties of wholes; see On the Elements according to Hippocrates, 1.3, 70.15–74.23. I owe this reference to Victor Caston.

MAKING SENSE OF EMERGENCE

9

to establish itself as a player in the mainstream philosophy of science. The main reason for this, I believe, is that philosophy of science during much of the middle half of this century, from the 1930s to the ’70s—at least, in the analytic tradition—was shaped by the positivist and hyper-empiricist view of science that dominated the Anglo-American philosophy at the time. Influential philosophers of science during this period—for example, Carl Hempel and Ernest Nagel4—claimed that the classic idea of emergence was confused and incoherent, often likening it to neo-vitalism, and what they saw as the only salvageable part of the emergence concept—the part that they could state in their own positivist/formalist idiom—usually turned out to be largely trivial, something that could be of little interest for serious philosophical purposes. But the idea of emergence refused to die, continuing to attract a small but steady stream of advocates from both the philosophical and the scientific ranks, and it now appears to be making a strong comeback. This turn of events is not surprising, given the nearly total collapse of positivistic reductionism and the ideal of unified science which was well underway by the early ’70s. The lowly fortunes of reductionism have continued to this day, providing a fertile soil for the reemergence of emergentism. Classic emergentists like Morgan and Alexander thought of themselves as occupying a moderate intermediate position between the extremes of “mechanistic” reductionism on one hand and explicit dualisms like Cartesianism and neo-vitalism on the other. For them everything that exists is constituted by matter, or basic material particles, there being no “insertion” of alien entities or forces from the outside. It is only that complex systems aggregated out of these material particles begin to exhibit genuinely novel properties that are irreducible to, and neither predictable nor explainable in terms of, the properties of their constituents. It is evident that emergentism is a form of what is now standardly called “nonreductive materialism,” a doctrine that aspires to position itself as a compromise between physicalist reductionism and all-out dualisms. It is no wonder then that we now see an 4 See Hempel’s “Studies in the Logic of Explanation” (with Paul Oppenheim), reprinted in his Aspects of Scientific Explanation (New York: The Free Press, 1965), and Nagel, The Structure of Science (New York: Harcourt, Brace & World, 1961), chap. 11. It is interesting to note that another early positivist philosopher of science, Karl Popper, became in the final stages of his career a strong defender of emergentism; see John C. Eccles and Karl R. Popper, The Self and Its Brain (Berlin & New York: Springer International, 1977).

10


increasing, and unapologetic, use of expressions like “emergent property,” “emergent phenomenon,” and “emergent law,” substantially in the sense intended by the classic emergentists, not only in philosophical writings but in primary scientific literature as well.5 Does this mean that emergentism has returned—as an ontological doctrine about how the phenomena of this world are organized into autonomous emergent levels and as a metascientific thesis about the relationship between basic physics and the special sciences? I think the answer is a definite yes. The fading away of reductionism and the enthronement of nonreductive materialism as the new orthodoxy simply amount to the resurgence of emergentism—not all of its sometimes quaint and quirky ideas but its core ontological and methodological doctrines. The return of emergentism is seldom noticed, and much less openly celebrated; it is no surprise, though, that the fortunes of reductionism correlate inversely with those of emergentism. It is no undue exaggeration to say that we have been under the reign of emergentism since the early 1970s. I have argued elsewhere6 against nonreductive materialism, urging that this halfway house is an inherently unstable position, and that it threatens to collapse into either reductionism or more serious forms of dualism. But in this paper I am not primarily concerned with the truth or tenability of emergentism or nonreductive materialism; rather, my main concern is with making sense of the idea of emergence—the idea that certain properties of complex systems are emergent while others are not. Even if we succeed with the conceptual task of giving a coherent sense to emergence, it is another question whether any particular group of properties is emergent—for example, whether intentional or qualitative mental properties are emergent relative to neural/biological properties, or whether biological properties are emergent relative to physicochemical properties—or indeed whether there are any emergent properties at all. In trying to make emergence intelligible, it is useful to divide the ideas usually associated with the concept into two groups. One group of ideas is manifest in the statement that emergent properties are “novel” and “unpredictable” from the knowledge of their lower-level bases, and that they are 5 e.g., John Searle, The Rediscovery of the Mind (Cambridge, Mass.: MIT Press, 1992); Francisco Varella, Evan Thompson, and Eleanor Rosch, The Embodied Mind (Cambridge, Mass.: MIT Press, 1993). 6 In particular, “The Myth of Nonreductive Materialism,” reprinted in Supervenience and Mind (Cambridge: Cambridge University Press, 1993). First published in 1989.


11

not “explainable” or “mechanistically reducible” in terms of their underlying properties. The second group of ideas I have in mind comprises the specific emergentist doctrines concerning emergent properties, and, in particular, claims about the causal powers of the emergents. Prominent among them is the claim that the emergents bring into the world new causal powers of their own, and, in particular, that they have powers to influence and control the direction of the lower-level processes from which they emerge. This is a fundamental tenet of emergentism, not only in the classic emergentism of Samuel Alexander, Lloyd Morgan, and others but also in its various modern versions. Emergentists often contrast their position with epiphenomenalism, dismissing the latter with open scorn. On their view, emergents have causal/explanatory powers in their own right, introducing novel, and hitherto unknown, causal structures into the world. In this paper I will adopt the following strategy: I am going to take the first group of ideas as constitutive of the idea of an emergent property, and try to give a unified account of emergence on the basis of a model of reduction that, although its basic ideas are far from new, is significantly different from the classic Nagelian model of reduction which has formed the background of debates in this area. I will then consider the doctrines that I take to constitute emergentism, focusing on the claims about the causal powers of emergent properties, especially the idea of “downward causation.”

II The concepts of explanation, prediction, and reduction figure prominently at several critical junctures in the development of the doctrine of emergence. Most importantly, the concept of explanation is invoked in the claim that emergent phenomena or properties, unlike those that are merely “resultant”, are not explainable, or reductively explainable, on the basis of their “basal conditions,” the lower-level conditions out of which they emerge. This is frequently coupled with the claim that emergent phenomena are not predictable even from the most complete and exhaustive knowledge of their emergence bases. I believe that emergentists took the two claims to be equivalent, or at least as forming a single package.

12


Let us assume that every material object has a unique complete microstructural description: that is, any physical system can be exhaustively described in terms of (i) the basic particles that constitute it (this assumes classic atomism, which the early emergentists accepted); (ii) all the intrinsic properties of these particles; and (iii) the relations that configure these particles into a structure (with “substantial unity,” as some emergentists would say). Such a description will give us the total “relatedness” of basal constituents; it also gives us what we may call the total microstructural (or microbased) property of the system—that is, a macroproperty (macro since it belongs to the system as a whole) constituted by the system’s basic microconstituents, their intrinsic properties, and the relations that structure them into a system with unity and stability as a substance.7 I would expect most emergentists to accept mereological supervenience, in the following form: [Mereological supervenience] Systems with an identical total microstructural property have all other properties in common.8 Equivalently, all properties of a physical system supervene on, or are determined by, its total microstructural property. It is a central claim of classic emergentism that among these properties supervenient on a system’s total microstructural property, some have the special character of being “emergent”, while the rest are only “resultant.” What is the basis for this distinction? Lloyd Morgan says this: The concept of emergence was dealt with (to go no further back) by J.S. Mill . . . The word ‘emergent’, as contrasted with ‘resultant’, was suggested by G.H. Lewes . . . Both adduce examples from chemistry and from physiology; both deal with properties; both distinguish those properties (a) which are additive and subtractive only, and predictable, from those (b) which are new and unpredictable.9

7 Some will complain that this picture is inextricably wedded to the now defunct pre-quantum classical particle physics; that may be, but it is the picture the British emergentists worked with. Moreover, it is an open question, I believe, whether anything of substance would change if the issues were set in a quantum-mechanical framework. 8 Obviously extrinsic/relational/historical properties (e.g., being 50 miles to the south of Boston) must be excluded, and the statement is to be understood to apply only to the intrinsic properties of systems. There is also a tacit assumption that the intrinsic properties of a system determine its causal powers. 9 C. Lloyd Morgan, Emergent Evolution (London: Williams and Norgate, 1923), pp. 2–3.


13

There is no need to interpret the talk of “additivity” and “subtractability” literally; I believe these terms were used to indicate that resultant properties are simply and straightforwardly calculated and predicted from the base properties. But obviously ease and simplicity of calculation as such is of no relevance here; predictability is not lost or diminished if calculationally complex mathematical/logical procedures must be used.10 I believe that predictability is the key idea here, and that an appropriate notion of predictability must be explained in terms that are independent of addition or subtraction, or the simplicity of mathematical operations. In any case, resultant properties are to be those that are predictable from a system’s total microstructural property, but emergent properties are those that are not so predictable. Morgan’s (b) above introduces the idea of “newness,” or “novelty,” an idea often invoked by the emergentists. Is he using “new” and “unpredictable” here as expressing more or less the same idea, or is he implying, or at least hinting, that emergent properties are unpredictable because they are new and novel properties? I believe that “new” as used by the emergentists has two dimensions: an emergent property is new because it is unpredictable, and this is its epistemological sense; and, second, it has a metaphysical sense, namely that an emergent property brings with it new causal powers, powers that did not exist before its emergence. We will discuss the causal issue in the latter part of this paper. In speaking of predictability, it is important to distinguish between inductive predictability and theoretical predictability, a distinction that I believe the emergentists were clearly aware of. Even emergent properties are inductively predictable: Having observed that an emergent property, E, emerged whenever any system instantiated a microstructural property M, we may predict that this particular system will instantiate E at t, given our knowledge or belief that it will instantiate, M, at t.11 More generally, on the basis of such empirical data we may have a well-confirmed “emergence 10 What I believe to be more appropriate here is William C. Wimsatt’s notion of aggregativity defined in terms of certain invariance conditions; see his “Emergence as Non-Aggregativity and the Biases of Reductionisms”, Foundations of Science 5 (2000): 269–297. However, I bypass these considerations here in favor of a simpler and more straightforward notion of predictability. See also Paul Humphreys’ interesting paper, “How Properties Emerge,” Philosophy of Science 64 (1997): 1–17. I cannot discuss here Humphreys’ interesting proposals, but I believe everything of any significance I say here is consistent with his views. 11 Cf. Morgan: “Lewes says that the nature of emergent characters can only be learnt by experience of their occurrence; hence they are unpredictable before the event.” Emergent Evolution, p. 5.

14


law” to the effect that whenever a system instantiates basal condition M it instantiates an emergent, E. What is being denied by emergentists is the theoretical predictability of E on the basis of M: we may know all that can be known about M—in particular, laws that govern the entities, properties, and relations constitutive of M—but this knowledge does not suffice to yield a prediction of E. This unpredictability may be the result of our not even having the concept of E, this concept lying entirely outside the concepts in which our theory of M is couched. In cases where E is a phenomenal property of experiences (a “quale”), we may have no idea what E is like before we experience it.12 But this isn’t the only barrier to predictability. It may be that we know what E is like—we have already experienced E—but we may be powerless to predict whether or not E—or whether E rather than another emergent E*—will emerge when a complex is formed with a novel microstructure M* that is similar to M in some significant respects. In such a case the emergence law “Whenever a system instantiates M, it instantiates E” would have to be taken as a basic law, stating a brute correlation between M and E. It is clear that we can inductively predict—in fact, we do this all the time—the occurrences of conscious states in the sense just explained, but, if the emergentists were right about anything, they were right about the phenomenal properties of conscious experience: these properties appear not to be theoretically predictable on the basis of a complete knowledge of the neurophysiology of the brain. This is reflected in the following apparent difference between phenomenal properties and other mental properties (including cognitive/intentional properties): We can imagine designing and constructing novel physical systems that will instantiate certain cognitive capacities and functions (e.g., perception, information processing, inference and reasoning, and using information to guide behavior)—arguably, we have already designed and fabricated such devices in robots and other computer-driven mechanisms. But it is difficult to imagine our designing novel devices and structures that will have phenomenal experiences; I don’t think we have any idea where to begin. The only way we can hope to manufacture a mechanism with phenomenal consciousness is to produce an appropriate physical duplicate of a system that is 12 See, e.g., what Michael Tye calls “perspectival subjectivity,” in his Ten Problems of Consciousness (Cambridge, Mass.: MIT Press, 1995). The situation here reminds one of Frank Jackson’s much discussed super-vision scientist Mary confined in a black-and-white room.


15

known to be conscious. Notice that this involves inductive prediction, whereas theoretical prediction is what is needed to design new physical devices with consciousness. The emergentists were wrong in thinking that sundry chemical and biological properties were emergent;13 but this was an understandable mistake given the state of the sciences before the advent of solid-state physics and molecular biology. The interest of the ideas underlying the emergentist’s distinction between the two kinds of properties need not be diminished by the choice of wrong examples.

III As was noted at the start of our discussion, another idea that is closely related to the claimed unpredictability of emergents is the doctrine that the emergence of emergent properties cannot be explained on the basis of the underlying processes, and that emergent properties are not reducible to the basal conditions from which they emerge. These two claims can be combined into one: Emergent properties are not reductively explainable in terms of the underlying processes. Some may wish to distinguish the issue of reduction from that of reductive explanation;14 we will address this issue later. I will now turn to the task of describing a model of reduction that connects and makes sense of these three ideas, namely that emergent properties are not predictable from their basal conditions, that they are not explainable in terms of them, and that they are not reducible to them. Let me begin with an example—an idealized, admittedly somewhat simplistic example. To reduce the gene to the DNA molecule, we must first prime the target property, by giving it a functional interpretation—that is, by construing it in terms of the causal work it is to perform. Briefly, the property of being a gene is the property of having some property (or being a mechanism) that performs a certain causal function, namely that of transmitting phenotypic characteristics from parents to offspring. As it turns out, it is the DNA molecule that fills this causal specification (“causal role”), and we have a theory that explains just how the DNA molecule is able to 13 As noted by McLaughlin in “The Rise and Fall of British Emergentism.” 14 For example, David Chalmers, The Conscious Mind (New York: Oxford University Press, 1996), p. 43.

16


perform this causal work. When all of this is in, we are entitled to the claim that the gene has been reduced to the DNA molecule. We can now formulate a general model to accommodate reductions of this form. Let B be the domain of properties (also phenomena, facts, etc., if you wish) serving as the reduction base—for us, these contain the basal conditions for our emergent properties. The reduction of property E to B involves three steps:15 Step 1: E must be functionalized—that is, E must be construed, or reconstrued, as a property defined by its causal/nomic relations to other properties, specifically properties in the reduction base B. We can think of a functional definition of E over domain B as typically taking the following (simplified) form: Having E ¼ def Having some property P in B such that (i) C1, . . . , Cn16 cause P to be instantiated, and (ii) P causes F1, . . . , Fm to be instantiated. (We allow either (i) or (ii) to be empty.) The main point to notice is that the functionalization of E makes E nonintrinsic and relational—relational with respect to other properties in B. E’s being instantiated is for a certain property P to be instantiated, with this instantiation bearing causal/nomic relations to the instantiations of a specified set of properties in the base domain. We may call any property P in B that satisfies the causal specification (i) and (ii) a “realizer” or “implementer” of E. Clearly, multiple realizers for E are allowed on this account; so multiply realizable properties fall within the scope of the present model of reduction. A functionalization

15 The fundamental ideas for this view of reduction are present in David Armstrong’s A Materialist Theory of Mind (New York: Humanities Press, 1964), and David Lewis’s “An Argument for the Identity Theory,” Journal of Philosophy 67 (1970): 203–211. However, neither Armstrong nor Lewis, to my knowledge, explicitly associate these ideas directly with models of reduction. The idea of functional analysis of mental terms or properties is of course the heart of the functionalist approach to mentality; it is interesting, therefore, to note that most functionalists have regarded their approach as essentially antireductionist. For similar views on reduction see Robert Van Gulick, “Nonreductive Materialism and the Nature of Intertheoretic Constraint,” in Emergence or Reduction?; Joseph Levine, “On Leaving Out What It Is Like,” in Consciousness, ed. Martin Davies and Glyn W. Humphreys (Oxford: Blackwell, 1993). See also Chalmers’ discussion of “reductive explanation,” in The Conscious Mind, ch. 2. I discuss these issues in greater detail in Mind in a Physical World (Cambridge, Mass.: MIT Press, 1998). 16 For brevity we will often speak of a property causing another property—what is meant of course is that an instantiation of a property causes another property to be instantiated.


17

of property E in the present sense is to be taken as establishing a conceptual/ definitional connection for E and the selected causal role. An important part of this procedure is to decide how much of what we know (or believe) about E’s nomic/causal involvement should be taken as defining, or constitutive of, E and how much should be left out. We should keep in mind that such conceptual decisions can be and often are based on empirical knowledge, knowledge of the causal/nomic relations in which E is embedded, and can be constrained by theoretical desiderata of various sorts, and that in practice the boundary between what’s conceptual and what isn’t is certain to be a vague and shifting one. Step 2: Find realizers of E in B. If the reduction, or reductive explanation, of a particular instance of E in a given system is wanted, find the particular realizing property P in virtue of which E is instantiated on this occasion in this system; similarly, for classes of systems belonging to the same species or structure types. This of course is a scientifically significant part of the reductive procedure; it took many years of scientific research to identify the DNA as a realizer of the gene. Step 3: Find a theory (at the level of B) that explains how realizers of E perform the causal task that is constitutive of E (i.e., the causal role specified in Step 1). Such a theory may also explain other significant causal/nomic relations in which E plays a role. We presumably have a story at the microbiological level about how DNA molecules manage to code and transmit genetic information. When temperature, for gases, is reduced to mean translational kinetic energy of molecules (another over-simplified stock example17), we have a theory that explains the myriad causal/nomic relations in which temperature plays a role. Steps 2 and 3 can be expected to be part of the same scientific research: ascertaining realizers of E will almost certainly involve theories about causal/nomic interrelations among lower-level properties in the base domain. Notice how this functional conception of reduction differs from the classic Nagel model of intertheoretic reduction18—in particular, there is 17 See Lawrence Sklar, Physics and Chance (Cambridge: Cambridge University Press, 1993). 18 See Ernest Nagel, The Structure of Science (New York: Harcourt, Brace, and World, 1961), chap. 11.

18


no talk of “bridge laws” or “derivation” of laws. The question whether appropriate bridge laws are available that connect the domain to be reduced with the base domain—more specifically, whether or not there are bridge laws providing for each property to be reduced a nomically coextensive property in the base domain—has been at center stage in debates over reduction and reductionism. However, from the emergentist point of view, the bridge laws, far from being the enablers of reduction (as they are in Nagel reductions), are themselves among the targets of reduction. For it is these bridge laws, laws that state that whenever certain specified basal conditions are present a certain novel property is manifested, that the emergentists were anxious to have explained. Why is it that pain, not itch or tickle, occurs when a certain neural condition (e.g., C-fiber stimulation) obtains? Why doesn’t pain accompany conditions of a different neural type? Why does any phenomenal consciousness occur when these neural conditions are present? These are the kinds of explanatory/reductive questions with which the emergentists were preoccupied. And I think they were right. The “mystery” of consciousness is not dispelled by any purported reductive procedure that, as in Nagel reduction, takes these bridge laws as brute unexplained primitives. The philosophical emptiness of Nagel reduction (at least in contexts like mind-body reduction), if it isn’t already evident, can be plainly seen from the following fact: a Nagel reduction of the mental to the physical is consistent with, and sometimes even entailed by, many dualist mindbody theories, such as the double-aspect theory, the theory of preestablished harmony, occasionalism, and epiphenomenalism. It is not even excluded by the dualism of mental and physical substances (although Descartes’ own interactionist version probably excludes it). This amply shows that the antireductionist argument based on the unavailability of mind-body bridge laws—most importantly, the multiple realization argument of Putnam and Fodor—is irrelevant to the real issue of mindbody reduction or the possibility of giving a reductive explanation of mentality. Much of the debate over the past two decades about reductionism has been carried on in terms of an inappropriate model of reduction, and now appears largely beside the point for issues of real philosophical significance.


19

IV Let us now try to see how the functional model of reduction can meet the explanatory/predictive/ontological demands that reductions of genuine philosophical interest must meet. Let E be the property targeted for reduction, where E has been functionalized as the property of having some property P meeting causal specification C. 1. The explanatory question Why does this system exhibit E at t? Because having E is, by definition, having a property with causal role C, and the system, at t, has property Q, which fills causal role C (and hence realizes E). Moreover, we have a theory that explains exactly how Q manages to fill C. Why do systems exhibit E whenever they instantiate Q? Because E is a functional property defined by causal role C, and Q is a realizer of E for these systems. And there is a theory that explains how Q realizes E in these systems. Suppose that pain could be given a functional definition—something like this: being in pain is being in some state (or instantiating some property) caused by tissue damage and causing winces and groans. Why are you experiencing pain? Because being in pain is being in a state caused by tissue damage and causing winces and groans, and you are in neural state N, which is one of those states (in you, or in systems like you) that are caused by tissue damage and that cause winces and groans. Why do people experience pain when they are in neural state N? Because N is implicated in these causal/ nomic relations, and being in pain is being in some state with just these causal/nomic relations. It is clear that in this way all our explanatory demands can be met. There is nothing further to be explained about why pain occurs, or why pain occurs when neural condition N is present. But is this a reductive explanation? This question is connected with the question whether, and in what sense, the proposed model is a model of reduction, a question that will be considered below. It is of course another question whether pain can be functionalized. We will briefly return to this issue later, but our concern here is to give a clear sense to what it is to “reduce” pain.

20


2. The predictive question Will this system exhibit E at time t? Can we predict this from knowledge of what goes on in the base domain? Yes, because, given the functional definition of E, we can in principle identify the realizers of E for the system solely on the basis of knowledge of the causal/nomic relations obtaining in the base domain. Given this knowledge of E’s realizers for this system, we can predict whether or not the system will, at t, instantiate property E from our knowledge, or warranted belief, that it will, or will not, instantiate a realizer of E at t. Clearly, what enables the ascent from the reduction base to higher properties is the conceptual connections generated by the functionalization of the higher properties. This is in sharp contrast to Nagelian reduction with bridge laws taken as auxiliary premises. These laws are standardly conceived as empirical and contingent, and must be viewed as net additions to our theory about the reduction base, which means that the base theory so augmented is no longer a theory exclusively about the base domain. This is why bridge laws only enable inductive predictions, whereas functionalization makes theoretical predictions possible. These reflections seem to give us an answer to a question we raised earlier—why we seem to lack the ability to design novel physical devices that will exhibit phenomenal consciousness: it is because brute bridge laws may be all we can get to connect phenomenal properties with physical properties, whereas what is required is an ability to make theoretical predictions of qualia solely on the basis of knowledge of the base domain, namely physics, chemistry, biology, and the like. The functionalization of phenomenal experience would give us such an ability. 3. The ontological question In what sense is the functional model a model of reduction? What does it reduce, and how does it do it? Central to the concept of reduction evidently is the idea that what has been reduced need not be countenanced as an independent existent beyond the entities in the reduction base—that if X has been reduced to Y, X is not something “over and above” Y. From an ontological point of view, reduction must mean reduce—it must result in a simpler, leaner ontology. Reduction is not necessarily elimination:


21

reduction of X to Y need not do away with X, for X may be conserved as Y (or as part of Y). Thus, we can speak of “conservative” reduction (some call it “retentive” or “preservative” reduction), reduction that conserves the reduced entities, as distinguished from “eliminative” reduction, which rids our ontology of the reduced entities. Either way we end up with a leaner ontology. Evidently, conservative reduction requires identities, for to conserve X as Y means that X is Y, whereas eliminative reduction has no need for (in fact, excludes) such identities. Our question, then, is in what ways the model of reduction being recommended here serves the cause of ontological simplification. Two cases may be distinguished: the first concerns instances of property E; the second concerns property E itself. First, consider property instances: system s has E, in virtue of s’s instantiating one of its realizers, say Q. Now, s’s having E on this occasion just is its having some property meeting causal specification C, and in this particular instance, s has Q, where Q meets specification C. It seem perfectly natural, then, to identify s’s having E on this occasion with its having Q on this occasion. There seems no fact of the matter about s’s having E on this occasion over and above s’s having Q. Each instance of E, therefore, is an instance of one of E’s realizers, and all instances of E can be partitioned into Q1-instances, Q2-instances, . . . , where the Q’s are E’s realizers. Hence, the E-instances reduce to the Qi-instances. Suppose someone were to object as follows: There is no good reason to identify this instance of E with the instance of Q in virtue of which E is realized on this occasion. Rather, s’s having E should be identified with s’s having some property or other meeting causal specification C, and this latter instance is not identical with s’s having Q. For having some property or other meeting C is not the same property as having Q; that is, property E 6¼ property Q. How should we counter this line of argument? I think it will be helpful to consider the causal picture, and ask: What are the causal powers of this instance of E, namely s’s having E on this occasion? If s has E in virtue of E’s realizer Q, it is difficult to see how we could avoid saying this: the causal powers of this instance of E are exactly the causal powers of this instance of Q. This is what I have elsewhere called the “causal inheritance principle”:

22


If a functional property E is instantiated on a given occasion in virtue of one of its realizers, Q, being instantiated, then the causal powers of this instance of E are identical with the causal powers of this instance of Q. If this principle is accepted, the E-instance and the Q-instance have identical causal properties, and this exerts powerful pressure to identify them. What good would it do to count them as different? If they were different, the difference could not even be detected. Refusal to identify them could leave the E-instances high and dry without any causal powers, epiphenomenal entities that no one needs to care about. This means that on the present picture E-instances are conservatively reduced to Q-instances, instances of E’s realizers. Let us now turn to the reduction of E, the property itself. Here we need to come to terms with E’s having multiple realizers, Q1, Q2, . . . There are three possible approaches here. First, one may choose to defend E as a legitimate higher-level property irreducible to its realizers, the Q’s. This is the position taken by many functionalists: psychological properties are functional properties defined in terms of input/output correlations, with internal physical/biological properties as realizers, and yet they are irreducible to their realizers, constituting an autonomous domain for the special science of psychology (cognitive science, or whatever). Second, one may choose to identify E with the disjunction of its realizers, Q1 v Q2 v . . . 19 Notice, though, that this identity is not necessary—it does not hold in every possible world—since whether or not a property realizes E depends on the laws that prevail at a given world. The reason is that E is defined in terms of a causal/nomic condition, and whether something satisfies such conditions depends on the laws that are in force at a given world. This means that in another world with different laws, E may have a wholly distinct set of realizers, and in still others E may have no realizers at all. So the identity, E ¼ Q1 v Q2 v . . . is metaphysically contingent, although nomologically necessary, and “E” becomes nonrigid, although it remains nomologically rigid or “semi-rigid” (as I have called it elsewhere). For example, in a world with laws quite different from those prevailing in this 19 In “Reduction with Autonomy” (Philosophical Perspectives 11 (1997): 83–105), Louise Antony and Joseph Levine advance interesting arguments against the disjunctive approach.


23

world, molecules of another kind, not the DNA molecules, may perform the causal task of coding and transmitting genetic information.20 Third, we may give up E as a genuine property and only recognize the expression “E” or the concept E. As it turns out, many different properties are picked out by the concept E, depending on the circumstances—the kind of structures involved and the nomological nature of the world under consideration. One could argue that by forming “second-order” functional expressions by existentially quantifying over “first-order” properties, we cannot be generating new properties (possibly with new causal powers), but only new ways of indifferently picking out, or grouping, first-order properties, in terms of causal specifications that are of interest to us.21 As noted, the concept is only nomologically rigid: it picks out the same properties only across worlds that are similar in causal/nomological respects. Here I will not argue my points in detail. It is clear, however, that the second and third approach effectively reduce the target property E: the second is a conservative reduction, retaining E as a world-indexed disjunction of properties in the base domain. In contrast, the third is eliminative: it recommends the elimination of E as a property, retaining only the concept E (which may play a practically indispensable role in our discourse, both ordinary and scientific). The first approach, as I said, is one that is widely accepted: many philosophers, on account of multiple realization, want to argue that E is an irreducible property that nonetheless can be a property playing an important role in a special, “higher-level,” science. I believe, however, that this position cannot be sustained. For if the “multiplicity” or “diversity” of realizers means anything, it must mean that these realizers are causally and nomologically diverse. Unless two realizers of E show significant causal/ nomological diversity, there is no clear reason why we should count them as two, not one. It follows then that multiply realizable properties are ipso facto causally and nomologically heterogeneous. This is especially obvious when one reflects on the causal inheritance principle. All this points to the inescapable conclusion that E, because of its causal/nomic heterogeneity, is 20 This point is valid whether or not E has single or multiple realizers in the actual world. A property may have a single realizer here but multiple realizers in other worlds, and vice versa. For further issues with the disjunction strategy, see Essay 10 of this volume. 21 For more details on this approach see my “The Mind-Body Problem: Taking Stock After 40 Years,” Philosophical Perspectives 11 (1997): 185–207, and Mind in a Physical World.

24


unfit to figure in laws, and is thereby disqualified as a useful scientific property. On this approach, then, one could protect E but not as a property with a role in scientific laws and explanations. You could insist on the genuine propertyhood of E as much as you like, but the victory would be empty.22 The conclusion, therefore, has to be this: as a significant scientific property, E has been reduced—eliminatively. What I hope I have shown is this: Functionalization of a property is both necessary and sufficient for reduction (sufficient at least as a first conceptual step, the rest being scientific research). This accords well with the classic doctrines of emergentism: as I argued, it nicely explains why reducible properties are predictable and explainable, and correlatively it explains why irreducible properties are neither predictable nor explainable on the basis of the underlying processes. I believe this makes good sense of the central ideas that make up the concept of emergence. However, emergentism may yet be an empty doctrine. For there may not be any emergent properties, all properties being physical properties or else functionalizable and therefore reducible to physical properties. Physical properties include not only basic physical magnitudes and the properties of microparticles but microstructural properties of larger complexes of basic particles. So are there emergent properties? Many scientists have argued that certain “self-organizational” properties of organic, living systems are emergent. But it is not clear that these are emergent in our sense of nonfunctionalizability.23 And, as I said earlier, the classic emergentists were mostly wrong in putting forward examples of chemical and biological properties as emergent. It seems to me that if anything is going to be emergent, the phenomenal properties of consciousness, or “qualia,” are the most promising candidates. Here I don’t want to rehearse the standard arguments pro and con, but merely affirm, for what it’s worth, my own bias toward the pro side: qualia are intrinsic properties if anything is, and to functionalize them is to eliminate them as intrinsic properties.24

22 For more details see my “Multiple Realization and the Metaphysics of Reduction”, Philosophy and Phenomenological Research 52 (1992): 1–26; reprinted in Supervenience and Mind. 23 This point is argued by David Chalmers; see his The Conscious Mind, p. 129. 24 More details and an overview of the philosophical terrain involved, see Chalmers, The Conscious Mind, chap 3. Two early papers arguing this point are Joseph Levine, “Materialism and Qualia: the Explanatory Gap,” Pacific Philosophical Quarterly 64 (1983): 354–361, and Frank Jackson, “Epiphenomenal Qualia,” Philosophical Quarterly 32 (1982): 127–136.


25

V The doctrine of emergence has lately been associated quite closely with the idea of “downward causation.” It is not only that emergent properties are to have their own distinctive causal powers but also that they be able to exercise their causal powers “downward”—that is, with respect to processes at lower-levels, levels from which they emerge. The claim that emergents have causal powers is entirely natural and plausible if you believe that there are such properties. For what purpose would it serve to insist on the existence of emergent properties if they were mere epiphenomena with no causal or explanatory relevance? The very idea of downward causation involves vertical directionality—an “upward” direction and a “downward” direction. This in turn suggests an ordered hierarchy of domains that gives meaning to talk of something being located at a “higher” or “lower” or “the same” position in relation to another item on this hierarchy. As is familiar to everyone, positions on such a hierarchy are usually called “levels,” or sometimes “orders.” In fact, talk of “levels”—as in “level of description,” “level of explanation,” “level of organization,” “level of complexity,” “level of analysis,” and the like— has thoroughly penetrated not only writings about science, including of course philosophy of science, but also the primary scientific literature of many fields. The emergentists of the early 20th century were among the first to articulate what may be called “the layered model” of the world, although a general view of this kind is independent of emergentism and has been espoused by those who are opposed to emergentism.25 In fact, a model of this kind provides an essential framework needed to formulate the emergentist/reductionist debate. In any case, the layered model takes the natural world as stratified into levels, from lower to higher, from the basic to the constructed and evolved, from the simple to the more complex. All objects and phenomena have each a unique place in this ordered hierarchy. Most early emergentists, such as Samuel Alexander and C. Lloyd Morgan, viewed

25 See, e.g., Paul Oppenheim and Hilary Putnam, “Unity of Science as a Working Hypothesis,” in Minnesota Studies in Philosophy of Science, vol. 2, ed. Herbert Feigl, Michael Scriven, and Grover Maxwell (Minneapolis: University of Minnesota Press, 1958). For further discussion of the layered model, see Essay 2 of this volume.

26


this hierarchy to have evolved historically: In the beginning there were only basic physical particles, or just a space-time framework (as Alexander maintained), and these have evolved into increasingly more complex structures—atoms, molecules, unicellular organisms, multicellular organisms, organisms with consciousness and mentality, and so on. Contemporary interest in emergence and the hierarchical model is focused not on this kind of quasi-scientific and quasi-metaphysical history of the world, but rather on what it says about the synchronic structure of the world—how things and phenomena at different levels hang together in a temporal cross-section of the world, or over small time intervals. We want to know whether, and how, the emergentist ideas can help us in understanding the interlevel relationships between items at the adjacent levels on this hierarchy, and ultimately how everything is related to the items at the bottom physical level (if there is such a level). The layered model gives rise to many interesting questions: for example, how are these levels to be defined and individuated? Is there really a single unique hierarchy of levels that encompasses all of reality or does this need to be contextualized or relativized in certain ways? Does a single ladder-like structure suffice, or is a branching tree-like structure more appropriate? Exactly what ordering relations generate the hierarchical structures? But these questions go well beyond the scope of this paper. Here we will work with the fairly standard, intuitive notion of levels that is shared by most of us.26 This will not significantly compromise the discussion to follow. Although, as one would expect, there has been no universal agreement among the emergentists, the central doctrines of emergentism are well known. For our present purposes, we will take them to include the following claims: 1. Emergence of complex higher-level entities: Systems with a higher-level of complexity emerge from the coming together of lower-level entities in new structural configurations (the new “relatedness” of these entities).

26 For an informative discussion of the issues in this area see William C. Wimsatt, “Reductionism, Levels of Organization, and the Mind-Body Problem,” in Consciousness and the Brain, ed. Gordon G. Globus, Grover Maxwell, and Irwin Savodnik (New York: Plenum Press, 1976), and “The Ontology of Complex Systems: Levels of Organization, Perspectives, and Causal Thickets,” Canadian Journal of Philosophy, suppl. vol. 20 (1994): 207–274.


27

This claim is by no means unique to emergentism; it is completely at home with universal physical reductionism (what the early emergentists called “mechanism”), the view that all things and phenomena are physical, and are explainable and predictable ultimately in terms of fundamental physical laws. A characteristically emergentist doctrine makes its appearance in the idea that some of the properties of these complex systems, though physically grounded, are nonphysical, and belong outside the physical domain. The following three propositions unpack this idea. 2. Emergence of higher-level properties: All properties of higher-level entities arise out of the properties and relations that characterize their constituent parts. Some properties of these higher, complex systems are “emergent,” and the rest merely “resultant.” Instead of “arise out of,” such expressions as “supervene on” and “are consequential upon” could have been used. In any case, the idea is that when appropriate lower-level conditions are realized in a higher-level system (that is, the parts that constitute the system come to be configured in a certain relational structure), the system will necessarily exhibit certain higher-level properties, and, moreover, that no higher-level property will appear unless an appropriate set of lower-level conditions is realized. Thus, “arise” and “supervene” are neutral with respect to the emergent/resultant distinction: both emergent and resultant properties of a whole supervene on, or arise out of, its microstructural, or micro-based, properties. The distinction between properties that are emergent and those that are merely resultant is a central component of emergentism. As we have already seen, it is standard to characterize this distinction in terms of predictability and explainability. 3. The unpredictability of emergent properties: Emergent properties are not predictable from exhaustive information concerning their “basal conditions.” In contrast, resultant properties are predictable from lowerlevel information. 4. The unexplainability/irreducibility of emergent properties: Emergent properties, unlike those that are merely resultant, are neither explainable nor reducible in terms of their basal conditions. Earlier in this paper we saw how it is possible to give unity to these claims on the basis of an appropriate analysis of reduction. More specifically, by

28


identifying emergent properties with irreducible properties, on the functional model of reduction, it is possible to explain why emergent properties are neither explainable nor predictable on the basis of the conditions from which they emerge, whereas nonemergent (or resultant) properties are so explainable and predictable. Our present concern, however, lies with the question what emergent properties, after having emerged, can do—that is, how they are able to make their special contributions to the ongoing processes of the world. It is obviously very important to the emergentists that emergent properties can be active participants in causal processes involving the systems they characterize. None perhaps understood this better than Samuel Alexander, who made the following pointed comment on epiphenomenalism, the doctrine that mental properties are wholly lacking in causal powers: [Epiphenomenalism] supposes something to exist in nature which has nothing to do, no purpose to serve, a species of noblesse which depends on the work of its inferiors, but is kept for show and might as well, and undoubtedly would in time be abolished.27 We may, therefore, set forth the following as the fifth doctrine of emergentism: 5. The causal efficacy of the emergents: Emergent properties have causal powers of their own—novel causal powers irreducible to the causal powers of their basal constituents. In what ways, then, can emergent properties manifest their causal powers? This of course is where the idea of “downward causation” enters the scene. But when we view the situation with the layered model in mind, we see that the following three types of inter- or intra-level causation must be recognized: (i) same-level causation, (ii) downward causation, and (iii) upward causation. Same-level causation, as the expression suggests, involves causal relations between two properties at the same level—including cases in which an instantiation of one emergent property causes another emergent property to be instantiated. Downward causation occurs when a higherlevel property, which may be an emergent property, causes the instantiation of a lower-level property; similarly, upward causation involves the causation of a higher-level property by a lower-level property. I believe that, for the

27 Space, Time, and Deity, vol. 2 (London: Macmillan, 1927), p. 8.


29

emergentist,28 there is good reason to believe that downward causation is fundamental and of crucial importance in understanding causation. For it can be shown that both upward and same-level causation (except same-level causation at the ultimate bottom level, if there is such a level and if there are causal relations at this level) presupposes the possibility of downward causation. Here is an argument designed to show why this is so.29 Suppose that a property M, at a certain level l, causes another property M+, at level l + 1. Assume that M+ emerges, or results, from a property M* at level l (M* therefore is at the same level as M). Now we immediately see a tension in this situation when we ask: “What is responsible for this occurrence of M+? What explains M+’s instantiation on this occasion?” For in this picture there initially are two competing answers: First, M+ is there because, ex hypothesi, M caused it; second, M+ is there because its emergence base M* has been realized. Given its emergence base M*, M+ must of necessity be instantiated, no matter what conditions preceded it; M* alone suffices to guarantee M+’s occurrence on this occasion, and without M*, or an appropriate alternative base, M+ could not have occurred. This apparently puts M’s claim to have caused M+ in jeopardy. I believe that the only coherent description of the situation that respects M’s causal claim is this: M causes M+ by causing its base condition M*. But M’s causation of M* is an instance of same-level causation. This shows that upward causation entails same-level causation; that is, upward causation is possible only if same-level causation is possible. As an example, consider this: physical/mechanical work on a piece of marble (M) causes the marble to become a beautiful piece of sculpture (M+). But the beauty of the sculpture emerges from the physical properties (M*, consisting in shape, color, texture, size, etc.) of the marble piece. Notice how natural, and seemingly unavoidable, it is to say that the physical work on the marble caused the beauty of the marble piece by causing it to have the right physical properties. This of course is an instance of same-level causation.

28 As I have argued elsewhere, this holds for certain positions other than emergentism, e.g., the view that higher properties supervene on lower properties, and the view that higher properties are realized by lower properties. 29 I first presented this argument in “‘Downward Causation’ in Emergentism and Nonreductive Physicalism,” in Emergence or Reduction?

30


Another example: a bee sting causes a sharp pain. But pain emerges from a certain neural condition N (say, C-fiber excitation). I believe that we want to say, and must say, that the bee sting caused the pain by causing N (the firing of C-fibers). This again is same-level causation. An exactly similar argument will show that same-level causation presupposes downward causation. Briefly, this can be shown as follows: Suppose M causes M*, where M and M* are both at level l. But M* itself arises out of a set of properties M at level l 1. When we ponder the question how M* gets to be instantiated on this occasion, again we come to the conclusion that M caused M* to be instantiated on this occasion by causing M, its base condition, to be instantiated. But M’s causation of M is downward causation. This completes the argument. A general principle is implicit in the foregoing considerations, and it is this: To cause any property (except those at the very bottom level) to be instantiated, you must cause the basal conditions from which it arises (either as an emergent or as a resultant). We may call this “the principle of downward causation.”

VI Even the early emergentists were explicit on the importance they attached to downward causation, although of course it is unlikely that they were influenced by anything like the argument of the preceding section. The following statement by C. Lloyd Morgan, for example, is typical: Now what emerges at any given level affords an instance of what I speak of as a new kind of relatedness of which there are no instances at lower levels . . . But when some new kind of relatedness is supervenient (say at the level of life), the way in which the physical events which are involved run their course is different in virtue of its presence—different from what it would have been if life had been absent.30 Compare this with what Roger Sperry says: 30 Emergent Evolution (London: Williams & Norgate, 1927), pp. 15–16. Emphasis added.


31

. . . the conscious subjective properties in our present view are interpreted to have causal potency in regulating the course of brain events; that is, the mental forces or properties exert a regulative control influence in brain physiology.31 Both Morgan and Sperry are saying that life and consciousness, emergent properties out of physicochemical and neural properties respectively, have a causal influence on the flow of events at the lower levels, levels from which they emerge. That of course is downward causation. The appearance of an emergent property signals, for the emergentists, a genuine change, a significant evolutionary step, in the history of the world, and this requires emergent properties to be genuine properties with causal powers. They are supposed to represent novel additions to the ontology of the world, and this could be so only if they bring with them genuinely new causal powers; that is, they must be capable of making novel causal contributions that go beyond the causal powers of the lower-level basal conditions from which they emerge. But how do emergent properties exercise their novel causal powers? How is that possible? According to the argument presented in the preceding section, they can do so only by causally influencing events and phenomena at lower-levels—that is, through downward causation. That was what we called the principle of downward causation. But is downward causation possible? The idea of downward causation has struck some thinkers as incoherent, and it is difficult to deny that there is an air of paradox about it: After all, higher-level properties arise out of lower-level conditions, and without the presence of the latter in suitable configurations, the former could not even be there. So how could these higher-level properties causally influence and alter the conditions from which they arise? Is it coherent to suppose that the presence of X is entirely responsible for the occurrence of Y (so Y’s very existence is totally dependent on X) and yet Y somehow manages to exercise a causal influence on X? I believe a train of thought like this is behind the suspicions surrounding the idea of downward causation. But if downward causation is incoherent, that alone could cause serious damage to emergentism. For the principle of downward causation directly implies that if emergent properties have no downward causal 31 “Mental Phenomena as Causal Determinants in Brain Function,” in Consciousness and the Brain, ed. Gordon G. Globus, Grover Maxwell, and Irwin Savodnik (New York: Plenum Press, 1976), p. 165.

32


powers, they can have no causal powers at all, and this means that emergent phenomena would just turn out to be epiphenomena, a prospect that would have severely distressed Alexander, Morgan, and Sperry. But we need to analyze whether the kind of intuitive argument in the preceding paragraph against downward causation has any real force. For cases in which higher-level entities and their properties prima facie causally influence lower-level entities and their properties seem legion. The celadon vase on my desk has a mass of 1 kilogram. If it is dropped out the window of my second floor office, it will crash on the paved sidewalk, causing myriads of molecules of all sorts to violently fly away in every which way. Even before it hits the ground, it will cut a rapid downward swath, causing all sorts of disturbance among the local air molecules. And these effects are surely micro and lower-level in relation to the fall of an object with a mass of 1 kilogram. Note that we cannot think of this case as one in which the “real” causal process occurs at the microlevel, between the microconstituents of the vase and the air molecules, for the simple reason that no micro-constituent, in fact no proper part, of my celadon vase has a mass of 1 kilogram. There is no question that the vase, in virtue of having this mass, has a set of causal powers that none of its micro-constituents have; the causal powers that this property represents cannot be reduced to the causal powers of micro-constituents of its bearers. Of course, emergentists would not consider mass an emergent property; they would say that the mass of an object is a resultant property, a property that is merely “additive or subtractive.” But this simple example suffices to show that there need not be anything strange or incoherent in the idea of downward causation as such—the idea that complex systems, in virtue of their macrolevel properties, can cause changes at lower microlevels. However, the idea of downward causation advocated by some emergentists is stronger and more complex than what is suggested by our example. Here again is Sperry: The subjective mental phenomena are conceived to influence and to govern the flow of nerve impulse traffic by virtue of their encompassing emergent properties. Individual nerve impulses and other excitatory components of a cerebral activity pattern are simply carried along or shunted this way and that by the prevailing overall dynamics of the whole active process (in principle—just as drops of water are carried


33

along by a local eddy in a stream or the way the molecules and atoms of a wheel are carried along when it rolls down hill, regardless of whether the individual molecules and atoms happen to like it or not).32 Sperry has used these and other similar examples elsewhere; in particular, the rolling wheel seems to have been one of his favorites. What is distinctive about this form of downward causation appears to be this: Some activity or event involving a whole W is a cause of, or has a causal influence on, the events involving its own micro-constituents. We may call this reflexive downward causation, to distinguish it from the more mundane nonreflexive kind, involved in the example of the falling vase above, in which an event involving a whole causes events involving lower-level entities that are not among its constituents. But downward causation must be viewed in the context of the doctrine that emergent properties arise out of their basal conditions (claim 2. in Section V). For Sperry himself recognizes this in his claim that there is also upward determination in this situation. The paragraph quoted above from Sperry continues as follows: Obviously, it also works the other way around, that is, the conscious properties of cerebral patterns are directly dependent on the action of the component neural elements. Thus, a mutual interdependence is recognized between the sustaining physico-chemical processes and the enveloping conscious qualities. The neurophysiology, in other words, controls the mental effects, and the mental properties in turn control the neurophysiology.33 After all, an eddy is there because the individual water molecules constituting it are swirling around in a circular motion in a certain way; in fact, an eddy is nothing but these water molecules moving in this particular pattern. Take away the water molecules, and you have taken away the eddy: there cannot be a ghostly eddy swirling around without any water molecules! Thus, reflexive downward causation is combined with upward determination. When each and every molecule in a puddle of water begins to move in an appropriate way—and only then—will there be an eddy of water. But in

32 “A Modified Concept of Consciousness”, Psychological Review 76 (1969): 532–536. 33 Ibid.

34


spite of this, Sperry says, it remains true that the eddy is moving the molecules around “whether they like it or not.” Thus, reflexive downward causation is combined with upward determination. Schematically, the situation looks like this: a whole, W, has a certain (emergent) property M; W is constituted by parts, a1, . . . , an, and there are properties P1, . . . , Pn respectively of a1, . . . , an and a certain relation R holding for the ais. The following two claims make explicit what Sperry seems to have in mind (I do not want to rule out other possible interpretations of Sperry): (i) Downward causation: W’s having property M causes some aj to have Pj. (ii) Upward determination: Each ai’s having Pi and R holding for the ais together determine W to have M—that is, W’s having M depends wholly on (or is wholly constituted by) the ais having the Pis respectively and being related by R. The question is whether or not it is possible, or coherent, to hold both (i) and (ii).

VII As I said, downward causation as such presents us with no special problems; however, what Sperry wants (also there is a hint of this in the quotation from Lloyd Morgan above) is the reflexive variety of downward causation. But how is it possible for the whole to causally affect its constituent parts on which its very existence and nature depend? If causation or determination is transitive, doesn’t this ultimately imply a kind of self-causation, or self-determination—an apparent absurdity? It seems to me that there is reason to worry about the coherence of the whole idea. Let us see if it is possible to make reflexive downward causation intelligible. To sharpen the issues we should distinguish two cases: Case 1. At a certain time t, a whole, W, has emergent property M, where M emerges from the following configuration of conditions: W has a complete decomposition into parts a1, . . . , an; each ai has property Pi; and relation R holds for the sequence a1, . . . , an. For some aj, W’s having M at t causes aj to have Pj at t.


35

Note that the time t is fixed throughout, and both the downward causation and upward emergence (or determination) hold for states or conditions occurring at the very same time. We may, therefore, call this “synchronic reflexive downward causation.”34 A whole has a certain emergent property, M, at a given time, t, and the fact that this property emerges at t is dependent on its having a certain micro-configuration at t, and this includes a given constituent of it, aj, having Pj at t. That is, unless aj had Pj at t, W could not have had its emergent property M at t. Given this, it makes one feel uncomfortable to be told also that aj is caused to have Pj at that very time, t, by the whole’s having M at t. But what exactly is the source of this metaphysical discomfort? Why does this picture seem in some way circular and incoherent? Moreover, what is it about causal circularity that makes it unacceptable? One possible explanation, something I find plausible myself, is that we tacitly subscribe to a metaphysical principle like the following: For an object, x, to exercise, at time t, the causal/determinative powers it has in virtue of having property P, x must already possess P at t. When x is caused to acquire P at t, it does not already possess P at t and is not capable of exercising the causal/determinative powers inherent in P. If a name is wanted, we may call this “the causal-power actuality principle.” The reader will have noticed that this principle has been stated in terms of an object “acquiring” property P at a time. In Case 1 above, we said that the whole, W, causes one of its proper parts, aj, to “have” P. If there is real downward causation, from W’s having M to aj’s having Pj, this “having” must be understood as “acquiring.” For if aj already has Pj at t, what role can W’s having M at t play in causing it to have Pj at t? Obviously, none. In any case, it is now easy to see the incoherence involved in Case 1: the assumption that W’s having M at t causes aj to have Pj at t implies, together with the causal-power actuality principle that aj does not already have Pj at t. This means, again via the causal-power actuality principle, that aj cannot, at t, exercise the causal/determinative power it has in virtue of having Pj,

34 This case, therefore, involves the controversial idea of synchronic causation (where a cause and its effect occur at the same time). However, this is a general metaphysical issue, and in the present context it will be unproductive to focus on this aspect of the situation.

36


which in turn implies that the assumed emergence base of W’s having M at t has vanished and W cannot have M at t. Case 1, therefore, collapses. If you are willing to reject the causal-power actuality principle and live with causal circularity (perhaps even celebrate it in the name of “mutual causal interdependence”), then Case 1 could serve as a model of downward causation for you. Speaking for myself, I think there is a good deal of plausibility in the principle that says that for properties to exercise their causal/determinative powers they must actually be possessed by objects at the time; it cannot be that the objects are in the process of acquiring them at that time. So let’s try another model. Case 2. As before, W has emergent property M at t, and aj has Pj at t. We now consider the causal effect of W’s having M at t on aj at a later time t + ˜t. Suppose, then, that W’s having M at t causes aj to have Q at t + ˜t. This, therefore, is a case of diachronic reflexive downward causation. It is still reflexive in that a whole causes one of its micro-constituents to change in a certain way. Notice, however, that the mysteriousness of causal reflexivity seems to have vanished. The reason is obvious: the time delay between the putative cause and effect removes the potential circularity, and the causalpower actuality principle does not apply. W’s having M at t causes aj to have Q at t + ˜t. But aj’s having Q at t + ˜t is not part of the basal conditions out of which M emerges in W at t; so there can be no problem of circular reciprocal causation/determination. This becomes particularly clear if we consider the four-dimensional (or “time slice”) view of persisting things. On this view, W’s having M at t turns out to be W at t having M—that is, the time slice of W at t having M. Let us use “[x, t]” to denote the time slice of x at t (if t is an instant, [x, t] is a temporal cross section). Diachronic downward causation, then, comes to this: [W, t] having M causes [aj, t + ˜t] to have Q, where, of course, t < t + ˜t. The point to notice is that [aj, t + ˜t] is not a constituent of [W, t], and this gets rid of the hint of reflexivity present in Case 1. Examples falling under Case 2 are everywhere. I fall from the ladder and break my arm. I walk to the kitchen for a drink of water and ten seconds later, all my limbs and organs have been displaced from my study to the kitchen. Sperry’s bird flies into the blue yonder, and all of the bird’s cells and molecules, too, have gone yonder. It doesn’t seem to me that these cases


37

present us with any special mysteries rooted in self-reflexivity, or that they show emergent causation to be something special and unique. For consider Sperry’s bird: for simplicity, think of the bird’s five constituent parts, its head, torso, two wings, and the tail. For the bird to move from point p1 to point p2 is for its five parts (together, undetached) to move from p1 to p2. The whole bird is at p1 at t1 and moving in a certain direction, and this causes, let us suppose, its tail to be at p2 at t2. There is nothing mysterious or incoherent about this. The cause—the bird’s being at p1 at t1 and moving in a certain way—includes its tail’s being at p1 at t1 and moving in a certain way. But that’s all right: we expect an object’s state at a given time to be an important causal factor for its state a short time later. And it is clear that Sperry’s other examples, such as the water eddy and the rolling wheel, can be similarly accommodated. We must conclude then that of the two types of reflexive downward causation, the diachronic variety poses no special problems but perhaps for that reason it is rather unremarkable as a type of causation, but that the synchronic kind is problematic and it is doubtful that it can be given a coherent sense. This may be due to its violation of what I called the causalpower actuality principle, but apart from any recondite metaphysical principle that might be involved, one cannot escape the uneasy feeling that there is something circular and incoherent about this variety of downward causation.

VIII Emergentists like C. Lloyd Morgan will likely point out that the Sperrystyle cases do not really involve downward causation by emergent properties, since the motion of the bird as a whole is the same kind of event as the motion of its constituent parts. The properties implicated in causal relations in these cases are one and the same, namely motion, and this shows that these cases simply are not cases of emergent causation, whether downward or upward. (The same will be said about the example of the falling celadon vase.) It would seem, then, that contrary to what Sperry seems to suggest, emergent downward causation should not simply be identified with

38


causation from properties of the whole to properties of its own parts, that is, reflexive downward causation. One reason that downward causation is thought interesting and important is that mental-to-physical causation is commonly supposed to be a special case of it, the mental occupying a higher emergent level relative to the physical level. So let us turn to mind-body causation. Here again we may consider two varieties, synchronic reflexive downward causation and its diachronic counterpart. Can my experience of pain at a given time causally influence its basal neural process (C-fiber excitation, say) at the very same time? Here we encounter exactly the same difficulties that we saw in Sperry’s examples of the water eddy and the like (taken as cases of synchronic downward causation), and I do not believe that classical emergentists, like Alexander, Morgan, and C.D. Broad, would necessarily have insisted on it. Nor do I see why Sperry himself, as an emergentist, should need it; it isn’t at all clear that Sperry’s overall position on the mind-body relation requires a commitment to this dubious variety of emergent causation. This leaves diachronic downward causation as the only player on the scene—up to this point, at any rate. One might say that this is all that the emergentists need—the diachronic causal influence of emergent phenomena on lower-level phenomena. But the problem is that even this apparently unproblematic variety of downward causation is beset with difficulties. On my view, the difficulties boil down to a single argument to be sketched below. The critical question that motivates the argument is this: If an emergent, M, emerges from basal condition P, why doesn’t P displace M as a cause of any putative effect of M? Why can’t P do all the work in explaining why any alleged effect of M occurred?35 As you may recall, I earlier argued that any upward causation or same-level causation of effect M* by cause M presupposes M’s causation of M*’s lower-level base, P* (it is supposed that M* is a higher-level property with a lower-level base; M* may or may not be an emergent property). But if this is a case of downward emergent causation, M is a higher-level property, and as such it 35 I raised this question earlier in “‘Downward Causation’ in Emergentism and Nonreductive Physicalism”, in Emergence or Reduction?. The argument can be generalized to the supervenience and realization views of the mind-body relation. For more details see my “The Nonreductivist’s Troubles with Mental Causation” (reprinted in Supervenience and Mind) and Mind in a Physical World. See also Timothy O’Connor, “Emergent Properties”, American Philosophical Quarterly 31 (1994): 91–104, for an attempt to counter the argument.


39

must have an emergence base, P. Now we are faced with P’s threat to preempt M’s status as a cause of P* (and hence of M*). For if causation is understood as nomological (law-based) sufficiency, P, as M’s emergence base, is nomologically sufficient for it, and M, as P*’s cause, is nomologically sufficient for P*. Hence, P is nomologically sufficient for P* and hence qualifies as its cause. The same conclusion follows if causation is understood in terms of counterfactuals—roughly, as a condition without which the effect would not have occurred. Moreover, it is not possible to view the situation as involving a causal chain from P to P* with M as an intermediate causal link. The reason is that the emergence relation from P to M cannot properly be viewed as causal.36 This appears to make the emergent property M otiose and dispensable as a cause of P*; it seems that we can explain the occurrence of P* simply in terms of P, without invoking M at all. If M is to be retained as a cause of P*, or of M*, a positive argument has to be provided, and we have yet to see one. In my opinion, this simple argument has not so far been overcome by an effective counter-argument. If higher-level property M can be reduced to its lower-level base, M’s causal status can be restored. As may be recalled from our earlier discussion, however, if M is emergent, this is precisely what cannot be done: emergent properties, by definition, are not reducible to their lower-level bases. The conclusion, therefore, isn’t encouraging to emergentists: If emergent properties exist, they are causally, and hence explanatorily, inert and therefore largely useless for the purposes of causal/explanatory theories. If these considerations are correct, higher-level properties can serve as causes in downward causal relations only if they are reducible to lower-level properties.37 The paradox is that if they are so reducible, they are not really

36 C. Lloyd Morgan explicitly denies that emergence is a form of causation, in Emergent Evolution, p. 28. 37 Here I must enter some caveats. As the reader may recall, I earlier said that there is no special problem of downward causation, citing such examples as my celadon crashing on the pavement of the sidewalk. Cases like this are not the cases of downward causation that most emergentists have in mind, for like Sperry’s example of the flying bird they don’t seem to involve genuine “higher-level” properties. In general, complex systems obviously can bring new causal powers into the world, powers that cannot be identified with causal powers of more basic, simpler systems. Among them are the causal powers of microstructural, or micro-based, properties of a complex system. Note that these properties are not themselves emergent properties; rather, they form the basal conditions from which further properties emerge (for example, that consciousness is not itself a microstructural property of an organism, although it may emerge from one). If all this sounds too complicated, you could regard the argument in the text to be restricted to consciousness and other standard examples of emergent properties.

40


“higher-level” any longer. If they are reducible to properties at level l, they, too, must belong to l. Does this make the idea of downward causation useless? Not necessarily. For example, we may try to salvage downward causation by giving it a conceptual interpretation. That is, we interpret the hierarchical levels as levels of concepts and descriptions, or levels within our representational apparatus, rather than levels of properties and phenomena in the world. We can then speak of downward causation when a cause is described in terms of higher-level concepts, or in a higher-level language, higher in relation to the concepts in which its effect is represented. On this approach, then, the same cause may be representable in lower-level concepts and languages as well, and a single causal relation would be describable in different languages. The conceptual approach may not save real downward causation, and it brings with it a host of new questions; however, it may be a good enough way of saving downward causal explanation, and perhaps that is all we need to, or should, care about.

2 The Layered World: Metaphysical Considerations I Many familiar philosophical problems, notably in philosophy of mind, metaphysics, and philosophy of science, are apt to be formulated in terms of “levels.”1 One of the prominent issues in recent philosophy of mind and of science has centered on the question how we should understand the relationship between the “higher-level” properties and theories of the special sciences and the properties and theories at the “lower levels”— more specifically, whether the former are reducible to the latter or form autonomous domains irreducible to the more basic sciences. A related question that has lately received much attention is the problem of “downward” causation—whether and how “higher-level” events and phenomena, like belief, desire, and consciousness, can project their causal powers “downward” and alter the course of events at “lower levels,” like those involving limbs, organs, and cells. Philosophers have also worried about how macrolevel laws and explanations, in “higher-level” sciences, can be grounded in, or be reduced to, underlying microlevel laws and processes. And there is the long-standing problem of consciousness which has agitated scientists and philosophers alike: how events and phenomena at the biological/computational level can give rise to thought and consciousness, phenomena at what seems like an entirely different level. Talk of levels is An early version of this paper was presented at the Supervenience, Causality, Mind, and Action Conference, Aix-en-Provence, May, 1999. Pascal Engel was the commentator. I thank Nicholas White who made the actual presentation on my behalf. 1 Terms like “order” and “tier” have also been used; however, “level” seems to have taken over and is now the favored expression.

42

THE LAYERED WORLD

just as common for areas outside psychological phenomena. As some worry about how the mental level “emerges” out of biological processes, so others worry, or have worried, about how “vital” teleological phenomena can arise out of the processes at the physicochemical, or “mechanistic,” levels that only involve inert matter. The idiom of “level” is not limited to philosophy. It appears to have thoroughly permeated primary scientific literature as well. David Marr, whose work on vision has been very much a topic of philosophical interest, famously distinguished three levels at which the mechanism of vision could be analyzed; these are the computational level (“What does the system do?”, “What problems does it solve?”), the algorithmic level (“How does the system do it?”), and the implementational level (“What is the underlying physical mechanism?”). Zenon Pylyshyn begins chapter 2 of his book, Computation and Cognition,2 with this sentence, “In this chapter I attempt to elaborate on the idea of levels of explanatory principles,”3 and goes on to discuss the relationship between the representational level and the functional level. Expressions like “levels of description,” “levels of analysis,” “levels of explanation,” “levels of organization,” and “levels of complexity,” are routinely encountered—in fact, difficult to avoid—in scientific writings in various areas, including, especially, the cognitive sciences, the biological sciences, computer science, and systems theory. One gets the impression that those who are engaged in one or another of the so-called special sciences think of themselves as situated at one or another of these levels (or perhaps straddling a specific set of levels), levels higher than the bottom level. And it is assumed that physics is in charge of this bottom base level, and evidently that is one main reason we think of physics as our “basic” science. All this talk of levels may turn out to be only a figure of speech, a harmless but suggestive metaphor, which can easily be paraphrased away. But what often seems implicit is a certain overarching ontological picture of the world according to which the entities of the natural world are organized in an ascending hierarchy of levels, from lower to higher, from simpler to more complex. I propose that we begin by taking the levels talk and its attendant metaphysics seriously, more or less at face value, and try to make sense of it—or as much sense as we can. As we will see, it isn’t easy to come up 2 (2nd edn., Cambridge, Mass.: MIT Press, 1985). 3 Ibid 23.

THE LAYERED WORLD

43

with a neat and satisfying general model of levels that will serve useful philosophical purposes. But I think it is important, and also instructive, to be clear about what is involved in the levels picture and appreciate both its possibilities and limitations. The idea of “higher” and “lower” is essential to the levels picture: the levels are ordered from lower to higher, and as noted it is often thought that there is a bottom level, a level than which there is no lower. Unless one could meaningfully speak of “higher” and “lower,” there would be little point in talking about levels. Why not just talk about domains or spaces or fields? If you are a Cartesian substance dualist, you wouldn’t try to fit your basic ontology into a scheme of levels: the domain of mental substances is neither higher nor lower than that of physical substances, although of course within each there could be levels. Those who use the levels language today are, we may assume, concerned exclusively with the physical domain, and the shared imagery evoked by levels talk is a picture, somewhat fuzzy and unarticulated, of the physical world neatly stratified into a structure of discrete levels, with a bottom level of basic particles—perhaps leptons and quarks, or whatever our best physics tells us are the fundamental constituents of matter—and the rest as forming a vertically ordered system of levels each resting on the one below and all ultimately resting on the base level of microparticles. I am sure that to most of you this is a familiar picture; perhaps some of you have made use of this kind of scheme in your philosophical work; some of you may even have contributed to filling out and promoting a layered picture like this. Although the layered model, as we may call it, has formed the backdrop of much of our recent philosophical thinking in metaphysics and philosophy of mind, there has been little conscious effort to clarify it and make explicit the metaphysics that underlies it. I believe it is clear that the model gives rise to a host of questions. The foremost question is this: Exactly how are levels distinguished or generated, and what makes a given level higher or lower than another? Or, what comes to the same, what makes a given entity or phenomenon belong to a level higher than the level to which another entity or phenomenon belongs? That is, what is the ordering relation involved here? There are many other questions. To mention a few: Is there a fixed system of levels to which all things and phenomena of this world belong, or is this a matter of conventional classification dependent on a context of inquiry? Is it the case that for each object, or phenomenon, there is some unique level to which it belongs? It is an important part of the levels picture that properties,

44

THE LAYERED WORLD

too, form a hierarchy of levels, with mental/cognitive properties at a higher level than biological properties, which in turn are thought to be higher relative to physicochemical properties. But what makes a given property “higher” or “lower” than another in this ordering? And what is the relationship between the hierarchy of properties and the hierarchy of the objects? The discussion to follow will touch on some—not all—of these questions. Specifically, we will focus on the emergentist model of C. Lloyd Morgan and the model of Oppenheim and Putnam inspired by a vision of global microreductionism. As you will see, my discussion will be far from complete or conclusive, raising more questions than answering them. However, I hope to have initiated a serious discussion of the layered model, a picture that has so far managed to stay hidden in the background, away from scrutiny, in spite of the pervasive influence it has exercised on the way we think about many philosophical issues. But before getting on with the specifics, I would like to warn you that we should be prepared for a deflationary outcome: we might decide in the end that to speak of levels in regard to things, properties, and phenomena of the world is misleading and perhaps even not entirely coherent, and that it makes better sense to apply the levels picture to concepts, descriptions, languages, and the like—that is, to our representations of the world rather than to the world itself. Or we might come to the conclusion that it makes little sense, and serves no important philosophical or scientific purpose, to try to construct a global, all-inclusive model of levels in which every object and phenomenon of the world has a unique designated place. And we might conclude that a more useful strategy is to take a local approach that applies a levels picture to describe and analyze restricted domains in the context of a specific scientific/philosophical inquiry.

II I don’t know when the layered picture made its first appearance.4 The earliest modern manifestation of it that I know was in the writings of the 4 It is probably easy enough to read this picture into some of the ancient philosophical systems, e.g., Aristotle’s. But the ancient conceptions that might be found are not likely to be directly relevant to contemporary concerns.

THE LAYERED WORLD

45

British emergentists early in the 20th century. The layered model fits the emergentist doctrines to order; for the very idea of emergence encourages us to view what emerges as being on a level higher than the underlying processes from which it emerges. In Emergent Evolution, C. Lloyd Morgan writes “Let there be three successive levels of natural events, A, B, and C. Let there be in B a kind of relation which is not present in A; and in C a kind of relation, not yet present in B or in A.”5 For Morgan these are levels of emergence: level A is the lowest of the three and at level B there may emerge novel characteristics not predictable from knowledge of things and phenomena at A; and again at level C new properties may emerge that are not deducible from information solely about levels A and B. Unsurprisingly, an up-down vertical imagery comes naturally to Morgan; a few pages later he says, “Under naturalistic treatment, however, the emergence, in all its ascending grades, is loyally accepted, on evidence, with natural piety.”6 The term “ascending” recurs throughout Morgan’s presentation, and, helpfully, he offers a diagram depicting what he calls “a pyramidal scheme,” with space-time and matter anchoring the whole structure at the base level, life at a point about halfway up, and mind near the vertex, like this:

Mind

Life

Matter

Space-Time

5 C. Lloyd Morgan, Emergent Evolution (London: Williams and Norgate, 1923), pp. 5–6. 6 Ibid. 8. Emphasis added.

THE LAYERED WORLD

46

Morgan explains: Such a diagram . . . is, so to speak, a synoptic expression, or composite graph, of a vast multitude of individual pyramids—atom-pyramids near the base, molecules a little higher up, yet higher, ‘things’ (e.g. crystals), higher still, plants (in which mind is not yet emergent), then animals (with consciousness), and, near the top, our human selves. Classify how you will; but let every individual entity have its appropriate place in the synoptic pyramid. It is intended to embrace all natural entities from atoms . . . upwards.7 Note what Morgan says: each and every entity in the world is to belong to one unique level, “its appropriate place,” in this all-embracing system (notice, though, that Morgan implies that the levels represented in the diagram are not all the levels that there are, and that intermediate levels can be interpolated without end). We may take it then that the scheme Morgan is describing is intended as a sketchy but complete picture of the world. That is, he is presenting to us a comprehensive ontology of the entire natural order. So Morgan’s hierarchical system has the following salient levels: 5. 4. 3. 2. 1.

Human beings Animals Plants Molecules Atoms

Compare this with a more recent statement of the layered model by Paul Oppenheim and Hilary Putnam in 1958,8 in the only explicit discussion of the levels picture I know of in contemporary analytic philosophy. The six levels distinguished by Oppenheim and Putnam are these: 6. Social groups 5. Multicellular living things

7 C. Lloyd Morgan, Emergent Evolution. 11. The diagram is Morgan’s depiction of Samuel Alexander’s grand vision of emergent evolution. The baseline is supposed to represent “space-time”; Morgan has “Deity” at the apex of the pyramid, but he doesn’t seem to take that aspect of Alexander’s scheme seriously. 8 Paul Oppenheim and Hilary Putnam, “Unity of Science as a Working Hypothesis,” Minnesota Studies in the Philosophy of Science, vol. 2 (Minneapolis: University of Minnesota Press, 1958).

THE LAYERED WORLD

4. 3. 2. 1.

47

Cells Molecules Atoms Elementary particles

Obviously the Oppenheim-Putnam scheme is informed by the more developed science of the mid-20th century, but one is struck by a remarkable similarity between the two hierarchies. In fact, John Locke or Robert Boyle might have come up with something like Morgan’s scheme; the scientific knowledge available then, or at least during the 19th century, would have supported a hierarchical system like Morgan’s (except perhaps for the distinction between molecules and atoms). But, as we will shortly see, the Morgan hierarchy is structurally quite different from the OppenheimPutnam hierarchy. Oppenheim and Putnam refer to the six levels as “reductive levels,” the terminology revealing their commitment to a microreductionist program. However, the adoption of the layered scheme itself should be independent of reductionism—or so we should expect; after all, Morgan was an antireductionist emergentist. A scheme of this sort seems necessary for stating the general issue of reductionism; one might think that only against such a backdrop does a global reductionist/antireductionist debate make sense. Like Morgan’s, the Oppenheim-Putnam model is clearly intended to encompass the whole natural order (with a caveat to be mentioned below). If there is a single fixed hierarchical system for all of reality, of the kind proposed by Morgan, and Oppenheim and Putnam, then it must be the case that for any two arbitrary objects, one is higher or lower than the other, or else they are on the same level. And the same should be true of properties as well. But how do we determine whether an object is higher or lower than another? What criteria should govern such comparisons? Our judgments as to higher and lower should be motivated and principled, and this means not only that the criteria that regulate such judgments must be clear and metaphysically coherent but also that the ordering of objects and properties they generate must serve some useful purposes. Giving robust and intelligible meaning to the talk of higher and lower is perhaps the most pressing problem facing those advocating a layered picture of the world.

48

THE LAYERED WORLD

III Let us first look at Morgan. On the notion of “higher” and “lower,” he says this: But we must now ask: Higher in what sense? . . . When two or more kinds of events, such as I spoke of before as A, B and C, co-exist on one complex system in such wise that the C kind involves the co-existence of B, and B in like manner involves A, whereas the A-kind does not involve the co-existence of B, nor B that of C, we may speak of C, as, in this sense, higher than B, and B than A. Thus, for emergent evolution, conscious events at level C (mind) involve specific physiological events at level B (life), and these involve specific physicochemical events at level A (matter). No mind without life; and no life without a physical basis.9 As Morgan’s reference to “kinds of events” suggests, this appears better suited for characterizing an ordering of properties and kinds than individual entities. That is, as his examples of mind, life, and physical basis indicate, Morgan’s A, B, and C are best understood as standing for properties and kinds. But this is not necessarily a shortcoming of Morgan’s scheme; for it seems clear that an ordering of kinds and properties can generate an ordering of individual objects in a natural way. An object, we may say, is higher than another just in case the kind to which the first belongs is higher than the kind to which the second belongs. Or, if we wish to take cognizance of the possibility that objects can fall under multiple diverse kinds, we may say that an object is higher than another just in case the highest kind under which the first falls is higher than the highest kind under which the second falls. The same idea obviously can be put in terms of properties. As I take it, what Morgan is saying is something like this: Mind, or mentality, is higher than physiological/biological properties in that the former “involves” the latter—that is, mind exists only if certain “specific” configurations of the latter are present—whereas physiological/biological properties may be present without mentality being present. And similarly for physiological/biological properties in relation to physicochemical properties. Morgan would have added that the history of evolution shows this, namely that there was a time when physicochemical processes existed but no life existed, and a time when 9 Emergent Evolution, p. 15.

THE LAYERED WORLD

49

biological phenomena were present but no minds existed. It is only when appropriately complex and propitious physicochemical conditions came together that life made its appearance. However, physicochemical conditions could exist—and actually did exist—without there being life. Similarly for mental phenomena in relation to biological phenomena. It will be worthwhile to give this idea some thought. The key word in Morgan’s statement is “involve.” To capture what Morgan appears to have in mind, let us make explicit the rough reading given in the preceding paragraph of what makes biological properties higher than physicochemical properties: For any biological property to be instantiated in a system, the system must instantiate a special configuration of physicochemical properties, whereas any physicochemical property can be instantiated in a system without that system manifesting any biological property. More generally: Set M of properties is higher than set N of properties just in case whenever something instantiates a property in M, it must also instantiate some subset of N; but any property in N can be instantiated by something without it instantiating any property in M. In this sense, then, we might say that properties in M “involve” those in N, but not conversely. As it stands, there is a serious problem with this criterion. The problem concerns individuation of properties: if we allow conjunction as a property forming operation—more specifically, if we consider the set of biological properties to be closed under conjunction (so that if P and Q are biological properties, so is their conjunction, P & Q)—the criterion no longer certifies mental properties to be higher than biological properties. For let B be a specific configuration—that is, conjunction—of biological properties from which a certain mental property, m, emerges. By the closure of biological properties under conjunction, B itself is a biological property. This means, on the standard construal of emergence, B is sufficient, necessarily, for the instantiation of m. If so, it is not possible for B, a member of the set of biological properties, to be instantiated by a system without this system also instantiating mental property m. It will not be a good idea to get around this problem by denying that biological properties are closed under conjunction. The closure principle, I believe, is unexceptionable; in fact, the conjunctive closure principle is highly plausible for any “natural” family of properties, such as the mental, the physical, and the moral/normative.

50

THE LAYERED WORLD

Here is a more promising idea that is not far removed from Morgan’s. Take the standard idea of supervenience (or “strong” supervenience10): M (strongly) supervenes on N iff for any property m in M, if something x instantiates m, there is a property, n, in N such that x instantiates n, and, necessarily, if anything instantiates n, it instantiates m. I believe we can capture much of what Morgan appears to have had in mind in the following: M is higher than N iff M supervenes on N but N does not supervene on M. Thus, mental properties supervene on biological/physical properties but not conversely;11 and biological properties supervene on physicochemical properties but not conversely. As is well known, the idea of supervenience is usually taken to imply a relation of dependence or determination: That which supervenes is dependent on, and is determined by, the base on which it supervenes. This idea seems to fit quite well with the idea of emergence. I believe that emergentists in general will accept the thesis that what emerges supervenes on the “basal” conditions from which it emerges; I do not believe it would be acceptable to deny supervenience—that is, I don’t think the emergentist can say that mentality emerges from a given set of basal conditions and yet fails to emerge when the same basal conditions are repeated. Moreover, as Morgan’s use of “involve” clearly indicates, emergent properties require the presence of appropriate basal conditions, and this is reflected in the definition of strong supervenience. We should note that the relation of higher than, on this definition, has the desired properties of irreflexivity, asymmetry, and transitivity. Obviously, the relation is irreflexive and asymmetric. As for transitivity, suppose M is higher than N, and N is higher than L. We want M to come out higher than L—that is, (i) M supervenes on L, but (ii) L does not supervene on M. (i) follows from the transitivity of supervenience. As for (ii), suppose that it’s false—that is, suppose that L supervenes on M. But by assumption M

10 See, e.g., “Concepts of Supervenience,” reprinted in my Supervenience and Mind (Cambridge: Cambridge University Press, 1993). 11 For some philosophers, the acceptability of this supervenience claim will depend on the modal force attached to supervenience (that is, the interpretation of “necessarily” in the stated definition of supervenience). I believe that emergentists would have accepted mind-body supervenience in a very strong sense, with necessity interpreted as metaphysical necessity. Some philosophers would accept supervenience only in a weaker sense, with necessity understood as nomological, or “natural,” necessity.

THE LAYERED WORLD

51

supervenes on N. So, by the transitivity of supervenience, L supervenes on N, contradicting the assumption that N is higher than L, whence follows (ii). On this understanding of higher and lower, Morgan’s hierarchy turns out to be not quite right. Clearly, he wanted animals to rank higher than plants—that is, zoological properties to be higher than botanical properties. But neither kind supervenes on the other, and they are not comparable in respect of higher and lower. The same comment applies to minds and plants. (And minds would not supervene on biology—on the stated definition of supervenience—if there should be nonbiological systems, e.g., electromechanical robots, with mentality.) This means that the relationship we have defined for higher and lower is not connected: it is not the case that for any two sets of properties one is higher than the other. In general, if two property sets are such that neither supervenes on the other, then, on the suggested definition, neither is higher than the other, and the two are not comparable in respect of higher and lower. One might want to say that two property sets are on the same level if neither is higher than the other.12 But this would be ill motivated: it seems pointless to say, for example, that colors and shapes are “on the same level” just because colors don’t supervene on shapes nor conversely. However, the approach I am suggesting can yield a limited notion of “the same level”: we may say that if M supervenes on N and N supervenes on M, M and N are on the same level.13 The idea is that since there is supervenience, the two property sets should be comparable in regard to higher and lower, and that since the supervenience is mutual, neither should be higher than the other. No natural examples of M and N that are so related come to mind, and it is rather doubtful that this idea of “same level” will turn out to be very useful. If M and N mutually supervene, where the modal force of the supervenience relation is logical or metaphysical necessity, that would mean that M and N are necessarily equivalent, and this may well justify our identifying them. No wonder that they are “on the same level”! I believe that, on balance, the criterion of higher and lower formulated in terms of supervenience seems about the best that can be had. If this is right, 12 Similar results follow from Morgan’s criterion of higher and lower in terms of “involving”—at least, on my suggested reading of it. 13 Symmetric supervenience of this sort would of course be impossible if a robust relation of dependence or determination is built into the concept of supervenience.

52

THE LAYERED WORLD

we have to accept the fact that properties and kinds are not always comparable to each other in terms of higher and lower. And if individual objects are ranked in terms of the properties they have, as I have suggested, the same outcome has to be expected for them as well. When we consider all the myriads of properties and objects, this outcome should have been expected all along. Let us look at what else Morgan says. Here is another passage: Each higher entity in the ascending series is an emergent ‘complex’ of many entities of lower grades, within which a new kind of relatedness gives integral unity.14 Morgan is saying, as I take it, that a “higher” entity in the hierarchy is a composite thing made up of entities of “lower grades,” and moreover, it must be an organized, structured whole (showing “relatedness” of its components) with “integral unity.” It isn’t exactly clear what integral unity is supposed to be; but it is clear enough that a heap of apples or a mound of sand will not count as having integral unity, although each individual apple surely would and each particle of sand probably does. Also excluded would be “arbitrary” mereological sums like the sum of my left foot and my computer mouse. I think we can take Morgan’s entities to be, at least roughly, those that fall under significant kind terms in science and ordinary discourse. So two points seem to follow from this: first, that a higher entity is made up of parts that belong to lower levels, and, second, the higher entity is something with integral unity, an organized structure. But it’s obvious that these cannot be considered as a sufficient characterization of higher and lower. For one thing, not every “complex” of “lowergrade” entities will be a higher entity; there is no useful sense in which a slab of marble is a higher entity than the smaller marble parts that make it up. For another, Morgan’s statement taken as a characterization of higher and lower begs the question; it obviously will not do to say that something is higher than its parts if these parts are of a “lower” grade. We may also note the possibility that a whole could be lower in the intended hierarchy than the parts of which it is composed. Consider Ned Block’s Chinese nation organized and operating as a computing

14 Emergent Evolution, p. 13.

THE LAYERED WORLD

53

machine.15 Now just what mentality the Chinese nation so organized might possess is an arguable question; but it seems clear enough that the computing machine constituted by the Chinese nation cannot be credited with consciousness. That is, it is not at all obvious that the whole in this case should rank higher than the parts of which it is composed (perhaps it should be ranked lower). There is one more factor mentioned by Morgan: he says that the higher entity must be an “emergent complex” of entities at lower levels, and this presumably means that it has some property, an “emergent” property, or a new “relatedness,” not had by any of its proper parts. But it would not be enough just to say that the higher entity must have some property not had by its parts; for this would qualify the slab of marble as a higher-level entity in relation to its parts. So the distinctive property that qualifies a whole as a higher entity must be, for Morgan, a special property, a property that is “emergent” by his light. We see, therefore, that this characterization of “higher” and “lower” presupposes the concept of emergence: P is a higher property than Q just in case P emerges from Q (or from a set of properties including Q). This of course requires a prior understanding of emergence, and if the doctrine of emergentism is false and there are no emergent properties, the kind of hierarchy of levels we are looking for cannot even get started.16 Before moving on, let us note the following point concerning Morgan’s hierarchical scheme. Although the part-whole relation seems to play an important role in Morgan’s thinking, this relation cannot be what generates the hierarchy. It may be true that a higher entity is made up of things at the lower levels, higher and lower cannot be explained merely in terms of the part-whole relation. According to Morgan, human beings, with their rationality, are higher than animals, and animals, with sentience, are higher than plants. However, human beings obviously are not composed of animals as parts; nor are animals composed of plants as parts. All of them are made up of cells, molecules, and atoms, but in terms of the mereological relations they are not comparable to each other in respect of higher and lower. More importantly, even on Morgan’s own criterion considered above, for ranking properties as higher or lower, formulated in terms of properties “involving” 15 Ned Block, “Troubles with Functionalism” in Readings in Philosophy of Psychology, vol. 1, ed. Block (Cambridge, Mass.: Harvard University Press, 1980). 16 This is not intended as a criticism of Morgan. He would have said, I believe, that it is precisely the phenomenon of emergence that generates his pyramidal structure.

54

THE LAYERED WORLD

other properties, it is clear that the characteristic properties of animals as a group do not come out higher than the properties characteristic of plants. Presumably, zoological properties must come out higher than botanical properties if animals are to rank higher than plant life in a general hierarchy of levels of the sort sought by Morgan. In any case, what is clear is that in Morgan’s hierarchy it is not the case that the properties at a given level always emerge from those of the next lower level. Another aspect of this is that Morgan’s levels of emergence are not connected with respect to the relation of higher than, lower than, or at the same level as. Also the question raised earlier remains concerning the relative ranking of complex and sophisticated inanimate systems, like computers and robots, vis-a`-vis biological systems. We will return to this issue later.

IV Let us now turn to the model proposed by Oppenheim and Putnam. They explicitly invoke a mereological criterion to generate their hierarchy: each object at level n, above level 1, must have a complete decomposition into proper parts all of which are entities at the next lower level. Thus, every molecule must completely decompose into atoms and nothing else, and each cell must decompose completely into molecules and nothing else. As the wording suggests, the levels are ordered in such a way that it makes sense to speak of “the” next higher, or lower, level. This reflects their stipulation that there be a finite number of levels with a unique bottom level. The six levels in their system are selected in a way that is “natural” from the scientific point of view (whatever that may mean)—also natural in that each chosen level, except the bottom one, is likely to reduce to the next lower level in a global program of microphysical reduction. It is interesting to note that despite the fact that Morgan was selecting the levels with emergence in mind while Oppenheim and Putnam were doing so with reducibility as a guide, the models are quite similar in the actual levels chosen. It seems clear that both saw roughly the same significant divisions in nature, but where one saw emergence the other saw reducibility. But the similarity is only skin-deep. As we saw, on Morgan’s hierarchy, entities at a given level do not always decompose into those at the next lower level; in

THE LAYERED WORLD

55

fact, they need not have as parts any object at the level just below it, as witness the case of animals and plants, or humans and animals. This shows a basic difference between the two hierarchies: the Oppenheim-Putnam hierarchy is one that is based on the mereological relation of part-whole, and the Morgan hierarchy is not, at least not in a straightforward way. Oppenheim and Putnam further stipulate that, at each level n, all aggregates of things at n also belong to n.17 As they note, this condition in effect makes the bottom level of microparticles include all higher levels and all things that exist; the lowest level is the universal domain of physics, and this stipulation, they point out, reflects the view that the basic laws of physics apply to everything everywhere, with no exceptions allowed. In general, the stipulation, in conjunction with the earlier stipulation that each object must fully decompose into objects of the next lower level, has the consequence that any level includes objects belonging to all higher levels. Given this fact, how does the asymmetry of higher and lower arise for the Oppenheim-Putnam hierarchy? Apparently from the following fact: every object belonging to level n decomposes fully into parts that belong to level n1, whereas there are objects at level n1 that do not have things at level n as proper parts.18 Every object at the level of cells—that is, every cell and every object wholly made of cells—decomposes into molecules, but there are things at the molecular level, for example, individual molecules, which do not have cells as parts. The same applies to any pair of levels, not just to a pair of adjacent levels: if level n is higher than level m, every object at n fully decomposes into objects at m, whereas there are objects at m that do not decompose into parts that belong to n. “Higher than” understood this way is transitive as well as asymmetric and irreflexive, and it can be taken as the ordering relation generating the Oppenheim-Putnam hierarchy. The highest level to which an object belongs, Oppenheim and Putnam say, can be 17 “Any whole which possesses a decomposition into parts all of which are on a given level, will be counted as belonging to that level. Thus each level includes all higher levels.” Oppenheim and Putnam, “Unity of Science as a Working Hypothesis,” pp. 9–10. 18 Actually, Oppenheim and Putnam state as an independent condition that nothing at level n can have a part belonging to a level higher than n (“Nothing on any level should have a part on any higher level,” Ibid. 9). This, however, is too strong. Consider the molecular level: given their stipulation that any given level includes all higher levels, the molecular levels include biological organisms. Biological organisms of course belong to the level of biological organisms. But then from their requirement of full decomposability into the “next lower” level, biological organisms, which also belong to the levels of molecules, have as parts cells, which are on a higher level. Cells of course also belong to the molecular level; but this doesn’t remove the difficulty.

56

THE LAYERED WORLD

considered its “proper” level. Here, unlike in Morgan, mereology, applied to individuals, seems to do crucial work; there is no reference to properties or kinds of individuals. This is not to say that properties and kinds play no role in the Oppenheim-Putnam scheme. Indeed, they are crucial to the initial selection of levels. Why are cells, or organisms, singled out to constitute a level? Why aren’t arbitrary gerrymandered classes of objects considered? The reason evidently is that cells, as a class, are a significant nomic kind; that is, there is a cluster of significant predictive and explanatory laws and regularities for objects belonging to this kind. Laws and regularities pertain to kinds and properties—or to individuals qua having certain properties or falling under certain kinds. Laws probably aren’t all there is to the selection of levels for hierarchies of the sort under discussion; there are interesting laws about, say, uranium atoms, but we don’t consider them to constitute a level in their own right. Nevertheless, it seems clear that laws, and, therefore, properties and kinds are important to the initial selection of the levels in Oppenheim and Putnam, as they clearly were for Morgan as well. But not all is well with the Oppenheim-Putnam hierarchy of levels. Consider the requirement that anything at level n, higher than the bottom level, be composed wholly of things at level n1.19 This condition, as we saw, is what generates the asymmetry of higher and lower. However, it evidently has the undesirable consequence, probably unanticipated by them, of excluding most biological organisms from their hierarchical scheme; for surely most living organisms have as part of their functional/ organizational structure “free” molecules that are not part of cells (e.g., chemical molecules in bodily fluids). These free molecules may play indispensable causal roles in the biological functioning of such organisms. And we expect biological theory to make free use of laws and theories of chemistry and physics; in fact, it seems natural to consider physics and chemistry an integral part of biological theory quite independently of the question of reduction of biology to physics/chemistry. At any rate, if there are biological organisms with free molecules, they fail to decompose fully into cells, violating the Oppenheim-Putnam requirement. Similarly, I don’t 19 “Any thing of any level except the lowest must possess a decomposition into things belonging to the next lower level. In this sense each level will be as it were a ‘common denominator’ for the level immediately above it.” Ibid.

THE LAYERED WORLD

57

think that real-life social groups, say tribes and nations, could be considered as composed merely of individual human beings (or other organisms). Surely, the natural physical environment (e.g., climate, terrain, the availability of food and other resources) must be considered a constitutive element, if not a “member,” of a social group—as belonging to the domain of discourse of social theory—if we are to have a comprehensive explanatory theory.20 As we have seen, the cellular level includes things other than cells, namely aggregate structures made up of cells, but this obviously does not remove the difficulty; biological organisms with free molecules remain excluded from the Oppenheim-Putnam hierarchy. Oppenheim and Putnam note that there may be things that do not belong to any level, if the part-whole relation is construed in a certain way. They consider the possibility of interpreting this relation strictly so that x is part of y only if y is a whole in a “narrower” sense, something like a “structured organization of elements” (a term they borrow from Carnap). I take it that what they have in mind is similar to what Morgan means by a thing with integral unity. Oppenheim and Putnam write: For a trivial example, “a man in a phone booth” is an aggregate of things on different levels which we would not regard as a whole in such a narrower sense. Thus, such an “object” does not belong to any reductive level, although the “phone booth” belongs to level 3 and the man belongs to level 5.21 What Oppenheim and Putnam didn’t see is that perfectly natural objects, like biological organisms, may not have a homogeneous decomposition into lower-level elements either. Their reasoning here is not cogent; the fact that a man in a phone booth (in the sense of a person plus the phone booth he is in) is not an organized structure has nothing to do with its not meeting the requirement of homogeneous decomposition (as we may call it). Satisfaction of this condition is no guarantee for structured organization or integral unity. The condition is satisfied by things like the mereological sum of the phone booth and the left front tire of my car, or a heap of living cells.

20 This point can be finessed by talking about the relational/environmental properties of the group members. But this does not change the fundamental issues involved. 21 “Unity of Science as a Working Hypothesis,” p. 11.

58

THE LAYERED WORLD

So why not weaken the requirement of homogeneous decomposition, thus: Each object at any level higher than the lowest level must have a full decomposition into parts all of which belong to the lower levels. Obviously this accommodates living organisms with their free molecules. This relaxed condition will be satisfied by “a man in a phone booth” and arbitrary mereological sums. If we want to exclude such things from our scheme—I don’t see why we should worry about this issue in the present context—it seems to me that this should be done on independent grounds. The idea of an organized structure or an object with integral unity probably cannot be made precise; at least, it is clear that the concept prima facie has nothing to do with uniform decomposition. At any rate, the new condition seems to preserve the asymmetry of higher and lower in the following sense: each object at level n, higher than level 1, has a decomposition into parts all of which belong to levels lower than n, whereas at each level lower than n there are objects that do not have objects at n as a part. With this emendation, what should we think of the Oppenheim-Putnam hierarchy? One apparent shortcoming is that it does not make fine enough discriminations in terms of complexity of organization and structure. Presumably, objects with the same “proper” level belong to the same level. And an object whose proper level is higher than that of another is a higher-level entity than the latter. Consider the level of molecules: this level includes all inorganic, nonliving things, from molecules in isolation to rocks, mountains, planets, Michelangelo’s David, jet engines, and the most sophisticated computers. All of these things are on the same level; none is higher than any other. Moreover, a computer and its constituent parts—the CPU, memory chips, wirings, and all—are all on the molecular level. In connection with mentality and consciousness, we now seriously debate the question whether a computing machine, made up of inanimate, nonbiological components, could be conscious, or manifest other forms of mentality. (I believe a similar question is being debated about life—whether “living” things must be made up of organic cells.) Some inorganic structures may very well exhibit features that are characteristic of life and mind—purposive behavior, perception, information processing, deductive and inductive reasoning, decision making, intentional action, and perhaps even conscious thought and sensation. These aspects of such inorganic electromechanical systems may be best investigated by grouping them with selected biological and cogni-

THE LAYERED WORLD

59

tive/psychological systems, not with lifeless and mindless lumps of atoms and molecules. But they are not composed of cells and they are not organisms, and if they are to find a home in Morgan’s or OppenheimPutnam’s scheme, they have to be placed at the level of matter (in Morgan) or of molecules (in Oppenheim-Putnam). In Morgan’s scheme, mind emerges from life, or mentality emerges from biology; the possibility that mentality might emerge directly from nonbiological matter is excluded. But it is clear that there is nothing in the idea of emergence, or in the basic emergentist doctrines, that prohibits mind’s emergence out of nonbiological physicochemical processes.

V I think that this issue alone throws into doubt the idea of a single hierarchy of connected levels, from higher to lower, in which every object and phenomenon of the natural world finds its “appropriate” place. Even if some such idea can be fleshed out in a coherent way, its philosophical usefulness is by no means guaranteed. It does not seem to me that the question whether sophisticated industrial robots or powerful supercomputers, say, are higher or lower in comparison with sea slugs or birch trees makes sense or, even if it did, what possible significance the answer would have. Obviously no one would contemplate reducing or explaining the behavior of robots or computers in terms of cells or other biological organisms, or vice versa—unless perhaps we construct biological robots and computers with cells as the main building blocks. As some have noted, we may be biological computers, and it may be that our computational behaviors can be explained in terms of our biology. Even so, what of nonbiological, electromechanical robots that have human computational powers and that, moreover, possess the kind of behavioral/functional organization characteristic of humans or some other biological species? Does Commander Data, of Star Trek, an electromechanical creation functionally equivalent to us, rank higher or lower than simple biological organisms? On what possible ground would an answer to this question be based, and what possible use could the answer have? I hope it has become clear that these questions don’t have clear answers, and there is no reason why we should

60

THE LAYERED WORLD

seek them, to begin with. I think that attempts to construct an overarching levels ontology for the whole of the natural world in which every object has its “appropriate” place are rather pointless if not hopeless. I say not hopeless because, as I will sketch below, there does seem to be a way of developing such an ontology, although it isn’t going to have the kind of satisfyingly neat look that we get in Morgan’s pyramid or Oppenheim and Putnam’s scheme of reductive levels. Physics, as Quine aptly put it, is the only science that, by its very nature, aspires to “full coverage”—a comprehensive description and explanation of all phenomena of this world. Nothing is outside its concern and attention, and it’s not for nothing that we think of it as the most fundamental and general of all the sciences. The domain of physics includes all that there is, and the special sciences are so-called because their domains are specially restricted subdomains of the universal domain of physics. As we saw, this fact is appropriately reflected in the Oppenheim-Putnam hierarchy, by their requirement that each level includes all higher levels, which implies that the bottom level, of which physics is in charge, includes all higher levels. So let us begin with the universal domain, U, of physics. It is populated by bits of matter, their aggregates, aggregates of aggregates, and so on (it is not necessary to assume that there is an ultimate bottom level—some have suggested that the macro-micro hierarchy may have no bottom level22). U coincides with the whole space-time world. Objects in this domain have standard physical properties, such as mass, energy, momentum, charge, size, shape, and so on. What is to count as a “physical property” is a difficult question, as many have recognized; however, we can set this issue aside and move on. Physics is the science of this domain, and the fundamental properties and relations that it uses to formulate its basic laws are presumably those that it deems necessary, and collectively sufficient, for a complete description of the world. But keep this in mind: the properties recognized in theoretical physics may not be, and probably are not, all the physical properties of this world. Being a lump of pure gold with a mass of one million kilograms is not a property mentioned in theoretical physics, but we would want to count it among physical properties. The same goes for such properties as being a rock, being a mahogany table, and being a supernova. 22 See Ned Block, “Do Causal Powers Drain Away?”, Philosophy and Phenomenological Research 67 (2003): 133–150, and the references given therein.

THE LAYERED WORLD

61

Starting with U, it is easy to see how a scheme like Oppenheim and Putnam’s can be developed. First consider atoms: atoms and aggregates of atoms form a subdomain of U; call it UA. Here we will use “aggregate” in the sense of organized structure as in Oppenheim and Putnam, or in the sense of something with “integral unity” as in Morgan. Molecules and aggregates of molecules form a subdomain of UA, and hence of U. Call it UM. What are left out of UM are elementary particles and atoms that are not part of molecules. In a similar fashion we can take cells and their aggregates as a subdomain, UC, of UM. We then have the domain of multicellular living organisms, UL, as a subdomain of UC, and the domain of social groups US. Notice that U, UA, UM, UC, UL, and US form a nested series in which each domain that follows U is properly included in its predecessor. I think that putting the scheme in terms of domains and subdomains is less misleading than talk of levels; the latter seems inevitably to evoke a picture of a vertical hierarchy of levels arranged “lower” to “higher,” and this picture seems to carry with it certain evaluative associations—typically, the idea that what’s higher is in some ways more highly evolved and advanced and is richer and more valuable. Such associations are evident— in fact, they are part of—the emergentist worldview. In any event, what our considerations indicate is that a neat nested series like this is not a realistic picture. The problem of “free” molecules shows that the domain of living organisms, UL, cannot be considered a subdomain of UC (this domain is made up only of cells and aggregates of cells); rather, UL must be thought of as being generated out of cells and “free” molecules. Organisms are generated ultimately from molecules, of course, but it is important to recognize cells as an intermediate point because cells form a significant nomic kind, and the biological functions and behaviors of organisms seem perspicaciously explainable in terms of the properties and behaviors of cells. What our considerations about computing machines and the rest show is that there is no need to countenance a single linear series of domains and subdomains, or a single connected hierarchy of levels. All the domains of interest are subdomains of the universal domain U; but some subdomains may intersect one another without any of them wholly including any other, and there are likely to be subdomains that do not intersect with each other at all (e.g., plants and animals). If we want a hierarchical representation, the right picture seems to be a tree-like structure with multiple branches, not a single ladder-like system, with levels belonging to different branches not being

62

THE LAYERED WORLD

comparable in respect of higher and lower. Even this may be too neat and orderly: branches may merge as we ascend further on the tree (we will see an example that will illustrate this possibility). If a comprehensive levels ontology is wanted, a tree-like structure is what we should look for; it seems to me that there is no way to build a linear system like Morgan’s or OppenheimPutnam’s that will work. Concerning the choice of levels in the Oppenheim-Putnam hierarchy, we noted that these particular levels are selected because objects belonging to each are thought to constitute a significant nomic kind. Nomic kinds are kinds defined by a cluster of laws with significant explanatory and predictive powers.23 These are causally and often epistemically salient kinds, and they form the basic taxonomies of various sciences. In that sense, we may call them “scientific” kinds. With each such kind there is a set of properties, nomic properties, in terms of which laws over objects belonging to that kind are formulated. Biological organisms as a group form such a kind. But many subdomains of organisms—notably, species—are nomic scientific kinds as well. For each such kind, we may become interested in the possibility of explaining the properties and behaviors of objects falling under it, or the laws governing them, in terms of the properties and relations characterizing their parts. Not any mereological decomposition will serve such reductive explanatory purposes; when such explanations are possible, it will be because the parts selected are themselves nomic kinds. They are likely to be cells, large structures made up of cells, molecules, and so on. Even if you are an emergentist who does not believe in the possibility of microreductive explanation, you will still be interested in identifying the underlying microstructural states from which various life phenomena emerge. Reductionists and emergentists will in general agree on what these underlying states and processes are on which the macroproperties of a whole organism supervene. Both will look for the kind of mereological decomposition that will yield perspicacious nomic descriptions of the parts. Only such decompositions will enable the development of a fruitful and powerful theory about the underlying processes, and only against the background of such a theory, or the possibility of such a theory, could the question of reducibility, or reductive explainability, be meaningfully debated. The thing to note here 23 See W.V. Quine, “Natural Kinds,” in Quine, Ontological Relativity and Other Essays (New York: Columbia University Press, 1969).

THE LAYERED WORLD

63

is that this is a kind of top-down approach: we first select a kind of interest to us, say biological organisms or a selected group of biological species, and ask how their properties and behaviors can be explained in terms of their microstructure. What microentities and microproperties are to be invoked in such explanations are not fixed in advance; their choice is entirely opportunistic—whatever decomposition that yields a microtheory sufficient for microreductive explanation will be the one to go for—and there may be more than one such choice. (And, for the emergentists, whatever decomposition that will yield perspicuous descriptions of the “basal” conditions for their holistic emergent properties.) We should not expect that the decomposition that works will be uniform and homogeneous, as Oppenheim and Putnam stipulate, or as Morgan assumes with his claim that minds can arise only from biological processes. Let us return to computing machines. There may be a certain class of computers that share a range of function and behavior that is of interest to us. Though they are artifacts, they may form a nomic scientific kind, and again we may be interested in explaining their behavior and function in terms of the properties and relations characterizing their component parts. It may make sense to talk of lower and higher levels within this local picture. A complex computing system may exhibit powers and behaviors that are in some clear sense higher than those of its simpler constituent parts, and these parts, in turn, may have properties that are higher than the properties of their parts; and so on. But this talk of higher and lower can only be provisional: the question of reducing them to “lower-level” properties is open, and if the reduction goes through, there is no higher or lower. Under a successful reduction, the properties initially regarded as “higher” have been reductively absorbed into lower-level properties, and the hierarchy of properties with which we began collapses. One possibility, however, is to define “higher” and “lower” not for properties but instead for predicates and concepts. If we do this, it could make good sense to say that even though two predicates turn out to designate the same property, one of them is higher than the other. And by extension there is the possibility of giving a clear meaning to “higher” languages and “lower” languages. If this is done, a sort of hierarchy could be developed for computing machines, or for any thing else, in which objects could be ordered in terms of the highest-level predicate or description that applies to each. In any case, whatever is done, the domain, or level, demarcated here, namely the domain of certain

64

THE LAYERED WORLD

computing machines, is a branch that starts off from the level of molecules in complete separation from the biological branch which, too, starts off from the molecular level. And levels in the two branches are not connected; that is, they are not comparable with respect to higher or lower; nor does there seem any need to want them to be comparable. But the branches may merge as well as split. Suppose that humans are biological computers, and that we want to consider a class of computers that includes us. For certain scientific and philosophical purposes, it may make sense to group humans together with electromechanical computers and robots that have a certain specified set of computational and perhaps other powers (e.g., purposive behavior). If there were beings like Commander Data around us, it probably would make a lot of sense to try to see whether, say, a single psychological theory of motivation and action applies to both humans and Data and his friends. Although given the radical structural differences between humans and Data, it’s extremely unlikely that the same microreductive strategy will work for both Data and us. But this should not prevent us from viewing Data and humans as belonging to the same kind for certain purposes—as when we are mainly interested in observable macro input-output regularities. It is usually thought that Data has cognitive/intentional states like those that we have, although probably not our experiences, or experiences of any kind, and that he is a language speaker.24 There is every reason to think that the cognitive/intentional psychological regularities that hold for us hold for Data and his friends just as well.25 If this is the case, some psychological kinds will include both electromechanical systems and biological organisms, and these kinds are not locatable in systems like Morgan’s or Oppenheim-Putnam’s. So what we have here is this. There are groupings of objects in nature that are of interest to us, and in the context of scientific inquiry these are the nomic/causal kinds, kinds defined by clusters of significant laws. In regard to such kinds, we may be interested in the question whether their properties and behavior are explainable in terms of their microstructure. And this may fairly naturally give rise to talk of levels, “higher” and “lower” properties, 24 For interesting reflections on Commander Data’s possible inner life, see Ned Block, “The Harder Problem of Consciousness,” Journal of Philosophy 99 (2002): 391–425. 25 Although, given Data’s fundamentally different microstructure, whether the regularities are projectible laws is another question; see my “Multiple Realization and the Metaphysics of Reduction,” in Supervenience and Mind (Cambridge: Cambridge University Press, 1993).

THE LAYERED WORLD

65

and the like. However, such kinds may not belong to any single level, but rather cut across various levels, in the kind of monolithic overarching system of levels we see in Morgan or Oppenheim and Putnam. One way of making levels talk meaningful and helpful is to localize it and take a topdown approach, as I have suggested: we first pick a nomic kind of interest to us and go from there, rather than start with a comprehensive levels ontology and then try to locate each object, or kind, of interest to us at a particular level in that ontology. But don’t we need a comprehensive ontology of levels to pose and discuss certain philosophical issues, for example, the thesis of universal microphysical reductionism, a topic for Oppenheim and Putnam, and the claims of emergentism as giving a global ontology and history of the world? As for emergentism, it is likely that the emergentist will want a special dedicated hierarchy of levels that can be justified only if the basic doctrines of emergentism are correct and there are genuine emergent properties. No scheme that is neutral to the issue of emergence will likely serve the emergentist’s needs. As for the global thesis of microphysical reductionism, I do not believe we need a global levels ontology to make sense of it and debate its merits. Even in the kind of picture I have tried to present, without a single hierarchy of levels, the question can be raised for each kind that we recognize, or indeed for each object, whether its properties can be reduced, or reductively explained, on the basis of its microphysical structure. To argue about this, and, indeed, to make use of some of the considerations marshaled by Oppenheim and Putnam for “unity of science,” we do not need a single monolithic and all-inclusive hierarchy of levels.

3 Emergence: Core Ideas and Issues Since around 1990, the idea of emergence has been making a big comeback, from decades of general neglect and disdain on the part of mainstream analytic philosophers. Indications are that the emergence boom is going to continue, on an upward trajectory, for years ahead. What it is about emergence that makes it such an attractive idea to so many thinkers, with diverse and disparate backgrounds and agendas—philosophers, practicing scientists from a variety of scientific fields, and science writers—is itself an intriguing philosophical question, or perhaps a question in the sociology of science. The term “emergence” seems to have a special appeal for many people; it has an uplifting, expansive ring to it, unlike “reduction” which sounds constrictive and ungenerous. We now see the term being freely bandied about, especially by some scientists and science writers, with little visible regard for whether its use is underpinned by a consistent, tolerably unified, and shared meaning (and if so what it is). This has created situations in which those discussing emergence, even face to face, more often than not talk past each other. Sometimes one gets the impression that the only thing that the participants share is the word “emergence.” The intuitive associations this word evokes in us do not add up to a concept robust enough to do any useful work, or even to serve as helpful constraints on a theoretical account or construction of the concept. “Emergence” is very much a term of philosophical trade; it can pretty much mean whatever you want it to mean, the only condition being that you had better be reasonably clear about what you mean, and that your concept turns out to be something interesting and theoretically useful. Of course we do not start with a totally blank page when we now ponder how best to understand emergence. For there are signposts that should guide our reflections, at least in their initial stages. Any account of

EMERGENCE: CORE IDEAS AND ISSUES

67

emergence, I believe, should show significant continuity with the concept that the British emergentists of the early 20th century, like Samuel Alexander, C. Lloyd Morgan, and C.D. Broad, had in mind, and we should take care that our understanding of the concept should be as hospitable as possible to the central doctrines of the emergentism these philosophers espoused. This is not to say that these emergentists were in complete agreement with one another; however, there was a sufficient conceptual and doctrinal convergence among them to generate a shared philosophical perspective and a movement—what now goes by the name “British emergentism.” Nor does it mean that we must stay with the early emergentists to the end; at some point, new possibilities and opportunities may well prompt us to proceed in different directions and explore new options. As is well known, the intuitive idea of an emergent property stems from the thought that a purely physical system, composed exclusively of bits of matter, when it reaches a certain degree of complexity in its structural organization, can begin to exhibit genuinely novel properties not possessed by its simpler constituents. Questions and disputes arise when we try to make this idea more precise and turn it into something that is philosophically and scientifically useful. For the friends of emergence, to say that a given property is an emergent property of some system must be saying something significant and explanatory about the property and the system that has it. But exactly what does it mean to say that something is an emergent property? What does it mean to say that it is a “novel” property? What do emergent properties do after they have emerged? In this paper, I want to set out and discuss two conditions that I believe must be considered necessary components of any concept of emergence that is true to its historical origins in British emergentism. The conditions are supervenience and irreducibility. How irreducibility is to be understood in this context will require some discussion. I will then take up the question whether supervenience and irreducibility together could be considered sufficient for emergence, and argue for a negative answer. Finally, I will turn to the question what needs to be done if we are to turn emergence into a robust metaphysical concept. As we will see, the idea of “downward” causation will loom large in my discussion. The picture I will present is not going to be reassuring to the friends of emergence, for my conclusion will essentially be something close to this: Emergentism cannot live without downward causation but it cannot live with it either. Downward causation

68


is the raison d’eˆtre of emergence, but it may well turn out to be what in the end undermines it.

I. Emergence and Supervenience In order to help fix the concept of emergence, let us begin with the question whether emergent properties should be thought of as supervenient on their basal conditions—conditions at the lower level from which they emerge. For our purposes, it will be useful to take as our starting point a helpful survey article by Van Gulick.1 According to Van Gulick, three grades of emergence can be distinguished: (1) “specific value emergence”; (2) “modest kind emergence”; and (3) “radical kind emergence.” The first, specific value emergence, is a pretty tame affair exemplified in a situation like the following: a whole, say a bronze statue, has a mass of 1 kg whereas none of its proper parts has that particular value of mass, namely 1 kg. Whence the name “specific value emergence.” Plainly, this is not a case of what emergentists have in mind when they speak of emergent properties; Van Gulick probably wanted to recognize cases of this kind because they do fit the letter of the emergentist slogan “A whole can (and often does) have properties that none of its constituents have.” Van Gulick explains the second, “modest kind emergence,” as follows: “The whole has features that are different in kind from those of its parts . . . For example, a piece of cloth might be purple in hue even though none of the molecules that make up its surface could be said to be purple. Or a mouse might be alive even if none of its parts (or at least none of its subcellular parts) were alive.”2 The examples offered are the sort that has traditionally been claimed to be emergent. What Van Gulick does not say is just in what sense the purple color of a cloth “emerges” from the properties of its constituent molecules. But before we take up this question, let’s look at Van Gulick’s third and strongest kind of emergence. Van Gulick defines “radical kind emergence” as follows: 1. the emergent property is “different in kind from those had by its parts,” and 2. it is “of a kind whose nature and existence is not necessitated by 1 “Reduction, Emergence and Other Recent Options on the Mind-Body Problem,” Jourmal of Consciousness Studies 8 (2001): 1–34. 2 Ibid. 17.


69

the features of its parts, their mode of combination and the law-like regularities governing the features of its parts.”3 The second condition, which is what distinguishes this kind of emergence from its weaker siblings, asserts that an emergent property of a whole is not determined by the properties and relations characterizing its parts, or, to put it another way, an emergent property of this third kind does not supervene on the microstructure of the object that has it. Thus, two wholes may have identical microstructure (i.e., composed of identical basic physical constituents configured in identical structure) and yet differ in respect of their emergent properties. For example, two molecule-for-molecule identical systems may be such that one of them is a live mouse and the second is not—that is, if being a live mouse is emergent in the sense of “radical kind” emergence. Van Gulick is uncertain whether there are actual cases of radical kind emergence, saying that accepting it would violate “atomistic physicalism.” However, the real problem with this purported form of emergence is not its violation of the doctrine of microdeterminism (or mereological supervenience). A more immediate concern is whether it is a form of emergence at all. We should remember that classic emergentists accepted microstructural supervenience of emergent properties,4 and they had good reasons. Suppose that on a given occasion a mental phenomenon, say pain, emerges from a certain configuration of neural events. It is highly unlikely that an emergentist will deny the proposition that if the very same configuration of physiological events were to recur, the same mental phenomenon, pain, would emerge again. If the connection between pain and its neural substrate were irregular, haphazard, or coincidental, what reason could there be for saying that pain emerges “from” that neural condition rather than another? What would be the point of saying that pain is an emergent property rather than saying that it is a property randomly distributed over neural states? The very meaning of “neural substrate” indicates the presence of a regular determinative, or necessitating, relationship. If supervenience, or upward necessitation, is taken away, that takes away something essential 3 Ibid., emphasis added. 4 For example, Broad, arguably the most astute of the British emergentists, writes: “No doubt the properties of silver chloride are completely determined by those of silver and of chlorine; in the sense that whenever you have a whole composed of these two elements in certain proportions and relations you have something with the characteristic properties of silver chloride.” For Broad, the characteristic properties of silver chloride are emergent properties (C. D. Broad, The Mind and Its Place in Nature (London: Routledge & Kegan Paul, 1925), p 64).

70


to the meaning of “emergence” in Van Gulick’s “radical kind emergence.” We need not insist that the dependence of an emergent property on its “basal” condition be deterministic; there might well be a viable concept of statistical or stochastic emergence, which assigns a stable objective chance of the emergence of a property given that an appropriate basal condition is present. But stochastic emergence in this sense would require statistical laws grounded in the base-level laws, and such laws could warrant talk of stochastic supervenience and stochastic necessitation. It is clear then that we must consider supervenience as a component of emergence—that is, we need to accept the following proposition: Supervenience: If property M emerges from properties N1, . . . , Nn, then M supervenes on N1, . . . ,Nn. That is to say, systems that are alike in respect of basal conditions, N1, . . . ,Nn, must be alike in respect of their emergent properties.

II. Emergence and Reduction Morgan, a leader of British emergentism, said that “resultant” properties— that is, properties of a whole that are not emergent—are “additive and subtractive only, and predictable”5 from information concerning the basal conditions. The implication of course is that emergent properties are neither additive nor subtractive, and not predictable, on the basis of the lower-level properties from which they arise. It has also been said that emergent properties are not explainable in terms of their basal properties—that is, for example, we cannot explain why someone is experiencing pain (rather than itch or tickle) on the basis of the neural processes from which pain emerges. Let us consider some relations a property can bear to a set of other properties. The first, which we have just discussed, is supervenience: 1. Supervenience/determination: Property M supervenes on, or is determined by properties N1, . . . , Nn in the sense that whenever anything has N1, . . . , Nn, it necessarily has M.

5 C.L. Morgan, Emergent Evolution (London: Williams and Norgate, 1923), p. 3.


71

I argued earlier that if M emerges from N1, . . . , N2, these properties must meet the supervenience condition. According to classical emergentism— that is, British emergentism—this determination thesis must be sharply distinguished from each of the following claims (we will see below what Broad says on this issue): 2. Predictability: The occurrence of M—that is, whether M will be instantiated on a given occasion—can be predicted from the occurrence of N1, . . . , Nn; full information concerning whether N1, . . . , Nn are instantiated in a system suffices for the prediction of whether the system will instantiate M. 3. Explainability: Why a system instantiates M can be explained, understood, and made intelligible in terms of its instantiating N1, . . . , Nn. Emergentists will deny that predictability and explainability hold for an emergent property in relation to its basal conditions. When Morgan denied that emergent properties are “additive” or “subtractive,” what he had in mind presumably is that their occurrences cannot be mathematically calculated, or logically deduced, on the basis of the properties from which they emerge (so the whole is not a mere “sum” of its parts). Emergentists will say that the fact that pain emerges from a certain neural state (say, C-fiber excitation) is a brute fact that cannot be explained, and that full and ideally complete knowledge of the neurophysiology of the brain does not suffice for prediction of conscious states. Thus, determination is one thing; explainability and predictability are quite another. This naturally leads to the following question: What is required for explanation and prediction that goes beyond mere supervenience or determination? Emergentists were quite aware that there is a sense in which the occurrence of an emergent phenomenon can be predicted. Consider an inference like this: Jones’s C-fibers will be stimulated at t. Anyone whose C-fibers are stimulated experiences pain. Therefore, Jones will experience pain at t. This may be called an “inductive” prediction of pain—based on our inductive knowledge of the pain/C-fiber stimulation correlation. It should be clear why a prediction of pain of this kind will not impress an emergentist who asks “Can an emergent phenomenon be predicted on the basis of knowledge of its basal conditions?” What is wrong with the

72


above prediction of pain is that the evidence base, specifically the second premise, makes use of knowledge of facts going beyond those at the basal level; it assumes knowledge of the “emergence law,” or “trans-ordinal law” (in Broad’s terminology), linking pain with C-fiber stimulation. We should remember that the iconic emergentist question is “By knowing all about the base-level facts (e.g., facts about the brain), can we know anything about the higher-level phenomena (e.g., consciousness)?” Similar comments apply to the following “inductive” explanation of pain: Jones’s C-fibers were stimulated at t. Anyone whose C-fibers are stimulated experiences pain. That is why Jones experienced pain at t. What the emergentist had in mind, when he denied the explainability of emergents, is what we may call “reductive” explanation, or what some emergentists called “mechanistic” explanation—an explanation of why a whole has a certain property exclusively on the basis of its constituent microstructure. Such an explanation would be prohibited from using any information regarding the emergent level in its explanans; the emergentist is interested in the question why emergent properties emerge from their basal conditions, and assuming something like the second premise of the foregoing argument would be begging the question. Broad contrasted emergentism with “mechanism,” a view that we would now call “reductionism.” Consider the following passage in which the contrast is described: [The two approaches] differ according to the view that we take about the laws which connect the properties of the components with the characteristic behaviour of the complex wholes which they make up. (1) On the first form of the theory the characteristic behaviour of the whole could not, even in theory, be deduced from the most complete knowledge of the behaviour of its components, taken separately or in other combinations, and of their proportions and arrangements in this whole. This alternative . . . is what I understand by the “Theory of Emergence”. . . . (2) On the second form of the theory the characteristic behaviour of the whole is not only determined by the nature and arrangement of its components; in addition to this it is held that the behaviour of the whole could, in theory at least, be deduced from a


73

sufficient knowledge of how the components behave in isolation or in other wholes of a simpler kind. I will call this kind of theory “Mechanistic”.6 Both emergentism and reductionism—that is, Broad’s “mechanism”—agree in holding that “the behavior of the whole” is determined by—that is, it supervenes on—the properties and structural relationships characterizing its components. Where they differ, according to Broad, concerns the inprinciple deducibility—presumably, logical/conceptual deducibility—of the properties of the whole from facts at its basal level. Reductionism asserts, and emergentism denies, the deducibility of all properties of a whole from the properties at the basal level—that is, the properties and relationships for its constituents. Deducibility is crucial in this context, for when the British emergentists discussed predictability and explainability in connection with emergents, it is plain that they construed prediction and explanation as a matter of deduction or derivation. The question “Is emergent property M explainable in terms of basal facts F?” comes down to “Is the fact that M is instantiated deducible from F?” Similarly, and only more obviously, the question “Is the occurrence of M predictable from F?” is simply the question “Is the occurrence of M—that is, the fact that M will be instantiated—deducible from F?” It is instructive in this connection to consider Ernest Nagel’s model of bridge-law reduction,7 which has dominated discussions of reduction for many years. It can be seen that Nagelian reduction is irrelevant to the questions of reductive prediction and explanation in the present context. Bridge-law reduction, like the reduction envisaged by Broad, involves derivation; however, the distinctive feature of the Nagel model is that, as its name indicates, it allows as auxiliary premises of the derivation the use of “bridge laws” connecting properties to be reduced with base-level properties—these are Broad’s trans-ordinal laws, like the second premise of the two inferences above. This is expressly prohibited in reductive derivations as conceived by the emergentist: for the emergentist the real explanatory question is why these particular bridge laws—for example, one connecting pain with C-fiber excitation—hold. Why does pain arise from, or correlate with, C-fiber excitation, but not another kind of neural state? Why does 6 Broad, The Mind, p. 50; emphasis in the original. 7 E. Nagel, The Structure of Science (New York: Harcourt, Brace and World, 1961).

74


pain, not itch or tickle, correlate with C-fiber excitation? Nagelian reduction allows the use of psychoneural correlations as unexplained additional premises of reductive derivations, and this, by the emergentist’s light, begs the explanatory question at hand. There is another model of reduction—reduction through functionalization, or, briefly, functional reduction—which better suits our purposes.8 Bridge laws are the heart of Nagel reduction: in fact, it can be shown that the availability of bridge laws is necessary and sufficient for Nagel reduction.9 In contrast, functional reduction proceeds via the “functionalization” of the properties to be reduced in terms of properties at the base level. Suppose pain can be given a “functional” definition like this: To be in pain ¼ def. to be in a state that is typically caused by tissue damage and trauma and that typically causes aversive behavior. This definition connects pain conceptually with physical/behavioral properties. Reduction of pain is accomplished for a population of interest to us (say, humans, mammals) when we are able to identify a “realizer” of pain so conceived, namely a physical state that fits the functional definition, for that population. So suppose neuroscientific research has identified state N1 (say, the activation of a group of neurons in certain cortical areas) as the state that is typically caused by tissue damage and that in turn triggers aversive behavior. N1 would be what is called the neural “correlate” or “substrate” of pain (for humans, mammals). When we have such a neural state for the population of interest to us, we have a neural reduction of pain for this population. On this model, a neural reduction of pain, it should be noted, does not require a logical derivation of pain from a neural state—in particular, from its neural realizer; and it does not require logical or conceptual connections between pain and neural states. Trying to derive a pain statement from statements about neural states would be hopeless. What we should keep in mind is the fact that the mind–body problem involves three players, not two; they are pain (and other mental states), the brain, and behavior. Reduction requires conceptual connections, but these connections connect 8 For more details see J. Kim, Mind in a Physical World (Cambridge, Mass.: MIT Press, 1998) and “Making Sense of Emergence,” Philosophical Studies 95 (1999), pp. 3–36; Essay 1 in this volume. For some criticisms see A. Marras, “Kim on Reduction,” Erkenntnis 57 (2002): 231–257. 9 J. Kim, Mind in a Physical World, p. 91.


75

pain with behavior, not directly with the brain. Brain phenomena enter the picture as the realizers of functionally characterized mental phenomena. It is important to notice that the fact that N1 is a realizer of pain (for a given group of organisms), or that the brain is the realizer of mentality, is an empirical and contingent (though lawful) fact. What is not contingent is the relation between pain and pain behavior. I am not saying that pain can in fact be reduced this way; what I am saying is that if pain is to be reduced to a brain process, the following is what must be accomplished: pain must first be given a functional definition or interpretation and then we must identify its neural realizers. The first step involves conceptual work: Is the concept of pain functionally definable or interpretable and if so how should a functional definition of pain be formulated? The second step, that of discovering the realizers of pain, is up to empirical scientific research. It is in effect the research project of finding the neural correlates of conscious experiences. From a philosophical point of view, the crucial question, therefore, is whether pain can be given a functional characterization, in terms of physical input and behavioral output; the rest is up to science. Philosophical functionalism, still the orthodoxy on the mind-body problem, holds that pain, along with other mental phenomena, can be functionalized; if philosophical functionalism is correct, all mental phenomena will be functionally reducible and hence nonemergent. I am with those who do not believe pain and other sensory states (“qualia”) can be given functional characterizations.10 However, this does not change the fact that functionalizability is crucial to reduction and reducibility, and hence to understanding emergence (as we shall shortly see). As noted earlier, it would be futile to try to derive pain from its neural correlate. However, it is quickly shown that if pain can be functionalized, we can derive a statement to the effect that pain occurs from facts at the behavioral/neural level. And this will vindicate functional reduction as characterized here as the appropriate concept of reduction for the purposes of defining and clarifying emergence. So suppose that pain has been functionally reduced with neural state N1 identified as its realizer for a given population. A derivation of the occurrence of pain can proceed as follows:

10 e.g., D.J. Chalmers, The Conscious Mind (New York and Oxford: Oxford University Press, 1996).

76


System s is in neural state N1 at t. N1 is such that tissue damage causes s and systems like s to go into N1, and N1 causes these systems to emit aversive behavior. By definition, a system is in pain iff it is in some state P such that P is caused by tissue damage and P in turn causes aversive behavior. Therefore, s is in pain at t. The derivation is valid. Note that the third sentence, as a conceptual definition, does not count as an additional premise. It introduces no facts about pains or about how pains correlate with instances of N1; if it is about anything, it is about the meaning of the term “pain” or the concept pain; it does not state a fact about pains. In general, definitions come free in proofs; they do not count as premises. Consequently, the derivation delivers an affirmative answer to the emergentist question “Can the occurrence of pain be derived from information about lower-level facts alone?” This is the crucial difference between this derivation and the earlier derivation using Nagelian bridge laws as auxiliary premises. Both derivations invoke laws; the difference is that the law used in the Nagelian derivation is an interlevel law “bridging” the emergent level with the base level, whereas the law used in the functional derivation (i.e., the second line) concerns the base level alone. And that is a crucial difference. It is clear that the above derivation can serve as a reductive explanation of the occurrence of pain and also as a predictive inference to the occurrence of pain. Thus, a functional reduction of a mental property, M, guarantees the following: (1) Instantiations of M can be predicted on the basis of information concerning neural and behavioral processes alone (including laws concerning these processes). (2) Similarly, why an organism instantiates M at a time can be explained on the basis of information concerning facts at the lower level, namely neural and behavioral facts. This shows that functional reduction gives a unified account of the emergentist idea that an emergent property is irreducible to the basal phenomena and neither explainable nor predictable in terms of them. Moreover, a functional reduction of pain has the following causal and ontological implications:


77

(3) Each occurrence of pain has the causal powers of its neural realizer; thus if pain occurs by being realized by N1, this occurrence of pain has the causal powers of N1. In general, if M occurs by being realized by N1 on a given occasion, the M-instance has the causal powers of the N1-instance.11 If two individual events have identical causal powers, there is little reason to think of them as distinct. In fact, Davidson has proposed a causal criterion of event identity: “Events with the same causes and the same effects are one and the same event.”12 Even if we do not accept Davidson’s proposal as a “criterion” of event identity, it may nonetheless be true and can serve as a good metaphysical guide. What is of essential importance to an event may well be its place in the causal network of events of the world. If this is right, functional reduction also accomplishes ontological reduction: (4) If M is instantiated in virtue of the instantiation of its realizer N1 on a given occasion, the M-instance is identical with the N1-instance. We have, therefore, identified a second condition of emergence: Irreducibility of emergents: Property M is emergent from a set of properties, N1, . . . , Nn, only if M is not functionally reducible with realizers at the level of the N’s. Thus, supervenience and functional irreducibility are two necessary conditions of emergence. Are they together sufficient? I will return to this question in the section to follow and argue for a negative answer. I believe, however, that the two conditions together capture the concept as it was intended by the classical emergentists like Alexander, Morgan, and Broad. This means that the notion of emergence in British emergentism was undercharacterized.13 When we consider recent proposals concerning emergent properties in complex systems, in terms of such ideas as chaotic, nonlinear

11 This is what I called the causal inheritance principle in “Multiple Realization and the Metaphysics of Reduction,” reprinted in J. Kim, Supervenience and Mind (Cambridge: Cambridge University Press, 1993). 12 D. Davidson, “The Individuation of Events,” in N. Rescher et al. (eds.), Essays in Honor of Carl G. Hempel (Dordrecht: Reidel Publishing Co., 1969). Reprinted in Davidson, Essays on Actions and Events (Oxford: Oxford University Press, 1980). 13 See Essay 4 of this volume for an argument that there may well be an incoherence in the conjunction of these two conditions.

78


dynamics, the necessity of simulation (rather than computation), and so forth, we should, first of all, examine them to see whether they fit the classic conception of emergence encapsulated in our two conditions. Of course, to judge that one or another of these new proposals does not fit the classic conception should not be taken to imply that it is not an interesting and potentially fruitful concept. But the conjunction of supervenience and functional irreducibility can serve as a useful benchmark and reference point; any deviation from it is a deviation from the classic conception, and new proposals can be analyzed and compared with one another in terms of how far, and in what ways, they deviate from the reference standards.

III. Are Supervenience and Irreducibility Sufficient for Emergence? There are reasons for thinking that emergence would be under-characterized and under-explained if supervenience and irreducibility together were taken to constitute its full definition—that is, as a necessary and sufficient condition of emergence. At least, this would be the case if one expects emergence to be a robust and natural relation holding between a higherlevel property and its lower-level base properties. First, consider supervenience, or its core notion of dependence. Brief reflection will show that supervenience is not a homogeneous, “natural” relation. This becomes clear when one considers some sample cases of supervenience and asks why supervenience holds in each case. Take, for example, the supervenience of normative properties on factual, non-normative properties. Why does supervenience hold for these two families of properties? What grounds normative supervenience? Different metaethical theories give competing and mutually exclusionary answers. Ethical naturalists will say that the supervenience holds because ethical properties (e.g., being good) are fully definable in terms of non-normative, naturalistic properties (e.g., being desired, being pleasurable). Non-naturalists, or intuitionists, will say that there are synthetic necessary connections, which we intuit through our moral sense, between non-natural ethical properties and certain natural properties. Noncognitivists and moral antirealists will see normative supervenience as arising from certain consistency requirements on the use of


79

ethical language or from our moral and evaluative practices (e.g., the requirement of universalizability or generalizability). Each of these is a possible answer to the question what grounds normative supervenience, but they all advert to different grounding relations. Under each explanation normative supervenience holds; however, the only thing common to the three explanations is the fact that ethical/normative properties covary in a certain way with naturalistic properties. But property covariations are “phenomenological” relations—surface phenomena arising from, and requiring explanations in terms of, some deeper underlying relations. We see an analogous situation with mind–body supervenience. Many diverse mind-body theories accept supervenience; for example, type physicalism, functionalism, epiphenomenalism, emergentism, and the doubleaspect theory. (Token physicalism and even substance dualism are at least consistent with mind-body supervenience.) Each will give a different explanation of why the mental supervenes on the physical; for example, functionalism will advert to the fact that physically indiscernible systems have identical physical causal powers; epiphenomenalism will invoke the “same cause, same effect” principle; emergentism, rather like non-naturalism in regard to normative supervenience, will say that mind-body supervenience is grounded in brute and fundamental physical-mental connections (primitive “laws of emergence”), which Alexander counseled us to accept with “natural piety.” And so on. Again, this shows that supervenience, or dependence, is not a homogeneous relation; it is not a genuine, “natural” relation, but rather something that arises from natural relations holding at a deeper level, like causation and reduction (assuming of course that these count as natural relations). This means that the bare statement that a family of properties supervenes on another does not tell us much. For this to be philosophically informative and enlightening we must know the deeper relation that grounds and explains why supervenience holds between these two sets of properties. Now, according to classical emergentism, that is precisely the kind of information we cannot have: emergence is brute, and that means that the supervenience relation holding in cases of emergence, too, is brute and unexplainable. If we could explain why pain, not itch or tickle, supervenes on C-fiber stimulation (or, more correctly, why pain has C-fiber stimulation as its supervenience base), the emergence of pain from a neural state would no longer be a brute unexplainable fact. So the demands of emergentism

80


make the supervenience relation involved in emergence necessarily unexplainable; we cannot know what kind of dependence grounds and explains the supervenience relation involved in emergence. This impossibility of knowledge has nothing to do with our epistemic limits; rather, for classical metaphysical emergentists like Broad, there is nothing here, no fact of the matter, to be known. The upshot is that the supervenience condition on emergence simply amounts to the assertion that there is an in-principle unexplainable covariation between the putatively emergent properties and their base properties. This cannot be considered a substantive positive characterization of the emergence relation. Let us now turn to irreducibility. The first thing we notice is that it is a negative characterization. If we know that X is reducible to Y, we know something interesting and important about the relationship between X and Y. And if we also know that U is reducible to W, we know something common that the pairs [X,Y ] and [U,W ] share.14 I believe we can take reducibility as a genuine relation characterizing two domains of properties, or two theories. But this does not mean that irreducibility, namely the absence of reducibility, is also a genuine and informative relation. As has often been observed, being red is a property but that does not mean that being nonred is also a genuine property. There are too many diverse things that are nonred: green things, yellow things, transparent things, numbers, atoms and molecules, thoughts and ideas, and countless other things. The same should hold for relations and their negations. Number theory is irreducible to hydrodynamics and vice versa. Chemical properties are irreducible to biological properties; geological properties are irreducible to economic properties and vice versa. If emergent properties are irreducible to their base properties, does this instance of irreducibility have anything in common with those other cases of irreducibility? The answer has to be “none.” What we have in supervenience and irreducibility, therefore, are two seriously defective relations, and it is difficult to take them as providing an informative positive characterization of emergence. As we saw, supervenience is not a natural, homogenous relation; nor is irreducibility. It is difficult not to notice that the common slogans describing emergents are full of negative predicates, for example “irreducible,” “unexplainable,” 14 This needs to be qualified because there are different types of reduction. My remark is best taken to apply to cases of reduction falling under one type. For types of reduction, see Esssay 10 in this volume.


81

“unexpected,” “underivable,” “unpredictable,” and so on. I believe one pressing item on the emergentist agenda ought to be the provision of a robustly positive characterization of emergence. Some recent work on emergence,15 in fact, can be seen as attempts in that direction. It remains to be seen whether any of them will succeed. Success here includes at least two things. First, the proposed characterization of emergence must explain why emergents so characterized supervene on their base properties and why, in spite of the supervenience relation, the former are not reducible to the latter. And if supervenience is rejected as a component of emergence, as is sometimes done, we need a carefully argued motivation for it (including an explanation of why the proposed concept should count as a concept of emergence). Second, it must successfully cope with the problem of downward causation. We turn to this problem in the next section.

IV. The Problem of Downward Causation It is critically important to the emergentists that emergent properties have distinctive causal powers of their own, irreducible to the causal powers of their base properties. Properties that are lacking in causal powers—that is, whose possession by an object makes no difference to the causal powers of the object—would be of no interest to anyone; in fact, it was Alexander, an iconic emergentist, who equated the existence of an entity with its having causal powers, saying that epiphenomena “might as well, and undoubtedly would in time be abolished.”16 On the causal theory of properties,17 properties are simply clusters of causal powers. This means that if there are emergent properties, they must represent causal powers. Moreover, the causal powers that emergent properties bring with them must be genuinely 15 D.V. Newman, “Chaos, Emergence, and the Mind-Body Problem,” Australasian Journal of Philosophy 79 (1997): 180–196; P. Humphreys, “How Properties Emerge,” Philosophy of Science 64 (1997): 1–17; M. Bedau, “Weak Emergence,” Philosophical Perspectives 11 (1997): 375–399; A. Rueger, “Physical Emergence, Diachronic and Synchronic,” Synthese 124 (2000): 297–332; M. Silberstein, “Converging on Emergence,” Journal of Consciousness Studies 8 (2001): 61–98; T. O’Connor and H.Y. Wong, “The Metaphysics of Emergence,” Nouˆs 39 (2005): 658–678; and others. 16 S. Alexander, Space, Time, and Deity, vol. 2 (London: Macmillan, 1927), p. 8. 17 S. Shoemaker, “Causality and Properties,” in P. Van Inwagen (ed.), Time and Cause (Dordrecht: Reidel Publishing Co., 1980).

82


new powers (remember: emergent properties are supposed to be “novel”); if they were reducible to the causal powers of the base-level properties, they would have nothing new to contribute to the evolving causal structure of the world. Here is the trouble with emergent causal powers and relations.18 Suppose that an instance of an emergent property, M, causes another emergent property M* to instantiate. This, we might say, is an instance of “samelevel” causation. Now, M*, as an emergent, must have a basal (physical) property P* from which it emerges; M* cannot be instantiated unless some appropriate basal condition, say P*, is present; moreover, the presence of P* by itself guarantees that M* will be instantiated at that time, no matter what has preceded this occurrence of M*. That is, as long as P* is there at the time, M* will be there at the same time whether or not M*’s purported cause, M, had been there at all—unless, that is, M had something to do with P*’s presence at that time. In fact, the only way to save the claim that M caused M* appears to be to say that M caused M* by causing P*. It makes sense to think that in order to bring about an emergent phenomenon, you must bring about an appropriate basal condition from which it will emerge. If pain emerges from neural state N, then to cause pain you must bring about N. (And to extinguish pain you must get rid of N—that’s why we take ibuprofen.) In any case, if these considerations are correct, they show that same-level causation, from M to M*, entails “downward” causation from M to P*. This is downward causation because P* is a property at the base level; if M and M* are mental properties, P* would be a physical/neural property. This means that to understand the possibility of causation between emergent phenomena we must understand the possibility of downward causation. Apart from this, downward causation is of paramount importance to the emergentists. For they want to claim that the emergence of consciousness and rational thought has made a fundamental difference to the world at the physical level. It is because of our emergent mental powers that we have built cities and bridges, sent space probes to Jupiter and Saturn, destroyed rain forests, and burned holes in the ozone layer. With the argument in the preceding section, this shows that downward causation is 18 As readers will notice, what follows is a fairly standard form of the “exclusion” argument. For more details see J. Kim, Mind in a Physical World and Physicalism, Or Something Near Enough (Princeton: Princeton University Press, 2005).


83

fundamental to emergent properties, and that any causal relation that emergent properties enter into presupposes the possibility of downward causation. But can we understand downward causation? In our schematic argument above, we concluded that M causes M* by causing P*. So M causes P*. We notice that M, as an emergent, must itself have a base property; call it P. Now we face a critical question: if an emergent, M, emerges from basal condition P, why doesn’t P displace M as a cause of any supposed effect of M? Why can’t P do all the work in explaining why any alleged effect of M occurred? If causation is understood as nomological (law-based) sufficiency, P, as M’s emergence base, is nomologically sufficient for it, and M, as P*’s cause, is nomologically sufficient for P*. It follows that P is nomologically sufficient for P* and hence qualifies as its cause. The same conclusion follows if causation is understood in terms of counterfactuals—roughly, as a condition without which the effect would not have occurred. Or if we understand causation in terms of control, it seems plausible that the way to bring about an instance of P* is to bring about an instance of P; M, on the other hand, is not something we can bring about unless we bring about its emergence base P. Moreover, it is not possible to view the situation as involving a causal chain from P to P* with M as an intermediate causal link. The reason is that the emergence relation from P to M cannot properly be viewed as causal.19 This appears to make emergent property M otiose and dispensable as a cause of P*; it seems that we can explain the occurrence of P* simply in terms of P, without invoking M at all. If M is to be retained as a cause of P*, a positive argument has to be provided. If M is somehow retained as a cause, we are faced with the highly implausible consequence that every case of downward causation involves causal overdetermination (since P remains a cause of P* as well).20 Moreover, this goes against the spirit of emergentism in any case: emergents are supposed to make distinctive and novel causal contributions. If there is systematic causal overdetermination in all cases of downward causation, 19 Morgan explicitly denies that emergence is a form of causation (Emergent Evolution, p. 28). Moreover, there is little to recommend in the claim that a neural state causes pain and then pain in turn causes, say, my hand withdrawal. How can there be a causal chain from pain to the hand motion that is separate and independent from the physical causal chain from the neural state to the motion of the hand? 20 For an argument that systematic overdetermination is an acceptable option for mental causation, see E. Mills, “Interaction and Overdetermination,” American Philosophical Quarterly 33 (1996): 105–117. Mills does not specifically argue from the perspective of emergentism.

84


that would compromise the emergents’ causal contribution since anything they causally contribute can be, and is, contributed by a physical cause. That is, they do not make causal contributions over and beyond what their underlying physical bases contribute. This result, unless it is successfully rebutted, threatens the very raison d’eˆtre of emergentism. If downward causation goes, so goes emergentism.

V. Concluding Remarks In many ways, the idea of emergence is an attractive, and initially appealing, one, and it is not difficult to understand its popularity. But it is not easy to make the idea precise and give it substantive content. Two important unresolved items remain on the emergentists’ agenda, or wish list. The first is to give emergence a robust positive characterization that goes beyond supervenience and irreducibility. The second is to come face to face with the problem of downward causation. Somehow the emergentist must devise an intelligible and consistent account of how emergent properties can have distinctive causal powers of their own—in particular, powers to influence events and processes at the basal level.

4 “Supervenient and Yet Not Deducible”: Is There a Coherent Concept of Ontological Emergence? I C. Lloyd Morgan, a leading British emergentist of the early 20th century, describes the “emergent evolution” of the world, or how we got where we are and where we are headed from here, in these words: From [the ultimate basal phenomenon, space-time] first emerged ‘matter’ with its primary, and, at a later stage, its secondary qualities. Here new relations, other than those which are spatio-temporal supervene. So far, thus supervenient on spatio-temporal events, we have also physical and chemical events in progressively ascending grades. Later in evolutionary sequence life emerges—a new “quality” of certain material or physicochemical systems with supervenient vital relations hitherto not in being. Here again there are progressively ascending grades. Then within this organic matrix, or some highly differentiated part thereof, already “qualified” . . . by life, there emerges the higher quality of consciousness or mind. Here once more, there are progressively ascending grades. . . . As mental evolution runs its course, there emerge . . . ‘tertiary qualities’— ideals of truth, of beauty, and of the ethically right . . . And beyond this, at or near the apex of the evolutionary pyramid of which space-time is the base, the quality of deity—the highest of all—emerges in us . . . 1 1 C. Lloyd Morgan, Emergent Evolution (London: Williams and Norgate, 1923), pp. 9–10. Here Morgan presents himself as reporting Samuel Alexander’s views in Space, Time, and Deity (London: Macmillan, 1920). But there is little doubt that Morgan accepts the scheme he describes.

86 “ S U P E R V E N I E N T

AND YET NOT DEDUCIBLE”

It is plain that Morgan is presenting “emergent evolution” as an actual history of the world, though perhaps more than a little speculative. The sequential emergence of matter from space-time, life in inorganic systems, and mind from biological processes is claimed as a historical fact about this world. For Morgan, as well as many other emergentists, emergent phenomena like life and mind are genuinely novel features of reality, which make their distinctive causal contributions to the subsequent evolution of the world. There are bridges and building, works of art and electronic gadgets, nuclear bombs and ozone holes, because minds and consciousness have emerged. Emergence is an objective feature of the world, with powers to change, create, and destroy. In the fullness of time, Morgan assures us, if things continue to go right, we humans will achieve divineness.2 In his classic The Mind and Its Place in Nature,3 C.D. Broad distinguished three types of theories concerning biological phenomena: “Biological Mechanism” (what we would now call physical reductionism), “Substantial Vitalism” (Broad had in mind Hans Driesch’s neo-vitalism which posited “entelechies” to account for biological phenomena), and his own “Emergent Vitalism” (which takes biological phenomena as arising from, but not reducible to, physicochemical phenomena). These were clearly intended as three views making contrasting claims about biological organisms and their distinctively biological capacities, functions, and activities. These theories stake out mutually exclusionary positions on the nature of biological entities and processes. Unquestionably, the emergent nature of biological phenomena, for Broad, is an objective fact about them; it does not concern what anyone knows or believes about them. This conception of emergence as an objectively real fact about the world is now standardly called “metaphysical” or “ontological” emergence; some call it “strong” emergence. This is contrasted with an “epistemological,” or “weak,” conception which seems more common among the burgeoning ranks of emergentists on the current scene, especially those from scientific fields. Unlike metaphysical emergence, the epistemological conception focuses on certain supposed epistemic aspects of emergent properties and

2 Most emergentists seem incorrigible optimists; they are blissfully unmindful of the glaring fact that if good things have emerged, so have many unspeakably bad and evil things. 3 C.D. Broad, The Mind and Its Place in Nature (London: Routledge & Kegan Paul, 1925), p. 58.

“SUPERVENIENT


87

phenomena, emphasizing such features as their novelty, unpredictability, and our inability to “calculate,” or “compute,” them from information concerning the basal conditions from which they emerge. The main point then is that we, as cognizers, cannot get there from here—that is, get to higher-level emergent phenomena from information about the lower-level base phenomena. On the metaphysical conception, it isn’t just that emergent phenomena are unpredictable for us; that may be more of a commentary on our cognitive powers than the phenomena themselves. More importantly, the point is that they are objectively new, extra additions to the ontology of the world. Their newness, or novelty, does not consist in their ability to “surprise” us, or our inability to “predict” them; rather, their newness is meant to be metaphysical in import: before these emergents came on the scene, there had been nothing like them in the world and they were net additions to the world’s furniture. And what could be the point of these extra entities if it isn’t their bringing with them new causal powers, powers that go beyond the powers of the lower-level conditions in their emergence base? I don’t know when the distinction between the two types of emergence was explicitly recognized, although of course there have been weak and strong emergence theories and theorists for a long time. In their entry “Emergent Properties” in the Stanford Encyclopedia of Philosophy (http:// plato.stanford.edu/), Timothy O’Connor and Hong Yu Wong say that the emergence concept in J.S. Mill and C.D. Broad was the strong ontological variety, whereas Samuel Alexander’s emergence concept, though still metaphysical, had considerably weaker ontological significance4 (they liken Alexander’s position to contemporary nonreductive physicalism). In his “Weak Emergence”, Mark Bedau introduces an epistemological notion of emergence; a distinction between ontological and epistemological emergence is explicitly drawn in Silberstein and McGeever’s “In Search for Ontological Emergence”. Following Silberstein and McGeever, Van Gulick recognizes a similar distinction between epistemic and metaphysical emergence in his useful 2001 survey article on reduction and emergence. In “Strong and Weak Emergence”, Chalmers’ main theme, unsurprisingly,

4 See also Philip Clayton, “Conceptual Foundations of Emergence Theory,” in The Re-Emergence of Emergence (Oxford: Oxford University Press, 2006), ed. Philip Clayton and Paul Davies.



is the difference between the two kinds of emergence he distinguishes.5 This list is not intended to be exhaustive.6

II We will assume that the metaphysical emergentist would want to include supervenience as a component of his emergence concept; that is, if a property emerges from a set of basal properties, it supervenes on the latter. To put it another way, if the same basal conditions recur, the emergent property will necessarily recur as well (we will return to the question what sort of necessity is involved). C.D. Broad clearly recognized this; he writes: No doubt the properties of silver-chloride are completely determined by those of silver and of chlorine; in the sense that whenever you have a whole composed of these two elements in certain proportions and relations you have something with the characteristic properties of silver-chloride . . . But the law connecting the properties of silverchloride with those of silver and chlorine and with the structure of the compound is, so far as we know, an unique and ultimate law.7 And again: And no amount of knowledge about how the constituents of a living body behave in isolation or in other and non-living wholes might suffice to enable us to predict the characteristic behavior of a living organism. This possibility is perfectly compatible with the view that the characteristic behaviour of a living body is completely determined by the nature and arrangement of the chemical compounds which 5 Mark Bedau, “Weak Emergence,” Philosophical Perspectives 11 (1997): 375–399; Michael Silberstein and John McGeever, “In Search for Ontological Emergence,” Philosophical Quarterly 49 (1999): 182–200; Robert Van Gulick, “Reduction, Emergence and Other Recent Options on the Mind-Body Problem: A Philosophical Overview,” Journal of Consciousness Studies 8 (2001): 1–34; David J. Chalmers, “Strong and Weak Emergence,” in The Re-Emergence of Emergence, ed. Philip Clayton and Paul Davies. 6 Carl Gillett distinguishes three concepts of emergence, “strong,” “weak,” and “ontological,” in “The Varieties of Emergence: Their Purposes, Obligations and Importance,” Grazer Philosophische Studien 65 (2002): 95–121. All of these seem to be metaphysical notions of emergence in our sense, and Gillett’s distinction does not straightforwardly map onto the standard ontological/epistemological distinction. 7 The Mind and Its Place in Nature, pp. 64–65.

“SUPERVENIENT


89

compose it, in the sense that any whole which is composed of such compounds in such an arrangement will show vital behaviour . . . 8 Supervenience, or “upward determination,” may ultimately turn out to be detrimental to the emergence program; and yet, without supervenience, it would be difficult to make sense of the “from” in “property P emerges from basal conditions C,”9 David Chalmers says that an emergent phenomenon “arises (in some sense) from”10 lower-level phenomena, but he says nothing further about just what sense of “arising from” is involved here. Supervenience supplies a clear and robust sense to Chalmers’ “arises from.” We can explore whether some weaker relation might be able to serve the purpose,11 but it is clear that if the emergent phenomenon occurs randomly when its purported basal conditions are realized, it would be difficult to give a sense to the claim that the phenomenon emerges “from” these conditions, or that the lower-level phenomena are its “basal” conditions. It is clearly not enough to say that these basal conditions are necessary for the occurrence of an emergent. This would allow cases of emergence to proliferate beyond what even the most lavish and bountiful emergentist would want. In any case, as we will see, major early emergentists accepted the supervenience of the emergents as a condition of emergence.12

8 Ibid. 67–68. 9 For more on this, see my “Emergence: Core Ideas and Issues,” Synthese 151 (2006): 547–559. Essay 3 of this volume. In his “From Supervenience to Superdupervenience” (Mind 102 (1993): 555–586), Terence Horgan has an interesting footnote (n. 7) in which he quotes Arthur Lovejoy as countenancing a form of emergence which excludes supervenience. 10 “Strong and Weak Emergence,” p. 244. 11 The idea of probabilistic supervenience is clearly coherent and deserves consideration. But we will presumably need stable lawlike probabilities grounded in the laws at the basal level. Details of this approach need to be worked out. 12 It is interesting to note that several current philosophical advocates of emergence reject supervenience, for apparently different reasons, as a component of emergence; see, e.g., Paul Humphreys, “How Properties Emerge,” Philosophy of Science 64 (1997): 1–17; Timothy O’Connor and Hong Yu Wong, “The Metaphysics of Emergence,” Nouˆs 39 (2005): 658–678; Michael Silberstein, “In Defense of Ontological Emergence and Mental Causation,” in The Re-Emergence of Emergence. In the current postclassical period of neo-emergentism, the idea of emergence has become very fluid, plastic, and variegated. “Emergence” is a term of art, and one is free to define it as one wishes, the only constraint being that the resulting concept is a philosophically or scientifically useful one. However, in rejecting supervenience, these writers are making a radical (and, in my view, unwise) departure from the core concept held by the classic British emergentists like C.D. Broad and C. Lloyd Morgan. I believe that one should stay at least in the vicinity of these writers in order to justify the claim that they are dealing with emergence.



What else do we need to characterize ontological emergence? Some writers, for example, Silberstein,13 cite the capacity for “downward causation”—that is, the power to causally affect the events at the basal level—as a condition of ontological emergence. I think this is not a wise move; downward causation is a highly controversial and problematic issue, and building it into the very concept of emergence will make it more difficult to defend the claim that there are any real cases of emergence. A better course would be to characterize emergence, or ontological emergence, in less contentious terms, postponing the issue of downward causation to be threshed out another day. The most important and widely used strategy, which goes back to Morgan and Broad, has been to add the condition that an emergent, even though supervenient on, and determined by, its basal conditions, is not deducible from them. When Broad first introduces the idea of emergence in The Mind and Its Place in Nature, this is what he says: Put in abstract terms the emergence theory asserts that there are certain wholes, composed (say) of constituents A, B, and C in relation R to each other; that all wholes composed of constituents of the same kind as A, B, and C in relations of the same kind as R have certain characteristic properties; that A, B, and C are capable of occurring in other kinds of complex where the relation is not of the same kind as R; and that the characteristic properties of the whole R(A, B, C) cannot, even in theory, be deduced from the most complete knowledge of the properties of A, B, and C in isolation or in other wholes which are not of the form R(A, B, C). The mechanistic theory rejects the last clause of this assertion.14 So the difference between emergentism and its principal rival, mechanism, consists precisely in that, on mechanism, “the characteristic properties” of a whole are deducible from the facts about its parts and their relationships, whereas emergentism holds them to be not so deducible. Or, to put it another way, let F be a property of a whole which is determined by, or supervenient on, properties and structural relations characterizing its constituents: F is emergent if and only if F is not deducible from these facts about its constituents.15 13 Michael Silberstein, “In Defense of Ontological Emergence and Mental Causation,” p. 203. 14 p. 61. Emphasis added. 15 In speaking of “deducing” properties we follow the usual practice. When we say “F is deducible from conditions C” what is meant is that the fact, or proposition, that something has F, or that F is instantiated in something, is deducible from the proposition that conditions C hold for that thing. Similarly, when we speak about deducing a property from other properties.

“SUPERVENIENT


91

Emergentism, in its broadest form, would be the claim that there are properties like F, while mechanism, or reductionism, would deny that such exist. (There are also specific emergentist theses concerning selected classes of phenomena; notably, biological phenomena in relation to physicochemical phenomena, and mentality and consciousness vis-a`-vis the domain of the neural/biological sciences.) It is widely assumed that there is an important connection between logical deduction on one hand and explanation and reduction on the other. Thus, we might add: F is mechanistically explainable, or reducible, in case F is deducible from facts about the constituents of a whole. In the quote above, Broad speaks of “the most complete knowledge” of a whole’s constituents, but this seemingly epistemological aspect of his characterization is easily eliminated: instead of “the most complete knowledge,” we can refer to “all the facts” or “a complete set of truths.” This replacement would be appropriate because whether or not anyone “knows” these facts about the constituents is irrelevant. Further, we should understand “all the facts,” or “the most complete knowledge,” to include all the laws operative at the basal level. So, when we say biological phenomena are emergent from physicochemical phenomena, the latter is understood to include physicochemical laws as well as individual facts at this level. In any case, a property’s nondeducibility from base-level facts is, on Broad’s characterization, what separates emergent from nonemergent properties. Broad is not alone here. When C. Lloyd Morgan talked about nonemergent properties as “additive and subtractive only, and predictable,”16 he is naturally taken to be referring to something similar to deducibility. In a paper published in 1926, shortly after Morgan’s Emergent Evolution and Broad’s The Mind and Its Place in Nature, Stephen C. Pepper describes emergentism as follows: The theory of emergence involves three propositions: (1) that there are levels of existence defined in terms of degree of integration; (2) that there are marks which distinguish these levels from one another over and above the degree of integration; (3) that it is impossible to deduce the marks of a higher level from those of a lower level . . . 17

16 Emergent Evolution, p. 3. 17 “Emergence,” Journal of Philosophy 23 (1926): 241–245. (The quote is from p. 241; emphasis added.)



Later writers who have invoked nondeducibility as the pivotal criterion of emergence include Carl G. Hempel, Ernest Nagel, James Van Cleve, and David Chalmers.18 Broad, like many other writers, often resorts to epistemological terms to explain emergence. We have already noted this in regard to Broad’s statement that “no amount of knowledge” about the micro-constitution of a living thing is sufficient for the “prediction” of the biological features of the organism. Similarly, an emergentist about consciousness would sometimes put his claim by saying that a complete physical, physiological, and computational knowledge of our brain at a given time does not suffice to give us any knowledge about our consciousness—whether or not we are conscious at the time and if we are, what sort of consciousness is being experienced. There is also Frank Jackson’s Mary,19 the superstar vision scientist confined to a black-and-white room: we are to suppose that she has complete physical information about the physical/neural processes involved in the workings of our visual systems but, before her release from the room, she has no knowledge of color qualia. One could take such talk as referring to an epistemological conception of emergence, but I believe that would be premature. I think that the epistemological relationships being talked about are best explained in terms of deducibility. If, as Broad thought, physicochemical knowledge of an organism doesn’t yield knowledge of its biology, that would be so because biological truths are not deducible from physicochemical truths. If only truths about visual qualia were deducible from physical/neural truths, Mary could know, before her release, what it would be like to see a ripe tomato. In this way, metaphysical emergence characterized in terms of nondeducibility would appear to offer an explanation of epistemological emergence. We will soon see that the notion of deducibility is itself fraught with problems but, at least on a first pass, it seems like just what we need to characterize metaphysical emergence. Deducibility can fail on two levels. First, there is the idea that the novelty, or surprising character, of an emergent property is beyond our conceptual grasp before the property actually makes its appearance and we have a 18 Carl G. Hempel and Paul Oppenheim, “Studies in the Logic of Explanation,” Philosophy of Science 15 (1948): 135–175. Ernest Nagel, The Structure of Science (New York: Harcourt, Brace, and World, 1961). James Van Cleve, “Mind-Dust or Magic? Panpsychism versus Emergence,” Philosophical Perspectives 4 (1990): 215–226. Chalmers, “Strong and Weak Supervenience.” 19 “Epiphenomenal Qualia,” Philosophical Quarterly 32 (1982): 126–136.

“SUPERVENIENT


93

chance to observe or experience it. The pre-release Mary seems often taken to be in a situation of that sort in regard to color qualia (though I don’t believe she has to be, for Jackson’s purposes). According to Thomas Nagel,20 our epistemic position vis-a`-vis the experiences of a bat is precisely like that: we have no idea, no conception, of what a bat’s phenomenal experiences are like, and as a result we cannot even entertain propositions about their qualitative character. This would mean that we don’t even know just what propositions we should try to deduce about bat phenomenology from truths about bat physiology. Second, deducibility can fail even though we know what the supposed emergent properties are and know what propositions are to be considered for deduction. I believe a situation of this kind is what figures primarily in the emergentists’ claim that propositions involving emergent properties are not deducible from truths about the base-level processes. This is the sort of situation Broad considers in regard to the deducibility of biological truths from physicochemical truths. The availability of biological concepts is not at issue. To sum up, then, deducibility, or the absence thereof, is the key to the standard conception of metaphysical emergence. Properties of a whole are emergent just in case they are not deducible from properties and relations characterizing its constituent parts, even though they are determined by and supervenient on them.

III So then, do we now have a properly characterized ontological concept of emergence, a conception that makes emergence something objective in the world, not a phenomenon that has to do with our cognitive resources and powers? We have reached a concept that takes nondeducibility as the mark that distinguishes emergent properties from the rest. Now the question arises: Is deducibility, or nondeducibility, itself a wholly nonepistemic concept? If biological properties are emergent from physicochemical properties, we cannot deduce truths involving the former from those that only involve the latter—that is, we cannot deduce biological truths from

20 “What Is It Like to Be a Bat?”, Philosophical Review 83 (1974): 435–450.



physicochemical truths. But whom does this “we” refer to? Who is doing the deduction? How adept a logician is Jackson’s Mary supposed to be? No human person, we may assume, has unlimited logical powers. Although “we” cannot deduce biological truths from physical truths, why couldn’t a cognizer with vastly greater logical and mathematical powers produce the required deductions? Mustn’t we fix the level of deductive or logical competence we have in mind to give a clear meaning to “deducible”? If so, is there a right level to pick, and what makes it right? But, more importantly, don’t these questions show that the idea of deducibility, or nondeducibility, threatens to turn into an epistemic notion, making Broad’s emergence epistemic rather than metaphysical? Think about how we go about making deductions—how we reason from premises to a conclusion in practice. I believe we have something like the following picture in mind: we start off with a list of premises, and proceed from there, step by step, where each step is seen as obviously and directly implied by selected earlier steps (in the best cases, in accordance with simple formal rules known to guarantee implication, like modus ponens), and, with luck, finally reach the proposition to be deduced. In his Rules for the Direction of the Mind, Descartes described deduction as “a continuous and uninterrupted movement of thought in which each individual proposition is clearly intuited.”21 Further, he says, “the self-evidence and certainty of intuition is required not only for apprehending single propositions, but also for any train of reasoning whatever.”22 More recently, Gilbert Harman has given a similar characterization of reasoning. According to him, we have a fundamental disposition, or power, to recognize “immediate implications” and “immediate inconsistencies,” and this capacity is what guides us through reasoning processes. Harman recognizes that the idea is relative to individual cognizers, saying “I suggested that certain implications and inconsistencies are ‘immediate’ for a given person.”23 It is clear that the problem with this picture of deduction, or deducibility, is that it makes the notion of deducibility relative to the cognitive powers and dispositions of the deducers and thereby makes the concept of 21 The Philosophical Writings of Descartes, vol. 1, tr. John Cottingham, Robert Stoothoff, and Dugald Murdoch (Cambridge: Cambridge University Press, 1985), p. 15. 22 Ibid. pp. 14–15. 23 Change in View (Cambridge, Mass.: MIT Press, 1986), p. 19.

“SUPERVENIENT


95

emergence defined in its terms both epistemic and relative, whereas what we are seeking is an objective, ontological conception. To his credit, Broad was well aware of this problem. Observant readers have surely noticed that in the last quoted passage above, he says that the emergent properties of a whole “cannot, even in theory, be deduced” (emphasis added) from those of its constituents. Evidently in a similar spirit, Chalmers writes that “truths concerning [emergent phenomena] are not deducible even in principle” from truths about lower-level phenomena.24 But what does this mean? Here is where Broad summons his “mathematical archangel”: If the emergent theory of chemical compounds be true, a mathematical archangel, gifted with the further power of perceiving the microscopic structure of atoms as easily as we can perceive hay-stacks, could no more predict the behavior of silver or of chloride or the properties of silver-chloride without having observed samples of those substances than we can at present. And he could no more deduce the rest of the properties of a chemical element or compound from a selection of its properties than we can.25 The mathematical archangel,26 we may presume, is logically and mathematically omniscient. If a proposition is deducible, “in principle” or “in theory,” from a set of premises, it will know that it is, and be able to construct a step-by-step proof. If the archangel cannot produce a proof, it’s because there is no proof and the proposition is not deducible, in an absolute sense, from the premises. Thus, if a property is emergent from a set of basal conditions, there is no deduction of it from those conditions, and this has nothing to do with the epistemic powers of any cognizers. This idealization of deduction is Broad’s attempt to purge any epistemic and relativistic aspects from the notion of deduction and thereby objectify deducibility, or nondeducibility. For him, there being no deduction of an emergent property from its basal conditions is not an epistemological fact. It is not

24 “Strong and Weak Emergence,” p. 244. 25 The Mind and Its Place in Nature, p. 71. 26 Achim Stephan calls the mathematical archangel “a colleague of the Laplacian demon” in his “Emergence – A Systematic View on its Historical Facets,” in Emergence or Reduction?, ed. Ansgar Beckermann, Hans Flohr, and Jaegwon Kim (Berlin: de Gruyter, 1992). For interesting discussion of some issues formulated in terms of Laplace’s demon that are relevant to our concerns in this paper, see Terence Horgan, “Supervenience and Cosmic Hermeneutics,” Southern Journal of Philosophy 22, Spindel Supplement (1984): 19–38.



because we are not smart enough, or don’t have enough time or inclination, that we cannot deduce, say, geological truths from the truths of macroeconomics, or facts about the surface composition of the moon from facts about neurotransmitters in the human brain; not even the logically all-powerful archangel can do that, and that is because there are no deductions between these sets of truths. This surely seems like an objective fact about the relationships between sets of truths, or facts. Broad’s idealization strategy appears to remove from the notion of deducibility an apparent epistemic relativity, and the characterization of emergence as supervenience plus nondeducibility appears to stand as an ontological conception of emergence. At least, so it may seem at this point.

IV There remains, however, one further issue to deal with: deduction or proof makes sense only relative a specific set of rules of inference, or a proof system, which specifies permissible transformations of sentences in constructing proofs. That is to say, a sequence of sentences is a proof only relative to a system of deduction. Should we say, following Descartes and Harman, that the relevant system must include only those rules that give us “immediate implications”? No; that would bring back epistemic relativity. Evidently, what the mathematical archangel sees when it recognizes a sentence as “deducible” from a given set of premises is a proof in some proof system. But what proof system does the archangel use? Does it matter? The answer is that it does matter—and matters very much. The proof system must be a correct system in the following sense: If there is a proof of Q from P1, . . . , Pn, then P1, . . . , Pn must logically imply, or entail, Q. That is, the premises of a proof in a correct system must logically guarantee the truth of the conclusion proved. We are assuming that the language in which the issues of emergence are considered is provided with a semantics and that notions like validity and implication are available for sentences of the language. What matters from a metaphysical point of view is semantics, more specifically logical implication, not syntax; proofs are relevant only if they are proofs in a correct system, and that is so because that guarantees these proofs to have a right semantic property, that of preserving truth from premises to conclusions. Does it matter which correct proof system is used

“SUPERVENIENT


97

by the archangel? Some proof systems are more intuitive and perspicacious than others; the usual deduction systems we find in logic textbooks are formulated, we may assume, with an eye toward simplicity and perspicuity— that is, for the typical student. But that is epistemology; the archangel is hardly a typical student, and any system is as perspicuous to it as any other. Moreover, nothing needs to prevent it from working with a formal system in which all semantically valid sentences are taken as axiom; it can intuitively ascertain all such sentences, and all proofs in the system will consist of a single line, namely the sentence being proved! What this shows is that the important thing is semantics (logical implication), not syntax (deducibility). There also is a formal reason why for first-order logic (with identity), arguably the core of what we call “logic,” the choice of a system does not matter—because this is a “complete” system, that is, there are formalizations of this logic that are complete. A proof system is complete in the technical sense just in case there is a proof of a sentence from a set of premises if and only if the premises imply, or entail, that sentence. In consequence, it doesn’t matter which one of these complete deductive systems we use to define deducibility; for they are all equivalent in that something is provable in one system from a given set of premises if and only if it is so provable in each of the rest. The only thing that matters is the fact that there being a proof guarantees logical implication. And if there is no proof in the system, then there is no logical implication; that is, nondeducibility amounts to the absence of implication in languages whose logic is amenable to complete formalization. So the relativity of proofs to proof systems cancels itself out as cause for concern for us. Does this mean that we are finally home free with Broad’s metaphysical conception of emergence as supervenience plus nondeducibility? Unfortunately, the answer is no, at least not yet; in truth, we are now in far worse trouble than before. For look where we are: we have just seen that in speaking of deducibility and nondeducibility, what matters turns out to be logical implication, or entailment, or the absence thereof. Deducibility in a complete system of proof, or deducibility as the mathematical archangel sees it, is nothing but logical entailment; similarly, nondeducibility amounts to logical nonentailment. With this in mind, look again at Broad’s attempt to combine supervenience with nondeducibility to obtain a metaphysical relation of emergence. An emergent property supervenes on basal-level



conditions (including basal-level laws); this means that certain basal conditions, namely those that constitute a supervenience base for it, necessitate, or entail, the emergent property. We have just seen that the net effect of nondeducibility comes to absence of logical necessitation. Consequently, metaphysical emergence, as conceived by Broad and others, comes to this: Emergent properties are entailed but not logically entailed by their basal conditions. Or, to put it another way, emergent properties supervene, but do not logically supervene, on their basal conditions. On the face of it, this may seem like a coherent conception, because we think that logical supervenience is not the only kind of supervenience. There is another recognized variety that is weaker than logical supervenience, what is called “nomological,” or “natural,” supervenience: the bases of supervenient properties do not logically entail them but when combined with prevailing laws of nature, the entailment holds. Thus, the necessitation is nomological, not logical or metaphysical. As nomological supervenience appears to be the only alternative to logical supervenience, the Broad-style conception of emergence has now taken the following form: Emergent properties supervene nomologically, but not logically, on their basal conditions. And this may seem to fit in well with certain current forms of dualism, in particular, Chalmers’ “naturalistic dualism,” which consists in the claim that consciousness is naturally (that is, nomologically), but not logically, supervenient on physical phenomena.27 On the present construal, therefore, the classical British emergentism of Broad and others might seem to be an almost exact anticipation of Chalmers’ naturalistic dualism. But not so fast! We must at this point look a bit deeper into the nature of nomological supervenience that may be involved in the Broad-style metaphysical emergence. When we refer to the “basal conditions” of an emergent property, we have noted more than once that these conditions are taken to include not only particular facts—events, states, and processes—at the base level but also laws operative at that level. So as regards the emergence of mental properties on biological/physical properties, the appropriate basal conditions include all biological/physical laws—all laws applicable to biological, neurological, and physicochemical systems and phenomena. 27 The Conscious Mind (New York and Oxford: Oxford University Press, 1996). Chalmers says: “In general, B-properties supervene naturally on A-properties if any two naturally possible situations with the same A-properties have the same B-properties,” where a situation is naturally possible if “it could actually occur without violating any natural laws,” p. 36. On naturalistic dualism, see pp. 168ff.

“SUPERVENIENT


99

Recall the iconic emergentist question: Knowing all about the biological, neural, and physicochemical facts about the brain, can we predict, or know, what conscious experience, if any, will be present in that brain? Obviously, “knowing all about what goes on at the neural level” should be taken to include knowing the laws that hold at that level.28 This means that when we say that an emergent property is not logically implied, or necessitated, by its basal conditions, the latter include not only particular basal facts but also all basal laws, laws at the base level. So in a sense the logical supervenience we are talking about is a form of nomological supervenience, where the laws involved are base-level laws (we may assume that these laws include all laws holding at still lower levels). It is crucially important to keep in mind that the laws assumed to be included in the basal conditions are not all laws of nature; they are only laws that hold at the basal level.29 Thus, when we say emergent properties are not logically but only nomologically necessitated by the basal conditions, the additional laws needed to yield the necessitation are not base-level laws (we have them already) but laws that connect the emergents with specific basal conditions, namely those that Broad calls “trans-ordinal laws,” laws connecting events and states at different levels, or “orders.” These are also sometimes called “laws of emergence” or “supervenience laws,”30 and take the following form: When conditions C at the base level hold for system S at t, S instantiates, at t, emergent property E. These laws are the auxiliary premises we need to deduce statements about facts involving emergent properties from statements about their basal conditions. Trans-ordinal laws, like the bridge laws in Nagelian theory reduction,31 are not exclusively about the base level; they concern the relationships between the base-level phenomena and the phenomena at a higher level. In the case of the mind-body relation, these are psychophysical laws telling us under what neural/biological conditions, a given type of conscious experience occurs. On emergentism, every conscious state will be connected by 28 Chalmers’ notion of logical supervenience on physical facts is similar; the physical facts are stipulated to include all physical laws. See The Conscious Mind, p. 33. 29 I believe it should be assumed—and this is what I assume—that laws at a given level include all laws that hold at the lower levels as well. Thus, for example, laws at the biological level include all physical and chemical laws. 30 Chalmers’ term in The Conscious Mind, p. 127. 31 See Ernest Nagel, The Structure of Science (New York: Harcourt, Brace, and World, 1961), chap. 11.

100

“SUPERVENIENT


such an inter-level law with its underlying neural basal conditions; and there can be multiple emergence bases for a given type of conscious state. As these laws involve higher-level phenomena, they cannot be part of the basal conditions from which the deducibility of an emergent is considered. These considerations bring to light a deep difficulty—in fact, what may well be an incoherence—in Broad’s characterization of emergence as supervenience plus nondeducibility. For this combination threatens to turn into an outright contradiction: supervenience says that basal conditions entail, or imply, the emergent phenomenon; however, nondeducibility, which, as we saw, comes to nonentailment, or nonimplication, says that the basal conditions do not entail or imply the emergent phenomenon. Or equivalently, the emergent phenomenon both supervenes and does not supervene on its basal conditions. We tried to defuse this potentially disastrous situation by construing the combination as nomological supervenience plus the denial of logical supervenience; that is, emergent properties are nomologically, but not logically, supervenient on their basal conditions. We can now see that this rescue strategy fails. The reason is that when Broad claimed that an emergent is determined by, or supervenes on, its basal conditions, nomological determination or supervenience is not what he had in mind; rather, he was talking about logical supervenience. Broad well knew that trivially, emergents nomologically supervene on basal conditions, since nomological supervenience is modulo all laws of nature including his trans-ordinal laws. Look at an earlier quote from Broad again: No doubt the properties of silver-chloride are completely determined by those of silver and of chlorine; in the sense that whenever you have a whole composed of these two elements in certain proportions and relations you have something with the characteristic properties of silver-chloride . . . But connecting the properties of silver-chloride with those of silver and chlorine and with the structure of the compound is, so far as we know, an unique and ultimate law.32 What Broad is claiming is that when we have fixed the base-level conditions and laws, that determines what phenomena will, or will not, emerge, even though the latter are not deducible from the former. So an emergent property supervenes on its basal conditions plus basal laws but is 32 The Mind and Its Place in Nature, pp. 64–65.

“SUPERVENIENT


101

not deducible from them. It is crucial here that the supervenience base and deduction base are held identical. When we speak about nomological supervenience, we have in mind all laws prevailing in this world, which will include Broad’s trans-ordinal laws, or Chalmers’ “laws of supervenience,” or Nagel’s “bridge laws”—that is, laws connecting phenomena at different levels. Broad will cheerfully admit that if trans-ordinal laws are admitted as part of the deduction base (that is, among the premises), emergent properties or phenomena are easily deduced from the basal conditions. His point is that when we are limited to base-level laws and conditions, no deduction is possible, since there is none. And yet he maintains that the emergents supervene on, or are determined by, the base-level phenomena and laws. And this supervenience is logical or metaphysical supervenience. There is a nomological aspect to that because the supervenience base includes basal laws. But once these laws are considered as part of the base, the supervenience relation becomes the logical/metaphysical variety. This apparently puts Broad and his like-minded colleagues in an untenable position. He has been reduced to saying that emergent properties logically supervene on basal conditions and laws and yet they are not deducible from them. To repeat, nondeducibility, on Broad’s idealization, turns into the absence of logical entailment. So Broad’s conception of emergence turns into an apparent incoherence: logically supervenient on basal facts but not logically entailed by them—that is, to be up front, both logically supervenient and not logically supervenient on basal facts! I think the source of the problem is rather obvious: to avoid an epistemic interpretation of “deducible,” Broad tries to idealize it in terms of the mathematical archangel, as Laplace tried with his logically all-powerful demon (who, like the mathematical archangel, is also microphysically omniscient), but idealized deducibility turns into entailment, contradicting the first component of his metaphysical emergence, that is, supervenience. Is there a way out for Broad from this apparently incoherent situation?

V We have been assuming that the language in which emergence issues are considered is a first-order language with a complete proof system. Without



completeness, we cannot equate nondeducibility with nonentailment, or the absence of implication. So what about languages for which there are no complete proof systems? Elementary number theory, or arithmetic, is famously incomplete; that is, there is no consistent formalization of number theory in which all and only number-theoretic truths are provable. Surely, it might be said, number theory must be taken to be part of the deductive system when we consider the deducibility of emergent phenomena. However, it doesn’t seem to me that the emergentist can exploit the incompleteness of arithmetic to dispute our claim that nondeducibility boils down to nonentailment. The reason is that we can happily let the emergentist add all mathematical truths (including of course arithmetic truths) to her deductive system and ask her whether this would help her deduce the emergents from their basal conditions. I believe the emergentist would have to say no. After all, the mathematical archangel is mathematically omniscient and it has at its disposal all mathematical truths. Its deductive system has to be complete. If a truth is not deducible from another in such a system, it must be because the former is not implied by the latter. There is another concern we should address briefly. One might point out that “x is a male” is not deducible in first-order logic from “x is a bachelor” and yet obviously “x is a bachelor” logically implies “x is a male,” The reply is that definitions are free in deductions and don’t count as additional premises; one is entitled to use them at any point in a proof. So the deduction goes through with the definition “x is a bachelor iff x is an unmarried adult male.” We can go further and grant the same privilege to all conceptual, or analytic, truths; these can be taken as additional premises at any point in a proof. But this may not put away the concern entirely. For one might continue: Even so, “x is water” is not deducible from “x is H2O” since there is no conceptual definition linking “water” and “H2O.” And yet the proposition that Lake Michigan is filled with H2O entails that it is filled with water. Therefore, we cannot say that nondeducibility, even with the proviso concerning definitions, amounts to nonentailment. This point touches on various contested issues currently debated in philosophy of language and metaphysics—issues arising from (supposed) cases of a posteriori necessities and entailments. My reply is that “Lake Michigan is filled with water” is deducible, in an appropriate way, from “Lake Michigan is filled with H2O”, and that therefore this case is not a counterexample. I will simply state this

“SUPERVENIENT


103

response without detailed explanation or justification. This response is based on the views associated with a priori physicalism.33 True, there are no direct conceptual links between “water” and “H2O,” but this does not mean that appropriate deductive links cannot be forged between statements about water and statements about H2O. What do we mean by “water”? We reply: it means something like “the local watery stuff,” where “watery” is short for a conjunction of predicates designating the observable properties by which we ordinarily recognize and identify samples of water—properties like transparency, odorlessness, water’s characteristic viscosity (the way it “flows”), its ability to quench thirst and extinguish flames, and so on. We take that as a conceptual fact grounded in meanings. Now consider the following array of statements; (i) (ii) (iii) (iv)

Lake Michigan is filled with H2O. The local watery stuff ¼ H2O. Water ¼ the local watery stuff. Therefore, Lake Michigan is filled with water.

We claim that (ii) is deducible, or obtainable at any rate, from physical facts—physical laws as well as particular facts. Physical theory, we may assume, can show that the local stuff that is transparent, odorless, and so on is made up of H2O molecules; and it can explain why quantities of H2O molecules behave in the way water behaves. Step (iii) is a meaning-based conceptual truth, and (iv) then follows from (i) and (iii). Note that the deduction makes use of an additional premise, (ii), obtained from physical theory. This is appropriate since, as we have repeatedly noted, the deduction base in the emergence debate is taken to include base-level facts, including laws, and in this case these are physical facts. A full reply would have to consider whether a strategy like this can be applied to other cases of supposed a posteriori necessities, what the upshot would be if it were applied to mental states, in particular qualitative conscious states, and other questions. There clearly remain issues to be discussed. So I acknowledge that my arguments are not fully conclusive. 33 See David Chalmers and Frank Jackson, “Conceptual Analysis and Reductive Explanation,” Philosophical Review 110 (2001): 315–360; Frank Jackson, “A Priori Physicalism,” in Contemporary Debates in Philosophy of Mind, ed. Brian McLaughlin and Jonathan Cohen (Oxford: Blackwell, 2007). For views critical of this approach, see Ned Block and Robert Stalnaker, “Conceptual Analysis, Dualism, and the Explanatory Gap,” Philosophical Review 108 (1999): 1–46; Brian McLaughlin, “On the Limits of A Priori Physicalism,” in Contemporary Debates in Philosophy of Mind, ed. Brian McLaughlin and Jonathan Cohen.



But the emergentists need to ponder whether a posteriori necessities are something they can exploit to their advantage in the present context. My conjecture is that the answer is no.34

VI It seems, then, that our provisional conclusion regarding the incoherence of the standard characterization of ontological emergence stands. The pressing question is, What is to be done? How might we construct a viable metaphysical notion of emergence? This is a question not only for the friends of emergence but also for those who, although not enamored of emergence, are inclined to consider emergentism an empirically significant thesis, though perhaps not true. Some recent writers, including O’Connor, Humphreys,35 and Silberstein, have urged that supervenience be given up as a component of emergence. I do not believe this is a wise move; without supervenience, there is nothing on the table and we start from scratch. Perhaps, one should take seriously Silberstein’s suggestion that downward causal power should take center stage in developing a metaphysical concept of emergence. There are also proposals by O’Connor, Humphreys, and others. My own impression is that none of them give us what is needed—at least, their promises have yet to be proved. A wiser course at this point might be to abandon the search for metaphysical, or ontological, emergence, and be satisfied with an epistemic concept. The problem of course is that epistemic emergence does not serve as a vehicle for stating a metaphysical worldview of the sort that those who champion emergence would want to defend. Moreover, epistemic emergence, unlike metaphysical emergence, seems like a rather unremarkable, platitudinous idea, not something to get philosophically excited about. But it may well be that in our search for a coherent concept of metaphysical emergence, we are looking for something that does not exist. 34 It is worth noting that, for emergentists, many of these a posteriori necessities have the same status as trans-ordinal laws of emergence, in that they connect items at two levels, the basal and emergent levels. 35 Paul Humphreys, “How Properties Emerge,” Philosophy of Science 64 (1997): 1–17.

5 Reasons and the First Person I Late one night, you find yourself in the kitchen, holding the refrigerator door open and peeking inside, but you have forgotten what it is that you are looking for. Is it a piece of cheese, or some orange juice, or perhaps a bottle of beer that you wanted out of the fridge? Perhaps you wanted to check on the supply of milk and eggs. You are befuddled, and feel foolish—even a bit helpless. You feel sort of frozen, as if in a paralysis, in the middle of an action. Here is another such episode: As you are making your way down the stairway from your study, you suddenly realize that you have no idea why you are going downstairs. Are you getting the day’s mail? Are you going to pick up the evening paper? To take a break from your work and listen to some music? To feed the cat? You aren’t sure, and your steps slow down— perhaps you will come to a complete stop. Perhaps, you will continue to proceed downstairs, hoping that you will remember your reason for going down. Or you see no point in continuing your descent and decide to return to your study.1 Experiences like these, I believe, are not uncommon. In such cases, we are doing a certain thing—holding open the fridge door, looking for something in the fridge, and going downstairs—and also know that we are doing it. What we don’t know is why we are doing what we are doing. Because we lack this knowledge, we don’t know, in an important sense, what it is that we are doing. Am I looking for some cheese or checking on This paper derives in part from an earlier paper, “Self-Understanding and Rationalizing Explanations,” Philosophia Naturalis 21 (1984): 309–320. I am indebted to John Bellwoar, Josep Corbi, and David Henderson for helpful comments on earlier versions. 1 David Velleman discusses such cases, in detail and with subtlety, in his Practical Reflection (Princeton: Princeton University Press, 1989).

106

REASONS AND THE FIRST PERSON

the supply of milk? Am I getting the mail? And because of this lack of knowledge, I may even feel a bit estranged from my own action, no longer in control of it as its agent; it’s as though I am viewing myself at a distance, somewhat uncomprehendingly. I do not understand what I am doing, and I look at my action as though it is someone else’s, trying to figure it out, the way I would look at my wife’s when I see her opening the fridge door (“What is she looking for?”) or hear her footsteps down the stairway (“What is she doing so late at night?”). One way in which I can come to understand what I am doing is to know an appropriate want or desire that underlies my action: I want to get some cheese, and that’s why I am looking inside the fridge; I want to check the mail, and that’s why I am going downstairs. As we know, there is a tacit belief element here, too: it’s no accident that, given my desire for cheese, I have opened the fridge, not the dishwasher. It’s because of my belief that cheese is to be found in the fridge, not in the dishwasher. Let us adopt Davidson’s terminology of “primary reason,” in the following sense: a person S’s wanting x and believing that doing A is a way of securing X is a primary reason for S to do A.2 We can say this then: (1) One way in which we understand why we are doing something is to know a primary reason for which we are doing it. So a primary reason is a coordinated pair of a desire and a belief—a desire for something and the belief that doing a certain thing is an acceptable way of realizing that desire. Knowledge of primary reasons is not the only way to understand one’s own action. I yelled at the doorman because I thought he insulted my wife; I threw the sheaf of paper at the computer because I couldn’t get the thing to run all day. But it is also clear, although I will not argue the point, that primary reasons, in the present sense, have a special and central place in our understanding of actions. This is reflected in the fact that the relationship between a primary reason and an action for which it is a primary reason mirrors practical inference—reasoning from goals and means-ends relations to actions to be undertaken. Primary reasons “rationalize” actions in the sense that they provide agents with reasons for undertaking the actions for which they are reasons. And we understand our actions by knowing the primary reasons that rationalize them. 2 Donald Davidson, “Actions, Reasons, and Causes,” Journal of Philosophy 60 (1967); reprinted in his Essays on Actions and Events (Oxford and New York: Oxford University Press, 1980).


107

Experiences like your standing in front of an open refrigerator wondering what has brought you there are fairly common, as I said. They are common in that all of us—most of us, at any rate—have had experiences of this kind, and know what it’s like to have them. (The naturalness of the “what it’s like” locution here suggests the presence of a distinctive phenomenology in experiences of this kind.) But these experiences are definitely not common in the sense of being frequent occurrences in our daily lives. They are commonly known but comparatively rare. If episodes like these happened to you, say, dozens of times a day, you would think that something strange was happening to you, that you might be losing your mind. Worse still, suppose that this happened as a routine matter, that most of your actions are incomprehensible to you in this way. In such a situation it would be difficult for you to keep any confidence in yourself as an agent, someone who can deliberate and form intentions, and effectively execute an action plan. Your conception of yourself as an agent would be seriously undermined. You are likely to feel that your actions are no longer under your control, and this means that they will no longer seem to you like your own actions— a disturbing form of self-alienation. Something like this would be a true nightmare and might bring on a serious psychological breakdown. These thoughts seem to show how essential self-understanding is to agency. I am not saying that such understanding must consciously accompany our actions. We are not always actively aware of our reasons—and there may well be actions that we do for no particular reason—but we are in most instances able to say what our reasons are if anyone should ask, and if we are doing something for no particular reason, we should know that’s what we are doing. Our notion of agency demands that an agent—or, at any rate, a reflective agent of the kind that we are—is someone the majority of whose actions are rationalizable, and rationalizable by the agent herself. Moreover, it seems necessary that, when we are performing a rationalizable action, we are aware, consciously or tacitly, of its primary reason. (2) When an agent is performing a rationalizable action, she must know, consciously or tacitly, a primary reason for performing that action. I will not go into the question how the “must” in this statement is to be understood; it seems to me that awareness of primary reasons is at

108


least a psychological necessity for us, human agents, and that when we suddenly realize that we are lacking it on a given occasion, that realization itself makes us interrupt what we are doing (we stop on the way down the stairway). There appears to be a good deal of psychological evidence, in particular, in social psychology, that indicates the importance of self-understanding— the importance of the agent’s maintaining a coherent and intelligible scheme of rationalizations for her actions. The following example is typical.3 In a psychological experiment, some subjects were offered $1 to tell another person that a boring task they had just performed had been interesting; other subjects were offered $20 to do the same. It was found that the $1 subjects were appreciably more prone to have changed their minds about the boring nature of the task (many of them came to believe that the task was pretty interesting after all) than the $20 subjects. The psychologist Daryl Bem offered the following, quite plausible, explanation: the subject offered only $1 to lie realizes that she doesn’t like to tell lies, and that she had no particular reason to do so (because she took only $1); so she now forms the belief that she must not have particularly disliked the task and therefore must not have told a lie when she said it was interesting. On the other hand, the subject offered $20 tends not to reevaluate the task and change her mind because, while she does know that she doesn’t like to tell lies, she is willing to do so if given sufficient compensation. It is evident that the subject’s saying of what she had believed to be a dull task that it was interesting put pressure on her cognitive system to produce a rationalization, and the $1 subjects and the $20 subjects respond to this pressure in expected ways: each constructs a distinct rationalization that is consistent with the further given that she was offered so many dollars to tell the lie. This removes the apparent “dissonance” in their cognitive systems, and restores them to a state of equilibrium. But what does an agent know when she knows her primary reason for the action she is performing?

3 This is taken from Richard Nisbett and Stuart Valins, Perceiving the Causes of One’s Own Behavior (New York: General Learning Press, 1971).


109

II Suppose you are invited to spend a few weeks at a research center in Australia next summer, and you must decide soon whether to accept the invitation. You carefully weigh the pros and cons: you know you will enjoy the trip—meeting old friends and talking philosophy with them, the train rides from Sydney to Melbourne and then to Adelaide, the striking desert scenery, and all the rest. On the other hand, the trip will take you away, for several weeks, from many pressing obligations you have to meet at home: the papers you have promised, the family visits, students who need your advice, the serious inconveniences the trip is likely to cause for your spouse, and so on. It is a difficult decision to make, and after several days of inconclusive deliberation you feel totally frustrated. But wait! There is an easy way out! You realize that the same thing happened every year for the past several years. Every year around this time, a similar invitation came from Australia, and each time you went through the same difficult, drawn-out process of weighing the pros and cons, trying to make up your mind. Reasons pro and con were always pretty much the same: the prospects of an enjoyable trip, good philosophical company, and the like, on one hand, and the many obligations at home on the other. And every time the outcome was the same: “Go for it! Australia, here I come!” There is all the reason in the world to think that the outcome of your present deliberation will, again, be the same: you will end up deciding to accept the invitation, and it seems to you like a foregone conclusion that come July, you will find yourself on a flight to Sydney. So what is the point of all this agonizing deliberation, you ask. You can cut out all this frustration and waste of time by simply informing your hosts that you are coming, and calling up your travel agent to arrange the flights. And that’s just what you do. In changing your perspective this way, you are giving up your role as a deliberator to assume the role of a predictor; you are no longer acting as an agent, but as a spectator, viewing yourself as a third person—the way your spouse and friends would view you. You are no longer deciding what to do, but predicting what you will do. Taking a predictive stance of this kind is a perfectly natural thing to do with regard to a third person: you say, “He is going through all this agony trying to decide what to do, but it’s plain as day

110


what he will end up doing. He will turn down the California offer and stay in New York. He always does.” But there is something peculiar about taking such a stance about your own action, treating your preferences and beliefs as evidence, not as reasons, for what you will do. This means that merely knowing what your wants and beliefs are does not suffice for having, or knowing, a primary reason that rationalizes your action. Neither does merely having the wants and beliefs suffice for having rationalizing reasons. You know what your wants are and know what you believe about how they could be satisfied, what the consequences of satisfying them will be, and so on, and you know you are doing what those wants and beliefs would, or could, rationalize, but you are not doing the action for these reasons. In the case of our philosopher contemplating a trip to Australia, he is doing what he has predicted on the basis of his knowledge about his current wants and beliefs, not something he decided on their basis as reasons. What should he think of his action, as he is flying to Australia? Does he understand why he is going to Australia? I think the situation is actually pretty complex. There clearly is a sense in which he understands why he is going to Australia. Perhaps, he subsumes his action under an action principle “Do whatever you predict you will do,” and this affords him a rationale for his action. And he may even formulate for himself a sort of Hempelian explanation of why he is going to Australia: Whenever he is invited to give lectures in Australia during the summer, he goes. But does he understand his reasons for doing what he is doing? Consider the matter from his perspective: he may be saying to himself, “I am going to Australia because I always do that whenever I am invited.” But does he have a reason for doing what he is doing and understand his action by knowing that reason? He knows his wants and beliefs about the many attractions of the trip and his home-bound obligations, and may believe, or even know, that whenever he has these wants and beliefs, he goes to Australia. But this connection between these wants/beliefs and what he does is an inductive generalization; there may even be an underlying causal mechanism that is operative here. But he is not doing what he is doing because he has arrived at a decision, or intention, to go to Australia by reflecting on and weighing his reasons. The only reason he can cite is his desire to do what he predicts he will do and his belief that this was what he would do anyway. As reasons go, this is a highly impoverished one, “reason” in name only; for if this is the only kind of reason we ever have


111

(say, after we have successfully completed our “behavioral sciences”), we would hardly merit the status of agent. Consider the well-known causal account of the explanatory role of primary reasons. According to this influential and widely received view, whose currency is largely due to Davidson, a primary reason explains an action by being its cause. Not all primary reasons are explanatory reasons, on this view, for an agent could have reason, a good reason, for doing something, but does it for another reason or for no particular reason at all. What makes a reason an explanatory reason is the fact that it is a cause of the action for which it is a reason. In brief, explanatory reasons are the primary reasons that cause actions. Now, as is often pointed out, the word “cause” has a pretty plastic meaning, and in its generic sense, no one could object to the causal account. For it seems not much more than a truism to say that for a reason to explain an action, it must be the case that the agent did what she did because she had that reason, or that if she had not had that reason, she would not have acted in the way she did. But what our example shows is that there are cases for which such “because” statements or counterfactuals could be true but in which the agent does not do what she does for the indicated “reason,” namely cases in which the wants and beliefs cited do not function as reasons but as evidence. Let us look at the nomological version of the causal approach. On this view, an explanatory relation holds between reasons and actions in virtue of their joint subsumption under contingent empirical laws that connect mental states, in particular desire and belief, to action and behavior. Carl Hempel was perhaps the most notable advocate of this position.4 Hempel would say that the following basic principle connecting primary reasons with actions is an empirical lawlike generalization: [Desire-Belief-Action Principle (DBA)] If S wants X and believes that doing A is an optimal way of securing X, then S will do A; or, more simply, when S has a primary reason for doing A, S will do A. As stated, DBA is pretty rough; we are familiar with many ways in which it fails—for example, there may be conflicting wants whose satisfaction is considered by the agent to be incompatible with doing the action called

4 See his discussion of rational action in Aspects of Scientific Explanation (New York: The Free Press, 1965).

112


for, the agent may be physically unable to perform the action, and many others. We may regard the principle as qualified by a generous escape clause, or consider ways of weakening it, for example, by weakening its consequent into a “tendency” statement. But these issues won’t matter for our present purposes. For, as has been widely noted, most generalizations that we depend on, in much of science as well as daily life, in support of causal judgments and explanations appear to be pretty rough affairs, tolerant of exceptions and in need of escape clauses, and in these respects at least, DBA probably does not substantially differ from them. The main point is that, for Hempel, DBA is the kind of law that can underwrite nomological explanations of actions in terms of wants and beliefs, and, given the nomic-regularity account of causation that Hempel and many others accept, explanations generated by DBA will count as causal explanations. There are familiar difficulties with viewing DBA, or other similar principles, as contingent and empirical;5 but let us simply concede DBA as an empirical law—or, at any rate, the kind of rough lawlike generalization that can support explanatory/causal/counterfactual connections. What our considerations show is that DBA so construed will not be able to serve as a basis for the kind of self-understanding that we seek for our own actions. For DBA is consistent with the agent’s wants and beliefs serving merely as a predictive basis of her action—not just for the third person observing her as an object of inquiry but for the agent herself. Indeed, the Hempelian reading of DBA positively encourages us to view wants and beliefs in that way. The reason of course is that, on Hempel’s so-called “covering-law” model of explanation (which he claims to apply to rationalizing explanations), all explanations must pass a predictive test, in the following sense: any adequate explanation must provide, in its explanans, sufficient information to enable the prediction of the phenomenon to be explained. In any case, we may conclude that the Hempelian nomological interpretation of DBA does not give us a principle that underlies the self-understanding of actions, whatever its merits might be as an analysis of third-person prediction and causal explanation of actions. Our rejection of the nomological model, then, is independent of the Davidsonian objection based on “the anomalism of the mental,” or the so-called “logical connection argument” according to which DBA is a priori true in virtue of a “logical connection” between its 5 See, e.g., Davidson’s “Hempel on Explaining Action,” reprinted in his Essays on Actions and Events.


113

antecedent and consequent. We reject both Davidson’s causal account and Hempel’s nomological account because they are unable to distinguish wants and beliefs as reasons and wants and beliefs as a predictive basis.6

III The point about first-person understanding of action that we are belaboring here can be seen to have not much to do with the question whether there is a logical connection between the antecedent and the consequent of DBA— more generally, whether rationalizing action explanations based on some principle like DBA are causal explanations or logically based intentional explanation. In his Explanation and Understanding,7 Georg von Wright argues against the causal interpretation of rationalizations and defends an approach loosely based on the verstehen-hermeneutics tradition. My own approach, as will become clear, is closely related to the approach that stresses the importance of “empathetic understanding” in making sense of third-person actions, and for this reason I find von Wright’s approach congenial. However, von Wright thinks that his approach is closely tied to the view that DBA is an a priori logical truth, not an empirical law. For he thinks that his conception of rationalizing explanation is adequately captured by the following model: (PI) S intends to bring about p. S believes that she cannot bring about p unless she does action A. Therefore, S sets herself to do A. It is von Wright’s claim, as a self-professed “intentionalist” as distinguished from the “causalist” in regard to rationalizations, that instances of schema PI are logically valid,8 and that the logical dependence of its conclusion on its premises is the constitutive fact about rationalizations that sets them apart from law-based causal explanations. I think that von Wright is arguably right about the logical status of PI, or DBA; as stated, they may not be

6 For the same reason we must reject Daniel C. Dennett’s “intentional stance” account when this is applied to an agent’s self-understanding. See Dennett’s “Intentional Systems,” Journal of Philosophy 8 (1971): 87–106; reprinted in his Brainstorms (Montpelier, Vt.: Bradford Books, 1978). 7 Georg von Wright, Explanation and Understanding (Ithaca NY: Cornell University Press, 1971). 8 Actually, with some caveats and qualifications; see ibid. for details.

114


strictly valid logically, or logically true, but it seems pretty certain that they are a breed apart from sundry empirical laws. But it seems clear that the logical dependence relation itself has little to do with how reasons rationalize actions and make them intelligible. For consider an example like this: I intend to give a paper at Utrecht, and I consider that I cannot do this unless I fly to the Netherlands. This gives us a nice instance of PI: I can explain, and understand, why I am flying to Amsterdam on the basis of an intention and a belief. No problem here. But now that I know, or believe, that I will be in Amsterdam, I decide to visit the Van Gogh Museum in Amsterdam. This gives us another instance of PI, as follows: I intend to visit the Van Gogh Museum. I believe that unless I fly to Amsterdam I cannot visit the Van Gogh Museum. Therefore, I set myself to fly to Amsterdam. This appears to be a perfectly good instance of PI: the two premises are true, and if, as von Wright claims, PI represents its conclusion as logically dependent on its premises, the same kind of dependence must hold in this example as well. In spite of this, though, it is clear that this instance of PI has no explanatory force: its premises do not rationalize, and does not explain, why I am flying to Amsterdam. My intention, or desire, to see the Van Gogh Museum is not what rationalizes my action; what rationalizes it is my intention to attend a conference in Utrecht.9 Notice that the above instance of PI might have been a good explanation; for I might have flown to Amsterdam just to see some Van Gogh paintings. And I might have formed the intention to attend this conference because I believed I would be in the Netherlands on my museum trip. This shows that the presence of a logical relation between the premises of PI and its conclusion is irrelevant to the explanatory, or rationalizing, force of any particular instance of PI. That is, the question whether an intention/belief pair, or a want/belief pair, rationalizes an action, is not answered by saying that the former logically implies the latter. What then distinguishes explanatory instances of PI from nonexplanatory ones? What is the thing that is

9 I owe this example (reworded) to Nicholas Sturgeon.


115

present in my intention to attend the conference but absent from my intention to visit the Van Gogh Museum that explains the difference in their explanatory status? That there is here a significant difference is seen when we consider the following two counterfactuals: If I hadn’t intended to attend the Utrecht conference, I would not have flown to Amsterdam. If I hadn’t intended to visit the Van Gogh Museum, I would not have flown to Amsterdam. The first is true, the second false. One could invoke the counterfactual analysis of causation here and contend that these counterfactuals show that my intention to attend the conference, but not my intention to see Van Gogh paintings, was the cause of my flying to Amsterdam. However, this is not the heart of the issue; the crucial question is what makes the first counterfactual true and the second false—and what this might be connected to the presence of a rationalizing force in the first situation and its absence in the second.

IV I now offer a simple and straightforward answer to this question. My intention, or desire, to attend the conference, explains and rationalizes my flying to Amsterdam because it is the reason on the basis of which I chose, or decided, to do that action.10 Attending the conference was a goal I had in view, and I chose what I took to be a necessary, or optimal, means to realize it. As far as rationalization goes, this is all there is to it. I would have no objection if you wish to insist on giving a causal reading of the situation: it is true, uncontroversially, that if I had not had that reason, I would not have chosen to do as I did, and, as we might even say, that having that reason caused me to choose the means that I chose. The point to remember is that saying this does not commit me to the view that having the reason was a nomological cause of my decision or action, or that it somehow “naturally 10 My approach bears a certain similarity with the accounts of George Wilson in The Intentionality of Human Action (Stanford: Stanford University Press, 1989), and Carl Ginet in On Action (Cambridge: Cambridge University Press, 1990).

116


necessitated” it. I did what I did because, in light of my goals and the information I had about how they might be achieved, I chose to do what I did. In general, then, what I want to offer can be summarized thus: (3) An agent understands her action insofar as she knows the primary reason on the basis of which she chose, or would have chosen if she had deliberated, to do what she did. This admittedly is somewhat idealized; we don’t always consciously choose what we do, much less consciously deliberate about what to do. Many of our actions, including those that are rationalizable, are done more or less automatically, and it is only ex post facto that we could reconstruct primary reasons for them. I am saying, however, that our ability to do so is essential to self-understanding as reflective agents. Actions for which no amount of effort on our part yields a rationalization are those that we could not understand as our own actions, from our special perspective as their agents. Thus, self-understanding arises out of the context of deliberation, choice, and decision. The context of deliberation is necessarily a first-person context. For when you deliberate, you must call on what you want and believe about the world—your preferences and information—from your internal perspective, and that’s the only thing you can call on. The basis of your deliberation must be internally accessible, for the simple reason that you can’t use what you haven’t got. Reasons for action, therefore, are necessarily internal reasons, reasons that are cognitively accessible to the agent. That is one crucial respect in which reasons for actions differ from causes of actions: reasons must, but causes need not, be accessible to the agent. But there is the question whether the self-understanding of an action in this sense is compatible with a causal-nomological explanation of that action. This question divides itself into two: (a) If an agent understands his own action in terms of his reasons for choosing to do the action, can he also understand or explain his action in terms of his reasons taken as causes (that is, nomological, necessitating causes)?, and (b) in the same situation, can there be a third-person causal explanation of the action which takes the agent’s reasons as causes? It has often been argued that the answer to (a) is an obvious and emphatic no, on the basis of the fact, or what appears to be an evident fact, that one cannot be deliberating about what to do and also be predicting what one will do. But it would seem that if the answer to (b) is yes, which I think it is, the answer to (a) should also be yes. An objectively


117

valid explanation of my action from the third-person perspective must really be objective, that is, interpersonally sharable, and there is no reason why I cannot see and appreciate the point of such an explanation about my own action. What the incompatibility of deliberation and prediction rules out is that the agent cannot be doing both at the same time; he must either deliberate and decide—that is, take the role of an agent—or take the third-person point of view and try to predict what he will do, just as he would about others around him. The predictive stance and the deliberative stance are mutually exclusionary. But this does not mean that the agent cannot retrospectively explain his action causally or nomologically. Time puts a distance between his action and his current cognitive stance, and the action, like the actions of anyone else’s, should be open to him for causalnomological understanding. I will not here go into the question what the correct answer to (b) is.11 Some have argued that we must answer it in the negative, on the ground that if an action is caused by antecedent conditions, it is in principle predictable on the basis of those conditions. Moreover, if it is in principle predictable, it must be predictable by the agent herself, because the prediction does not use any information that is not in principle available to the agent. But it is a contradiction, the argument goes, to suppose that an agent is deliberating about what to do on the basis of her wants and beliefs, while believing at the same time that her decision is causally determined by her total present state including of course her wants and beliefs. This argument raises a host of complex questions that are tangential to my purposes here, and I merely set aside the issues connected with question (b). The question I want to address briefly here concerns the possibility that the agent is mistaken in thinking that he is choosing to do A because of a particular set of wants and beliefs which he takes to constitute his reasons for doing A. Perhaps the agent is deceived, or self-deceived, about his own motives, and his choice to do A is caused by some desires of which he is unaware, or desires of which he is aware but which he consciously rejects as his reasons for doing A on this occasion. Consider the following possibility: Agent S decides, or so he thinks, to do A on the basis of his reasons R. But we, the spectators, can produce a causal-nomological explanation of what 11 On this and related issues see Carl Ginet, “Can the Will Be Caused?” Philosophical Review 71 (1962): 49–55.

118


S does on the basis of his wants and beliefs (call them R*), which are different from R. On the basis of R*, we could have predicted, with great certainty, what S would decide to do, and there is, we are assuming, every reason to think that R* lawfully necessitates what S does. What should we say about S’s “self-understanding” of his action in such cases? Various possibilities can be distinguished depending on how R and R*, the agent’s reasons and the necessitating causes, are related to each other. For example, if R and R* largely overlap each other, or if R is a large subset of R*, there would be no special problem. However, if R and R* are entirely disjoint and R* itself consists of the agent’s intentional mental states, in particular those of which he is aware, then we would have to say that the agent is self-deceived about his action. His reasons had nothing to do with how he decided, and his belief that he did what he did for reason R is just mistaken. His deliberations were idlers; contrary to what the agent thinks, they did no work at all in the agent’s doing what he did. It seems possible that such cases can, and do sometimes, happen. I must set aside the large metaphysical questions this raises; instead, let me just note a few points that I believe are pertinent. First, a possibility of this kind also exists when we form a new belief as the result of reasoning from beliefs we already have. You may, for example, come to believe a certain mathematical proposition on the basis of a proof from a set of premises that you take to be uncontroversially true. You can recite the entire proof, step by step, and see exactly how you have arrived at the proposition, and this surely suffices to rationalize your acceptance of it. You understand why you now believe this proposition to be true. But here again the possibility of error and selfdelusion lurks: What if there exists a third-person causal-nomological explanation of why you now believe this proposition on the basis of certain prior states of yours, in particular a set of intentional states which are distinct from the premises of your proof? A possibility of this kind cannot be ruled out for theoretical reasoning, leading from belief to belief, any more than it can be in the case of practical inference, or deliberation, leading to a decision to act. We can be self-deceived about why we believe what we believe just as we can be self-deceived about why we do what we do. Bad faith can infect theoretical reason as well as practical reason. It seems clear that, although a possibility of this kind does exist, we must rule out the possibility that we could be massively wrong about why and how we come to a certain decision or belief as a result of deliberation


119

and inference. If in almost all decisions we reach after deliberation the deliberative process had no role in determining the outcome, or in almost all the beliefs we come to accept as the result of reasoning, the reasoning process were simply an idler, with the outcome necessitated by an entirely different set of conditions, then our conception of ourselves as an agent or cognizer would be irreparably undermined. To put it another way, our integrity as a reflective agent and cognizer depends on our confidence that the possibilities under discussion are largely just possibilities, and that our deliberation and reasoning do play a crucial role in our choice of a course of action or the beliefs we acquire. There is a further epistemic point here. The agent must, I think, have special first-person access to her reasons and why she has chosen to do as she has. I have already emphasized that reasons for decisions must of necessity be internal reasons—that is, they must be accessible to the agent. What I am saying here is that not only must they be accessible to her but accessible in a special first-person way. I don’t want to say that all of the wants and beliefs that form the basis of my deliberations must be so accessible, but they must by and large be so. What needs to be eliminated is the possibility that I come to know them, by and large, from the testimony of others or the analysis and inference based on observation of my own behavior. For my decisions to count as mine, as issuing from my reasons, it must be the case that I am in a specially intimate epistemic contact with my reasons. Moreover, I must have direct first-person knowledge of why I have decided as I have; I must have first-person access to my own deliberation and decision-making process. The issue of first-person knowledge of our own mental states has been much discussed in connection with epistemology. What I am saying here is this: the special character of first-person knowledge of our own mental states, however it is to be described, is just as crucial to our nature as agent as it is to our nature as cognizer. It is at least as important to practical reason as it is to theoretical reason.

V Let us now turn attention to our understanding of others’ actions. I think it is important to distinguish two different stances (to borrow a term of

120


Dennett’s12) we can take toward others. One is the third-person predictive stance of the sort we have already discussed: by taking all the relevant antecedent conditions, including the wants, beliefs, and the rest, that obtain for another person, and invoking all the applicable laws and generalizations, whether precise or rough and ready, we try to formulate a causal/nomological explanation of what she has done, or a prediction of what she will do. Something like DBA, relating belief/desire pairs with action, could be central to such an enterprise. A picture like this is what Hempel presents with his account of explanation of rational action. There are causal laws that hold for rational agents, and with their help we generate law-based explanations and predictions for a class of objects, namely rational agents. The same picture emerges from the so-called “theory theory” of commonsense psychology. We have a tacit psychological theory, much of it couched in the intentional vocabulary of belief, desire, and the rest, consisting of rough empirical generalizations about human behavior, and this is what we invoke in formulating explanations and predictions of our fellow humans’ behavior. The account is completely third-person; other people are the objects in the domain of our theory, and we explain and predict their behavior by appealing to the laws and generalizations we have formulated about them, largely on the basis of past observations of their behavior. But is this what we do when we try to understand what others do? It is clear that at least sometimes that is exactly what we do. But I submit that in most cases that isn’t what we do. It seems to me that if the third-person theorizing is the only thing we do about other people, we would indeed be “anthropologists on Mars,” to use an expression from Oliver Sacks who attributes it to Temple Grandin, a well-known autistic animal scientist.13 Grandin uses this expression to describe the feeling she has when trying to navigate her way among “normal,” nonautistic people, that is, her “Martians.” Let me read to you a particularly interesting and revealing passage from Sacks’ article: “Much of the time,” she said, “I feel like an anthropologist on Mars.” She was at pains to keep her own life simple, she said, and to make everything very clear and explicit. She had built up a vast library of 12 Dennett, “Intentional Systems.” 13 Oliver Sacks, An Anthropologist on Mars (New York: Alfred A. Knopf, 1995). See the chapter with the same title.


121

experiences over the years, she went on. They were like a library of videotapes, which she could play in her mind and inspect at any time— “videos” of how people behaved in different circumstances. She would play these over and over again and learn, by degrees, to correlate what she saw, so that she could then predict how people in similar circumstances might act. She had complemented her experience by constant reading, including reading of trade journals and the Wall Street Journal—all of which enlarged her knowledge of the species. “It is strictly a logical process,” she explained.14 Apparently, Grandin is a person of high intelligence—an excellent memory combined with an astonishing ability for reasoning and computation; she describes herself as having a Sun workstation in her head. She observed how people behaved under what circumstances, stored a vast amount of these data in her head, and made careful generalizations, which afforded a basis for anticipating other people’s behaviors and enabled her to engage in limited social interactions with them. But, as Sacks makes clear, her interpersonal interactions were painfully limited, and she was very much aware of this fact; outside her professional life she had none of the usual social intercourse with other people. An astonishing fact is that with certain animals, especially cows and pigs, she had the kind of intuitive empathy and understanding that most of us have for other humans. Grandin says: “When I’m with cattle, it’s not at all cognitive. I know what the cow’s feeling . . . It’s different with people . . . Studying the people there, trying to figure out the natives. But I don’t feel like that with animals.”15 Sacks was struck by “the enormous difference, the gulf, between Temple’s immediate, intuitive recognition of animal moods and signs and her extraordinary difficulties understanding human beings, their codes and signals, the way conduct themselves.”16 As he points out, it isn’t the case that Grandin lacks feeling or empathy; it’s just that she lacks it for humans, and has to approach her transactions with fellow humans completely “cognitively,” by careful observation and generalization.

14 Ibid. 259–260. 15 Ibid. 268–269. 16 Ibid. 269.

122


It is evident that if the “theory theory” about folk psychology is correct, we would be doing what Temple Grandin does, in trying to understand and predict the behavior of others. Grandin’s strategy is clearly what we have called “the third-person predictive stance.” Here I do not want to get into the current controversy between the “simulationists” and the “theory theorists” about commonsense psychology. Although what I want to say here may seem to sit more comfortably with the simulation theory (at least some versions of it), I would not like to appear as though I am endorsing this approach. One reason is that in my view the approach has not been worked out in detail with sufficient coherence for an overall assessment; in particular, the way the first-person stance is involved in the simulation approach is not entirely clear to me, and some versions of the theory seem just as thirdperson as the theory theory. In any case, it is clear enough that Temple Grandin’s way is not the way we normally understand other people’s behavior—that is, by the application of psychological generalizations obtained from careful observation of their behavior. I am not saying that the theory approach is never used, or that it cannot be useful or even essential. And I do not here want to get into a discussion of exactly how we do manage our psychological affairs with other people (it would at least be farfetched to use the Grandin example to show that “simulation” is what we do). The only point I want to make concerns the issue of rationalizing understanding of other people’s actions. What I want to suggest here is this: [The projection thesis] we understand the actions of others by projecting to them the way we understand our own actions. In speaking of “projection” I don’t mean anything that is philosophically loaded or controversial; in particular, I am not proposing projection as a method of finding out what others believe, desire, feel, and so forth, as some simulation theorists and friends of “empathetic understanding” and verstehen would insist. Nor am I suggesting projection as a way of extending one’s own psychological/behavioral patterns to others. What I have in mind is very simple: In understanding what others do, we consider them to be agents like ourselves, persons who deliberate, make decisions and form action plans, and take steps to carry them out, all from their own first-person


123

point of view. And we consider that self-understanding, in our present sense, is just as essential to their status as agents as it is to ours, and that their self-understanding, like ours, consists in knowing the primary reason for which they have chosen, or would have chosen, to do what they in fact do. And when we know the primary reason for which they did what they did and know it as such, we, too, understand what they did. Finally I want to say something about the explananda of rationalizing explanations: [The autonomy thesis] Actions explained by rationalizing explanations are not actions in the sense of overt physical or bodily motions, or in a sense in which such motions are involved. Rather, what is rationalized is an intention or decision. What actual physical movement issues from our intentions and decisions, whether or not our actions succeed, is beyond the purview of rational intentional psychology. Insofar as commonsense psychology treats bodily actions, it goes beyond “pure” psychology; it is a mixture of folk psychology and folk physics/physiology. As far as rational understanding of agents goes—that is, as far as the question of the intelligibility of action goes— the proper explanandum is decision and intention. The actual success in the execution of a decision depends on factors beyond the agent; nature must cooperate. In a derivative sense, we can of course “make sense of,” or “make intelligible,” a bodily action that is a successful execution of an intention. But in the same sense we can make sense of bodily actions that are failures—those that fail to implement an intention—in terms of the agent’s calculation and decision and the natural circumstances. How intentions and decisions issue in bodily actions is part of the mind-body problem, not a problem concerning the rationality of agents. When we speak of “rational action,” what is involved is the rationality of decision and intention.17 I have not said anything about the general issues of rationality; my concern has been solely with rational self-understanding of actions. About this, I have advocated a first-person perspective, and in that sense, my account views reasons as subjective. On the other hand, I believe the ability

17 For a similar view see William P. Alston’s “Conceptual Prolegomena to a Psychological Theory of Intentional Action,” in Philosophy of Psychology, ed. S.C. Brown (London: The Macmillan Press, 1974).

124


to view things from a first-person perspective is a sine qua non condition of self-understanding for all reflective agents, agents who make decisions and plan actions on the basis of reasons. So in this sense my account is committed to the universality of rationality—rational understanding of the kind I have talked about is universal to all reflective agents.

6 Taking the Agent’s Point of View Seriously in Action Explanation I On the first page of his landmark paper “Actions, Reasons, and Causes,” Donald Davidson writes: A reason rationalizes an action only if it leads us to see something the agent saw, or thought he saw, in his action—some feature, consequence, or aspect of the action the agent wanted, desired, prized, held dear, thought dutiful, beneficial, obligatory, or agreeable.1 This leads us to expect Davidson’s account of rationalizations—that is, reason-based explanations of actions—to be in some clear and robust sense agent-centered and first-personal; for Davidson appears to be saying that we can rationally make sense of what an agent did only if we see, or can see, the situation—the choices open and their probable consequences—in the way the agent did. Such an account seems to make the agent’s perspective primary and central, superseding an “objective” third-person perspective on the agent and his actions. But this expectation is quickly disappointed; Davidson’s causal theory of rationalizations is nothing of the sort. Causal explanations, by their very nature, appear to be “objective” and third-personal—there is nothing specifically first-personal or “subjective” about causal relations, or explanations that invoke them. Although causal judgments may be sensitive to certain contextual factors, we don’t think of causal relations as varying in relation to the perspectival differences between the first-person and third-person points of view. The objective 1 Donald Davidson, “Actions, Reasons, and Causes,” in Essays on Actions and Events (Oxford: Oxford University Press, 1970). The quotation is from p. 3. First published in 1963.

126

THE AGENT’S POINT OF VIEW

third-personal aspect of Davidson’s account of action explanation becomes particularly clear in his insistence that all causal relations, including those involved in rationalizations, must instantiate “strict laws,” laws that, according to him, can be found not at the level of reasons (that is, the level that deals in things like beliefs, desires, and emotions) but at lower levels—the levels of biology and physiology, and perhaps as far down as fundamental physics.2 Surely, the existence of these laws is not a matter that depends on taking a first-person or third-person stance. The significance of perspectival differences for action theory has been noted by other philosophers. For example, Christine Korsgaard has this to say: It is easy to confuse the criteria of explanatory and normative adequacy. Both, after all, concern questions about how people are motivated to do the right thing and why people care about moral issues so deeply. . . . Nevertheless the issue is not the same. The difference is one of perspective. A theory that could explain why someone does the right thing—in a way that is adequate from a third-person perspective—could nevertheless fail to justify the action from the agent’s own, first-person perspective, and so fail to support its normative claims.3 Here, the first-person perspective is distinguished from the third-person perspective, and the normative/justificatory issue is assigned to the former, the explanatory/motivating issue to the latter. From this paragraph we need not conclude that in Korsgaard’s view, the first-person perspective has nothing to do with the understanding of an action; after all, she says that the normative question, like its explanatory counterpart, concerns “how people are motivated to do the right thing” and knowing the answer, from the agent’s first-person point of view, must be expected to contribute to our understanding of why the agent did what she did. Nonetheless, it is interesting that according to her, the distinction between justification and explanation rests on a perspectival difference, and that the first-person perspective is essentially tied to justification, not explanation. This contrasts with an alternative view according to which justification and explanation 2 In “Actions, Reasons, and Causes,” Davidson says that the laws grounding rationalizations may have to be sought at the “neurological, chemical, or physical” level. The claim that strict laws are found only in “developed physics” is made in his later “Thinking Causes,” in Mental Causation, ed. John Heil and Alfred Mele (Oxford: Clarendon, 1993). 3 Christine Korsgaard, The Sources of Normativity (Cambridge: Cambridge University Press, 1996), p. 14.


127

are not separable and the justifying reason for an action is crucial to making it intelligible. A view of this kind will be at center stage of our discussion later in this paper. Thomas Nagel is another philosopher who takes the first-person perspective seriously in connection with agency. In The View from Nowhere, he describes the agent’s “internal” perspective: From the inside, when we act, alternative possibilities seem to lie open before us: to turn right or left, to order this dish or that, to vote for one candidate or the other—and one of the possibilities is made actual by what we do.4 But there is also the third-person, external point of view: . . . it seems possible that many of the alternatives that appear to lie open when viewed from an internal perspective would seem closed from this outer point of view, if we could take it up. And even if some of them are left open, given a complete specification of the condition of the agent and the circumstances of action, it isn’t clear how this would leave anything further for the agent to contribute to the outcome—anything that he could contribute as a source, rather than merely as the scene of the outcome—the person whose act it is. If they are left open given everything about him, what does he have to do with the result? From an external perspective, then, the agent and everything about him seems to be swallowed up by the circumstances of action; nothing of him is left to intervene in those circumstances. This happens whether or not the relation between action and its antecedent conditions is conceived as deterministic.5 Nagel is claiming that from an external, third-person point of view, action and agency are invisible. There are only happenings, including our limbs moving this way and that way, but no agents or actions. Interesting too is what Nagel says about intentional explanations of actions, that is, explanations that invoke the agent’s beliefs, desires, and other intentional states. The intuitive idea of autonomy includes conflicting elements, which imply that it both is and is not a way of explaining why an action was 4 The View from Nowhere (Oxford: Oxford University Press, 1986), p. 113. 5 Ibid. 113–114.

128


done. A free action should not be determined by antecedent conditions, and should be fully explained only intentionally, in terms of justifying reasons and purposes.6 As the opening sentence of this quotation indicates, Nagel thinks that, from another perspective, intentional explanations do not fully explain—by their very nature, autonomous actions exclude full explanations, intentional or otherwise. What interests us about this quotation, though, is the fact that, on Nagel’s view of autonomy, actions are explained, for the agent, in terms of “justifying reasons and purposes.” Normative reasons that justify are also explanatory reasons, from the agent’s internal perspective. As this brief survey indicates, the importance of “points of view” or “perspectives” is often emphasized in discussions of agency and action explanation. Taking one perspective rather than another is thought to make a crucial difference to such issues as justification, explanation, and prediction of actions, and, according to some, the very possibility of agency and freedom. Here, my focus will be on explanation and understanding of actions, and what I propose to do is to explore the possibility of developing an account of action explanation that takes the agent’s first-person perspective as primary and central. When an agent asks “What shall I do?”, her concern is essentially normative, and I want to see how taking an agent’s normative perspective can yield an understanding of why the action is done. Korsgaard and Nagel tie the agent’s perspective to justification and explanation, and Nagel makes the claims that from the agent’s internal perspective, justification yields explanation and understanding. I hope to explore these suggestions, especially the connection between justification and explanation of actions. I am going to approach this task in a somewhat roundabout way, by returning to an interesting debate, in the mid-20th century, between Carl G. Hempel and William Dray on the nature of explanations in history. This debate, as we look back from today’s perspective, had nothing specifically to do with “historical” explanations—or even with explanations of actions by historically significant individuals. It was really a debate about explanations of human actions in general, and it brings into sharp relief the contrast between Hempel’s objective third-person approach grounded in scientific

6 Ibid. 115.


129

laws of behavior and Dray’s agent-oriented account that puts normative considerations at front and center.

II Hempel arguably was the leading theoretician on the nature and form of scientific explanation during the second half of the last century. According to his “covering-law” approach to explanation, explanations of individual events require these events to be “covered” by, or “subsumed” under, general laws—statements of empirical regularities of the sort investigated in the sciences.7 The laws could be deterministic, in which case we have deductive-nomological (D-N) explanations; when they are statistical, the explanations are inductive-statistical (I-S). Formally, explanations of both types take the form of an argument or inference, with laws and antecedent conditions as the explanatory premises (the “explanans”) and the statement of the event to be explained as conclusion (the “explanandum”). It is an important tenet of Hempel’s views on explanation that this nomologicalinferential conception of explanation has universal validity: it applies to explanations of events and phenomena in all areas—explanations of human actions, or “rational actions,” no less than explanations of natural phenomena like eclipses of the moon and the fall of an object released from the top of a building. How, on Hempel’s covering-law model, are human actions to be explained? Let us focus on rational actions, actions performed by rational agents qua rational agents; these actions result from these agents’ exercise of their rational agency. Hempel offers the following schema to represent the general form of explanations of such actions.8 Schema H.

At time t, agent S was in a condition of kind C. S was a rational agent at t. Rational agents in a condition of kind C do action A. Therefore, S did A at t.

7 See the title essay in Hempel’s Aspects of Scientific Explanation (New York: The Free Press, 1965). 8 See Hempel, Aspects of Scientific Explanation, pp. 463–487, where he deals in detail with explanation of actions by reasons. Schema H is a slight variation on Hempel’s Schema R, p. 471.

130


The first two statements describe the antecedent circumstances; the third premise is a law about how rational agents generally behave in a condition of a specified kind. The condition referred to may—will likely though this is not required—include the agent’s beliefs and desires—that is, the information (or misinformation) at her disposal and her preference structures. On Hempel’s view, to say that an agent is rational is to attribute to her a set of complex dispositions—dispositions to think, reason, deliberate, and make choices in certain broadly specifiable ways. Hempel will say that laws about the behavior patterns of rational agents are empirical laws confirmed on the basis of observation and experiment, like other similar regularities investigated in social psychology, motivation and action theory, and the like. Such laws are likely to be statistical, not universal/deterministic, laws; this would render instances of Schema H inductive-statistical explanations. Or we can do a bit of idealization and think of the law as deterministic; this will make no difference to the discussion to follow. Something like the following would serve as an instance of Schema H: At t, Jones believed that it was going to rain in the afternoon; and Jones had an umbrella in her closet. At t, Jones was a rational agent. Rational agents who believe it is going to rain and have an umbrella available will take an umbrella with them when they go out. Therefore, when Jones went out at t, she took an umbrella with her. 9 It is clear that there is nothing first-personal, or subjective, about action explanations conceived this way; on Hempel’s view, each statement comprising the explanans is an objectively verifiable empirical statement, and there is no one in a uniquely privileged position—and that includes the agent himself—to gain explanations of this form. For Hempel, this will be so even if a “condition of kind C” includes the agent’s own mental states.10 9 One thing to be noticed about Schema H is that an explanation conforming to it does not necessarily imply that what the rational agent did was a rational action. As an empirical matter of fact, there may well be conditions in which rational agents do something that is not necessarily an expression of their rational agency. If so, H could not be regarded as a model of explanation of rational actions, though it could be a model of the actions of rational agents. 10 Works by cognitive scientists that cast doubt on privileged first-person knowledge of one’s own mind have become pretty widely known to philosophers. For example, see Richard E. Nisbett and Timothy DeCamp Wilson, “Telling More Than We Can Know: Verbal Reports on Mental Processes,” Psychological Review 84 (1977): 231–259; Alison Gopnik, “How We Know Our Minds: The Illusion of First-Person Knowledge of Intentionality,” Behavioral and Brain Sciences 16 (1993): 11–14.


131

Such explanations are part of objective scientific psychology—or, at least, can be refined and sharpened into explanations in systematic psychology. The implication is that for me to understand why I did a certain thing, say, took my wife’s car instead of mine to come to work this morning, I need a psychological generalization about agents, or rational agents, in the kind of condition I was in this morning—and to be able to identify and separate out the relevant factors in that condition. To understand what I did, I have to know what all agents like me would do in similar circumstances. Prima facie, that seems impractical and unrealistic, if not outright impossible. Moreover, it just doesn’t seem the way an agent’s self-understanding works—having that kind of nomological information seems neither necessary nor sufficient— indeed it seems entirely irrelevant—for my understanding of my own actions. This strong counter-intuitiveness of Hempel’s model leads us to raise a fundamental question about Hempel’s Schema H: what is it about the information contained in the explanans of Schema H that makes it explanatory of the action to be explained? That is, what is the source of explanatory insight and illumination offered by explanations instantiating H? Let us use the term “explanatory relation” to refer to the relation an explanans bears to its explanandum in virtue of which it serves to explain the explanandum and enables us to understand it. Explanatory relations, in short, are what makes explanations explanatory.11 Hempel’s general view on explanatory relations is expressed in what he calls “the requirement of explanatory relevance,” which he explains as follows: Thus, even if we happened never to have seen a rainbow, the explanatory information provided by the physical account [of rainbows] would constitute good grounds for expecting or believing that a rainbow will appear under the specified conditions. We will refer to this characteristic by saying that the physical explanation meets the requirement of explanatory relevance: the explanatory information adduced affords good grounds for believing that the phenomenon to be explained did, or does, indeed occur. This condition must be met if we are to be entitled to say “That explains it—the phenomenon in question was indeed to be expected under the circumstances!”12 11 For more details on explanatory relations see my “Explanatory Knowledge and Metaphysical Dependence,” Essay 8 of this volume. 12 Philosophy of Natural Science (Englewood Cliffs, NJ: Prentice-Hall, 1966), p. 48.

132


Since, for Hempel, the explanatory information, or the explanans, must include laws, we may say that his explanatory relation consists in nomic expectability, namely that it is the fact that the explanans makes the explanandum expectable on the basis of laws that makes the former explanatory of the latter. And this is supposed to apply across the board, to explanations of human actions as well as explanations of natural, physical events and phenomena. Let us assume the condition C in which the agent finds herself includes facts about her beliefs and desires, so that we can speak of these mental states as her reasons in acting as she did. On Hempel’s schema, then what makes her reasons explain what she did is the fact that on the basis of information about her beliefs, desires, and perhaps other conditions, in combination with laws, or regularities, about how agents like her behave when they have these beliefs and desires, we could have correctly predicted—or, at any rate, inferred—what she did. Or, to put it in general terms, beliefs and desires of an agent provide reasons that explain an action because, in conjunction with empirical psychological laws, they enable us to expect the agent to have acted in the way she did. We have already discussed the implausibility of Hempel’s model when applied to the agent’s understanding of his own actions. Let me add another consideration along the same line. Consider a deliberative action—an action that an agent performs by acting on an intention, or decision, reached through a conscious deliberation about reasons pro and con. When an agent asks himself “Why did I do that?”—or when another person asks him the same question—he will likely try to retrace the process of deliberation in which he engaged and which resulted in the formation of decision that led to the action. One thing that is clear is that we engage in a deliberative process about the choice of an action precisely because we do not know what we will do, and what gives meaning to the deliberation— indeed, what conceptually makes deliberation possible—is the fact that we do not know, or have a firm belief13 about, what we will do.14 On Hempel’s nomic model, when we retrospectively try to understand a past action of ours, the retracing of our deliberative process that led to it is irrelevant; 13 To borrow a term from Timothy Williamson, we don’t have an “outright belief” about what we will do. Williams defines “outright beliefs” as those beliefs whose contents the believer is prepared to use as premises in practical deliberations. See his Knowledge and Its Limits (Oxford: Oxford University Press, 2000), p. 99. 14 See Carl Ginet, “Can the Will Be Caused?” Philosophical Review 71 (1962): 49–55.


133

rather, the question we should ask is “Did I do what rational agents generally do in a similar condition?” This seems just wrong; it renders practical deliberations irrelevant to understanding the decisions and actions to which they lead. We should note that Hempel does distinguish between two aspects of rationalizations: a critical/evaluative aspect and an empirical explanatory aspect. By the former, he means our (or the explainer’s) judgment that “in light of the agent’s beliefs, the action he decided upon constituted a reasonable or appropriate choice of means for achieving his end.”15 The empirical aspect is the claim that the agent’s action was motivated by his goals and beliefs, in particular instrumental beliefs about the means-ends relationships. For Hempel, the empirical aspect does all the explanatory work; he writes “the explanation [of a human action] will aim at showing that the action was to be expected in view of [the agent’s] objectives and beliefs.”16 The normative/critical dimension of a rationalization has nothing to do with its status or effectiveness as an explanation. Writing about William Dray’s account which takes the opposite stance on the issue (as we will see shortly), Hempel says “That the appraising function which Dray considers essential for rational explanation has no explanatory import is shown also by this consideration,”17 where the “consideration” is the claim that unlike the fact that the agent has a general disposition to act in a certain way, or that the agent in the circumstances was in fact inclined to regard a certain action as the right thing to do, the judgment as to the appropriateness or rightness of the action for the agent has nothing whatever to do with the question whether the agent actually performed the action. Conceding only that Dray-style rational explanation can have “evaluative overtones,” Hempel concludes: “what I maintain is only that whether a critical appraisal is included in, or suggested by, a given account, is irrelevant to its explanatory force; and that an appraisal alone, by means of what Dray calls a principle of action, cannot explain at all why A did in fact do x.”18 It is clear that Hempel faithfully hews to nomic expectability as the essential explanatory relation for rationalizations.

15 16 17 18

Hempel, Aspects of Scientific Explanation, p. 463. Original emphasis. Ibid. 469. Ibid. 471. Ibid. 472.

134


Before we move on to Dray, it is worth emphasizing again that, like Davidson’s, Hempel’s is a completely third-person, objective account: an agent’s explanation of her own action has exactly the same form as a thirdperson’s explanation of the action. The agent is perhaps in a better situation to obtain some of the needed explanatory information, but the content of the information is identical in both explanations. In that regard, Hempel’s account does not differ significantly from Davidson’s causal account—they differ only in that Hempel believes that there are predictive/explanatory psychological laws, at the level where we speak of beliefs, desires, and decisions, whereas, on Davidson’s view, we have to reach much further down, to neurophysiology, or perhaps even to physics and chemistry, before we can find “strict” laws that can underwrite the causal relations between reasons and actions.19 Both accounts are saddled with an implausible epistemology of rationalizations—how we are able to formulate or confirm rationalizations, especially concerning our own actions. Hempel’s model requires knowledge of behavioral-psychological laws; the problem is that, one, we don’t seem to have many such laws, and there may in fact be few such laws if any; two, at least, ordinary folk like most of us don’t have access to them, although their capacity to explain and understand actions doesn’t seem severely hampered by this lack. Davidson’s account of rational causation, in terms of strict laws at neural, biological, or physicochemical levels, is frankly unbelievable; though he does not require rationalizers to have these laws in their possession, they must know that appropriate laws at these level exist. Where would such knowledge come from? I doubt that the best current science would be able to come up with strict laws supporting a single case of rational causation. If we have such knowledge, it must be based on broad metaphysical considerations. It seems obvious that such knowledge is neither available to, nor needed by, those who are capable

19 “Actions, Reasons, and Causes,” p. 17. However, it would be inappropriate to regard Hempel’s account as a causal account. It would be a causal account if causation is analyzed as subsumption under laws. Hempel’s attitude toward causation seems to have been deeply ambivalent; Wesley Salmon has said that Hempel “casually identifies” D-N explanations with causal explanations, but this reading of Hempel is unwarranted. There are places where Hempel explicitly rejects the view that D-N explanations are causal—the view that the causal relation is what makes D-N explanations explanatory. We would do well to keep in mind that, in his heart of hearts, Hempel was a deeply committed positivist epistemologist and didn’t believe in any metaphysical relations underlying his “theoretical systematizations,” of which D-N explanations were a species.


135

of formulating apt rationalizations of actions, their own or those of their fellow humans. We now turn to Dray, who, as I said, takes an approach to action explanation radically different from Hempel’s in its fundamental perspective. Dray’s conception of what it is to explain an action is not merely another, perhaps a better or corrected, version of models like Hempel’s. For, unlike Hempel, Dray considers the normative appraisal of the agent’s choices and actions as an essential component of action explanation; it is what explains the agent’s choices and actions and makes them intelligible. It will be useful to approach the issues through Hempel’s critique of Dray’s account.

III According to Dray, “rational” explanation makes actions intelligible by revealing the agent’s “rationale” for doing what was done. He writes: The goal of such explanation is to show that what was done was the thing to have done for the reasons given, rather than merely the thing that is done on such occasions, perhaps in accordance with certain laws (loose or otherwise). The phrase “thing to have done” betrays a crucially important feature of explanations in terms of agent calculations—a feature quite different from any we have noticed so far. For the infinitive “to do” here functions as a value term. I wish to claim therefore that there is an element of appraisal of what was done in such explanations; that what we want to know when we ask to have the action explained is in what way it was appropriate. In the ordinary course of affairs, a demand for an explanation is often recognized to be at the same time a challenge to the agent to produce either justification or excuse for what was done.20 Dray’s commitment to a normative model of rationalizations is explicit and total. It might be that some law or other exists that connects the agent’s reasons, or calculations, to the action performed, thus yielding a Hempelian D-N argument. A D-N argument of this sort would only show that in these circumstances persons generally do what the agent did. For Dray, however, 20 William Dray, Laws and Explanation in History (London: Oxford University Press, 1957), p. 124.

136


that is not what makes intelligible, and enables us to see, why the agent did what she did; what does the explanatory work is the fact that in the circumstances the thing done was the appropriate thing to do. To set the stage for his criticisms of Dray, Hempel sets up the following schema to model Dray’s rational explanation: Schema D. Agent A was in a situation of type C. In a situation of type C, the appropriate thing to do is X. That is why A did X. Hempel’s dissatisfaction with this schema is predictable. His main criticism is this: the purported explanans of Schema D “cannot possibly explain why A did X. For according to the requirement of [explanatory relevance set forth above], any adequate answer to the question why a given event occurred will have to provide information which, if accepted as true, would afford good grounds for believing that the event did occur.”21 As Hempel points out, the first two lines of Schema D do not entail that A did X; all that Dray’s explanans shows is that X would have been the appropriate thing for A to do. Platitudinously, we often fail to do the appropriate thing. Hempel then goes on to offer Schema H as a replacement. Obviously, not everyone will be impressed by Hempel’s criticism. For it is a restatement of his conviction that nomic expectability is the only explanatory relation for all explanations. Hempel simply does not consider whether there might be a viable alternative explanatory relation that connects reasons with actions, a relation that is in part but essentially normative. He simply rejects Dray’s claim that knowing that what was done was the appropriate thing to do leads to an explanatory understanding of the action—namely that a normative relation can underwrite an explanatory relation. In any case, whatever one thinks of Hempel’s complaint, Dray’s explanatory schema has other problematic aspects that can make it look rather hopeless. For one thing, we can see that the explanatory premises of Schema D can be true and yet it is possible for the agent not even to know that she is in a situation of kind C. Perhaps, she believed that she was in a different sort of situation, C*, and she reached the conclusion that in a situation of kind C*, the appropriate thing for her to do was X. And she proceeded to do X. 21 Hempel, Aspects of Scientific Explanation, pp. 470–471.


137

Or perhaps, she did believe she was in a situation of kind C but concluded that the right thing to do in C was X*, not X. But she inadvertently did X. In such cases, it would be plainly wrong to take Schema D as explaining why S did X on that occasion. A related issue concerns the status of the second premise, the statement that in a situation of kind C, the appropriate thing to do is X. For Dray, this is the crucial normative-evaluative premise that carries the explanatory burden. The question this premise prompts is “Appropriate from whose point of view?” That is, who is making the normative judgment? The statements comprising the explanation are of course the explainer’s, from which it follows that it is the explainer who makes the normative judgment. This means that for an explainer to succeed in rationalizing, or understanding, another person’s action, he must judge that the thing done was the “appropriate” thing for the agent to do in her circumstances. This conclusion seems both obvious and unavoidable. But this cannot be right—at least it cannot be the whole story. The reason is that, on this interpretation, it isn’t at all clear why a rationalization conforming to Dray’s model can be an explanation of the agent’s action in terms of her reasons—her perceptions, goals, and judgments of what is right and wrong, of what works and what doesn’t, and so on. One might say that not only must the explainer make the judgment but it must also be the case that the judgment is true—that is, it must be true that in situations of the sort the agent is in, X is the right thing to do. But this doesn’t make the need of an intelligible tie between the normative judgment and the agent’s motivation go away. Dray often speaks of the agent’s “calculations,” emphasizing how important it is for the explainer to see, or reconstruct, these calculations—that is, see how things seemed to the agent and how the agent reasoned and calculated to determine the appropriate action to take. All this talk about “empathetically identifying” with the agent, “imaginatively projecting” oneself into the situation as perceived by her, “re-living” and “re-enacting” her calculations, and “re-experiencing” her hopes and fears and desires, suggests that the agent’s perspective is primary and the explainer’s interpretation is secondary and derivative. This reinforces the thought that the appropriateness judgment that does the explanatory work must be the agent’s, not the explainer’s. A moment’s reflection shows that this doesn’t work either. If the appropriateness of the action is solely from the agent’s point of view, the second

138


premise of Schema D must be restated to read “Agent A judges that in circumstances of kind C, the appropriate thing to do is X.” But this is a descriptive report, from the explainer’s third-person point of view, of what the agent believes and judges, and this would turn Schema D into a descriptive, nonnormative schema. For Dray, and anyone else who takes the normative model seriously, the explainer’s judgment of what action is, or isn’t, appropriate in a given situation cannot be eliminated; it is an essential part of the approach that in offering a rationalization of an agent’s action, the explainer himself is making an “appraisal” of the agent’s choices and actions. That is to say, explaining an action by invoking the agent’s reasons is a normative process involving the explainer’s evaluation of the rationality of the action. I believe, however, that there is a way to make sense of these odd and puzzling features of Schema D. The key to understanding this schema is to look at it from the agent’s first-person point of view—the perspective of someone who is considering what to do when she believes she is in circumstances of kind C. Think of yourself as the agent involved. You begin by noting that you are in circumstances of a certain kind, which requires you to take some sort of action. You then ask: Given that my circumstances are like these, what is the right thing to do? You have to consider the options open to you and evaluate their likely consequences— that is, as Dray puts it, you do “calculations”—that is, engage in deliberation and practical reasoning. After some thinking, you reach the conclusion that in these circumstances, the right thing to do is X. It is thus that you reach a decision, or form an intention, to take a course of action. So from your firstperson point of view as an agent, the following represents your practical deliberation: Schema I. I am in circumstances of kind C. In circumstances of kind C, the appropriate thing to do is X. So I ought to do X. (And I do X.) The crucial thing to notice is this: We can no longer complain that the agent may not be aware that she is in a situation of kind C. The practical inference is performed by the agent, and she takes as a premise the first line “I am in a


139

situation of kind C” because the statement expresses her belief.22 (It’s important to see that she doesn’t begin with the premise to the effect that she believes that she is in a situation of kind C.) The transition from the first to the second line represents the agent’s normative deliberation, or her “calculations,” and it is here that the substance of her practical reasoning takes place. Of course, the second line, the normative judgment, is the agent’s; it is her judgment that X is the right thing to do in her circumstances. Why did I do X? Because I was in circumstances of kind C, and the appropriate thing to do in C was X. That’s why I did X. I believe this is a basic pattern of the agent’s understanding of his own actions. My general claim then is this: (1) To know, or be able to reconstruct, the practical deliberation that led me to decide to do something (and actually do it) is a fully satisfactory way of understanding why I decided, and acted, as I did. I can only say that to me (1) is self-evident and wholly compelling; I find it hard to know what I can say in its defense. If anyone should disagree, I would ask him to tell me how he understands his decisions and actions. Actually that might not help; if he told me something very different (say, a Hempelian story) from what I know about my own case, I am not sure how I could or should respond. In defense of his causal account of action explanation, Davidson writes: One way we can explain an event is by placing it in the context of its cause; cause and effect form the sort of pattern that explains the effect, in a sense of ‘explain’ that we understand as well as any.23 Mimicking Davidson, why can’t we say: One way we can explain an action is by placing it in the context of deliberation leading to it; deliberation and the action to which it leads form the sort of pattern that explains the action, in a sense of ‘explain’ that we understand as well as any. It is no simple matter to say what it is about the causal relation that gives it its explanatory power, the power to make things and events intelligible and enhance our understanding. We can expect this to be a deep and difficult 22 That is, the statement expresses an “outright belief” of hers, in Timothy Williamson’s sense. See n. 13, above. 23 “Actions, Reasons, and Causes,” p. 10.

140


problem. An explanation of why (1) holds may be equally difficult and complex, and it has to be set aside. Why (1) holds may be an issue; that it does hold seems to me beyond reasonable dispute, and I find (1) at least as compelling as the claim that causal relations explain. Before we move on, though, I want to consider an argument that can be considered an objection to (1). I am thinking of an argument advanced by Nagel in The View from Nowhere. The argument goes like this: Consider an agent deliberating between two possible actions, X and X*. From her perspective, there are reasons, R and R*, R favoring X and R* favoring X*, and as a result of her deliberation, the agent chooses to do X. When asked for an explanation of why she chose X, she says she did it because of reason R. Nagel argues that this fails as an explanation because if she had chosen X* rather than X, she could have explained her choice equally well: She did X* on account of reason R*. This is inevitable given the fact that the agent made the choice: neither R nor R* necessitated the action each favors, and it was the agent’s burden qua agent to make the choice. Nagel says that a rationalization “may render the action subjectively intelligible, but it does not explain why this rather than another equally possible and comparably intelligible action was done.”24 It seems to me that there can be a simple answer here: the agent did X rather than X* because she judged, or decided, that X, not X*, was the right thing to do. Why did she judge that way? Well, she found R more compelling than R*. Why did she do that? There could be many kinds of answers here: in comparing R and R*, perhaps she found that X would have consequences more appealing than the expectable consequences of X*, or that X was easier and less costly to do than X*, or that she just felt more like doing X than X*, and so on. Of course, she could have chosen to do X*, in view of R*, and if she had, she could tell a similar story. But the fact is that she chose to do X, not X*. What more should we expect to know? Nagel himself admits that this account makes the choice and action “subjectively intelligible.” Subjective intelligibility, if it means intelligibility to the agent, is all we need or should want. Nagel goes on to complain that this explanation does not explain why this action, rather than another one, was chosen. But that is to complain that the agent’s way of making her

24 The View from Nowhere, p. 116.


141

action intelligible to herself doesn’t yield an objective, perhaps Hempelian, explanation. Well, so be it!

IV Given this view of the agent’s self-understanding, what should we say about the third-person understanding of actions, actions of other agents? I would like to propose two further general rules to guide our thinking: (2) The way I understand why I did something must be pretty much like the way you understand why you did something. (3) The way I understand why you did something must be pretty much like the way you understand why you did the thing. With (2), there is an issue about who this “you” is—that is, the class of agents “you” ranges over. We might take a large view, a sort of Kantian view, and say that all rational agents, agents who have the capacity to act for reasons, come under its purview. There is a sense of agent under which all agents count as rational agents; but if you count some animals among agents, not all agents are likely to be reckoned as rational agents. Perhaps what this means is that we should regard rational agency as admitting of degrees. In any case, I would want to include all human agents, and think of (2) as merely affirming belief in the common nature of human beings as agents and cognizers. I hope that readers will find a reading of (2) that they find acceptable. (3) is what is important for our purposes; it lays out what I, as a third-person observer, need in order to understand another agent’s understanding of her own action. Given (2), (3) seems plausible; however, (2) does not seem to entail (3), and I will not attempt here to justify (3) on the basis of (2). Apart from that, there is an issue about interpreting (3): (1) says that I understand my action by recapitulating or reconstructing my practical inference that led to it, in accordance with something like Schema I. In conjunction with (1), (2) tells us that you, another agent, understand your action by recapitulating or reconstructing your practical inference that led to your action. Now, (3) tells us that for me to understand why you did something, I need to recapitulate or reconstruct your practical inference which led to your action. As we saw, the practical inference represented by Schema I is conducted, as all instances of practical inference must be, from

142


the agent’s first-person point of view. How then can I, as a third-person observer, “recapitulate” or “reconstruct” a first-person practical inference? We can put this issue another way. Suppose I succeed in understanding your action in accordance with (1), (2), and (3). If I have an understanding, or explanation, of your action, the explanation must have a certain cognitive content. If so, just what is this content? How can I “write down” this explanation? What does it look like? In general, what is the cognitive content of a third-person understanding of an action, an action whose basic and primary explanation is from the agent’s first-person point of view, as represented in Schema I? Let us try to see what is involved. Suppose you want to know why you are carrying an umbrella on a clear sunny afternoon. You recall the practical reasoning that led you to take the umbrella with you when you left home this morning, perhaps something like this: (D1) It’s going to rain this afternoon. Unless I have an umbrella with me, I will get wet. I don’t want to get wet. In this situation, the right thing to do is to take an umbrella. Therefore, I ought to take an umbrella. (And you take an umbrella.) This is your practical inference, not mine. For me to understand why you took an umbrella, I must somehow recapture or recapitulate it. Difficulties are seen to arise when we ask: What does “recapitulating” something like (D1) mean exactly, and just what must I do to do it? As noted, I am a third-person observer of the first-person goings-on involving you and (D1). So I might try a third-person observer’s version of your practical reasoning like this: (D2) Mary believes that it’s going to rain this afternoon. She believes that unless she takes an umbrella, she will get wet. She doesn’t want to get wet. She deliberates and judges that in her situation the right thing to do is to take an umbrella. So she judges that she ought to take an umbrella. So she decides to take an umbrella. (And Mary takes an umbrella.) Is (D2) explanatory? Perhaps it is; if we have the information given in (D2), there is perhaps a sense in which we understand why Mary took an umbrella


143

when she left for work in the morning. But it is not an instance of normative explanation of the sort that Dray seeks and we want to understand. The reason is that (D2) is completely bereft of normativity; none of the statements comprising it are normative—they are all descriptive reports of what Mary believes and does. True, the statements report Mary’s normative activity but they are not themselves normative. As will be recalled, one essential component of Dray’s model of rational explanation is that such explanations must reflect, and incorporate, the explainer’s normative appraisal of the action being explained. This might lead us to try something like the following, which results from (D1) when “I” is replaced by “Mary”: (D3) It’s going to rain this afternoon. Unless Mary has an umbrella with her, she will get wet. Mary doesn’t want to get wet. In her situation, the right thing to do is to take an umbrella. Therefore, she ought to take an umbrella. (And Mary takes an umbrella.) Unlike (D2), (D3) is normative. But does (D3) capture my understanding of why Mary did what she did? There are obvious problems with (D3). If it represents my explanation of Mary’s action, I am asserting, or consenting to, each of the propositions that make it up, and that may well not be the case. For example, I may not believe that it’s going to rain this afternoon; perhaps I believe that she will actually enjoy getting wet. In consequence I may not believe, and not consent to, the proposition that in the situation she is in, the right thing to do is to take an umbrella. In general, I may disagree with the agent’s appropriateness judgment, and that will torpedo any chance of my understanding her action by recapitulating her practical inference (D1) in a form like (D3). This issue is the same as one that was raised before in regard to the second premise of Dray’s Schema D, the crucial proposition to the effect that in the agent’s situation, the appropriate thing to do is X. The question was “Appropriate from whose point of view? The agent’s or the explainer’s?” It should be clear that (D3) is similar to Schema D, and faces the same problems. In any case, (D3) is clearly inadequate to represent the content of my third-person understanding of Mary’s action, where this understanding is to be achieved in accordance with the rules (1), (2), and (3).

144


So then, what must I, the third-person explainer, do to gain an understanding of why Mary did what she did? I suggest that we take seriously, perhaps literally, the metaphors we find in Dray and like-minded thinkers like putting ourselves in the agent’s shoes, identifying with the agent, reenacting the agent’s calculations, and so on. We should abandon attempts like (D2) and (D3) and go back to (D1), the agent’s first-person practical inference. And here I believe we can borrow some idioms from the simulation theory of mind reading without committing ourselves to one side or the other in the current simulation vs. theory theory debate. What I should do to recapture the agent’s practical inference is to put myself in the agent’s place and run (D1) for myself “off-line.” That is, I simulate the agent’s practical deliberation by imagining myself in the agent’s situation and asking myself what is the appropriate thing to do—appropriate for me to do. Now here is the crucial point: For me to succeed in understanding Mary’s action on the present approach, I must come out where Mary came out, or at least nearly so, about what is the appropriate thing to do in the given situation. That is, the outcome of my practical deliberation must match Mary’s if I am to achieve a third-personal understanding of what Mary did. This understanding is third-personal in the sense that I am a thirdperson observer; but my understanding arises from my simulation, or reenactment, of Mary’s first-personal practical deliberation. The condition that the outcome of my simulated deliberation must coincide with Mary’s, which we might call the convergence condition, is a substantive requirement, and its satisfaction is not a priori guaranteed. At least, we must understand the process of recapitulation or re-enactment in such a way that the convergence requirement is not automatically met. To insure this, we should stipulate that in his simulated deliberation the re-enactor retain his own “principles of action,” to use Dray’s term; that is, he is not to adopt the principles of action of the agent being simulated, although of course the simulation begins with his imagining himself into the agent’s particular situation including her beliefs, desires, and goals. To allow, or require, him to also adopt her rules of deliberation and decision would make the process of simulation trivial and meaningless: if both the agent and the simulator used the same principles of deliberation, and used them correctly, on the same set of initial data, they would logically be guaranteed to reach the same outcome. What this means is that the simulator is no longer simulating in a


145

substantive sense but becoming one with the agent, and there is not explanatory value in exactly recapitulating the agent’s deliberation as the agent herself did. The simulator already knows where the agent came out; the informative, explanatorily valuable insight comes from the fact that when the simulator applies his own principles of action to the initial data available to the simulator, his deliberation reaches an outcome that is identical with, or closely similar to, the agent’s. It is because his deliberation is carried out on the basis of his action principles that his judgment of the appropriateness of actions counts as his appraisal of the action; if the explainer were to use the agent’s action principles, he would not be making his own independent assessment of the action. The simulator’s understanding of another’s action lies precisely in the fact that his deliberation, using his deliberative principles, reaches the same conclusion as the agent’s deliberation based on her action principles. That is how he can say “Now I see why she did it; I would have done the same in her situation!” Dray speaks of understanding “the actions of beings like ourselves.”25 What I am suggesting then is that this talk of “beings like ourselves” be understood to refer to agents who by and large share our principles of action. It is the actions of such agents that we are able to understand, on my interpretation of Dray’s view of rational explanation. What precisely these principles of action are is a question that must be set aside. But if there were, or could be, agents who habitually violate the principle “I desire G; doing A is required to bring about G; therefore, ceteris paribus, A is the right thing for me to do,” we would be unable to rationally explain and understand their actions. For the expression “beings like ourselves” to have a significance, we must first fix the reference point, namely “ourselves”; I am suggesting that “ourselves” should be taken to refer to our principles of action among other things. If we totally “empty ourselves” and become contentless, “beings like ourselves” would have no meaning. The conclusion that we can only understand actions of agents like ourselves, in sharing the same action principles, should not surprise us: if we take seriously things like empathetic identification, imaginative projection, re-enacting, and re-experiencing, as more than picturesque metaphors, what we might call “the circle of rational intelligibility” is only to be expected. This line of thinking seems to mesh well with the 25 Laws and Explanation in History, p. 110.

146


doctrine, much debated in the 19th century, that the “human” sciences are essentially different from the natural sciences in that, in the former but not in the latter, the investigators are creatures essentially like the subjects being investigated, and that this difference, in addition to introducing both hazards and opportunities, leads to some significant methodological differences, including modes of explanation and understanding.26 In this respect, the normative model explicated here differs fundamentally from both Davidson’s causal model and Hempel’s nomological model. These models do not have the implication that the explainers of actions must themselves be agents, or that we can only understand the actions of agents similar to ourselves. The explainer could be any cognizer, not necessarily a full-blown agent, and Martian scientists visiting this planet could produce Davidsonian or Hempelian explanations of our actions, regardless of what their action principles are like.27 I believe it is worthwhile to note that our outcome bears an interesting resemblance to a well-known thesis of Davidson’s concerning radical interpretation, a process whereby we interpret our subjects’ speech and attribute to them intentional mental states, only with observable behavior and physical environment as our evidence base. The thesis I have in mind holds that we can radically interpret only those speakers with whom we share most of our beliefs, at least beliefs about observable states of affairs, and that the beliefs of interpretable subjects must for the most part be true. Davidson turns this into an argument with the conclusion that there are no uninterpretable schemes of concepts and beliefs.28 This argument is not uncontroversial, but it is interesting in the present context, since if it works for Davidson, a parallel argument could be attempted to show that there are no unexplainable, or unintelligible, actions. It seems to me that the similarity between Davidson’s thesis about radical interpretation and our result about understanding action cannot be a mere coincidence, and that it may well arise from a shared conception of what it is for us to understand 26 For a helpful survey of the 19th-century debate on this issue see G.H. von Wright, Explanation and Understanding (Ithaca, NY: Cornell University Press, 1971), chap. 1. 27 Would Martians have belief-desire psychology in order for them to attribute beliefs and desires to us? If they had belief-desire psychology, wouldn’t they be agents like us? Perhaps, they would bypass belief-desire psychology and formulate explanations of human actions at the levels of physiology and physics. 28 See Donald Davidson, Inquiries into Truth and Interpretation (Oxford: Oxford University Press, 1984), especially the essays under “Radical Interpretation” and “Language and Reality.”


147

others—understand what they mean, what they believe and desire, and why they act the way they do.

V The essential task for someone like Dray who advocates a normative model of action explanation is to show how knowing that what was done was the right thing to do can make the action understandable and intelligible. This has been my task; my aim has been to give a plausible reconstruction of Dray’s views that shows how normativity can generate explanatory understanding—or, how the normative relation of justification can be an explanatory relation. Here is a brief summary of my overall line of reasoning. My starting point is the thesis that an agent’s self-understanding of his own actions consists in knowing, or being able to reconstruct, the practical deliberation that led to the action. I have tried to make this assumption appear plausible; but I am happy to accept it as my starting premise. I then argued that a third-party’s understanding of an agent’s action is to be able to recapitulate, or re-enact, the agent’s deliberation leading to the action. Much of the previous section was devoted to exploring what such reenactment could mean and what it concretely requires. A noteworthy implication of the normative model sketched here is that we can understand the actions of only those agents who are “like ourselves” in that they share with us “principles of action,” the rules and principles regulating practical deliberation.

7 Explanatory Realism, Causal Realism, and Explanatory Exclusion I Explaining is an epistemological activity, and “having” an explanation is, like knowing, an epistemological accomplishment. To be in need of an explanation is to be in an epistemologically imperfect state, and we look for an explanation in an attempt to remove that imperfection and thereby improve our epistemic situation. If we think in terms of the traditional divide between knowledge and reality known, explanations lie on the side of knowledge—on the side of the “subjective” rather than that of the “objective,” on the side of “representation” rather than that of reality represented. Our explanations are part of our knowledge of the world. Knowledge implies truth: we cannot know something that is not the case. On a realist view of knowledge, every bit of knowledge has an objective counterpart, the thing that is known which is itself not part of knowledge—at least, not part of that particular bit of knowledge. But exactly what is it that we know when we have an explanation? Exactly in what does explanatory knowledge consist? If explanations constitute knowledge, it makes sense to ask, for each explanation that we “have,” exactly what it is that we know in virtue of having that explanation. And when we gain a new explanation, precisely what change takes place in our body of knowledge? We usually think of knowledge as consisting of a set of “propositions,” thought to represent

I am indebted to David Benfield, Paul Boghossian, Brian McLaughlin, Joseph Mendola, and Michael Resnik for discussions of some of the issues taken up in this paper.

REALISM AND EXPLANATORY EXCLUSION

149

“facts” of the world. These propositions are discrete items, although they form a complex network of logical and evidential connections. How are we to represent explanatory knowledge within such a picture? Where do we locate explanations in a scheme of propositions? We can think of an explanation as a complex of propositions or statements divisible into two parts, explanans and explanandum proposition.1 Since explanations can take a variety of linguistic forms, this division is rough; in particular, it is not to be taken to imply that an explanation is an argument or inference, with the explanans as premise and the explanandum as conclusion. Let us focus on explanations of individual events. Such explanations typically explain an event (why a given event occurred) by reference to another event (or set of events). Let E be the explanandum, to the effect that a certain event e occurred. Let C be an explanans for this explanandum. C, let us assume, is the statement that event c occurred. Suppose then that we “have” this explanation—that is, C and E are related as explanans to explanandum in our body of knowledge (call this the “explanans relation”). What is the relationship between events c and e? What I want to call explanatory realism takes the following position: C is an explanans for E in virtue of the fact that c bears to e some determinate objective relation R. Let us call R, whatever it is, an “explanatory relation.” (The explanans relation relates propositions or statements; the explanatory relation relates events or facts in the world.) The explanatory relation is an objective relation among events that, as we might say, “ground” the explanans relation, and constitutes its “objective correlate.” On the realist view, our explanations are “correct” or “true” if they depict these relations correctly, just as our propositions or beliefs are true if they correctly depict objective facts; and explanations could be more or less “accurate” according to how accurately they depict these relations. Thus, that c is related by explanatory relation R to e is the “content” of the explanation consisting of C and E; it is what the explanation “says.” Although the attribution of truth or correctness to explanations is essential to explanatory realism, it by itself is not sufficient; those who reject explanatory realism in our sense, too, can speak of the truth or falsity of an explanation—for example, in the sense that propositions constituting the 1 I shall sometimes use “explanandum” as short for “explanandum proposition (or statement).” This should cause no confusion.

150


explanans are all true. What matters to realism is that the truth of an explanation requires an objective relationship between the events involved. By an “objective relation,” I have in mind a relation that at least meets the following condition: that it is instantiated does not entail anything about the existence or nonexistence of any intentional psychological state—in particular, an epistemological or doxastic state—except, of course, when it is instantiated by such states. I am not suggesting that the explanatory relation holding for events is all there is to explanations, or to the explanans relation. Just as knowledge requires more than truth, explanations presumably must meet further requirements (“internal” conditions—perhaps logical and epistemic ones), although exactly what these are does not concern us here. What could such an R be in virtue of which an event is correctly cited in the explanation of another? The obvious first thought is this: R is the causal relation. Perhaps there are noncausal explanations of individual events; however, few will deny that the causal relation is at least one important special case of R. And there are those who hold that the causal relation is the only explanatory relation—at least the principal one.2 Explanatory irrealism, on the other hand, would be the view that the relation of being an explanans for, as it relates C and E within our epistemic corpus, is not, and need not be, “grounded” in any objective relation between events c and e. It is solely a matter of some “internal” relationship between items of knowledge. Perhaps, there are logical, conceptual, or epistemic relationships among propositions in virtue of which one proposition constitutes an explanans for another, and when that happens, we could speak of the events represented as being related by an explanatory relation. That is, given the explanans relation over propositions, a relation over the events they represent could be defined: c explains (is related by R to) e just in case C is an explanans for E. But an R so defined would fail to be an objective relation, as required by realism, for it would depend crucially on what goes on within our body of knowledge and belief. 2 e.g., Wesley C. Salmon, Scientific Explanation and the Causal Structure of the World (Princeton, NJ, 1984); David Lewis, “Causal Explanation,” in Philosophical Papers, vol. 2 (Oxford, 1986). However, there are relations other than causation one might want to consider: e.g., the relation of supervenience, the micro-reductive relation. Whether or not these possible explanatory relations require the same explanandum as the causal relation is another question; see Robert Cummins’s distinction between “explanation by subsumption” and “explanation by analysis” in his The Nature of Psychological Explanation (Cambridge, Mass., 1983), chap. 1. See also Peter Achinstein, “A Type of Non-Causal Explanation,” Midwest Studies in Philosophy 9 (1984): 221–243.


151

In the following passage Wesley Salmon gives a clear and forceful expression to the realist view of explanation: We need not object to [the purely psychological conception of explanation] merely on the ground that people often invoke false beliefs and feel comfortable with the ‘explanation’ thus provided. . . . We can, quite consistently with this approach, insist that adequate explanations must rest upon true explanatory bases. Nor need we object on the ground that supernatural ‘explanations’ are often psychologically appealing. Again, we can insist that the explanation be grounded in scientific fact. Even with those restrictions, however, the view that scientific explanation consists in release from psychological uneasiness is unacceptable for two reasons. First, we must surely require that there be some sort of objective relationship between the explanatory facts and the fact-to-be-explained.3 However, merely to hold that C is an explanans for E just in case c is a cause of e is not necessarily to espouse explanatory realism. Whether that is so depends on one’s conception of causation. Consider, for example, Hanson, who writes: The primary reason for referring to the cause of x is to explain x. There are as many causes of x as there are explanations of x.4 Causes certainly are connected with effects; but this is because our theories connect them, not because the world is held together by cosmic glue. The world may be glued together by imponderables, but that is irrelevant for understanding causal explanation. The notions ‘the cause x’ and ‘the effect y’ are intelligible only against a pattern of theory, namely one which puts guarantees on inference from x to y.5 For Hanson, causal relations essentially depend on an appropriate conceptual interlocking of our descriptions as provided by the theories we accept. He makes it evident, in the quoted passages, that he views the causal relation between x and y as derivative from an inferential relation from x to y, and the inferential relation as intimately associated with explanation; it is also 3 Scientific Explanation and the Causal Structure of the World, p. 13 (emphasis in the original). Explanatory realism, as I have characterized it, appears closely related to what Salmon calls “the ontic model” of scientific explanation. 4 Norwood Russell Hanson, Patterns of Discovery (Cambridge, 1958), p. 54. 5 Ibid. 64. See, for a view similar to Hanson’s but worked out in greater detail, William Ruddick, “Causal Connection,” Synthese 18 (1968): 46–67.

152


evident that he does not take the dependence of causation on inference and explanation to be merely epistemological. If one accepts this view of causation and causal explanation, there is nothing realist about the position that causal explanations hold just in case the causal relation holds. For causal relations, on such an approach, depend on inferential-explanatory connections which are primary and more basic. More generally, if one wants to analyze causation itself in terms of explanation,6 one would be rejecting explanatory realism—unless one could identify an objective relation other than causation as the explanatory relation. But what could such a relation be? One might wish to propose the nomological relation as a candidate. The idea is this: that two events, c and e, are “subsumed under,” or “instantiate,” an appropriate law is the objective correlate of the explanans relation for C and E. Giving an account of “subsumption under a law” without presupposing causal notions is not an easy task, but let us not press this issue.7 The point to consider is how we understand the notion of “law.” If a law is taken as “mere Humean constant conjunction,” with no modal or subjunctive force intimating some tie of “necessitation,” this approach would give us realism. But it is highly dubious that a conception of an explanatory relation based on such a notion of “law” could provide a basis for an adequate account of explanation; it is even more dubious that an analysis of causation based on such a conception of explanation will come close to capturing our concept of causation. On the other hand, if laws are endowed with sufficiently strong modal force, it is doubtful whether the nomological relation will be distinguishable, in any meaningful way, from the causal relation.8 Indeed, the nomological account of causation is one of the more influential approaches to the analysis of the causal relation. An analysis of causation in terms of a conception of explanation that in turn is based on the nomological relation as the explanatory relation will essentially be just a nomological analysis, possibly with some psycho-epistemological embellishments. It would be difficult to see why 6 See, for example, Michael Scriven, “Causation as Explanation,” Nouˆs 9 (1975): 3–16. I discuss Scriven’s account in “Causes as Explanations: A Critique,” Theory and Decision 13 (1981): 293–309. Some of the present material has been drawn from this paper. 7 For some general difficulties in explaining “subsumption under a law,” see Donald Davidson, “Causal Relations,” Journal of Philosophy 64 (1967): 691–703, and my “Causation, Nomic Subsumption, and the Concept of Event,” Journal of Philosophy 70 (1973): 217–236. 8 Where the nomological and the causal relation do not match up, the former also fails to yield the explanatory relation.


153

one should not just go for a direct nomological analysis of causation, and use the causal relation as one’s explanatory relation. We must conclude that any attempt to analyze causation as explanation will result in a form of explanatory irrealism. For an analysis of causation to be a genuine explanatory analysis, the concept of explanation assumed as the basis of analysis must be a robustly epistemological and psychological notion whose core is constituted by such notions as understanding and intelligibility, not some pale, formal reconstruction of it. If, for example, the Hempelian deductive-nomological conception of explanation is used to explain causation, the result is not a genuine explanatory analysis of causation but rather the old standby, the nomological-subsumptive, or quasi-Humean, analysis. Thus, on a real explanatory approach to causation, causation will turn out to be a nonobjective psycho-epistemological relation and, therefore, fail to serve as an objective correlate of the explanans relation.

II It is plausible to conclude, therefore, that explanatory realism requires the causal relation as an explanatory relation. As I said, we may leave open the question whether the causal relation is the only explanatory relation. But at least this much is certain: in both everyday and most scientific contexts,9 explanations of individual events are predominantly causal explanations in the sense that the events cited in the explanation of an event are its causes and, further, their explanatory efficacy is thought to stem from their causal status. And when each of a class of events can be given a similar causal explanation, we may have a causal explanation of a regularity. We shall in this paper focus exclusively on the causal relation as our explanatory relation; our general metaphysical points should be valid, mutatis mutandis, for other explanatory relations if any exist. Explanatory realism says this about causal explanations: a causal explanation of event e in terms of event c (“e occurred because c caused it”) is correct, or true, just in case c did as a matter of objective fact cause e. That the causal relation holds between the two events constitutes the “factual content” of the explanation. This may sound obvious and trivial. 9 By the qualification “most,” I intend to leave out consideration of what some tell us goes on at the deepest and most abstract levels of theoretical physics.

154


Perhaps it sounds obvious only because we take explanatory realism for granted. But it certainly is not trivial. It requires, for its intended realist purposes, that causality itself be an objective feature of reality. This doctrine, which we may call causal realism, has not gone unchallenged. Hume’s celebrated critique of “necessary connection” as an objective relation characterizing events themselves was perhaps the first—clearly the most influential— expression of a systematically articulated, irrealist position on causation. He wrote, “Upon the whole, necessity is something that exists in the mind, not in objects.”10 Hume well understood the causal realist’s sentiments: But though this be the only reasonable account we can give of necessity, the contrary notion is so riveted in the mind from the principles above-mentioned, that I doubt not but my sentiments will be treated by many as extravagant and ridiculous. What! the efficacy of causes lie in the determination of mind! As if causes did not operate entirely independent of the mind, and would not continue their operation, even though there was no mind existent to contemplate them, or reason concerning them. Thought may well depend on causes for its operation, but not causes on thought. This is to reverse the order of nature, and make that secondary, which is really primary.11 Hume was understanding, but in the end dismissive: I can only reply to all these arguments that the case here is much the same as if a blind man should pretend to find a great many absurdities in the supposition that the colour of scarlet is not the same with the sound of a trumpet, nor light the same with solidity. If we have really no idea of a power or efficacy in any object, or of any real connection betwixt causes and effects, it will be to little purpose that an efficacy is necessary in all operations.12 Hume regarded the other ingredients he identified in the causal relation, namely temporal precedence, spatiotemporal contiguity (or connectability), and constant conjunction, as objective and mind-independent features of causally connected events;13 but evidently he thought that necessitation, too, 10 A Treatise of Human Nature, ed. L. A. Selby-Bigge (Oxford, 1888), p. 165. 11 Ibid. 167. 12 Ibid. 168. 13 For discussions of this and other matters concerning Hume on causation and necessity, see Barry Stroud, Hume (London, 1977), chaps. 3, 4, and Tom L. Beauchamp and Alexander Rosenberg, Hume and the Problem of Causation (New York and Oxford, 1981), especially chap. 1.


155

was an essential element in our philosophically unenlightened (by his light) concept of causation. Most philosophers will now agree that an idea of causation devoid of some notion of necessitation is not our idea of causation—perhaps not an idea of causation at all. According to most conceptions of causation now current, at any rate, Hume was a causal irrealist par excellence. Hume, our original causal irrealist, had some illustrious followers. Russell ridiculed causation as “a relic of a bygone age,” recommending the “extrusion” of the word “cause” from the philosophical vocabulary;14 Wittgenstein said, “Belief in causal nexus is superstition.”15 The positivist-inspired suspicion of modalities, counterfactuals, and the like, which characterized much of analytic philosophy during the first two-thirds of this century, is of a piece with Hume’s causal irrealism in their fundamental philosophical motivation, and it seems that many prominent philosophers in the analytic tradition during this period consciously avoided serious discussion of causality, making little use of it in their philosophical work.16 More recently, Hilary Putnam has attacked the idea of “non-Humean causation” as a physically real relation.17 I think it is more difficult than one might at first suppose to find philosophers who have consciously advocated an unambiguously realist conception of causality.18 According to causal realism, therefore, causal connections hold independently of anyone’s intentional states—in particular, epistemological or doxastic states—except, of course, when the causal connections concern such states. The realist believes, as Hume observes in the quotation above, that causal relations—the same ones—would hold even if there were no conscious beings to “contemplate them, or reason concerning them.” This means that according to causal realism every event has a unique and determinate causal

14 Bertrand Russell, “On the Notion of Cause,” Proceedings of the Aristotelian Society 13 (1913): 1–26. 15 Ludwig Wittgenstein, Tractatus Logico-Philosophicus (London, 1922), 5.1361. However, Wittgenstein may have had in mind by “causal nexus” something much stronger than what we would now understand by “causal necessity.” 16 C. J. Ducasse and Hans Reichenbach were among the exceptions. 17 “Is the Causal Structure of the Physical Itself Something Physical?,” Midwest Studies in Philosophy 9 (1984): 3–16. 18 Some possibilities among recent writers: Salmon, Scientific Explanation and the Causal Structure of the World; J. L. Mackie, The Cement of the Universe (Oxford, 1974), especially chap. 8. Quine seems to have studiously avoided discussing causation or making use of it in his philosophical work. The uses to which Donald Davidson has put the concept of causation indicate a realist attitude; consider, for example, his commitment to an event ontology and his causal criterion of event individuation in his Essays on Actions and Events (Oxford, 1980). But he may reject the terms in which I have formulated the positions.

156


history whose character is entirely independent of our representation of it. We may come to know bits and pieces of an event’s causal history, but whether we do, or to what extent we do, and what conceptual apparatus is used to depict it, do not in any way affect the causal relations in which events stand to other events. This entails that the existence and character of events themselves must be an objective and determinate fact; that is, causal realism makes sense only in the context of global realism. Earlier, we raised the question of how explanatory knowledge is represented in our body of knowledge—that is, what it is that we know when we have an explanation of an event. The explanatory realist appears to have a simple answer: To “have an explanation” of event e in terms of event c is to know, or somehow represent, that c caused e; that is, explanatory knowledge is causal knowledge, and explanations of individual events are represented by singular causal propositions. Thus, explanatory knowledge is propositional knowledge of a certain kind, and to gain an explanation of an event is to learn a further fact about that event. But is there an alternative to representing explanations as additional bits of propositional knowledge? Isn’t explanatory knowledge a kind of knowledge, and isn’t all knowledge, in an epistemologically relevant sense, a matter of knowing that? Although I do not know whether anyone has held a view like this, it is possible to hold, I think, that explanations are essentially a matter of how a body of knowledge is organized or systematized—a matter of there being certain appropriate patterns of coherence among items of knowledge. That is, to “have” an explanation of why E in terms of C—that is, to “have” C as an explanans for E—is simply for the two propositions C and E to be appropriately related within our epistemic corpus; it is not a matter of there being a further proposition within it. According to this view, therefore, explanatory knowledge supervenes on nonexplanatory knowledge: if you and I know exactly the same first-order, factual propositions (roughly, propositions that can serve as explananda and elements of an explanans), we would share the same explanations. Various considerations might lead us to qualify this conclusion; for example, one might construe the notion of “having an explanation” in such a way as to require the subject’s awareness that the explanans is appropriately related to the explanandum. Thus, one might want to suggest that the presence of the two propositions in our body of knowledge is not enough, even if they in fact instantiate a required explanatory pattern, and insist that


157

we must somehow mentally “bring them together” and “see” that they do so. Caution is required, however; pursuing this line may take one back to the propositional view of explanatory knowledge. At any rate, the nonpropositional, “pattern” view of explanatory knowledge differs from the propositional view on the following point: gaining a new explanation, on the pattern view but not on the propositional view, does not necessarily involve acquiring new information about facts of the world. It seems clear that explanatory realism leads to the propositional view of explanatory knowledge; it makes “having” an explanation a matter of knowing a certain proposition to be true. On the other hand, explanatory irrealism, although it has an affinity for the pattern view, is not committed to it; it appears consistent with the propositional view. One might hold, for example, that a certain conceptual-epistemic relation between an explanandum and its explanans is what is fundamentally constitutive of an explaining relation, there being no independent objective relation characterizing the events represented by the explanans and the explanandum that grounds it,19 but that “having” an explanation is a matter of knowing that this relationship does in fact hold for the explanans-explanandum pair. This, however, may not be a plausible view; it is naturally construed as requiring anyone who “has” any explanation of anything at all to know what the explanans relation is, something that few philosophers would confidently claim to know. In any case, those who find the propositional view of explanatory knowledge too simplistic, or otherwise unpalatable, would have to settle for explanatory irrealism; explanatory realism is not an option for them. What difference does the choice between explanatory realism and irrealism make? We have already seen that explanatory realism plausibly entails causal realism. Does explanatory irrealism entail causal irrealism? There evidently is no strict inconsistency in holding both explanatory irrealism and causal realism. However, the combination seems somewhat incongruous and difficult to motivate: though acknowledging causation as a genuine relation in the world, the position denies it any essential role in explanation, severing the intuitive and natural tie between causality and explanation.20 What, then, would be the point of the causal relation? The concept of 19 Recall our earlier discussion of Hanson; he seems to have held a view like this. 20 Peter Achinstein’s views in The Nature of Explanation (New York and Oxford, 1983), chap. 7, seem to approximate this position.

158


causation, of course, has many roles to play, but it seems that its explanatory role is a central one, being closely tied to its other important roles. I think that the combination of explanatory irrealism and causal realism, though logically consistent, is not a plausible position. We have also seen that explanatory realism entails the propositional account of explanatory knowledge, whereas explanatory irrealism, again, seems consistent with each of the two alternatives, the propositional view and the nonpropositional, pattern view. I think that the issue of causal realism versus irrealism and that concerning the nature of explanatory knowledge are significant issues, both interesting in themselves and important in what they imply for other philosophical problems. Problems about what explanatory knowledge consists in—that is, what “understanding” something amounts to—have been almost entirely neglected within traditional epistemology; this is surprising in view of the centrality of explanation in philosophy of science, which, by and large, is the epistemology of scientific knowledge. The issue of causal realism is obviously important: whether causal relations are real and objective, or mere projections of the cognizing mind, is an issue that directly affects the significance of causation within both science and philosophy. If it is an objective relation characterizing physical events in the world, is it physically reducible, or physically based in some sense, as we expect of other physical properties and relations? If not, what accounts for its special status? Which of the special sciences are responsible for investigating the properties of the causal relation itself?21 As for the philosophical implications of the choice between causal realism and irrealism, it is an interesting question, for example, whether any of the so-called causal theories (of perception, memory, knowledge, action, event-identity, reference, time, persistence, properties, and no doubt many others) will be able to retain, under an irrealist conception of causation, what plausibility it enjoys. It is also an interesting question whether a substantive version of global realism can be combined with causal irrealism. I suspect that if all causal facts are taken away from the world, not much of interest may remain—the world would become so impoverished, a pale imitation of a world, that we may not care much whether it is real or only “ideal.” (If all those “causal theories” mentioned above are true, a world devoid of causal relations would be one in which there is no perception, 21 See Putnam, “Is the Causal Structure of the Physical Itself Something Physical.”


159

no knowledge, no naming or referring, no intentional action, no time, no persisting object, and none of the rest. It would also be a world in which there are no killings, no breakings, no pushings or pullings, and so on.) Some may consider it a disadvantage of explanatory realism that it comes only in a package with causal realism, whereas explanatory irrealism can in principle be purchased separately. However, others may consider that an advantage: causal realism gives more content to explanatory realism, and as a result explanatory realism can do work that its rival cannot. Moreover, there is a certain satisfying unity in the combination of explanatory and causal realism. In any case, there seems to be some incongruity, as we saw, in combining explanatory irrealism and causal realism, so that an explanatory irrealist may in effect have no real choice but to embrace causal irrealism as well. In what follows, I will explore the implications of the realist view of explanation for the issue of “explanatory exclusion” and the irrealist (or “internalist”) implications of the Hempelian inferential view of explanation.

III I have argued elsewhere22 that proffered explanations of a single event, with mutually consistent explanantia, can exclude one another in the following sense: there can be no more than a single complete and independent explanation of any one event, and we may not accept two (or more) explanations of a single event unless we know, or have reason to believe, that they are appropriately related—that is, related in such a way that one of the explanations is either not complete in itself or dependent on the other. This constraint on explanations, which we may call the principle of explanatory exclusion, has two clauses: the first is about the existence of explanations, the second about the acceptance of explanations. The first clause, I shall argue, can be seen as a plausible thesis if we assume explanatory realism. We shall not discuss the second clause here. Suppose, then, that each of C1 and C2 is claimed to be a causal explanans for E. Let c1, c2, and e be the events represented by C1, C2, and E. According to explanatory realism, it follows that c1 caused e and also that c2 caused e. How are we to understand this situation? There are various possibilities: 22 “Mechanism, Purpose, and Explanatory Exclusion,” Philosophical Perspectives 3 (1989): 77–108.

160


(1) It turns out that c1 ¼ c2. A single event is picked out by nonequivalent descriptions. Here, there is in reality only one pair of events related by the explanatory relation (that is, the causal relation), and this gives sense to the claim that there is, here, one explanation, not two. The exclusion principle makes sense only if a criterion of individuation is assumed for explanations—that is, only if we can make sense of “same” and “different” as applied to explanations. Now, explanatory realism yields a natural way of individuating explanations: explanations are individuated in terms of the events related by the explanatory relation (the causal relation, for explanations of events).23 For on realism it is the objective relationship between events that ultimately grounds explanations and constitutes their objective content. This provides us with a basis for regarding explanations that appeal to the same events standing in the same relation as giving, or stating, one explanation, not two—just as two inequivalent descriptions can represent the same fact. Thus, on explanatory realism, we can make good sense of the idea that logically inequivalent explanations can represent the same explanatory relation, and therefore state the same explanation. To the explanatory irrealist, this way of individuating explanations would be unmotivated: explanations would be more appropriately individuated in terms of descriptions or propositions and their internal logical, conceptual, and epistemic relationships. Nothing needs to prevent the explanatory realist from accepting this “internal” individuation criterion as well, as defining another useful sense in which we can count explanations. The point is only that explanatory realism motivates an “objective” individuation of explanations, which is both intuitively plausible and well suited for the exclusion principle. (2) c1 is reducible to, or supervenient on, c2. This sort of relationship might obtain, for example, on some accounts of the mind-body relation, which, though eschewing an outright psychophysical identification, nonetheless recognizes the reductive or supervenient dependency of the mental on the physical. In such a case, the causal relation involving the supervenient or reduced event must itself be thought of as supervenient or reducible to

23 If relations other than the causal relation can serve as explanatory relation, they can also be considered as a basis for individuation; however, that probably would be redundant. It is unlikely that when the explanatory relation is different, exactly the same events would be involved.


161

the causal or nomological relation involving the “base” event.24 In this sense, the two explanations are not independent; for the one involving the reduced causal (that is, explanatory) relation is dependent on the one representing the “base” causal relation. This, again, is an example of realist thinking: dependency between explanations is understood in terms of the dependency between the objective explanatory relations that they represent. (3) c1 and c2 are only partial causes, being constituents in a single sufficient set of causal conditions. Example: You push the stalled car and I pull it, and the car moves. In this case, neither explanation is complete: each gives only a partial picture of the causal conditions that made up a sufficient cause of the effect. This sense of explanatory completeness, understood in terms of sufficient cause, is again entirely natural within the realist picture. For, according to the realist view, the causal relation between events constitutes the objective correlate, or content, of the explanans relation; where a particular causal relation gives us a cause event that is only a partial cause, or one among the many constituents of a sufficient cause, the corresponding explanans, too, can be thought to be only partial and incomplete. Conversely, when the causal relation provides a sufficient cause, the explanans can also be said to be complete and sufficient. The realist scheme also yields a more global sense of “complete explanation,” one in which a complete explanation of an event specifies its entire causal history in every detail (as we noted earlier, under explanatory realism each event has a unique determinate causal history). This is an idealized sense of completeness, and no explanation can be complete in that sense (the notion of an ideally complete explanation, however, may be useful in explicating the concept of explanation).25 Obviously, in this idealized sense there is at most one complete explanation of any given event; again, though obvious and uninteresting, this is not trivial, unless causal realism is trivial. (4) c1 and c2 are different links in the same causal chain leading to event e. But they are not independent: the later event is, then, causally dependent on

24 For further discussion see my “Epiphenomenal and Supervenient Causation,” Midwest Studies in Philosophy 9 (1984): 257–270. I believe that the case in which c1 “generates” c2 in Alvin Goldman’s sense (see his A Theory of Human Action (Englewood Cliffs, NJ, 1970)) can be handled in a similar way, although the details may have to be somewhat different. 25 Compare Peter Railton’s notions of “ideal explanatory text” and “ideal causal D-N text” in his “Probability, Explanation, and Information,” Synthese 45 (1981): 233–256; see also David Lewis, “Causal Explanation,” Philosophical Papers, vol. 2.

162


the earlier one, and, therefore, the two explanations are not independent. This, too, reflects realist thinking: two explanations are thought to be nonindependent because the explanatory relations represented by them are not independent. (5) c1 is part of c2.26 The explanations, then, are not independent; nor can they both be complete. (6) c1 and c2 are independent, each a sufficient cause of e. This, then, is a standard case of “causal overdetermination.” Do we in this case have a counterexample to the explanatory exclusion principle? Why aren’t both explanations, “e happened because c1 caused it” and “e happened because c2 caused it,” sufficient and independent explanations? This is an interesting case from the point of view of both explanatory exclusion and the question of explanatory realism versus irrealism, and we shall discuss this in some detail. Hempel has called cases like this “explanatory overdetermination”:27 Suppose that a copper rod is heated while simultaneously being subjected to longitudinal stress. As a result, its length increases. Two deductivenomological (hereafter “DN”) arguments can be formulated: the first would invoke the lawlike premise that copper rods lengthen when they are heated, and the ‘initial condition’ that this particular copper rod was heated on this occasion; the second would appeal to the law stating that copper rods increase in length when subjected to longitudinal stress, and the initial condition that this copper rod was subjected to that kind of stress. The two arguments share the same conclusion, the statement that the rod’s length increased on this occasion. According to the standard DN account of explanation, therefore, each argument counts as an explanation. It is not surprising that Hempel rejects the view that these DN arguments are not “complete” as explanations. He writes: It might be objected that—even granting the truth of all the premises—both accounts are unacceptable since they are “incomplete”: each neglects one of the two factors that contributed to the lengthening. In appraising the force of this objection it is again important to be clear about just what is to be explained. If as in our example, this is simply the fact that Lr, i.e., that r lengthened, or that there was some 26 Karl Pfeifer brought this case to my attention. 27 Carl G. Hempel, Aspects of Scientific Explanation (New York, 1965), pp. 418–420.


163

increase in the length of r, then, I think, either of the two arguments conclusively does that, and the charge of incompleteness is groundless.28 Here he seems simply to affirm that, as an explanation of why the rod lengthened, “each of the two arguments conclusively does that.” But why does he say this? The use of the term “conclusively” suggests that he was moved by the consideration that each DN argument provides a premise-set that is deductively conclusive for the truth of the explanandum statement. This is not surprising. For, fundamental to the DN conception of explanation is the idea that explanations are inferences or arguments of a certain form. Given this assumption, a natural sense of “completeness” or “sufficiency” emerges for explanations: when an argument has the correct DN form, it is complete and sufficient. Hempel writes: I think it is important and illuminating to distinguish such partial explanations . . . from what might be called deductively complete explanations, i.e., those in which the explanandum as stated is logically implied by the explanans; for the latter do, whereas the former do not, account for the explanandum phenomenon in the specificity with which the explanandum sentence describes it. An explanation that conforms to the D-N model is, therefore, automatically complete in this sense; and a partial explanation as we have characterized it always falls short of being a D-N explanation.29 As explanation is conceived under the DN model, there is nothing one can do to a DN argument to improve it in regard to its “completeness” as an explanation. One may be able to make it deeper, more perspicuous, more systematic, and so on; but what could one possibly do to make it “more complete”? The DN conception of explanation does not seem to leave room for any other sense of explanatory completeness than deductive conclusiveness. These considerations suggest that a preoccupation with the deductive or inferential character of explanation leads to a form of explanatory irrealism (“explanatory internalism,” perhaps, is more appropriate), and this is certainly what we see in Hempel. This internalist tendency is evident also in 28 Ibid. 418–420. 29 Ibid. 416–417.

164


Hempel’s well-known emphasis on the predictive character of explanation, and in one of his two conditions of adequacy on explanations, that is, the requirement of “explanatory relevance” to the effect that “explanatory information adduced affords good grounds for believing that the phenomenon to be explained did, or does, indeed occur.”30 Hempel’s idea that explanations are arguments, his condition of “explanatory relevance,” and his emphasis on the predictive aspect of explanations go hand in hand: they all point to explanatory irrealism—at least, point away from explanatory realism with the causal relation serving as the explanatory relation. Hempel’s primary focus in analyzing the structure of explanation is on the logical and conceptual characteristics of statements making up an explanation (the “internal” properties, as I have called them), not on the events or other entities these statements describe and their interrelations. In fact, we get from Hempel a precise and elaborately constructed definition of what an explanation is, but only a very intuitive and unanalyzed idea of what it is that a given explanation is an explanation of.31 Hempel’s treatment of causal explanation and causation is also symptomatic of this attitude:32 the idea of a DN argument, an essentially internal notion, is primary in the characterization of explanations, and the idea of causal explanation falls out of this characterization as a not-so-clearly-defined special case. Hempel evidently does not regard the concept of causal explanation, or that of causation, as at all crucial to a theory of explanation; his discussion of causal explanation often comes across as a concession to the popular practice of referring to causes and causal explanations, not something that he sees as essential to the development of his theory. From such an internalist perspective, it is entirely natural that each of the two DN arguments about the expanding copper is regarded as complete and sufficient in itself as an explanation. What does explanatory realism say about the expanding copper rod? If the heating and the stress are each an independent sufficient cause of the

30 Philosophy of Natural Science (Englewood Cliffs, NJ, 1966), pp. 47–49. See also his Aspects of Scientific Explanation, pp. 367–368. The other adequacy condition is the unexceptionable requirement of “testability,” to the effect that the explanatory premises must be capable of empirical test in a broad sense. 31 For further elaboration of this point and some suggestions, see my “Events and Their Descriptions: Some Considerations,” in Essays in Honor of Carl G. Hempel, ed. Nicholas Rescher et al. (Dordrecht, 1969). 32 Hempel, Aspects of Scientific Explanation, pp. 347–354.


165

rod’s lengthening, we have a standard case of causal overdetermination. Moreover, if, as explanatory realism seems to suggest, explanatory completeness is to be understood in terms of sufficient cause, it follows that in the present case we have two independent and complete explanations. Thus, explanatory realism seems to yield the same result as Hempel’s irrealism: both seem to contradict the explanatory exclusion principle. The explanatory realist who wants to save explanatory exclusion might deny that the rising temperature and the stress were each a sufficient cause of the event to be explained, and deny, more generally, that genuine instances of causal overdetermination exist. Peter Unger has claimed that each event has a single unique cause (at most),33 and if this is right, then not both the heating and the stress can be a cause of the lengthening. Therefore, there could be at most one causal explanation here. But Unger’s thesis is a radical one, too strong to be plausible: he construes it to entail the denial of transitivity of causation, and hence the impossibility of causal chains with more than two links. And, his arguments rely exclusively on a certain kind of linguistic evidence whose point I find difficult to evaluate. Martin Bunzl, too, has argued that there are no genuine cases of causal overdetermination.34 His basic point is that the usual examples, when closely scrutinized, turn out to be either cases of causal preemption or of joint cause. That is, one of the two alleged overdetermining causes preempts the other (by “getting there first”) so that the second, in fact, is not a cause of the effect in question, or else the two causes together make up a single sufficient cause, neither of them alone being sufficient. I think Bunzl’s arguments, on the whole, are plausible, though not conclusive.35 Thus, when applied to the case of the copper rod, his analysis would probably give this diagnosis: the particular lengthening that took place was caused by the single joint cause made up of the heating and the stress. Neither of the

33 “The Uniqueness in Causation,” American Philosophical Quarterly 14 (1977): 177–188. 34 “Causal Overdetermination,” Journal of Philosophy 76 (1979): 134–150. 35 Bunzl, however, says that his considerations depend essentially on a certain view of the nature and individuation of events associated with Donald Davidson, and that they are ineffective if we assume the sort of view of events that I myself have advocated, namely, one that takes events as propertyexemplifications. See Bunzl, “Causal Overdetermination,” p. 150. However, I am not convinced of this; I think Bunzl may have been misled by just the kind of consideration that lead Hempel to believe in explanatory overdetermination. It is interesting to note that Bunzl accepts explanatory overdetermination in Hempel’s sense (p. 145). On causal overdetermination, see also Louis E. Loeb, “Causal Theories and Causal Overdetermination,” Journal of Philosophy 71 (1974): 525–544.

166


two events was, in itself, a sufficient cause of it. Thus, a complete explanation of the lengthening must refer to both the heating and the stress as a single sufficient cause. We must set aside the question whether genuine instances of causal overdetermination exist. What is of interest to us here is that under explanatory realism, the causal relation can be made to do some real work, in characterizing and constraining explanations. As we saw, the association between causation and explanation, underwritten by explanatory realism, yields a principle of individuation for explanations and a notion of “complete explanation,” both essential to interpreting the principle of explanatory exclusion. We also saw that if causal overdetermination is not possible, that takes away one potential case of explanatory overdetermination. It seems to me that we are inclined to take these considerations involving causation as both natural and relevant in discussing the nature of explanation. What accounts for this inclination, I think, is our tacit acceptance of explanatory realism: for a causal explanation to hold, the explaining event must be a cause of the event explained. Given this connection between causal explanations and causal relations, we are able to use facts about the latter to say something about the former. To return briefly to the matter of explanatory exclusion: if our considerations are generally right (especially in the treatment of the six cases in which two causal explanations are offered for one event), explanatory realism is seen to provide a sense, as well as support, for the explanatory exclusion principle– except, perhaps, in the case of causal overdetermination, which we set aside without a clear-cut resolution.36 I believe it is more difficult, though not impossible, to interpret and argue for explanatory exclusion if by embracing explanatory irrealism we lose the causal handle on explanation.37 36 From the point of view of explanatory exclusion, causal overdetermination is not crucial; the exclusion principle has content of sufficient interest even if causal overdetermination is simply exempted. 37 I think explanatory exclusion can hold under explanatory irrealism as well; however, unlike explanatory realism, irrealism does not, I think, provide a positive basis for explanatory exclusion. For some considerations favoring explanatory exclusion that are not based on explanatory realism, see my “Mechanism, Purpose, and Explanatory Exclusion.”

8 Explanatory Knowledge and Metaphysical Dependence I There is a famous remark Aristotle made about knowledge: “Men do not think they know a thing unless they have grasped the ‘why’ of it” (Physics II, ch. 3; see also Metaphysics V, ch. 2). This remark is often quoted by writers on scientific explanation to underline the importance of explanation to scientific knowledge, and why, as philosophers, we should concern ourselves with understanding what explanation is—that is, to show that “analyzing” scientific explanation, or building a “model” of explanation, is a reputable philosophical enterprise. It is not uncommon to invoke Aristotle not only to stress the centrality of explanation to science but also to motivate a distinction between explanatory knowledge and descriptive knowledge—knowing why from knowing that.1 We are told, for example, that to know that the spectral lines of light received from distant celestial bodies show a shift toward red is one thing, but quite another to know why these red shifts occur. The former, it is suggested, is merely knowing a fact, but the latter involves something deeper and more valuable, an understanding of why the fact obtains. The mark of a theoretical science, it is often claimed, is precisely that it aspires to go beyond “phenomenological descriptions”

My thanks to Enrique Villanueva, Marcelo Sabates, and Margarita Santana, each of whom presented an instructive set of comments on this paper at the 1993 SOFIA conference in Tenerife, Spain, and also to others at the conference who made useful suggestions and comments—in particular, Richard Foley and Peter Railton. Many thanks also to Paul Humphreys and Wesley Salmon for helpful and clarifying comments. 1 For example, Wesley Salmon in his Scientific Explanation and the Causal Structure of the World (Princeton, NJ: Princeton University Press, 1984), p. 4.

168

EXPLANATORY KNOWLEDGE

of observed regularities and provide us with explanatory insights, an understanding of why things are the way they are. From this one would expect theorists of explanation to go into detailed discussions of what explanatory knowledge consists in, and how it differs from knowledge that is merely descriptive; you would, in short, expect a theory of explanation to be an epistemological theory, a theory of explanatory knowledge, which presumably is a special kind of knowledge. And you would expect such a theory to be developed as part of a general theory of knowledge. But, as everyone knows, this is not what happens: once the writers get past the obligatory paragraphs about the importance of explanation and understanding, they quickly launch into highly technical constructions heavily laden with an array of terms like “law,” “derivation,” “causality,” “probability,” “simplicity,” and their forbidding technical kins, and we never see a serious discussion of just what these concepts have to do with understanding something or making something intelligible. The actual theories of explanation that we have accumulated to date, such as Hempel’s covering-law theory, the causal theories of Salmon, Lewis, and Humphreys, the pragmatic theories of Bromberger, Achinstein, and Van Fraassen2— don’t look much like theories of understanding or accounts of a type of knowledge. Although explanatory understanding is sometimes mentioned, as we shall see below, in the writings of some writers who take the simplification-unification approach to explanation, it seems quickly to be lost sight of when serious theory construction begins, and terms like “understanding,” “intelligibility,” and “explanatory knowledge” seldom make an appearance once the initial stage-setting is over. The distortion introduced by this trend into the debate over the nature of explanation is reflected in the fact that the standard rule of the game, when the adequacy of some proposed “model of explanation” is at issue, used to be, and still largely is, the method of counterexamples: you test the model against your intuitive judgments about examples and nonexamples of explanation.3 2 See Salmon, Scientific Explanation and the Causal Structure of the World; David Lewis, “Causal Explanation,” in Lewis, Philosophical Papers II (Oxford: Oxford University Press, 1986); Paul Humphreys, The Chances of Explanation (Princeton: Princeton University Press, 1989); Sylvain Bromberger, “An Approach to Explanation,” in R.S. Butler, ed., Analytic Philosophy, Second Series (Oxford: Basil Blackwell, 1962); Peter Achinstein, The Nature of Explanation (Oxford: Oxford University Press, 1983); Bas Van Fraassen, Scientific Image (Oxford: Oxford University Press, 1980). 3 As Peter Railton observes in a related context; see his “Explanation and Metaphysical Controversy,” in Philip Kitcher and Wesley C. Salmon, eds., Scientific Explanation, Minnesota Studies in the Philosophy of Science, vol. 13, (Minneapolis: University of Minnesota Press, 1989).


169

The models are seldom evaluated by explicitly considering whether they give a satisfactory account of understanding, and the stock examples and counterexamples that have played a key role in the rise and fall of models of explanation have not been seriously analyzed on principled grounds—exactly why they provide, or fail to provide, genuine explanatory understanding or knowledge. To be sure, our preanalytic and untutored judgments (“intuitions”) about them must reflect certain inexplicit and unarticulated general principles we hold that interrelate an amorphous group of concepts including understanding something, answering why-questions, knowing a cause of something, dispelling puzzles, and so on. But the central task of a theory of explanation must be to bring into the open these tacit principles, formulate them in precise and systematic statements, and provide them with a rationale in the context of a general theory of knowledge. We see in retrospect that somehow the concepts that ought to have been at center stage in the theory of explanation had been shunted aside. Think of a theory of justice that never mentions justice, or a theory of number that never mentions numbers. The anomaly would be exactly the same. In a nutshell, then, the trouble is that we have had models, but not theories, of explanation. But philosophy of science, in which we normally place theories of explanation, is not the only field to blame; theory of knowledge is equally at fault. The traditional theory of knowledge has focused very narrowly on propositional knowledge, and its chief project has been the analysis of knowing that p, for a single, stand-alone proposition p. The mainstream analytic epistemology has paid little attention to the concept of understanding, or how bits and pieces of propositional knowledge can come to constitute an intelligible epistemic structure. The concept of explanation or of understanding surely does not undergo a radical, discontinuous transformation when we move from ordinary knowledge to scientific knowledge, any more than the concept of knowledge does. Explanations in the developed sciences may be more precise, more systematic and unified, calling on resources richer and more varied than what is available outside them, but surely the root concept of understanding must be much the same. It is only that understanding can be realized in a variety of ways. There is every reason then why explanation, that is, understanding, should be among the central concerns of general epistemology. The point of going over all this is to remind ourselves that explanation is an epistemic affair, and that explanations are among our important epistemic

170


achievements. The idea of explaining something is inseparable from the idea of making it intelligible; to seek an explanation of something is to seek to understand it, to render it intelligible. These are simple conceptual points, and I take them to be untendentious and uncontroversial.

II If we begin with these ideas firmly in mind, we can see that the following must be recognized as the central question of any theory of explanation: The Epistemological Question: What is it that we know—that is, what exactly is our epistemic gain—when we have an explanation of p? Typically, when we seek an explanation of why p we already know, or at least believe that we know, that p is the case. We want an explanation of p presumably because we are in an epistemically imperfect state vis-a-vis the fact that p, and there is something we need to know, or at any rate something epistemic we must acquire, if we are to ameliorate our epistemic situation. It may be that there is a proposition, q, about some particular matters of fact such that if we were to know that q, we would have an explanation of p; that is, the deficit may be one of insufficient information about specific matters of fact. But an explanation may be more than the possession of pieces of propositional knowledge; it may involve, or require, the seeing of a pattern, a certain relationship or configuration, in disparate propositions. Let E be a set of propositions that makes up an explanation of p (we will follow the standard terminology and call it an “explanans” of p). The general epistemological question then is this: when we have E as an explanation of p, what is it that we know? What exactly is the epistemic gain we have achieved? So far the issues have been put in general terms so that they may in principle arise for any explanation involving an explanandum represented by a proposition. These issues could, it seems, be generalized to apply to all kinds of explanation, including, e.g., explanations of concepts and meanings, the rules of a game, the significance of a social ritual, the function of a biological organ or social institution, etc.4 From this point on, though, we

4 This of course doesn’t mean that there will be an informative uniform answer that covers all varieties of explanations.


171

will focus, more or less exclusively, on explanations of events, as has been customary in philosophical discussions of explanation. In such an explanation, the explanandum is an event, or the occurrence of an event if you prefer, and it is represented in an explanation by a statement to the effect that the event occurred. Let us use lower-case letters, “e,” “f ,” etc. to stand for individual events, and the corresponding upper-case letters, “E,” “F ,” etc. for statements affirming that these events occur or occurred.5 Suppose then that an explanation of why an event, e, occurred is being sought, and that another event, g, has been invoked to explain it. Let us see how the situation might be represented under the Hempelian covering-law models, in particular, the deductive-nomological (“D-N”) model. On this model, an explanation of event e in terms of event g is a D-N argument, with E (the statement to the effect that e occurred) as the conclusion and G (the statement to the effect that g occurred) as a premise. In addition, the explanans will include a general law, L, which, together with G, deductively yields E. Thus, the proffered explanation is a linguistic-propositional structure, something that belongs in our epistemic system, the total body of our knowledge at a given time. To evaluate whether a proposed explanation is a genuine, correct explanation, we would, on Hempel’s account, need to determine various things: that L is a law, that E is logically derivable from (or entailed by) L and G but not from either alone, and that each statement in the explanans is true. Notice that apart from the determination of the truth of the explanatory premises, the evaluation of an explanation is a purely internal affair, in that the objects involved are statements and they are examined in respect of their forms and mutual logical relationships. In particular, there is no need to ascertain whether any special relationship obtains between the two events, the event to be explained and the event being invoked as its explanation. For Hempel, for g to explain e is nothing more, or less, than for G to nomologically imply E; the explanatory relation between g and e collapses into a logical relation between their descriptions. The basic relation that generates an explanatory relation is a logico-linguistic one that connects descriptions of events, and the job of formulating an explanation consists, it seems, in merely re-arranging appro-

5 Alternatively we could formulate our discussion in terms of “facts,” rather than events, without affecting any of the essential points.

172


priate items in the body of propositions that constitute our total knowledge at the time. In explaining something, then, all action takes place within the epistemic system, on the “subjective” side of the divide between knowledge and the reality known, or between representation and the world represented. The foregoing, as I take it, is the official—that is, Hempel’s—story about covering-law explanations. There is, however, a very different story that can be told of what’s going on in a D-N explanation: one could say that the very point of a D-N derivation of E from G is to demonstrate that g is a cause of e, and that a D-N argument is explanatory only because, and to the extent that, it succeeds in depicting a causal relation between g and e. Ultimately it is this objective causal relation, something that holds outside our epistemic system, that grounds the fact that the D-N argument with G and L as premises and E as conclusion is an explanation.6 This brings us to the second fundamental question for theories of explanation: The Metaphysical Question: When G is an explanans for E, in virtue of what relation between g and e, the events represented by G and E respectively, is G an explanans for E? What is the objective relation connecting events, g and e, that grounds the explanatory relation between their descriptions, G and E? One possible response is to deny that any such objective relation exists, or needs to exist, to “ground” the explanatory relation between their descriptions. This, I believe, is the official stance of the covering-law theorist: there is no further fact about an explanation beyond the fact that it is a D-N argument showing that the event to be explained is nomologically expectable on the basis of the antecedent circumstances as described in the explanans. The scientific understanding produced by the explanation consists precisely in the grasping of the nomic expectability of the event to be explained. Consider the following interestingly revealing remarks by Hempel, in one of the few places where he explicitly mentions the issue of understanding in connection with his models of explanation:

6 Hempel’s position on causal explanations is complex and ambivalent. He would grant that many D-N explanations of individual events are causal explanations in the sense that the initial conditions represented by the singular premises are a cause of the event represented by the explanatory conclusion. However, he denies that all singular explanations are causal, and would deny that a given D-N argument constitutes an explanation in virtue of the fact that it correctly depicts a causal relation between the explanandum and the events invoked in the explanans.


173

Thus a D-N explanation answers the question “Why did the explanandum phenomenon occur?” by showing that the phenomenon resulted from certain particular circumstances, specified in C1, C2, . . . , Ck, in accordance with the laws L1, L2, . . . ,Lr. By pointing this out, the argument shows that, given the particular circumstances and the laws in question the occurrence of the phenomenon was to be expected; and it is in this sense that the explanation enables us to understand why the phenomenon occurred.7 Notice Hempel’s use of the causative “resulted” in his first sentence; this shows, we may speculate, that Hempel was intuitively pulled in the direction of giving a causal answer to the Metaphysical Question: a D-N explanation explains by showing that the event to be explained “resulted from,” that is, caused by, the events invoked in the explanans. But he immediately pulls back, and, in the sentence that follows, gives an entirely unrelated answer, the “official” position of the D-N theory: a D-N explanation explains by showing that the event being explained was to be expected given the events and laws cited in the explanans. As I take it, this represents Hempel’s real position; for him, nomic expectability is of the essence of scientific understanding. On a view like Hempel’s, then, explanation is essentially an activity internal to an epistemic corpus: whether or not something is an explanation—a good, “true” or “correct” explanation—depends on factors internal to a body of knowledge, not what goes on in the world—except, of course, for the truth of the statements comprising the explanans. We may call an approach of this kind “explanatory internalism.” If, on the other hand, you are disposed to look for a nontrivial answer to the Metaphysical Question, attempting to find a relationship between the explanandum event and the event invoked as its explanation, you are taking the stance of “explanatory realism” (or “explanatory externalism”), as we may call it. A realist about explanation believes that some objective relation between the events underlies, or grounds, the explanatory relation between their descriptions. That is, statement G constitutes a correct explanation of statement E in virtue of the fact that a certain relationship obtains between events g and e. What could such a relationship be? One strong traditional favorite of course is the causal relation: G is a correct explanans of E because 7 Carl G. Hempel, “Aspects of Scientific Explanation,” in his Aspects of Scientific Explanation (New York: The Free Press, 1965), p. 337 (the italics are Hempel’s).

174


g is a cause of e. On this view, what warrants the use of G in framing an explanation of event e, and thus an answer to the question “Why did e occur?,” is the fact that the event represented by G, namely g, is a cause of e. The fact that G is an explanans of E, therefore, depends on something outside one’s epistemic system; just as the claim to know that p requires an “objective correlate,” namely the fact that p, the claim that G is an explanans for E requires an objective correlate, namely the fact that g is a cause of e.8 But is the causal relation the only objective relation capable of grounding the explanans relation? Is it the only relation that is explanatory? David Lewis has defended an affirmative answer, at least for explanations of individual events.9 According to him, all explanations of individual events are causal explanations, each of them telling something about “the causal history” of the event being explained. Is Lewis right? We will touch on this question later, but for now Lewis’s position serves to illustrate the realist approach to explanation. For Lewis, every explanatory claim is a causal claim, and what makes an explanation of e in terms of g a correct explanation is the fact that an appropriate causal relation obtains between e and g, and the proffered explanation is informative as an explanation to the extent that it is informative about the causal history of e. It bears repeating that a proponent of the D-N model need not be an explanatory internalist, although Hempel pretty clearly was an internalist. It is clear that the D-N model is exactly what an explanatory realist who believed in causation as the “explanation-grounding relation” (as we might call it) would embrace, if he also believed in the nomic-subsumptive conception of causation, the essentially Humean view that events are connected as cause and effect in virtue of being subsumed by an appropriate law. For such a person, the D-N model is, more directly, a model of what it 8 In Scientific Explanation and the Causal Structure of the World Salmon distinguishes among “three basic conceptions” of scientific explanation, the “epistemic,” the “modal,” and the “ontic” conceptions. The epistemic conception, under which he includes Hempel’s covering–law theory, appears to be a form of what I am calling “explanatory internalism” (but not exhaustive of it); both the modal and the ontic conceptions appear to be forms of explanatory realism, the former grounding the explanatory relation in the relation of necessitation among events and the latter grounding it in the causal relation. The term “ontic,” therefore, seems a bit of a misnomer: the modal approach is as “ontic” as the ontic approach. And it would seem that the term “epistemic,” as used by Salmon, is also misleading. I think it should be granted on all hands that explanation is epistemic (what else could it be?); the proper question to raise is whether there is something “ontic,” out there in the world, that grounds or underwrites (“corresponds to,” if you prefer) the explanatory relation. 9 In “Causal Explanation.”


175

is for events to be subsumed under a law,10 and a D-N argument is a perspicuous way of exhibiting events in a causal–explanatory relation. One might argue that if you embrace the kind of causal–explanatory realism just sketched, there is little reason to accept the D-N model. For if the causal relation is what at bottom generates explanatory understanding, why should we restrict ourselves, one might ask, to a single way of representing that relation in an explanation? Why, that is, should we limit ourselves to D-N arguments to exhibit the causal relation and not avail ourselves of other ways of providing information about the causal facts involving the explanandum? (Notice that a parallel question can be raised in regard to any explanatory realist who believes in a “model” of explanation.) It would seem that the best course for the causal realist is to follow Lewis, who has said that “to explain an event is to provide some information about its causal history.”11 I believe that the D-N theorist has a plausible line for an answer: she could say that a D-N argument is a particularly perspicuous way of representing the causal information involved. A successful D-N argument, in virtue of the requirement of nomological derivation, exhibits those features of the supposed cause that are efficacious or relevant in the production of the explanandum event, and this makes it explanatorily informative in a way in which a mere tagging of the cause, by a proper name or a description based on causally irrelevant features of the situation, is not. Consider: “Why was the entire crop of the island wiped out?” “It was caused by the event reported on p. 12 of the last Sunday Times”. Although this is not an implausible response, it is not conclusive; granted that appropriate D-N arguments are particularly perspicuous as a vehicle for conveying causal information, there is no reason to exclude all other ways of representing such information. Understanding and explanatoriness are matters of degree, and there seems little reason to insist that all explanations take a single logical form, or a small set of such forms. The D-N model makes explanation an all–or–nothing affair, and it must be granted that this does not sit well with explanatory realism. The same comment applies to most other “models” of explanation. 10 During the heyday of logical positivism causation apparently came under the positivist strictures against metaphysics, and there was little overt discussion of issues related to causation. It is possible to think of the Hempelian D-N account of explanation as a covert causal theory of explanation dressed up in a positivistically correct internalist guise. 11 “Causal Explanation,” p. 217.

176


III According to Salmon, another causal explanatory realist, to explain an event is to exhibit its place in the causal structure of the world.12 As explanatory realists, therefore, Lewis and Salmon have a simple answer to the Epistemological Question. What it is that we know when we have an explanation is a causal fact; explanatory knowledge is causal knowledge. In general, then, the explanatory realist has a ready answer to the Epistemological Question: if R is identified as the explanation-grounding relation (as we might call it), what we know when we have an explanation is the fact that R obtains between appropriate events. Thus, explanatory knowledge is just additional bits of propositional knowledge: on causal explanatory realism, for example, to have an explanation of e in terms of g is to know, or accept, the proposition that g is a cause of e. But this still leaves our initial epistemological issues about explanation unresolved; we still need an epistemology of understanding, for a theory of explanation, we argued, must be a theory of understanding. Take the causal theorist like Lewis and Salmon; we still need him to answer the following question: what does causal knowledge have to with understanding? That is, why, and in what way, does the knowledge that some event, g, caused e produce, or enhance, our understanding of e, or help make e intelligible? In general, then, the explanatory realist has the task of explaining why and how knowledge that the grounding relation R is instantiated in a given situation promotes our understanding of that situation. Some explanatory realists seem to reject, or at any rate try to avoid, this as a responsibility of a theory of explanation. For example, Paul Humphreys thinks that considerations of understanding will lead to “a relativization of explanations to an individual,” since the epistemic situation differs from one individual to the next; he proposes to “set aside that whole issue of what constitutes or promotes understanding.”13 Salmon does raise the issue of explanation and understanding, writing:

12 See Scientific Explanation and the Causal Structure of the World, pp. 19–20. 13 The Chances of Explanation, p. 127.


177

Perhaps the most important fruit of modern science is the understanding it provides of the world in which we live, and of the phenomena that transpire within it. Such understanding results from our ability to fashion scientific explanations.14 We can agree with what Salmon is saying here. We prize explanations, especially those in science, because understanding, as Salmon puts it, “results from our ability to fashion explanations.” That is almost tautological; what isn’t tautological, and in fact far from obvious, is why and how understanding results from our ability to fashion X, where X is what a given theory of explanation says explanations are. That is, it is far from obvious just how understanding springs from our ability to construct D-N arguments, or to discover and formulate causal judgments of the sort that Salmon would consider explanatory. And the section entitled “Explanation and Understanding,” from which the foregoing quotation has been drawn, contains nothing further on understanding, and the present issue is never faced, at least not directly.15 The explanatory realist must exercise special caution in dealing with this problem. For, on explanatory realism, explanatory knowledge turns out to be just additional bits of propositional knowledge. For the causal theorist like Salmon and Lewis, an explanation of an event is merely a causal proposition, to the effect that a certain event caused the explanandum event. To “have an explanation,” on this view, is to have such a proposition of that kind in one’s epistemic corpus. This puts the distinction between “explanatory knowledge” and “descriptive knowledge” in potential jeopardy; the distinction as it is standardly drawn faces an imminent collapse. Why isn’t the knowledge that the breaking of the dam caused the town to be flooded any less descriptive than the knowledge that the bridge collapsed at midnight. What the explanatory realist must say, it seems, is that although all knowledge is descriptive, some descriptive knowledge has a special explanatory role in virtue of the particular content it has. His job then is to explain why the particular type of descriptive content (e.g., causal facts) he has identified as explanatory serves to enhance understanding. On the other hand, explanatory internalism leaves open the possibility of giving an 14 Scientific Explanation and the Causal Structure of the World, p. 259. 15 Salmon’s criticism of “psychologism” about explanation earlier in the book, and his rejection of “the epistemic model” of explanation, can be read as a repudiation of the concept of understanding as an element in a theory of explanation, although this might not have been his real intention.

178


account of a type of knowledge that does not consist in propositional knowledge, something that supervenes on, but distinct from, such knowledge. This is an intriguing possibility, and in some ways an exciting one, to which we will recur below.

IV Michael Friedman begins his influential paper, “Explanation and Scientific Understanding,”16 with some insightful remarks on the importance of understanding in a theory of explanation. He makes much the same point we have made at the outset, when he demands that “a theory of scientific explanation tell us what it is about the explanation relation that produces understanding.”17 And Friedman has no trouble showing that traditional conceptions of explanation, including Hempel’s D-N model, fail to provide an account of the relation between explanation and understanding. Friedman himself sees the key to understanding this relation in the idea of unification: scientific explanation produces understanding by effecting unification in our epistemic system. He writes: Once again, we have reduced a multiplicity of unexplained, independent phenomena to one. I claim that this is the crucial property of scientific theories we are looking for; this is the essence of scientific explanation— science increases our understanding of the world by reducing the total number of independent phenomena that we have to accept as ultimate or given. A world with fewer independent phenomena is, other things equal, more comprehensible than one with more.18 For example, the explanatory power of Newton’s mechanics consists in its unifying power over such diverse, and apparently unrelated, laws as Kepler’s laws, Galileo’s law of falling bodies, the law of simple pendulum, and countless others. By deriving these laws from Newton’s system, we show that they are not independent phenomena, and that they could be brought

16 Journal of Philosophy 71 (1974): 5–19. Reprinted in Joseph C. Pitt, ed., Theories of Explanation (New York and Oxford: Oxford University Press, 1988). All references to this article are to the version in the Pitt volume. 17 “Explanation and Scientific Understanding.” p.189. 18 Ibid. p.195.


179

under a single set of regularities. In this way, a significant reduction in the number of independent assumptions is achieved, and the result is a more unified, simpler system that makes the world a more intelligible, more comprehensible place. Philip Kitcher, finding faults with Friedman’s account of unification, offers another account of explanatory unification.19 In contrast to Friedman, who is interested in the reduction of the number of independent laws, Kitcher focuses on the idea of reducing the number of independent argument patterns. What Kitcher finds significant in such theories as Newton’s mechanics and Darwin’s evolutionary biology, both universally hailed for their great explanatory power, is the fact that they employ relatively few argument patterns to derive, and explain, large numbers of phenomena. Darwin’s theory, Kitcher notes, does not attempt to give a detailed derivation of how certain particular characteristics of species have emerged; rather, what’s impressive about it is that it produces a derivation schema that can be applied repeatedly to particular characteristics of all species to explain how they have emerged. The schema uses the principle of natural selection as the basic premise in conjunction with other specific premises about ancestral forms and the nature of their environment, and laws of variation and inheritance (largely unknown to Darwin). Kitcher’s development of this idea is rather involved, and its details will not concern us here. What is noteworthy is that Kitcher fully endorses Friedman’s demand that a theory of explanation must give an account of how explanation is conducive to the enhancement of understanding. He is also in agreement with Friedman in holding that unification is the key to understanding. They differ only in their analysis of unification—or about the kind of unification that makes a difference to understanding. It is useful to view Friedman and Kitcher as explanatory internalists.20 They appear to follow Hempel in taking explanations to be closely tied to law-based derivations. Neither tells us explicitly what explanations, in the sense of “products,” are, but it is clear enough that, like Hempel, they seem to take explanations as involving nomological derivations. The crucial point of interest for us here is this: what makes these derivations explanatory, on 19 In “Explanatory Unification,” Philosophy of Science 48 (1981): 507–531. Reprinted in Pitt, Theories of Explanation. 20 Salmon would probably agree; he thinks that they adopt the epistemic conception of explanation, which roughly corresponds to what I am calling explanatory internalism.

180


Friedman’s and Kitcher’s accounts, is their relationship to other items within our epistemic system, not some objective facts about external events or phenomena. On Kitcher’s account, for example, what makes a given D-N argument explanatory is the fact that it is a member of a class of arguments (a “systematization,” as Kitcher calls it, of our belief system) which best unifies our belief system. And the measure of degrees of unification depends solely on factors internal to the epistemic system, such as the number of argument patterns required to generate the given class of arguments, the “stringency” of the patterns, etc., not on any objective relations holding for events or phenomena involved in the putative explanations. For Friedman, whether or not a given law explains another21 is determined crucially by the unifying power of the explaining law, and the concepts in terms of which the latter is explained are exclusively logical ones (equivalence, implication, etc.) and evidential ones (“independent acceptability”). In general, such virtues as simplicity and unity are primarily properties of theories and belief systems; they are epistemic virtues that concern properties of our representations of the world, but not of the world itself. Let us now consider how these internalist theories answer the Epistemological Question about explanation: What is it that we know when we have an explanation? It is clear that, on either Kitcher’s or Friedman’s view, there is no particular proposition that we come to know when we gain explanatory knowledge. Explaining is not, for them, a matter of discovering, or imparting, more propositional knowledge; explanatory activity consists in constructing derivations whose structure and steps are logically or epistemically related in certain specified ways to the rest of the belief system. To put it somewhat crudely, explanation is a matter of the shape and organization of one’s belief system, not of its content. Both Kitcher’s and Friedman’s accounts make explanation a holistic affair: whether or not a given derivation is an explanation cannot be determined locally, just by looking at the derivation; it depends on facts about the whole belief system.22 This means that the epistemic gain represented by an explanation cannot be measured or represented precisely. The fact that a particular derivation is an explanation relative to a belief system says something about the structure 21 Friedman is concerned only with explanations of laws, and therefore doesn’t quite fit our scheme of classification; however, this does not obscure the internalist character of his account. 22 It seems that any unification approach to explanation will be holistic, although this isn’t true of all internalist theories (compare, e.g., Hempel’s covering–law theory).


181

and organization of the system and the place and role of the derivation within it, and this makes it impossible to say anything precise and useful about the epistemic gain to be associated with individual explanations. This holistic picture is appealing in some ways. We are apt to think that explanatory understanding cannot just be a matter of having more information, more propositional knowledge; somehow, it seems like something that overlays, or supervenes on, but is distinct from, knowledge of particular propositions. On the other hand, the holistic picture doesn’t fit certain other salient aspects of our explanatory practice. It may be that, as Friedman says, “a world with fewer independent phenomena is . . . more comprehensible than one with more,” and that, as Kitcher holds, a better unified theory is more explanatory. On such a holistic conception, though, it is not even clear that the idea of explaining some individual item, whether this is an event or a law, can have a determinate meaning. In particular, causal explanations of individual events and actions seem to make no sense within such a picture. It seems obvious, however, that we construct and verify such explanations locally, and derive explanatory understanding from them. To ascertain a derivation as an explanation, Kitcher asks us to consider our whole belief system and whether or not the derivation is a member of the class of derivations that can be generated by a set of argument patterns which collectively have a certain unifying property. This seems wildly unrealistic: it isn’t something we do or need to do; it probably isn’t something that any of us can do! But what does the unification approach tell us about explanation and understanding? Surprisingly little, I think. Kitcher complains about the Hempelian account by asking, “Why should it be that exactly those derivations which employ laws advance our understanding?” Good question! He then goes on to say, “As Friedman points out, we can easily connect the notion of unification with that of understanding.”23 But how is the connection supposed to go? Just how does unification manage to advance understanding? Kitcher himself says nothing further; so what does Friedman have to say? Against Hempel, who, as we saw, takes nomic expectability as the source of explanatory understanding, Friedman points out that this concept and that of understanding are “quite distinct notions,”24 and that Hempel 23 “Explanatory Unification,” p. 168 (in the Pitt volume). 24 “Explanation and Scientific Understanding,” p. 190.

182


fails to provide an intelligible linkage between them. Friedman is of course right; but then is it so clear that the concept of unification, whether taken preanalytically or in either of the senses explicated by Friedman and Kitcher, is any closer to understanding? The preanalytic notion of unification does not seem to have anything significant to do with the idea of understanding. To be sure, a better unified, or simpler, theory is easier for us to “understand,” in the senses of learning or making use of it.25 Given our finite intelligence, we do better with simpler concepts, fewer axioms and rules, and so on. But not always: as anyone who has studied axiomatic systems of logic knows, the simplest systems, simplest in terms of the number of primitives, axioms, and rules of inference, severely tax our ability to understand. We generally do better with middlingly complex systems with intuitive concepts and rules, and some redundancies built into them. In any case, this notion of understanding, roughly in the sense of learnability (for humans), does not seem to be what is relevant to the explanatory understanding of events and facts. Kitcher’s concept of unification is built essentially from logical concepts, but Friedman’s makes use of epistemic concepts as well. The principal epistemic concept in Friedman is that of “independent acceptability,” which seems to belong to the same general genre of evidential concepts as that of “expectability,” which Friedman finds “quite distinct” from understanding. And it is not obvious just at what stage Friedman’s analysis of unification is supposed to make contact with understanding. If unification is to be intelligibly connected with understanding, the connection must be consciously forged, and to do that one would have to say a lot more about understanding. It would be unrealistic to expect that somehow understanding will magically materialize out of simplicity and unification, concepts that appear to be—and, on Kitcher’s and Friedman’s accounts, are—largely formal and logical.

V In this final section, I want to sketch a realist approach to explanation which has a certain affinity with the simplification–unification approach. My 25 When Friedman writes “a world with fewer independent phenomena is . . . more comprehensible than one with more” (ibid. p.195), he sounds as though he has in mind something like this sense of “understanding.”


183

approach will show that the unification approach is not intrinsically internalist, and that, on the contrary, it can be plausibly motivated in a realist setting; in fact, I claim that explanatory realism can make better sense of the idea that explanations unify and simplify. How unification or simplification is to be connected to explanatory understanding is a difficult question, and I will have nothing useful to add to what has already been said by Friedman, Kitcher, and others. My main object is to describe a broadly realist approach that exploits the idea that explanations explain by unifying and simplifying. The key concept I will make use of is that of dependence, and my claim will be that dependence relations of various kinds serve as objective correlates of explanations. Dependence, as I will use the notion here, is a relation between individual events and states; however, it can also relate facts, properties, regularities between events, and even entities. We speak of the “causal dependence” of one event or state on another; that is one type of dependence, obviously one of central importance. Another dependence relation, orthogonal to causal dependence and equally central to our scheme of things, is mereological dependence (or “mereological supervenience,” as it has been called): the properties of a whole, or the fact that a whole instantiates a certain property, may depend on the properties and relations had by its parts. Perhaps, even the existence of a whole, say a table, depends on the existence of its parts. We can think of the principle of mereological supervenience as the metaphysical principle underwriting the research strategy of microreduction and the method of microreductive explanation, in the way in which the principle of causal determination provides a metaphysical basis for the research strategy of causal investigation and causal explanation. Just as the latter rationalizes our attempt to look for the diachronic, temporally antecedent determinants of phenomena, the former urges us to search for their synchronic micro-determinants. Causal dependence and mereological dependence may well be the two most basic, and important, relations of dependence that can serve as explanation-grounding relations. Are there other kinds of dependence relations that can ground explanations? Although this is not the place to try to be exhaustive, there are some cases to consider. Socrates expires in the prison, and Xanthippe becomes a widow. The widowing of Xanthippe depends on the death of Socrates, and

184


this is perhaps still another kind of dependence.26 There is a widespread belief in mind-body supervenience: the mentality of a creature depends on its physical nature.27 Belief in evaluative supervenience, the idea that evaluative and normative properties supervene on factual, or nonevaluative properties seems even more widely accepted.28 These and other dependence relations may generate explanations—explanations of mental phenomena in terms of their underlying neurobiology, and of evaluative facts (e.g., why some act is morally right, why some work of art is beautiful) in terms of the nonevaluative facts on which they supervene. We think of the world as a system with structure, not a mere agglomeration of unconnected items, and much of the structure we seek comes from the pervasive presence of dependence relations. Dependence is asymmetric and transitive, and can generate relational structures of dependent events, states, and properties. The ontological contribution of dependence relations lies exactly in this fact: they reduce the number of independent events, states, facts, and properties we need to recognize. And that is precisely the unifying and simplifying power of dependence relations. Unity and structure go hand in hand; dependence enhances unity by generating structure. These points apply to lawful regularities as well as individual events and states; in particular, we think of macrophysical regularities as arising out of microphysical regularities and hence explainable by them. The principle of micro–macro determination—that is, the doctrine of mereological supervenience—is applicable to laws and regularities as well as individual states and events. Here too, underlying dependency relations support explanations. My main proposal, then, is this: explanations track dependence relations. The relation that “grounds” the relation between an explanans, G, and its explanatory conclusion, E, is that of dependence; namely, G is an explanans of E just in case e, the event being explained, depends on g, the event invoked as explaining it.

26 See my “Noncausal Connections,” in Supervenience and Mind (Cambridge: Cambridge University Press, 1993). 27 There is the possibility of construing mind–body supervenience as a form of mereological supervenience. See “Postscripts on mental causation” in my Supervenience and Mind. 28 Moral irrealists will reject formulations of supervenience in terms of “properties;” however, there are formulations that should be acceptable to them, e.g., in terms of predicates or judgments.


185

On this proposal, therefore, the simplifying effect of an explanation is seen both in our belief system and in the world: by showing an event to be dependent on another, the explanation reduces the number of independent events in the world, and also the number of independent assumptions we need to accept about the world. This approach differs importantly from the internalist approaches of Kitcher and Friedman, in two crucial respects: simplicity and unity are features of the events and facts of the world as well of our beliefs and propositions, and they are, in the first instance, local features of the propositions and events involved in individual explanations, rather than holistic, global features of the whole belief system or world. But the account does not ignore the holistic aspect; far from ignoring it, it gives an account of how scientific understanding at the global level arises out of local explanations. For the dependency structures generated by individual explanations are cumulative, and contribute to the overall unity and simplicity of the system. In consequence, our account allows explanatory understanding both at the local and the global level, and relate them in a natural way, whereas on Friedman’s account, scientific understanding is exclusively a global feature of a whole theoretical system, and there is no obvious way of making sense of scientific understanding contributed by individual explanations.29 Within the general scheme of realism, explanatory realism is appealing and perhaps even compelling.30 I am in agreement with Salmon when he says, “we must surely require that there be some sort of objective relationship between the explanatory facts and the fact-to-be-explained.”31 The present approach does justice to this realist requirement. Moreover, it shows that the idea of unification and simplification can be combined with explanatory realism in a natural and plausible way, and opens the possibility of an account of understanding within a realist scheme. Producing a usable account of understanding is the hard part; one scarcely knows where to begin. The difficulty does not seem to lie on the side of unity and simplicity; I think we have a serviceably clear understanding of these notions. What we 29 Friedman, “Explanation and Scientific Understanding,” p. 197. 30 David–Hillel Ruben’s Explaining Explanation (London and New York: Routledge, 1990), chap. 7, contains a perceptive and enlightening discussion of the realist perspective on explanation, to which the present paper is indebted. I first discussed explanatory realism in “Explanatory Realism, Causal Realism, and Explanatory Exclusion,” Midwest Studies in Philosophy 12 (1988): 225–240. Essay 7 in this volume. 31 Scientific Explanation and the Causal Structure of the World, p. 13.

186


lack is an understanding of understanding that is clear and rich enough for useful theorizing. To build a bridge from unity and simplicity to understanding and intelligibility, we need an epistemology of understanding, something that has by and large been neglected by contemporary analytic epistemology and philosophy of science.

9 Hempel, Explanation, Metaphysics Through his long and distinguished career, Carl Hempel contributed to a variety of philosophical areas. To mention some of the topics on which he made notable contributions, the following comes quickly to mind: the coherence theory of truth, logical behaviorism, the verifiability criterion of meaning, philosophy of history, confirmation theory, concepts and concept formation in the sciences, the observational content of theoretical terms, functional analysis in biology and the social sciences, the concept of rational action, and the structure of scientific theories. But everyone will agree that it is his work on the nature of scientific explanation that was both the heart and pinnacle of his philosophical work. Clearly, it was, and still is, the best known and the most influential of his many contributions. With his “The Function of General Laws in History”1 published in 1942, Hempel arguably invented the problem of scientific explanation as a central problem of philosophy of science. One could say that the basic idea of the deductivenomological model, or the covering-law model, of explanation goes back to John Stuart Mill; one could also say that the idea was present in Karl Popper’s Logic of Scientific Discovery2 and in the writings of Ernest Nagel around the same time, but it was Hempel’s work, at once systematic, detailed, and comprehensive, that elevated scientific explanation to the status of a central philosophical problematic in the philosophy of science during the second half of this century. As many of you will recall, Hempel’s theory of explanation was the target of wave after wave of critical attention, from all quarters, some of it rather 1 Reprinted in Hempel, Aspects of Scientific Explanation (New York: The Free Press, 1965). 2 (London: Hutchinson, 1959).

188

HEMPEL, EXPLANATION, METAPHYSICS

jejune and picayune but a substantial portion of it positive and constructive, leading to new insights and perspectives. In retrospect, it is amazing to see how well Hempel’s basic insights have withstood the onslaught and endured. Later in this talk, I will argue that there is something fundamentally and importantly right about Hempel’s insistence that explanation is logical deduction—more specifically that an explanation has the structure of a logical argument with laws among its indispensable premises. But first I would like to show how certain basic positivist philosophical commitments that Hempel had embraced early in his career shaped, in a profound way, the surface contours of his theory of explanation, and how a different perspective on philosophical methodology might have led to a more satisfactory, though not fundamentally different, account of scientific explanation. But before I go into these matters, I would like to take this occasion to pay my personal tribute to Hempel as my teacher. When in the spring of 1958, I was waiting to hear from the graduate philosophy programs to which I had applied, I received a personal letter from Hempel in which he very gently urged me to come and study at Princeton. That was the most exciting thing that had ever happened to me, and I read and reread the letter many times over. Hempel’s carefully inscribed signature at the bottom of the page, in blue ink and with a hint of formal daintiness, was deeply burned into my visual memory, and I can still vividly see it in my mind. With that letter, Princeton shot to the top of my list. Actually at the time I was ill prepared for graduate-level work in philosophy, but Hempel was always kind, understanding, and patient. Above all, he treated everyone, including the greenest novices and the most petulant critics, with friendliness, honesty, and respect. Of course, he was a vigorous and effective debater, especially against some of his obsessive critics on explanation, but he discussed the issues and never attacked the critics. His respect for both his opponents and the truth was always evident. He was indeed a model of what a philosopher, a lover of knowledge and truth, ought to be and can be.

I When one looks at Hempel’s work during his early period, roughly the 1930s through the ’50s, one cannot fail to notice two very prominent


189

themes that shaped his philosophical approach. The first is his view of philosophy as the study of the “logical syntax” of the language of the sciences and his commitment to “syntactic analysis of language” as the proper method of philosophy. The second is his commitment to objectivity in both science and philosophy and his rejection of anything that smacked of psychologism or involved an appeal to subjective feelings and judgments. Let us call these “syntacticalism” and “anti-psychologism” respectively. Here I will mainly focus on syntacticalism, and will not have the time to discuss Hempel’s anti-psychologism. Hempel’s conception of philosophy as syntactical analysis of course stems from Rudolf Carnap’s early work, in particular his Logical Syntax of Language.3 In this and other works during the 1930s, Carnap promoted the idea that philosophy is the investigation of the syntax of language, especially that of the language of science, and that this is the only proper and defensible conception of philosophy. According to the orthodox positivist doctrine, science has a preemptive, monopolistic dominance over both empirical knowledge and a priori logical and mathematical knowledge, and this meant that the only sliver of the epistemic domain left over for philosophy is the metalinguistic/metalogical task of analyzing the language of science. For Carnap during this pre-Tarski period, syntax exhausted the logic of science (“The logic of science is syntax” is the title of one of the sections of the Logical Syntax4). Carnap’s fear and loathing of semantics was not fully allayed until the ground-breaking work of Tarski on truth. The odd thing, from the present perspective, is that Carnap’s Logical Syntax was rife with concepts that are now standardly regarded as semantic rather than syntactic— such as analyticity, implication, and equivalence—a fact Carnap explicitly acknowledged in his later work, Introduction to Semantics.5 Even the slogan “The logic of science is syntax” was changed to “The task of philosophy is semiotic analysis,” where “semiotic” is short of “syntactic and semantic.” In this later work, the notion of designation (what we would now call reference) plays an essential role, Carnap taking both individual constants and predicates as designators for individuals and properties respectively. And Carnap defines truth for atomic sentences in the by-now familiar schema: “Fa” is true iff the object designated by “a” has the property designated by “F.” 3 Logische Syntax der Sprache (Vienna: Springer, 1934). English edition, Logical Syntax of Language (London: Kegan Paul, 1937). 4 p. 331 of the English edition. 5 (Cambridge, Mass.: Harvard University Press, 1942).

190


Hempel wrote several papers during his stay in Belgium in the 1930s, before he moved to America. These mainly concerned the positivist concept of truth, and the point of his papers—in particular, “On the Logical Positivists’ Theory of Truth” and “Some Remarks on ‘Facts’ and ‘Propositions’”6—was to argue against Moritz Schlick’s correspondence theory of truth. For Hempel, the correspondence theory was founded on the incoherent idea that propositions, or sentences, can be compared with facts. Hempel argues that talk of propositions “expressing ‘facts’” is misleading because the term “indicates something which is once for ever fixed with all its characteristics, whilst it is essential for the system of scientific statements that it may always be changed again.”7 The following paragraph is revealing: Science is a system of statements which are of one kind. Each statement may be combined or compared with each other statement, e.g. in order to draw conclusions from the combined statements, or to see if they are compatible with each other or not. But statements are never compared with a “reality,” with “facts.” None of those who support a cleavage between statements and reality is able to give a precise account of how a comparison between statements and facts may possibly be accomplished, and how we may possibly ascertain the structure of facts. Therefore, that cleavage is nothing but the result of redoubling metaphysics, and all the problems connected with it are mere pseudoproblems.8 Statements can be compared only with other statements.9 When we try to describe a situation in which we are supposedly comparing a statement with a fact, it turns out, Hempel argues, that we are only comparing the statement with another statement.10 Fact talk takes you outside the confines of language, and that is not where Hempel thought we should venture. In contrast, all the concepts we need to do science, Hempel claimed, are purely syntactic and intra-linguistic affairs, and do not require us to look outside the language, at

6 Both in Analysis 2 (1935): 49–59 and 93–96 respectively. 7 “Some Remarks on ‘Facts’ and ‘Propositions’,” p. 95. 8 “On the Logical Positivists’ Theory of Truth,” Analysis 2 (1935): 49–59; the quotation is from p. 51. 9 Views like this are fairly widely held by philosophers outside the positivist circles; e.g., Wilfrid Sellars and Donald Davidson. 10 Here Hempel seems to anticipate considerations later put forth by some philosophers for the coherence theory of truth/justification; see, e.g., Donald Davidson, “A Coherence Theory of Truth and Knowledge,” in Davidson, Subjective, Intersubjective, Objective (Oxford: Oxford University Press, 2001).


191

“reality” or “the facts.” Outside the language lies the treacherous land of metaphysics. Hempel’s view during this period was that the acceptance of observation statements was essentially a matter of convention, or community practice, and that the proper theory of truth was the coherence theory. I am not sure exactly how Hempel accommodated himself to the newly found respectability of semantics—in particular, the early Carnapian semantics that seemed tailor-made for a correspondence approach to truth, going clearly beyond the kind of syntactic, coherence view of truth he had earlier advocated. In 1943, Hempel published “A Purely Syntactical Definition of Confirmation”11 and the better-known “Studies in the Logic of Confirmation” came out in 1945.12 As the title of the first of these works makes clear, Hempel’s methodology here was wholly syntactic. As many of you will recall, his well-known and much-discussed “Raven Paradox” was first presented in the second paper, in the course of developing an account of how hypotheses of the form “All Fs are G” are confirmed by observation of “positive instances,” i.e., objects that are F and G. The paradox arises from two general principles: “the Nicod criterion” to the effect that a hypothesis of this form, “All Fs are G,” is confirmed by things that are both F and G, that is, things that satisfy both the antecedent and consequent predicates, and the equivalence principle to the effect that logically equivalent hypotheses are confirmed, or disconfirmed, by the same evidence. Hempel noticed that “All ravens are black” is logically equivalent to its contrapositive “All nonblack things are nonravens,” and that the latter is confirmed, on the Nicod criterion, by observation of nonblack nonravens, e.g., observations of brown shoes, green cabbages, and red herrings. So by the equivalence principle, observations of these sundry non-ornithological items around us should confirm “All ravens are black.” The prospect of armchair ornithology was too good to be true. Hempel’s own solution, in terms of syntactic constructions, was not widely accepted, and there were many proposals by others. The most convincing, and perhaps simplest, solution was provided, I believe, by W.V. Quine, in his “Natural Kinds.”13 Quine pointed out that the Nicod criterion is plausible only when “F” and “G” are kind terms, terms denoting 11 Journal of Symbolic Logic 8 (1943): 122–143. 12 Mind 54 (1945): 1–26, 97–121. 13 In Essays in Honor of Carl G. Hempel, ed. Nicholas Rescher (Dordrecht: Reidel, 1969).

192


real kinds in nature. “Raven” and “black” are kind terms, but “nonblack” and “nonraven” are not. There is not sufficient “similarity” among nonblack things, or among things that aren’t ravens, to support inductive projections on these classes. In short, nonravens and nonblack things are not projectible kinds. Accordingly, observations of black ravens confirm both “All ravens are black” and “All nonblack things are nonravens,” but observations of nonblack nonravens confirm neither. Note that Quine’s solution brings in semantics, and also metaphysics, in a serious way. On this approach, the Nicod criterion must be given a semantical and metaphysical statement in terms of what “F” and “G” refer to and the nature of the items referred to.

II I have dwelt on the Raven Paradox to illustrate one of my main points in this talk, namely to exhibit the restrictive constraints Hempel’s syntacticalism placed on his work, and how a more freewheeling approach that doesn’t shy away from semantics and metaphysics could have yielded more satisfying results. I will later argue this point in connection with Hempel’s work on explanation. I will not here try to define precisely what Hempel’s “syntacticalism” amounted to. For it seems clear that Hempel’s metaphilosophical views underwent transformation during the ’40s and ’50s much in the way Carnap’s did, and that he began using semantical concepts, at least “semantic” in the current sense of semantics or model theory of formal systems. And truth was no longer a taboo concept for Hempel when he began work on scientific explanation. Indeed, one of Hempel’s conditions on scientific explanation is the requirement that the statements comprising the explanans be true. So semantics in the attenuated sense was very much in play. But in the seminal paper with Paul Oppenheim, “Studies in the Logic of Explanation,”14 formal logic, or the syntax of first-order logic, was still the main vehicle of theoretical developments. Their formal definitions of law and explanation, in Part III of the paper, are entirely syntactic; although the notion of truth does make an appearance in the definition of “(T, C) is an explanans for E,” where T is a theory, C a set of singular sentences, and

14 Philosophy of Science 15 (1948): 135–175. Reprinted in Aspects of Scientific Explanation.


193

E a singular sentence (since sentences in a “true” explanation must be true), they manage to avoid even this minimal semantic involvement by switching their explicandum (as some philosophers used to say) to “(T, C) is a potential explanans for E” which forgoes the truth requirement. One might argue that semantics lurks just under the syntactic surface— that even such syntactic concepts as “singular term,” “quantifier,” “connective,” and so on are not intelligible apart from their intended semantic interpretations. That may be true, but still there is a difference between syntax and semantics, and it would be of substantive philosophical interest to know whether or not there is some shared logical form for all (potential) explanations of individual events that could be captured exclusively in syntactic terms. Be that as it may, the unavoidability of metaphysics in theory of explanation is clearly seen when the formal theory of explanation is brought into the real world, to connect with events and phenomena of the world which after all are what our explanatory activities are all about. And Hempel does provide connections, though informally and somewhat grudgingly, between formal theory and the world. First, Hempel identifies the thing that is explained in an explanation, thus: “By the explanandum, we understand the sentence describing the phenomenon to be explained (not that phenomenon itself); by the explanans, the class of those sentences which are adduced to account for the phenomenon.”15 So “phenomena” are introduced as the real-world things that are explained when we explain. (But notice how Hempel reserves his important theoretical terms, “explanandum” and “explanans,” for linguistic items, not phenomena in the world.) Elsewhere, Hempel also uses the terms “event” and “condition” for the same purposes, to refer not only to the things given, or “described,” by the explanandum but also the things described by singular sentences of the explanans—what Hempel often refers to as “antecedent conditions.” But, except in one surprising paragraph that I will come to later, Hempel never raises serious questions about the nature of entities belonging to these metaphysical categories. In any case, it is clear that the informal semantics of the formal theory of explanation that Hempel was constructing postulates an ontological domain of events, conditions, and phenomena, whatever these are.

15 Aspects of Scientific Explanation, p. 247.

194


Another way in which metaphysics intrudes into the Hempelian picture is through his implied suggestion—although, as I will later argue, he remained fundamentally ambivalent about this—that in building his theory of explanation in “Studies in the Logic of Explanation,” he is giving an account of causal explanation. At the very outset of the article, after giving an informal example of an explanation, he writes: Thus, the event under discussion is explained by subsuming it under general laws, i.e., by showing that it occurred in accordance with those laws, in virtue of the realization of certain specified antecedent conditions.16 He goes to say in the next paragraph: Thus, here again, the question “Why does the phenomenon occur?” is construed as meaning “according to what general laws, and by virtue of what antecedent conditions does the phenomenon occur?”17 Expressions like “in virtue of” and “by virtue of” unmistakably signal a dependency relation, and reference to “antecedent conditions” indicates that the kind of dependency involved here is causal dependency. These explanations explain the occurrence of an event by specifying its temporally antecedent causes. Hempel is more explicit about all this when he later writes: The type of explanation which has been considered here so far is often referred to as causal explanation. If E describes a particular event, then the antecedent circumstances described [by the singular sentences of the explanans] may be said jointly to “cause” that event . . . 18 And in the course of arguing that the familiar desire-belief explanation of actions conforms to his D-N paradigm, Hempel says: The determining motives and beliefs, therefore, have to be classified among the antecedent conditions of a motivational explanation, and there is no formal difference on this account between motivational and causal explanation.19 For most of us, I think, it is precisely this kind of causal association and background that constitutes the intuitive content of Hempel’s theory and 16 Ibid. 246. 18 Ibid. 250.

17 Ibid. 19 Ibid. 254.


195

accounts for its plausibility and appeal. In fact, we can think of Hempel’s D-N model of explanation (when applied to individual events) as giving a sort of Humean nomological analysis of causal relations. The ironic thing is that the official theory, not only the formal logical account of Part III of “Studies in the Logic of Explanation” but also the less formal but more important exposition in Part I, abjures mention of cause or causal explanation, and does not appear seriously motivated or constrained by causal considerations. Causation and causal explanation remain out of the picture as far as the official story is concerned. There is no requirement that the singular sentences of the explanans “specify” or “describe” events or conditions temporally antecedent to the event specified by the explanandum, or that the explanatory laws be, in some sense, “causal laws” (and not, for example, regularities between collateral effects of a single causal process). There is not even the requirement that the singular sentences of the explanans be “event describing”—that is, that they be sentences of the sort that can specify the occurrence of an event. Actually, Hempel explicitly refuses to require, for explanations of individual events, that the explanans include any singular sentences at all, on the ground that it is possible to formulate a singular explanation by universally instantiating a law. Here Hempel appears to be thinking that the following would be a perfectly good singular explanation: Every F is G. Therefore, if a is an F, a is G, where a is any individual constant. Now, one wants to ask: In what sense is this an explanation? In what sense is there here an event, or phenomenon, to be explained? The explanandum is a conditional sentence, “If a is an F, a is G.” It does not, in any sense that we can understand, assert the occurrence of an event or state or phenomenon; rather it says that if a certain condition, namely a’s being F, obtains, then another state or condition, a’s being G, will obtain. This conditional itself is not, at least need not be countenanced as, a condition or state or event or phenomenon in its own right. You might say: There is the fact, if not an event or state, that if a is F, a is G, and this fact is explained by the law “Every F is G.” But this is not convincing. For this “explanation” would hold for anything, whether or

196


not it has property F—that is, every object whatever in this world. This means that the explanandum is only a reiteration of the explanatory law, and it is hard to see how the supposed explanation sheds any explanatory illumination on anything. As far as I know, the first wave of counterexamples against Hempel’s D-N model was inspired by Sylvain Bromberger’s Empire State Building case (sometimes called “the flagpole case”) in the late ’50s, or perhaps early ’60s. Bromberger pointed out that just as we could derive the length of the shadow of the Empire State Building from its height and the laws of optics (and the subtended angle), we could also deduce the height of the building from the length of the shadow. But the former deduction, intuitively, counts as an explanation but the latter surely does not; that is, we can explain the length of the shadow in terms of the height of the building, but not the other way around. Hempel’s formal theory had no resources to account for this apparent explanatory asymmetry. Now, from a metaphysical point of view, the nature of this asymmetry is quite obvious: it is an instance of causal asymmetry. That is, the height of the building, with the position of the sun fixed, causes it to cast a shadow of a certain length, but it is not the case that the length of the shadow causes the building to have the height it has. A number of purported counterexamples to the D-N model made the same point; just to mention some of the better known ones, you may recall the barometer reading/oncoming rainstorm case and the case of the length and period of swing of the pendulum. Further, there are nonexplanatory laws, statements of lawful regularities that are incapable of supporting explanatory connections. These laws may correlate two properties, or physical magnitudes, that are symptomatic aspects of a third, more fundamental, property or process. One often-cited example is the Wiedemann-Franz Law correlating electric and thermal conductivity in metals.20 It is plain that we cannot explain why a given piece of metal has the electric conductivity it has, or the thermal conductivity it has, by invoking this law. Again, the apparent reason for this is the perceived lack of an appropriate dependency relation between the two magnitudes. Classic gas laws seem to be similar to the Wiedemann-Franz Law in this respect.

20 I believe I first heard this example from Larry Sklar in the early 1960s.


197

To appreciate the pervasiveness of the problem involved, consider the following: the D-N account tells us that we can explain why this rod is expanding by pointing out that it is a metallic object and is being heated, and invoking the law that metals expand when heated. However, the D-N model also sanctions the following as an explanation, just as good as the one we just saw: Metals expand when heated. This metal rod is not expanding. Therefore, it is not being heated. But this hardly strikes us as any kind of explanation. Notice that this pattern of counterexamples guarantees that virtually every D-N argument that is a potential explanation can be turned, by using the explanatory law in the contrapositive direction, into another D-N argument that won’t be an explanation, not even potentially. A simple way of handling this range of counterexamples is to specify that the events and conditions invoked in the explanans be, jointly, a cause of the event specified by the explanandum, and/or that laws invoked in the explanans be causal laws. For all these cases fail to be explanatory on account of a single reason: an appropriate causal relation is either absent, or the purported explanation goes in a direction that is the reverse of the causal relation.21 I wonder whether a philosopher now working on explanation would, or should, avoid this route, namely making use of causal concepts in building a theory of explanation. It is of course a further question how we ought to analyze causation. One might feel that causation and explanation are so closely related that using one to explain the other is not doing much in the way of bringing philosophical illumination to either. Perhaps, philosophers like Wesley Salmon and Paul Humphreys22 feel this way, since they both make crucial use of causal concepts in their theories of explanation but go on to offer us substantial theories of causality. I do not believe that taking causation among one’s basic resources in building a theory of explanation

21 Actually, the causal relation, I believe, is only a special case; more broadly, what is involved is an asymmetric dependency relation whereby the event being explained depends, causally or otherwise, on the events that do the explaining. For details, see my “Explanatory Knowledge and Metaphysical Dependence,” Philosophical Issues 5 (1994): 51–69. Essay 8 of this volume. 22 Salmon, Scientific Explanation and the Causal Structure of the World (Princeton: Princeton University Press, 1984). Humphreys, The Chances of Explanation (Princeton: Princeton University Press, 1989).

198


makes the project trivial, or uninformatively easy, by any means. First of all, a full theory of scientific explanation must be above all a theory of scientific understanding, and that it is by no means obvious—in fact, this, I believe, is one of the really deep questions in the epistemology of science—why and how causal knowledge enhances understanding. Second, not all explanations, not even those that concern individual events and occurrences, are causal explanations; causation, though important, is clearly not the only venue for achieving understanding. Hempel would have resisted taking causation as a primitive for his theory. At this point in his career, I don’t think that he would have rejected causation on the ground that it is an illegitimate metaphysical concept. What is more likely, he would have said that doing so would be explaining explanation by a concept that is at least as much in need of clarification as explanation. Actually, he might have said something more deflationary, something similar to what he said about the notion of understanding or verstehen, namely that it is a psychologistic and anthropomorphic concept (rooted in our intuitive notions of agency and power) with no precise intersubjective criteria of application. No doubt, Hempel would have been horrified by the idea of using the concept of understanding as a primitive in a theory of scientific explanation. I believe Hempel’s attitude was that for a philosophical understanding of science, there is no need for the concept of causation, and that the salvageable part of this concept is fully captured in the idea of bringing a pair of events under a general law—that is, by the D-N, or nomic-subsumptive, model of explanation. This in fact seems to have been Hempel’s later view; he wrote: In the explanatory or predictive use of a deterministic theory, then, the notion of a cause as a more or less narrowly circumscribed antecedent event has been replaced by that of some antecedent state of the total system, which provides the “initial conditions” for the computation, by means of the theory, of the later state that is to be explained.23 And Hempel dismisses cases like the Empire State Building by pointing to the vagueness and indeterminateness of the commonsense notion of what can explain what. He wasn’t about to take seriously metaphysical notions like causal dependency, or dependency in general, as a metaphysical 23 Aspects of Scientific Explanation, p. 351.


199

underpinning of explanatory relations. For him, subsumption-under-a-law remained the fundamental core idea of explanation—fundamental in the sense that it requires no further support or backing from any other source, especially something metaphysical. As we saw, Hempel talks of “antecedent conditions” in connection with the singular sentences in the explanans. He also refers to them as “events,” and I was surprised to find that he had something to say about events that is quite close to what is now called “the property exemplification theory.” In the Hempel-Oppenheim paper, we read: When we spoke of the explanation of a single event, the term “event” referred to the occurrence of some more or less complex characteristic in a specified spatio-temporal location or in a certain individual object, and not all the characteristics of that object, or to all that goes on in that space-time region.24 I don’t see anything objectionably metaphysical in this notion of an event, or, for that matter, in the common preanalytic notion of an event. If Hempel had taken this concept of an event seriously as a philosophical resource, he could have easily handled another class of vexing counterexamples to his account. Consider the following argument: All ravens are black. Wilbur is a nonblack raven, or he has the flu. Therefore, Wilbur has the flu. (If you don’t think “All ravens are black” is a law, you can run this argument with your favorite law.) This is another perfectly nice D-N argument; in fact, it satisfies the elaborate formal requirements developed in Part III of “Studies in the Logic of Explanation,” and we may assume all of the sentences in this argument are true. But the argument is obviously and wholly nonexplanatory. If this argument were allowed to stand as an explanation, everything under the sun could be explained on the basis of

24 Ibid. 253. I noticed this paragraph for the first time when I was reading the Hempel-Oppenheim paper to prepare the present paper. As some will recognize, the idea here is the core of the property exemplification theory of events, something I had a hand in developing (see “Events as Property Exemplifications,” reprinted in my Supervenience and Mind (Cambridge: Cambridge University Press, 1993)). I was shocked to realize that the basic idea of this approach had been present in Hempel all along! But who knows—I was perhaps subliminally influenced by this paragraph, since I must have read this paper at least a dozen times before I began thinking about events in the mid-1960s.

200


but one law: for any arbitrary explanandum p, we have the following D-N argument: All ravens are black. Wilbur is a nonblack raven, or p. Therefore, p. Again, a metaphysical remedy is close at hand: each singular premise in the explanans must specify an event, or an antecedent condition if you wish. The disjunctive premise in the above explanation fails to do so.25 These examples are a bit far-fetched and may strike you as the inventions of a philosopher hell-bent on finding counterexamples. In thinking about mental causation, I have recently had to consider the explanatory import of the following kind of situation: Suppose you want to explain why a certain person, call her Mary, is experiencing pains in her joints. Mary has been diagnosed to have either lupus or rheumatoid arthritis, and it is known, let us suppose, that patients with either of these diseases have pains in their joints. Consider then the following argument: Mary has either lupus or rheumatoid arthritis. Patients with lupus have pains in their joints. So do patients with rheumatoid arthritis. Therefore, Mary has pains in her joints. Another perfectly nice D-N argument! I would urge, though, that here we do not as yet have an explanation of pains in Mary’s joints. We do not as yet know what is causing her pains. True, we have narrowed the possible explanations to two, and know that one or the other must be the correct explanation. But to have an explanation of Mary’s condition, to know its cause, we have more work to do: we must find out whether Mary has lupus

25 It is interesting and instructive to note that David Kaplan, in his revision of the HempelOppenheim model of deductive explanation, requires that C, the initial condition sentence in the explanans, be a conjunction of basic sentences, where a basic sentence is either an atomic sentence or the negation of one. (See “Explanation Revisited,” Philosophy of Science 28 (1961): 429–436.) If we think of the primitive predicates of a scientific theory as denoting kinds or properties recognized in that theory, it is plausible to think of an atomic sentence as affirming the occurrence of an event that consists in the instantiation of a property in the domain of the theory and the negation of an atomic sentence as denying the occurrence of an event. This means that, although Kaplan’s requirement was motivated exclusively on formal grounds, it can be taken as a syntactic enforcement of the metaphysical constraint of the sort I have in mind.


201

or she has rheumatoid arthritis. Here, we have a disjunction of two (possible) explanations, not a single disjunctive explanation. The reason for this is simple: the singular premise, the disjunctive sentence “Mary has either lupus or rheumatoid arthritis” fails to specify the occurrence of an event or state. It says one or the other of the two events occurs, but that is not the same thing as stating the occurrence of a “disjunctive event.” I believe there are no such things as disjunctive events any more than disjunctive persons or tables. “Mary or Sarah came to the party” doesn’t mean that a disjunctive person, “Mary or Sarah,” came to the party! Similarly, there are no disjunctive diseases like “lupus or rheumatoid arthritis.” This is related to the metaphysical principle that disjunctions of kinds are not necessarily kinds, or that kindhood, or propertyhood, is not closed under disjunction. This shows, I hope, what the infusion of a little serious metaphysics into Hempel’s conceptual arsenal could have done to streamline his theory of explanation. Unless you appeal to metaphysics there seems no obvious way of accounting for the last example. If your method is wholly syntactic, it would be difficult to justify the banning of certain disjunctions, but not conjunctions, from singular premises of the explanans. Of course I am not urging that we banish all disjunctive singular sentences from explanations; it may be that some disjunctive sentences can serve to specify events and states, and that some disjunctive predicates pick out genuine kinds.

III I now turn to another aspect of the theory of explanation—to praise Hempel’s D-N model, not to complain about it. In insisting that explanation be logical derivation—moreover, derivation from premises including laws—Hempel’s D-N model lays stress on one essential aspect of explanation that is missed by other theories that overtly deny, or neglect, the derivational aspect of explanation. Even some very early critics of Hempel excoriated him for requiring that explanations be logical derivations. Michael Scriven was one of Hempel’s most persistent critics, contesting him on virtually every point. Amazingly, Scriven even took issue with Hempel’s assumption that the explanandum of an explanation be true, arguing that even events that are fictional or

202


nonexistent can be explained. One of the more notable claims Scriven made was that explanations are not arguments or derivations, and that a single statement standing alone, like “The bridge collapsed because a bomb exploded,” can be a full explanation. Further, he argued that laws cannot properly be viewed as part of an explanation, and that inclusion of laws in an explanation is to confuse explanations with “grounds” for explanations.26 If laws are expunged from explanations, explanations could no longer be derivations. Scriven has not been alone in claiming that explanations are typically not arguments or derivations, and that singular statements standing alone can fully serve as explanations. In a backhanded compliment, Salmon, one of the most influential theorists of explanation, has dubbed the view that explanations are arguments “the third dogma of empiricism.”27 Let us look at what James Woodward, one of the more recent advocates of the nonderivational approach, has to say. He argues that the following statements can be full explanations in their own right: (1) The short circuit caused the fire. (2) The blow of the hammer caused the chestnut to shatter. Woodward goes on to state his basic position: According to the view I defend, singular causal sentences and singular causal explanations do not differ fundamentally in structure in this way. A singular explanation will simply consist of a sentence which reports a causal connection like (1) or (2) above. Singular causal explanations are extensional on both the cause-side and the effect-side . . . 28 This view, the doctrine of singular explanations, is particularly common among philosophers who associate explanations closely with causation. David Lewis is another such philosopher: according to him, to give a causal

26 Michael Scriven, “Explanations, Predictions, and Laws,” Minnesota Studies in the Philosophy of Science 3 (1962): 170–230. 27 Wesley Salmon, “The Third Dogma of Empiricism,” in Basic Problems in Methodology and Linguistics, ed. Robert Butts and Jakko Hintikka (Dordrecht: Reidel, 1977). I have no idea why Salmon thinks this view of explanation has anything to do with empiricism. 28 James Woodward, “A Theory of Singular Causal Explanation,” Erkenntnis 21 (1984): 231–262.


203

explanation of an event is to give information about its causal history, that is, the series of events that is the causal ancestry of the event to be explained.29 On this approach, there need be no significant connection between explanation and derivation; there is no reason to think that derivations, or arguments, are the only way, or even the main way, of providing causal information. A single statement standing alone, or sets of statements not configured as a derivation, can clearly do the job. Any theorist who is an extreme realist about explanations is especially apt to embrace this approach. Suppose that you believe that explanatory relations, at least those between individual events, must be backed, or supported, by causal relations. That is, for the occurrence of an event c to explain why another event, e, occurred, it must be the case that c caused e. On this view, the objective content of the statement that “c explains e” is exhausted by the fact that c caused e, and the information about the world that is carried by the explanation coincides with the causal information, information about what caused the event to be explained. If all this is the case, it is only natural to suppose that a statement of the form “c caused e” is an explanation of e in terms of c. What more can one expect? Of course, one could go on and supply further information as to why c qualifies as e’s cause, by what process c caused e, and so on, but this is information that goes beyond explaining e. Explanation of e is done when we have in hand the true singular causal statement “c caused e.” We know the cause of e, and hence we have an explanation, causal explanation, of e. But let us look at this more closely. Consider Woodward’s (1) and (2) above. He claims that these statements, of the form “X caused Y,” are referentially transparent at both positions “X” and “Y,” on account of the fact that causation is an extensional relation. But consider the result of replacing the referential terms in, say, the first of these statements via the following identities (which we assume to be true): (3) The short circuit ¼ the event that Mildred blamed on Harry. (4) The fire ¼ the top news event of the year. What we get is this causal statement:

29 David Lewis, “Causal Explanation,” in Philosophical Papers II (New York and Oxford: Oxford University Press, 1986).

204


(5) The event that Mildred blamed on Harry caused the top news event of the year. Given that (1) and (2) are true, (5) is no doubt a true causal statement, and “the fact” reported by it, we may concede, is the very same causal fact reported by the original statement. But clearly we have lost something in moving from (1) to (5). It is also clear what has been lost: we have lost crucial explanatory information. What (5) doesn’t tell us is what it is about the purported cause that can help us understand why this particular event occurred. In fact, we don’t even know what our explanandum event really is. We might even consider proper names for events: call the short circuit “Charlie” and the fire “Dave.” What explanation do we get from being told that “Charlie caused Dave”? What explanatory insight or illumination do we gain when we are told this? Obviously none.30 To my mind, the crucial virtue of the D-N model is the fact that it forces an explanation to bring into the open the explanatorily salient properties of the phenomena involved—properties that do the explanatory work. When the explanation is a causal explanation, it must point us to the causally efficacious or relevant properties in virtue of which the purported cause brought about the event to be explained. And for this to happen, the descriptions, or referential expressions, that are used must tell us something about these causally (therefore, explanatorily) salient properties. This is why (5) carries little explanatory information whereas (1) clearly does. The point is that although causation may be an extensional relation, explanation is not.31 This is why bare singular causal statements can be wholly unsatisfying, and totally useless, as explanations. “X caused Y” may tell us no more than that there is a causal explanation of Y somewhere. The virtue of Hempelian D-N explanations consists in the fact that they wear this crucial explanatory information on their sleeve. It makes evident what it is about the event, or events, invoked in the explanans that causally explains the explanandum event. The reason it does this is that laws, or causal laws, are actually used,

30 This may be one reason why we do not normally use proper names for events and find such designators as “the top news event of the year,” “the event I was just thinking about,” as in some sense deviant and unusual. 31 A point stressed by Davidson, in “Causal Relations,” reprinted in his Essays on Actions and Events (New York: Oxford University Press, 1980).


205

not simply named or assumed to exist, in an explanation. When we know what the explanatory laws are, we know what the explanatorily, or causally, efficacious properties are, and to make the derivation possible, we are forced to choose descriptions of events that pick out their active causal properties. Without this knowledge we have no explanation; we do not understand why the event we want to have explained occurred. To summarize, then, explanations must tell us something about the explanatorily or causally efficacious/relevant properties, and that is what the D-N model guarantees. To be sure, the D-N model may be asking too much: it requires that we be told about all the properties that are relevant— at least, all such properties holding at a given time. But often we look for an explanation to fill a particular gap in our understanding and all we need may be one small bit of relevant information. Does this mean that the D-N model is an overkill? Hardly. What we should do is to take a D-N explanation as an explanatory ideal and consider real-life explanations as approximations to this ideal.32 In this sense, the D-N model can serve as an indispensable reference point. This way of looking at Hempel’s theory of explanation is consistent with my earlier promotion of greater metaphysical involvement on the part of theories of explanation. We may require that explanations invoke causal laws, and that the singular premises of the explanans specify individual events and states. But this is consistent with the further requirement that explanations must ideally be deductive and nomological. This insistence is what guarantees that our explanations provide us with the explanatory information and illumination that we seek. And I take this to be the crucial contribution of Hempel’s D-N theory of explanation.33 In closing, I want to note Hempel’s insistence that in doing science, or seeking knowledge, we are not merely in search of more facts but rather we are seeking an understanding and illumination of the multitude of phenomena encountered in our experience. It is true that the particular way in which Hempel developed his theory of explanation was heavily influenced by his positivist commitments, and that he did not give full play to the important metaphysical, epistemic, and psychological components of 32 I believe Peter Railton has advocated a similar approach. 33 Michael Friedman and Philip Kitcher are among the few who have stayed with Hempel’s claim that explanations involve arguments or derivations.

206


knowledge and understanding. That is entirely understandable given his intellectual and philosophical context. But then we are working in our own historical-cultural context. It gives me a wonderful feeling to know that we continue to learn from Peter Hempel’s work, a body of work that teaches us what genuine intellectual honesty and integrity mean.

10 Reduction and Reductive Explanation: Is One Possible Without the Other? I A certain picture seems widespread and influential in recent discussions of issues that involve reduction and reductive explanation—especially, in connection with the mind-body problem. The same picture is also influential in the way many think about the relationship between the “higherlevel” special sciences and “basic” sciences. What I have in mind is the idea that reducing something is one thing and reductively explaining it is quite another. There supposedly is a vital difference, from both the scientific and philosophical point of view, between reducing psychological phenomena to biological/physical phenomena and reductively explaining the former in terms of the latter. The significance of the difference, on this line of thought, derives from the purported fact that reductive explanation is often an achievable scientific goal whereas reduction is an overreaching metaphysical aspiration that is seldom, if ever, realized. To see what this picture is and appreciate its appeal, consider two domains (or “levels,” if you like) of phenomena, M and P. (For concreteness, we may think of M as “mental” and P as “physical.”) To reduce M to P, we must show, to use J.J.C. Smart’s suggestive phrase, that the M-phenomena are “nothing over and above” the P-phenomena.1 A proposed reduction might be “eliminative”—that is, it consists in showing that 1 J.J.C. Smart, “Sensations and Brain Processes,” Philosophical Review 68 (1959): 141–156. Reprinted in numerous anthologies, e.g. Philosophy of Mind: A Guide and Anthology, ed. John Heil (Oxford: Oxford University Press, 2004).

208

REDUCTION AND REDUCTIVE EXPLANATION

there really are no such things as M-phenomena (“there really are no such things as caloric fluids; there is only molecular motion”). If such a reduction goes through, there trivially are no M-phenomena over and above P-phenomena. Whether eliminative reduction is a serious form of reduction can be debated, but we should keep in mind that “reduction” is a term of art and there need be no harm in the idea of eliminative reduction. A more central form of reduction is “conservative” (or “preservative,” “retentive”) reduction whereby the reduced phenomena survive as legitimate entities of the world. It’s only that they now turn up as entities in the base domain. Heat was conservatively reduced—it survives as molecular kinetic energy—whereas caloric fluids were eliminated. Genes were conservatively reduced; vital forces and entelechies were eliminated. If M-phenomena are to be conservatively reduced to P-phenomena, they must be shown to be “nothing over and above” the P-phenomena, and it is hard to see how this could be done unless each M-phenomenon is claimed, and shown, to be identical with a P-phenomenon. That is, reduction appears prima facie to require the identification of M-phenomena with P-phenomena, and this means that M is turned into a subdomain of P. This is no surprise: reduction must reduce, and if M is reduced to P, M-phenomena no longer exist as something extra, something in addition to P-phenomena. In contrast, when we think about reductive explanation—that is, explaining M-phenomena on the basis of P-phenomena (including P-laws)— a natural train of thoughts seems to lead to a considerably different picture. Suppose we explain an M-phenomenon in terms of P-phenomena. We now understand why, and how, this M-phenomenon arises from certain P-phenomena: it is because these particular P-phenomena constitute an underlying mechanism whose operations yield phenomena of kind M. Prima facie, this doesn’t seem to undermine, or affect in any way, the ontological status of the M-phenomenon vis-a`-vis P-phenomena. It apparently remains an entity with a legitimate, independent standing in its own right; it’s only that its existence and character has now been made intelligible in light of the underlying phenomena and mechanisms. There seems no reason to think such an explanation of an M-phenomenon carries any commitment, explicit or implicit, to the claim that it is “nothing over and above” the underlying P-phenomena; nor does it appear to imply, or suggest, that the M-phenomenon must be identified with a P-phenomenon. As an analogy, think about causal explanation: we do not think that a


209

causal explanation of an event adversely affects its status as an entity. The effect remains an independent entity ontologically distinct from the cause. If something like this is right, reductive explanation should be possible even where there is no reduction. Or so it seems at first blush. Conversely, it is difficult to see why we should expect reduction to yield reductive explanation: M-phenomena may be nothing over and above P-phenomena, but that in itself says nothing about explanation, something that has essential epistemic dimensions. If an M-phenomenon is identical with a P-phenomenon, there seems no phenomenon here which is specifically an M-type phenomenon that needs to be, or can be, explained, reductively or otherwise. The M-phenomenon is a P-phenomenon after all, and it is prima facie incoherent to speak of “reductively” explaining a P-phenomenon in terms of other P-phenomena. It seems as though reduction might actually preclude reductive explanation: if a phenomenon has been reduced, that seems to take away not only our need for a reductive explanation of it but also its very possibility. In any case, it is clear that there are issues about reduction and reductive explanation that need to be sorted out and clarified. Jerry Fodor’s “Special Sciences”2 was the canonical source of the antireductionist arguments in the latter half of the 20th century. As is widely known, Fodor’s antireductionist argument, based on the so-called multiple realizability of psychological and other special-science properties, played a pivotal role in creating what Ned Block has aptly dubbed “the antireductionist consensus.”3 The consensus perhaps no longer enjoys the pervasive hold it once did, but it is still a widely shared orthodoxy with considerable reach and influence.4 However, few commentators seem to have noticed the following surprising paragraph in Fodor’s paper: It seems to me (to put the point quite generally) that the classical construal of the unity of science has really badly misconstrued the goal of scientific reduction. The point of reduction is not primarily to find some natural kind predicate of physics coextensive with each kind predicate of a special science. It is, rather, to explicate the physical 2 “Special Sciences: The Disunity of Science as a Working Hypothesis,” Synthese 28 (1974), 97–115. 3 Block, “Antireductionism Slaps Back,” Philosophical Perspectives 11 (1997): 107–132. 4 For a vigorous defense, see, for example, Don Ross and David Spurrett, “What to Say to a Skeptical Metaphysician: A Defense Manual for Cognitive and Behavioral Scientists,” Behavioral and Brain Sciences 27 (2004): 603–627.

210


mechanisms whereby events conform to the laws of the special sciences [emphasis added]. I have been arguing that there is no logical or epistemological reason why success in the second of these projects should require success in the first, and that the two are likely to come apart in fact wherever the physical mechanisms whereby events conform to a law of the special sciences are heterogeneous.5 Here, Fodor is saying that although the “bridge laws” required for Nagelian reduction (more on bridge laws below) are unavailable (since, as he says, special science predicates in general have no coextensive physical predicates) and so reduction is impossible, this does not exclude reductive explanations of special-science laws in terms of “physical mechanisms” at the lower levels. In fact, he is suggesting something bold and new, namely that the idea of reduction be reconstrued as, or be discarded in favor of, reductive explanation. So, for Fodor, reduction is not possible anywhere; yet, reductive explanation is a legitimate scientific procedure which presumably is often successfully executed. Unfortunately, Fodor drops the matter here and says nothing further about how he conceives reductive explanation, or why explanation in terms of “physical mechanisms” is an appropriate replacement for reduction as traditionally conceived. The idea that reductive explanation can thrive even where reduction fails appears to reflect a natural way of thinking about the interlevel relations in the sciences, and it reappears, more than twenty years later, in the following remarks by David Chalmers: A reductive explanation of a phenomenon need not require a reduction of that phenomenon . . . In a certain sense, phenomena that can be realized in many different physical substrates—learning, for example— might not be reducible in that we cannot identify learning with any specific lower-level phenomena. But this multiple realizability does not stand in the way of reductively explaining any instance of learning in terms of lower-level phenomena.6 Chalmers has evidently bought into the Putnam-Fodor multiple realization argument against reduction, or type-identity reduction at any rate, but he claims that this in no way affects the possibility of reductively explaining higher-level phenomena in terms of phenomena and mechanisms at lower levels. 5 “Special Sciences,” p. 107. 6 Chalmers, The Conscious Mind (Oxford: Oxford University Press, 1996), p. 43.


211

But how is that possible? Chalmers doesn’t address this question directly, but it isn’t difficult, I believe, to discern an idea that could motivate us to take a stance like Chalmers’. Reduction requires type identities, which are ruled out by the phenomenon of multiple realization; however, reductive explanation can target individual instances instead of types. Any instance of a higher-level phenomenon occurs by being realized by a lower-level phenomenon, and it can therefore be explained in terms of its underlying realizer. This means that two instances of the same higher-level phenomenon (as a type) may receive two distinct reductive explanations, each in terms of its own realizing mechanism. This is an initial thought; we will explore below how it might be fleshed out. A related thought is that physicalism itself can, and perhaps should, be understood in a new way. In order to secure physicalism, we do not need a reduction of all phenomena to a physical base; if every ostensibly nonphysical phenomenon can be reductively explained in terms of physical phenomena, that should be good enough for physicalism. If all mental phenomena are shown to be explainable on the basis of physical phenomena and physical laws, why isn’t that physicalism enough? For that would mean that physical phenomena suffice for the understanding of everything about mentality— why the phenomena of the mind occur in the way they do, why they interrelate among themselves and relate to physical phenomena as they do, and all the rest. It is perhaps no accident that what many regard as the most important obstacle to physicalism is called the problem of “explanatory gap.”7 The problem, as everyone knows, is that of explaining—presumably, reductively explaining—phenomenal consciousness, or qualia, in terms of physical/biological phenomena. The idea is that once such an explanation is achieved, or shown to be achievable, the gap is closed and physicalism is home free. Conversely, if the gap resists closure, that should defeat physicalism. But can we separate reduction and reductive explanation so easily and neatly? Is it really possible to reductively explain a mental phenomenon, say pain, in neural terms and for this phenomenon to remain “over and above” neural phenomena, as a distinct and separate entity? In considering this question, we need to keep in mind what the qualifier “reductive” adds to 7 Joseph Levine, “Materialism and Qualia: The Explanatory Gap,” Pacific Philosophical Quarterly 64 (1983): 354–361.

212


explanation—that is, how reductive explanations might differ from ordinary, nonreductive ones.8 Conversely, if mental phenomenon M has been shown to be “nothing over and above” a physical/biological phenomenon P, will that give us a reductive explanation of M in terms of P? In what follows, we will consider these and related questions in regard to three models of reduction currently on the scene—bridge-law reduction, identity reduction, and functional reduction. As will be seen, answers to our question vary depending on the kind of reduction involved. Bridge-law reduction, I will argue, is an oxymoron: it yields neither reduction nor reductive explanation. In contrast, identity reduction gives us reduction but no reductive explanation. Finally, functional reduction can be seen to yield reductive explanation and, arguably but not unproblematically, reduction as well.

II Bridge-law reduction was devised by Ernest Nagel in the 1950’s and ’60s9 as an account of intertheoretic reduction in science. The model as it has been generally understood in the debate over reduction and reductionism in the decades that followed is a somewhat simplified version of the model actually formulated by Nagel,10 and here we will use this simpler and more familiar version. Let T1 and T2 be two theories where theories are construed as deductively closed sets of laws, with some laws designated as “basic” and the rest being logically derivable from them. According to the bridge-law model: T2 is reducible to T1 ¼ def. (1) [the bridge-law condition] for each primitive predicate F of T2 there is a T1-predicate G such that a “bridge law” of the biconditional form “Fx $ Gx” holds, and (2) [the derivability condition] each law of T2 is logically derivable from the laws of T1, with the bridge laws taken as auxiliary premises.

8 We will not consider this question explicitly in this paper. For details see Kim, Physicalism, Or Something Near Enough (Princeton: Princeton University Press, 2005), chap. 4. 9 Nagel, The Structure of Science (New York: Harcourt, Brace and World, 1961). 10 Ibid., and “Issues in the Logic of Reductive Explanations” (1970), in Nagel, Teleology Revisited (New York: Columbia University Press, 1979).


213

On this model, then, the reduction of one theory to another amounts to a deductive absorption of the former into the latter augmented with the bridge laws. Since the proprietary vocabularies of the two theories must be expected to be disjoint, or at least not to completely overlap (for example, thermodynamics and statistical mechanics, classical and molecular genetics), Nagel thought that principles connecting, or “bridging,” the two vocabularies are needed to enable the derivation, and that these would typically be empirical scientific laws correlating phenomena of the two domains involved.11 It is easy to appreciate the centrality of bridge laws to reductions of this form. For we can quickly see that if the bridge-law requirement (1) is met, the derivability condition (2) is automatically met as well—with a small caveat. Let L be any law of T2, the theory being reduced, and assume that the bridge-law condition has been satisfied. We can now use these biconditional laws as definitions and rewrite L entirely in the vocabulary of T1, the base theory. Let L* be this T1-rewrite of L. Either L* is derivable from T1-laws or it is not. If it is, then L can be derived from the T1-laws (derive L* first and then derive L from L* using the bridge laws) and the derivability condition is met. If L* is not derivable from T1-laws, add L* to T1 as an additional basic law. This is permissible because in missing L*, T1 is missing an important truth, a lawlike truth, in its domain (here we are assuming T2 and the bridge laws to be true). In this case, L is derivable from this augmented base theory T1 + L*, again with the help of the bridge laws. In either case, therefore, the derivability requirement is met, either in regard to T1 or with respect to an augmented version of T1 stated in the same theoretical vocabulary.12 We can see the importance of bridge laws in another way. Checking the derivability condition in any given case requires us to look at whole theories. For this to have real significance, the theories we consider must 11 Nagel’s final formulation of his model (The Structure of Science) does not require the bridge laws to be biconditionals in form; however, in discussions of reduction they are standardly taken to be biconditionals; see, e.g., Fodor’s talk, in his quotation above, of classical reduction requiring “some natural kind predicate of physics coextensive with each kind predicate of a special science” (added emphasis), and we follow this practice here. In any case, this is one reason to call the model as presented here “bridge-law reduction,” not “Nagel reduction” (another, and more important, reason is the centrality of bridge laws to the model as will shortly be seen). Also, condition (1) is simplified in that it only refers to monadic predicates. 12 Note: if T1 is physics, so is T1 + L*; if T1 is biology, so is T1 + L*.

214


be completed theories, or nearly completed ones. This makes it difficult to discuss, say, whether psychology is reducible to neuroscience because we have nothing like a completed psychology—or a completed science of the brain for that matter. However, the question whether or not we can expect to find bridge laws connecting psychological states with neural/physiological states has seemed more amenable to discussion—even at the present state of knowledge. Many well-known antireductionist arguments about psychology aim at showing that there can be no such psychoneural laws, quite independently of what completed psychology or neuroscience will look like. The Putnam-Fodor multiple realization argument13 is one such example; Donald Davidson’s anomalist argument is another.14 These arguments are largely independent of the current state of the psychological or neural sciences, and simply bypass consideration of the derivability requirement. Suppose that we have a theory about pain and that this theory includes the following law: (L) Pain causes distress. Suppose further that this pain theory has been bridge law reduced to neurophysiology, with the help of psychoneural bridge laws, including these: Pain occurs to x $ neural state N1 occurs in x. Distress occurs to x $ neural state N2 occurs in x. If (L) has been derived from neurophysiological laws with these bridge laws as extra premises, neurophysiology must include the following law: (L*) N1 causes N2. To reduce (L) to neural theory, then, we would derive (L) from (L*) using the two bridge laws. We may assume other laws about pain are similarly derivable from laws of neural theory. By carrying out a bridge-law reduction in this way, have we reduced pain? Do we have a reductive explanation

13 Hilary Putnam, “Psychological Predicates” (1967), reprinted with the title “The Nature of Mental States,” in Putnam, Mind, Language, and Reality: Philosophical Papers, vol. 2 (Cambridge: Cambridge University Press, 1979); Fodor, “Special Sciences.” 14 Donald Davidson, “Mental Events,” reprinted in Davidson, Essays on Actions and Events (Oxford: Oxford University Press, 1980).


215

of pain? And do we have a reduction, or reductive explanation, of psychological law (L) on the basis of a neural law? I believe the answer is no to each of these questions. On the standard conception of bridge-law reduction, bridge laws are a posteriori and contingent, and this means that the properties or states connected in a bridge law remain distinct. The bridge law connecting pain with N1 does not entitle us to say: pain is “nothing over and above” neural state N1. We do not think that nomological equivalence or coextensiveness guarantees identity, or warrants “nothing over and above” or “nothing but” talk. We might say that N1 is the neural correlate, or substrate, of pain, and that’s all this bridge law entitles us to say. Does our bridge-law reduction of pain give us a reductive explanation of pain in neural terms—more specifically, in terms of N1? The answer, again, is no. The reduction takes the pain–N1 correlation as an unexplained, underived, premise in the derivation of the pain theory from neural theory. Within a Nagelian bridge-law reduction, bridge laws are taken as fundamental premises. The “explanatory” gap between pain and N1 remains untouched. In fact, what creates the explanatory gap is exactly the pain–N1 bridge law. The gap arises because we are apt, or perhaps destined, to ask questions like the following: why does pain correlate with N1 rather than another neural state?; why doesn’t itch correlate with N1?; why does any qualitative experience correlate with N1?; and so on. Clearly, what is in need of explanation—reductive explanation—is why the bridge law correlating pain with N1 holds—what it is about the physical nature of N1 that explains why it correlates with a conscious state with the phenomenal character constitutive of pain. Finally, do we get from this bridge-law reduction an explanation of the psychological law (L)—that is, why pain causes distress? Again, the answer is no. All we can conclude from (L*) and the two bridge laws is that pain correlates with a neural state which causes the neural state with which distress is correlated.15 That doesn’t come anywhere near an explanation of why pain causes distress. The conclusion is unavoidable: Bridge-law reduction gives us neither reduction nor reductive explanation. At least in this case, reduction and reductive explanation go together—by both being absent. 15 To mimic a felicitous sentence from Ned Block and Robert Stalnaker, “Conceptual Analysis, Dualism, and the Explanatory Gap,” Philosophical Review 108 (1999): 1–46.

216


III A natural step to take at this point is to consider upgrading bridge laws into something stronger and more robust, something that could support reductive claims and perhaps also handle the explanatory issues. And it didn’t escape philosophers’ attention that identities might be just the needed replacements for the bridge laws. The idea goes back to the early mindbody identity theorists like Herbert Feigl and J.J.C. Smart. Feigl famously called these psychoneural bridge laws “dangling” laws, and Smart’s explicit aim in promoting psychoneural identity theory was to eliminate these “nomological danglers” and replace them with psychoneural identities.16 So in place of the likes of: Pain occurs $ N1 occurs we would now have identities like: Pain ¼ N1. These identities are stronger than the corresponding correlations and can do their work in inferential contexts. So theory reduction construed, a` la Nagel, as logical derivation of the reduced theory from the reducer can be based on identities as well as correlations.17 Early psychoneural identity theorists considered these identities contingent and a posteriori. Things have changed in our post-Kripkean modal paradise: now the identities are generally taken to be necessary truths (if true), though they are allowed to retain their a posteriori character. Their epistemic and theoretical status, according to some influential latter-day identity theorists,18 is supposed to be the same as, or at least similar to, that of 16 Feigl, “The ‘Mental’ and the ‘Physical’,” Minnesota Studies in the Philosophy of Science 2 (1958): 370–497; Smart, “Sensations and Brain Processes.” It should be noted that Feigl didn’t see eye to eye with Smart on this issue; see Feigl, “Postscript after Ten Years,” in Feigl, The “Mental” and the “Physical”: The Essay and a Postscript (Minneapolis: University of Minnesota Press, 1967), pp. 136ff. Here my discussion focuses on Smart’s views. 17 Lawrence Sklar, “Types of Intertheoretic Reduction,” British Journal for the Philosophy of Science 18 (1967): 109–124; Robert Causey, “Attribute Identities in Microreductions,” Journal of Philosophy 69 (1972): 407–422. Nagel himself later recognized identities as a form of bridge laws; his examples include “water ¼ H2O” and “light waves are electromagnetic waves” (“Issues in the Logic of Reductive Explanations”). 18 For example, Christopher Hill, Sensations (Cambridge: Cambridge University Press, 1991); Block and Stalnaker, “Conceptual Analysis”; Brian McLaughlin, “In Defense of New-Wave Materialism,” in Physicalism and Its Discontents, ed. Carl Gillett and Barry Loewer (Cambridge: Cambridge University Press, 2001).


217

scientific identities like “water ¼ H2O,” “heat ¼ molecular kinetic energy,” and “genes ¼ DNA molecules.” So it isn’t surprising that some have claimed psychoneural identities to be justifiable by the same sort of evidence and consideration that warrant acceptance of these familiar scientific identities.19 We may call this mode of reduction “identity reduction”: reduction is accomplished by identifying phenomena and properties being reduced with appropriate items in the base domain. The return of psychoneural type identity theory, during the 1990s, is one of the more interesting developments in the recent debate on the mind-body problem.20 There is no question about the reductive import of identity reduction. If pain ¼ N1, there is no pain over and above N1; if mental states are identical with brain states, there are no mental states over and above brain states. This is an open-and-shut affair if anything in philosophy ever is: Identities do reduce. For reduction nothing works as magically as identities, and it may well be that identities of some sort are required for any genuine reduction. It is of course an independent question where we can get these identities, in particular psychoneural identities. We must earn our entitlement to them; it is not acceptable to argue that the identities are warranted because they would give us psychoneural reduction. I believe this is the biggest remaining hurdle for identity reduction; I do not myself believe it can be overcome.21 But does identity reduction yield reductive explanation? Does the identity “pain ¼ N1” help close the explanatory gap between pain and N1? If the identity holds, there is here only one thing, not two, and, to push the “gap” metaphor a bit, at least two distinct items are needed to create a gap. If psychoneural identities hold, there isn’t any mind-brain gap to be closed and there never was. If we stay with the correlation “pain occurs $ N1 occurs,” we face explanatory challenges of the sort the emergentists have raised: Why

19 Block and Stalnaker, “Conceptual Analysis.” This suggestion would seem to turn the mind-body problem into a scientific problem, one that can be resolved by scientific research. We should keep in mind Smart’s remark to the effect that while the choice between the brain-state theory and the kidneystate theory is an empirical scientific issue, the choice between the brain-state theory and epiphenomenalism is not (Smart, “Sensations and Brain Processes”). I believe that Block and Stalnaker’s argument for their proposal is flawed (Kim, Physicalism). 20 How does the multiple realizability argument affect the new identity theory? This question has not received the attention it deserves. For some useful discussion see Hill, Sensations, pp. 101ff., and Block, “Antireductionism Slaps Back.” 21 Kim, Physicalism.

218


does pain, not itch or tickle, correlate with N1? Why doesn’t pain correlate with a different neural state? And so on. As I believe Ned Block remarked somewhere, the problem of the explanatory gap is to answer the question “Why do phenomenal states correlate with the neural states with which they correlate?” On behalf of the identity theory, Block and Stalnaker decisively dismiss such explanatory requests: If we believe that heat is correlated with but not identical to molecular kinetic energy, we should regard as legitimate the question of why the correlation exists and what its mechanism is. But once we realize that heat is molecular kinetic energy, questions like this can be seen as wrongheaded.22 In general, there are two ways of responding to an explanatory request “why p?” The first is to provide a correct answer to the question, by offering an explanation of why p. The second is to show that the presupposition of the request, namely that there is here something to be explained, is incorrect, and that in consequence no explanation is needed, or even possible. When p is false, the question “why p?” obviously has no correct answer (consider “Why does oil dissolve in water?”). In the present case, we could say either that the identity of pain with neural state N1 shows that there is here no correlation between the two states, and that this makes the presupposition of the question “Why does pain correlate with N1?” false, or we could say that the identity trivializes the question into “Why does pain correlate with pain?” or “Why does N1 correlate with N1?” In either case, there is nothing to be explained here, and there is no gap to be closed.23 That, however, is not the end of the story. There is an important further point that has to hold if psychoneural identities are to resolve the explanatory gap problem, and it is this: Identities like “heat ¼ mke” and “pain ¼ N1” are immune to further explanatory challenges. It isn’t enough that, as Block and Stalnaker say, “heat ¼ mke” renders the question “Why does heat correlate with mke?” wrongheaded; it must also be the case that 22 Block and Stalnaker, “Conceptual Analysis,” 24. 23 We should take note of a different take on the issue of identities and the explanatory problem. According to Hill, Sensations and McLaughlin, “In Defense of New-Wave Materialism,” psychoneural identities provide explanations for psychoneural correlations—that is, “Why does pain correlate with N1?” is correctly answered, and explained, by saying that “pain ¼ N1.” I do not think this view is correct; for discussion see Kim, Physicalism, chap. 5. In any case, I don’t believe that either Hill or McLaughlin would claim that identities deliver “reductive explanations” of the correlations.


219

“Why is heat identical with mke?” is also a wrongheaded question. If this is a legitimate question requiring an answer—a “correct” answer—then, a “gap” or no “gap,” the identities like “pain ¼ N1” will fail to free us from the burden of explaining psychoneural relations, and the emergentists’ explanatory challenges cannot be stopped. In order to put an end to them, we must assume that identities are absolute terminal points of regressive explanatory challenges “Why p? Because q. Why q? Because r. Why r? . . . ” When we are finally able to come up with an answer in the form of an identity “Because x ¼ y,” that will, it is hoped, stop the why-questions in their tracks. The reason, as the thought runs, is that it makes no sense to ask for an explanation of why x ¼ y—namely that identities are not proper explananda. But is this correct? Prima facie, there seem to be any number of identities for which we can sensibly ask for explanations—and find them if we are clever or lucky. Consider: Michael Jordan ¼ the most valuable player of the Chicago Bulls 32˚ F. ¼ the freezing point of water Black ¼ the color of my true love’s hair It surely makes sense to ask why Michael Jordan is the most valuable player of his team, why 32˚ F. is the freezing point of water, and so on, and receive informative answers that explain the facts in question. However, we also notice that, unlike “Cicero ¼ Tully” and “water ¼ H2O,” these identities don’t seem to be genuine identities—they are easily paraphrased into equivalent predicative statements like “Michael Jordan is a more valuable player than anyone else on the team,” “Water freezes at 32˚ F.,” and “My true love’s hair is black.” It surely makes sense to ask why Jordan is a more valuable player than any of his teammates, why water freezes at 32˚ F., and so on. Further, these identities are all contingent, each with a nonrigid designator flanking the identity sign. In contrast, identities like “heat ¼ mke” and “pain ¼ N1” are taken, in this context, to be necessary truths. What does the contingency or necessity of an identity have to do with the question whether it is a fit object of explanation? If p is a contingent truth, we can always ask the question “What is it about this world that makes it the case that p?”—that is, “Why is this world one in which p is true rather than one in which p is

220


false?” If p is a necessary truth, p is true everywhere and the question “What is it about this world that makes it the case that p?” either receives a putdown answer “Nothing special—p holds everywhere,” or can be charged with having a wrong presupposition, to the effect that there are certain special features of this world, not present in every world, which are responsible for p’s holding here. In either case, the question “why p?”, where p is a necessary truth, does not receive a “correct” answer; rather, the response it elicits dissipates the point of the question.24 What we have just said applies to all necessary truth, not just necessary identities. Is there anything specific to be said contra identities as possible explananda? There is this train of thought: If necessarily x ¼ y, saying that x ¼ y says no more, and no less, than that x ¼ x, or that y ¼ y. Therefore, asking “why x ¼ y?” makes no more sense than asking “why x ¼ x?” One might protest: saying that Cicero ¼ Tully is saying more than that Cicero ¼ Cicero, or that Tully ¼ Tully. We can grant that. “Cicero ¼ Tully” carries information that “Cicero ¼ Cicero” does not. What could this information be? As far as I can tell, it is the fact that the names “Cicero” and “Tully” name the same object. And of course this fact can be a proper explanandum. But “Cicero ¼ Tully” is not a statement about the names “Cicero” and “Tully,” and it does not report a fact other than the fact (if there is a fact here) reported by “Cicero ¼ Cicero.” It is difficult to see any fact here that can be a suitable object of explanation. In contrast, if it is only contingent that x ¼ y (e.g., Benjamin Franklin ¼ the inventor of bifocals), we can no longer say that “x ¼ y” says nothing more than “x ¼ x” or “y ¼ y,” and there can be an explanation of the fact that Benjamin Franklin ¼ the inventor of bifocals. I am not certain exactly what these considerations add up to show; I cannot say I find them, singly or together, wholly compelling or satisfying, though not entirely vacuous or irrelevant either. There undoubtedly is more, probably much more, to be said on the role of identities in explanations, either as elements in an explanans or as possible explananda.25 This is a 24 One pertinent issue here concerns the possibility of explanations of mathematical truths. One could argue, it seems, that proofs, especially constructive proofs, explain why the theorems are true. I have nothing interesting or informative to say about this, but it seems to me that mathematical explanations, if there are such things, must be radically diverse in kind from explanations of natural events, facts, laws, and the like. 25 In Kim, Physicalism, I discuss the role of identities as part of an explanans.


221

large and complex topic in its own right, and what I have said is pretty much all that I have to say. In any case, we can safely conclude that identity reduction does reduce but yields no reductive explanations. Rather, it deals with the explanatory questions of the sort the emergentists raised in a different way: it asserts that there is something seriously defective about the presuppositions of those questions, and that this renders these explanatory requests improper and misguided. If this is right, we have a case in which reduction and reductive explanation do not go hand in hand— identity reduction does reduce but generates no reductive explanations.

IV We now turn to functional reduction, the last of the three models of reduction. Reduction of this form proceeds in three steps. Suppose that property M is to be so reduced: Step 1. Functionalize M by providing a functional definition/characterization of the following form: For x to have M (or to be an M) ¼ for x to have some property or mechanism P such that C(P), where C(P) states a causal specification that P must meet. Step 2. Identify—that is, ascertain—the P, or Ps, that meet causal specification C in the system, or a population of systems, of interest in a given research program. Such Ps are called the “realizers” of M. Step 3. Develop a theory that explains how these realizers of M perform the causal task specified by C in the systems in question.26 Steps 2 and 3 are likely to go concurrently, hand in hand. As a simplified schematic example, consider this: (i) For x to be a gene ¼ for x to have, or be, a mechanism that encodes and transmits genetic information.

26 A more detailed description of functional reduction should say something about the language in which causal specifications at step 2 are to be formulated, and also the predicates/concepts in terms of which realizers are to be identified and described at step 3.

222


(ii) DNA molecules are the mechanisms that perform the task of encoding and transmitting genetic information (in biological organisms of interest to us). (iii) Molecular biology explains how DNA molecules perform this task. When all this is in, we may say that genes have been reduced—functionally reduced—to DNA molecules in the population of interest to us, presumably various sorts of terrestrial organisms. Philosophically, step 1 is the critical step. Steps 2 and 3 are obviously matters of scientific research. The question whether M can be defined functionally is a conceptual issue, though not necessarily a wholly a priori or philosophical one—in particular, just how a functional definition for a given M is to be formulated is likely to have to be informed by empirical findings and theoretical needs. The philosophical interest and importance of the issue of functionalizability stems from the fact that once M has been functionalized (or shown to be functionalizable), we know that M is functionally reducible—whether, and how, M is going to be reduced, that is, whether and how steps 2 and 3 will be carried out, is an issue of scientific research. Given that M is a functional property, if something has M, it logically follows that it has one or another of the realizers of M, though we may not have a scientifically perspicacious lower-level description of the realizer involved. From a metaphysical point of view, what matters is reducibility, not actual reduction. Let us first take up the issue of reductive explanation. Suppose that pain has been given the following functional definition—along the lines urged by the early functionalists: For x to be in pain ¼ for x to be in some state P such that tissue damage is apt to cause P and P is apt to cause winces, groans, and aversive behavior. (This is for illustrative purposes only; I am not suggesting that pain can be functionally defined or reduced—on the contrary, I believe pain is not functionally definable.) Supposing neural state N1 to be the realizer of pain in humans, let us consider explanatory questions like these: Why is Jones in pain at t? Why did Jones experience pain when he stepped on a thumbtack? Why is neural state N1 invariably accompanied by pain (in humans)?


223

Can we formulate explanations as responses to these questions, explanations in terms of neural laws and neural facts about Jones? I believe the following is a possible neural explanation of why Jones is in pain at t: Jones is in neural state N1 at t. Tissue damage is apt to cause N1 in Jones, and N1 is apt to cause Jones to wince, groan, and engage in aversive behavior. Being in pain ¼ being in a state apt to be caused by tissue damage and apt for causing winces, groans, and aversive behavior. Therefore, Jones is in pain at t. The argument is clearly valid, and it derives a pain fact from neural/physical facts alone. If anything could count as closing the explanatory gap between pain and its neural correlate, this explanation should. Please note that the third sentence is a definition; it is not a “fact” about pain’s correlation with any neural/physical fact; if it is about any fact, it is about a semantic/ conceptual fact about the term “pain.” Definitions don’t count as premises in a proof; they are free. Notice one more thing: at the second sentence, the argument invokes a nomological fact. It is an empirical lawful regularity that this particular neural state, N1, has the specified causal role in Jones and creatures like him. It has the form of a Hempelian deductive-nomological explanation. To see the reductive credentials of this explanation, note that the premises are exclusively about neural/physical facts, and that the explanation consists in the derivation of a psychological fact from neural/physical facts. We now turn to formulating a reductive explanation in answer to the second explanatory question “Why was Jones in pain when he stepped on a thumbtack?”: Jones stepped on a thumbtack. This caused Jones to suffer tissue damage. This in turn caused neural state N1. N1 is apt to be caused in Jones (and like individuals) by tissue damage and is apt to cause Jones to wince, groan, etc. Being in pain ¼ being in a state apt to be caused by tissue damage and apt for causing winces, groans, etc. Therefore, Jones was in pain.

224


We may assume that the second line is derivable from the first via physiology, and that the third line is similarly derivable from the second; or they can be taken as independent premises. I believe the argument again is a plausible reductive explanation of why stepping on a thumbtack in the way Jones did caused Jones pain. Again, a pain fact is derived from premises concerning neural/physical facts alone. An explanation can be formulated, on a similar pattern, in reply to the third explanatory request above. Plausibly, then, functional reduction delivers reductive explanations. But does it reduce? Take a given instance of a functionally reduced property, M—say, x’s having M at t. If x has M at t, it follows that x instantiates at t a realizer of M, say Pk. I believe that the identification of this instance of M, x’s having M at t, with the instance of the realizer involved on this occasion, x’s having Pk at t, is all but compelling. This for at least two reasons. First, the definition of M says that for x to have M at t is for x to instantiate, at t, a property that plays causal role C (for x); and Pk is exactly the property that x has at t which plays causal role C. There is no further fact about x’s having M at t than the fact that x has Pk at t. Second, there are causal considerations: What are the causal powers of this instance of M? The only reasonable answer seems to be that they are exactly the causal powers of this instance of Pk. (This is what I have called elsewhere the causal inheritance principle.27) These considerations point to a form of token identity thesis: this instance of M is identical with the instance of the realizer of M on this occasion. In general, then, if mental properties are functionally reducible, instances of mental properties are identical with the instances of their physical realizers. That, at any rate, is the proposal. We may call it token reductionism. Let us first deal with a possible objection. It might be argued that token reductionism of this form doesn’t sit well with the earlier claim that functional reduction yields reductive explanation. Consider the first of the two displayed derivations above. As may be recalled, the claim was that it is an explanation of why Jones has pain at a time on the basis of why Jones is in neural state N1 at that time. In this example, N1 is the realizer of pain on that occasion, so on the token reduction thesis, Jones’ having pain at t ¼ Jones’ in neural state N1 at t. If so, this would turn supposed the explanation into 27 Kim, “Multiple Realization and the Metaphysics of Reduction” (1992), reprinted in Kim, Supervenience and Mind (Cambridge: Cambridge University Press, 1993).


225

one in which something is explained in terms of itself—a self-explanation. But the very idea of explaining something by simply re-invoking it under another name seems foolish and incoherent.28 There may be a general lesson here: there is a prima facie incompatibility between reducing something x to y and reductively explaining x in terms of y—a point noted at the outset of this paper. Once x has been reduced to y, x ¼ y, and any attempt to explain x in terms of y would only result in an absurd self-explanation. Since the derivations that were claimed to be reductive explanations do seem to be explanatory, one might argue, the upshot is that we must reject the identity claim for pain-instances and N1-instances and, more generally, reject token reductionism. This is the objection to be considered. I believe that we can make good sense of the explanatory claim in a way that is consistent with the token identity thesis. Let us begin by reminding ourselves that the relationship between x’s having pain at t and x’s being in neural state N1 at t is contingent, not necessary. This stems from the contingency of the realization relation: that N1 is a realizer of pain in x and like systems is contingent, not necessary. N1 realizes pain in this population in virtue of satisfying the causal specification definitive of pain. That these particular causal relations hold for state N1 in systems like x is a contingent fact, a fact that depends on what laws prevail in our world. In worlds in which different laws hold (we are assuming that laws, or causal laws, are contingent), different causal relations will hold, and N1 might no longer meet the causal specification associated with pain. In some such worlds, N1 will fail to realize pain. Thus, the identity “x’s having pain at t ¼ x’s having N1 at t” is contingent. In some worlds where x has pain at t, it might be that x’s having pain at t ¼ x’s having Q at t (Q 6¼ N1), where Q realizes pain in x in that world. The contingency of these token identities is the key to seeing how “self-explanations” apparently involved in our reductive explanation can be perfectly explanatory. By providing a reductive explanation of the sort displayed earlier, we show how the description “x is in pain at t” applies to something as a causal-nomological consequence of the fact that the description “x is in neural state N1 at t” applies to it. Or think of it this way: the identity “x’s being in pain at t ¼ x’s having N1 at t” is contingent, so it makes sense to ask: What is it about this world that makes it so? Why is it that in this world this identity holds whereas in certain other 28 Compare: Why is Cicero wise? Because Tully is wise and Cicero ¼ Tully!

226


worlds it does not? The answer: it is because in this world these laws hold, enabling N1 to fill the causal role that defines pain, whereas in those other worlds different laws hold and they do not confer similar causal powers on N1. This is what the suggested reductive explanation does; it invokes laws holding in this world and shows how N1 possesses the causal powers required to realize M. This certainly is informative and there seems no reason to doubt its explanatoriness. We now turn to another question: Might the contingency of the identity “x’s having M at t ¼ x’s having Pk at t” undermine its reductive import? I don’t see why it should. The reductive claim is this: x’s having M at t reduces, in this world, to its having Pk at t—that is, x’s having M at t is “nothing over and above” its having Pk at t. In another world, x’s having M may reduce to x’s having Pj (k 6¼ j), and so on. Moreover, given that M is a functional property, in every world in which something has M at t, there is a realizer of M such that the object’s having M at t reduces to its having that realizer at t. So then, there is no world in which something’s having M is “over and above” its having a realizer of M. That is, in no world are there instances of M that are unidentified with instances of M’s realizers. Once you have all actual and possible instances of M’s realizers, you’ve got all instances of M, actual and possible; M-instances add nothing ontologically to the instances of its realizers. This seems reduction enough for all instances, or tokens, of M, actual or possible. The contingency of the token identities, therefore, appears to be consistent with the efficacy of these identities as vehicles of reduction; necessary identities are not needed for reduction. So token reductionism takes care of pain instances. But what of pain itself? That is, once a functional reduction of pain has been achieved, what happens to the type, or property or kind, being in pain? Is this property reduced and if so, to what? If not, aren’t we still stuck with an unreduced, and irreducible, nonphysical property? These are the questions we must now face.29 Let P1, P2, . . . be all the realizers of M at this world; this means that for something to have M in this world, it must have one of the P’s, and if 29 I have discussed this question elsewhere, in particular in Kim, Mind in a Physical World (Cambridge, Mass.: MIT Press, 1998). What I am going to say here is similar to what I have said before but not exactly identical.


227

something has one of the P’s in this world, it has M. Consider the (possibly infinite) disjunction P1 v P2 v . . . (or UP for short). Then anything has M at this world if and only if it has UP. So can we say that M ¼ UP? Is this a way of physically reducing M? In considering this question, we must first recognize the proposed type identity as a contingent truth. As in the case of token identities, the identity holds at this world, and worlds like this one in respect of causal laws, but in worlds in which different laws and causal relations obtain, M may have a different, perhaps a wholly disjoint, set of realizers, say Q1, Q2, . . . , and M ¼ UQ at those worlds. On the present proposal, then, “M,” or “having M,” is not a rigid designator; it refers to different properties in different worlds—to UP in this one, to UQ in certain other worlds, to UR in still others, and so on. To give a property designator a functional definition in terms of a causal specification is to make it nonrigid. M so defined is no longer a unitary property that can be tracked from world to world; “M” can designate one property in this world and a different property in another, and perhaps nothing at all in some worlds. So should we go with M ¼ UP? Accepting this identity would commit us to the token identity “x’s having M at t ¼ x’s having UP at t.” This contrasts with our earlier recommended token identity “x’s having M at t ¼ x’s having Pk at t.” Thus, the identification of M with the disjunction of its realizers at a world yields a competing token identity thesis, an alternative form of token reductionism. Which of these two token identity claims is preferable? We should remember that UP is, or can be, an extremely heterogeneous and unmanageably huge disjunction; this makes it unclear what causal-nomological import UP can have. Consider two properties, each with a distinctive set of causal powers, say a temperature of 100˚ C. and a mass of 1 kilogram. What causal powers should we associate with the disjunctive property of having a temperature of 100˚ C. or having a mass of 1 kilogram? What causal powers does an object have in virtue of having this disjunctive property? It isn’t clear what we should say. All we can say seems to be that if an object has this disjunctive property—that is, if it either has a temperature of 100˚ C. or has a mass of 1 kilogram—then it either has the causal powers associated with the temperature or those associated with the mass. The last “or” in the preceding sentence is sentence disjunction, not a special operator designating some kind of “disjunction” operation on properties or causal powers. That is, to say that something has causal powers C1 or causal powers C2 is to say only that either it has C1 or it has C2; there

228


is no need to posit a disjunction of C1 and C2, which one might denote as [C1 v C2], and say that the thing has this disjunctive causal power [C1 v C2]. If such disjunctions are to be posited, we will need an explanation of what the disjunctions stand for in terms of what each of their disjuncts stands for. But such an explanation is exactly something we don’t have. In consequence, we are without an understanding of what causal powers are to be associated with disjunctive properties, or with their instances. What this means is that if we identify x’s having M with x’s having UP, we are putting the causal status of x’s having M in jeopardy—or rather in a limbo; it seems that we can say nothing clear and motivated about what causal powers this token event should be credited with. In contrast, Pk is, by assumption, a specific causal-nomic property, and identifying x’s having M with its having Pk gives the M-instance robust causal reality and a specific causal profile. It seems to me that this is a sufficient reason for rejecting the proposal that we identify M with UP. One might suggest that if something has the disjunctive property [P1 v P2] in virtue of having P1 (that is, the truthmaker of “x has [P1 v P2 ]” is “x has P1”), we identify the causal power of x’s having [P1 v P2] with the causal power of x’s having P1. And likewise if x has [P1 v P2] in virtue of having P2. So, in the case of M, supposing that x has UP in virtue of having Pk, the suggestion is that we identify the causal powers of this instance of M with the causal powers associated with Pk. And if another thing y has UP in virtue of having Pj, y’s having UP has the causal powers of y’s having Pj; and so on. This means that no distinctive set of causal powers is associated with UP, and this disjunctive property drops out of the causal picture in favor of its disjuncts. I think we might as well be up front and identify x’s having M with x’s having Pk, y’s having M with y’s having Pj, and so on, and entirely bypass UP and its ilk. For these reasons, we may set aside the possibility of identifying a functionally reduced property with the disjunction of its realizers. What then? I believe there are two further options to consider: what we may call functional property realism and functional property conceptualism. Let us begin with functional property realism (it corresponds to what some have called role functionalism). On this approach, if M is a functional property, with the kind of functional characterization as indicated earlier, M is a robust property in its own right with a clear identity as a unitary property from world to world. Take pain: being in pain is the property of being in some state with


229

such-and-such inputs and such-and-such outputs. This property has diverse realizers from world to world, from species to species, from an individual at one time to the same individual at another time, and so on. But it is a single, unitary property with its own integrity as a property; it’s just that this one property has different realizers along various dimensions. The important thing to remember is that, on this view, the functional property only “has” realizers, and that it remains ontologically distinct from them, individually. Whence the name functional property realism. Those who hold this view will reject the claim that functional reduction in our sense gives us reduction, since there is no physical property with which M can be identified. One mark of this is the fact that, on this view, “M” is a rigid designator which tracks the same nonphysical property world to world; in this way, the view contrasts with the disjunction approach (M ¼ UP) earlier considered which makes “M” nonrigid. The position is strongly antireductionist with regard to psychological properties; its proponents include those, like the original functionalists such as Putnam and Fodor, who take functionalism as an essentially antireductionist and antiphysicalist view.30 I believe there are two plausible arguments against functional property realism, both of them based on causal considerations. The first argument goes like this: token reductionism, as we earlier urged, is highly plausible, but if token reductionism is true, functional property realism has little to recommend itself. We have already argued why token reductionism, of the sort we urged, should be accepted. So assume token reductionism. Functional property realism asserts that a functionally characterized property M is a real property in its own right. If so, it must represent a specific set of causal powers; in each world, M must confer on each and every object that has M some specific and uniform set of causal powers. (On some views, these causal powers go toward defining M; that is, they are constitutive of the very identity of M as a property.) If, as token reductionism claims, all actual and possible instances of M are instances of M’s realizers, M cannot have causal powers that go beyond the causal powers of its diverse realizers; there are no new causal powers that M brings to the world other than those contributed by its realizers. This seriously undermines the claim that M is a 30 Don Ross and David Spurett, in “What to Say to a Skeptical Metaphysician,” are recent advocates of this version of functionalism; there are others, including Ned Block (I believe; see Block, “Antireductionism Slaps Back”). For more on role functionalism, see Brian McLaughlin, “Is Role-Functionalism Committed to Epiphenomenalism?”, Journal of Consciousness Studies 13 (2006): 39–66.

230


genuine property in its own right, distinct from its physical realizers. At best, M’s causal powers are going to be very diverse and heterogeneous—as diverse and heterogeneous as the causal powers of its many diverse realizers—whereas we would expect genuine properties to show at least a certain degree of nomological and causal unity.31 M cannot be the kind of property in terms of which productive scientific theorizing could be conducted; M’s causal profile is too heterogeneous and fragmented for it to be a projectible nomic property, the kind of property that can be considered a causally efficacious property on its own.32 When M is invoked in a causal claim or explanation, this should be understood in terms of a tacit reference to a realizer of M which is doing the actual causal work.33 Invoking M rather than one of its realizers masks either our ignorance of the details of the situation or our laziness. Psychological properties on this view seem to form a badly gerrymandered taxonomy overlaid on the underlying domain of physical/biological properties. The second causal argument against functional property realism is the familiar exclusion argument: if x’s having M at t 6¼ x’s having Pk at t, where Pk is M’s realizer on this occasion, the M-instance’s causal role is threatened with preemption by the Pk-instance, or else we would have a case of spurious causal overdetermination.34 To insist on M as a real property only to have its causal status undermined and usurped by its realizers, anywhere and everywhere it is instantiated, should strike us as an empty and futile gesture. Since the basic considerations on the exclusion argument are well known, there is no need to rehearse them here. I believe that all these considerations convince us that functional property realism is highly problematic. We now turn to functional property conceptualism, our final option on the status of functionally reduced properties. The proposal is that we should

31 I realize that I am here touching on some general issues about properties, causality, realization, and other related topics. They obviously require more extended discussion and consideration than what I can do here. 32 Kim, “Multiple Realization and the Metaphysics of Reduction”; for replies see Block, “Antireductionism Slaps Back” and Fodor, “Special Sciences.” 33 That is, when we say “x’s having M caused E,” we should be understood as saying something like “there is a realizer P of M such that x had M on this occasion in virtue of having P, and x’s having P caused E.” 34 Kim, Mind in a Physical World; see also McLaughlin, “Is Role-Functionalism Committed to Epiphenomenalism?” for recent discussion.


231

take the causal and nomological disunity of the functionalized M seriously and abandon M as a genuine, unitary property. Ned Block once asked what all pains have in common in virtue of which they are instances of pain.35 If pain is a functional property definable in terms of physical inputs and behavioral outputs and realized in many diverse neural/biological/physical structures, then what all pain instances have in common is merely the fact that they all fall under the concept of pain as given by its functional characterization—no more and no less. That is to say, pains are pains because they conform to the definition of pain, not because they all share some hidden essence, like C-fiber stimulation or a pain quale. So there is the concept of pain, a concept given by its functional definition, but no property of pain, or being pain, that all pain instances have in common. There simply is no property in the world with causal and nomological unity required of genuine properties which answers to our concept of pain, and which is shared by all instances of pain (pains in humans, pains in reptiles, pains in Martians, and the rest). More generally, psychological functionalism may be characterized as the view that psychological kinds have no real essences, only nominal essences. If you feel that this doesn’t do full justice to pain, and that there obviously is a genuine property there—well, all pains hurt!—you must reject the functionalization of pain. Remember our question “What happens to pain, as a kind or property, if it has been functionally reduced?” Functional property conceptualism may sound like a form of eliminativism, and perhaps rightly so. M as a property is gone; it has been eliminated. It remains true that, in Smart’s idiom, M as a property is nothing “over and above” its realizing properties for the trivial reason that M is nothing. If mental properties are functionally reduced, we may well have to live with mental eliminativism and irrealism.36 But we should note a few things. First, saying that M is not a genuine property is no worse, if no better, than saying that being a table is not a genuine property. There may be no such property as being a table, but that doesn’t mean there are no tables. Similarly M may not be a property but that need not stop us from recognizing instances of M. Second, M as a concept stays, and individual instances falling under M are perfectly legitimate entities with causal-explanatory efficacy. The situation 35 Block, “Introduction: What is Functionalism?” in Readings in Philosophy of Psychology, vol. 1, ed. Block (Cambridge, Mass.: Harvard University Press, 1980). 36 Terry Hogan, I believe, was the first to remind me of this implication of functional reduction.

232


with the mental eliminativism of the sort advocated by Paul Churchland37 is quite different. As I take it, Churchland’s eliminative materialism discards psychological concepts—the concepts of propositional attitudes, like those of belief, desire, and intention—as well as psychological states and properties. The concept of belief suffers the same fate that befell the concepts of phlogiston and caloric fluid. We should remember that functionalism about the mind as originally formulated by Hilary Putnam in 1967 was a thesis not about mental states or properties but about psychological expressions and concepts. Putnam’s seminal 1967 paper, which first introduced psychological functionalism, carried the title “Psychological Predicates,” which later took an ontological turn and became “The Nature of Mental States.” It shouldn’t surprise us that functional reduction of the mental, at least on one reading, ends up not as a reduction of mental properties but as a thesis about mental concepts. We have reviewed three ways of dealing with psychological properties in functional reduction, the disjunction approach, functional property realism, and functional property conceptualism. I believe that reasons for rejecting the first are quite compelling, and that there are nearly as compelling reasons for rejecting the second. The third smacks of psychological antirealism and, for that reason, is not very appealing. But it seems to me that it is the best of the unappetizing lot, the only one that is free of major philosophical difficulties. I am willing to admit that this whole scene may well be worth revisiting and reconsidering.

V To conclude, I have argued three main points. First, bridge-law reductions deliver neither reductions nor reductive explanations. The source of the trouble is the use of bridge laws as unexplained, unreduced auxiliary premises of reductive derivations. Second, identity reductions, in which bridge laws are replaced by identities, give us reductions but no reductive explanations. Rather, such reductions eliminate the need for—indeed, the possibility of—such explanations. Instead of “closing” the explanatory gap, reductions of this type show that no such gaps exist in the first place. That

37 “Eliminative Materialism and the Propositional Attitudes,” Journal of Philosophy 78 (1981): 67–90.


233

is a perfectly effective way of dealing with the supposed explanatory gap problem. Finally, I argued that functional reductions give us reductive explanations of the sort we expect and help close the gap, and that it arguably gives us reductions as well, though there remains room for further debate as to the exact nature of the reductions involved. One thing that our discussion reveals is something not unexpected: as far as reduction goes, nothing beats identities. That appropriate identities achieve reduction is intuitively obvious and beyond any philosophical second thoughts. That, unsurprisingly, is the chief attraction of the psychoneural identity theory as a form of reductionist physicalism. Unsurprisingly again, the main issue about the reductive significance of functional reduction comes down to the question whether and how functional reduction can yield appropriate identities, for psychological properties/kinds and their instances. All this goes toward reconfirming the point, perhaps an obvious one when we think about it, that identities are absolutely central to reduction.38 Finally, we should note that in this paper we have not touched on the important question whether or not reductionism of either form can be plausibly held—that is, the question whether and how we may earn our entitlement to psychoneural identities or to the functional definability of mental kinds.39 38 Unless, that is, the reduction being promoted is wholly eliminative. Carl Gillet defends the view that compositional relations, rather than identities, are fundamental to much scientific reduction in “Understanding the New Reductionism: The Metaphysics of Science and Compositional Reduction,” Journal of Philosophy 104 (2007): 193–216. 39 For further discussion of this issue see Block and Stalnaker, “Conceptual Analysis”; Kim, Physicalism; and Block, “Functional Reduction,” in Supervenience in Mind, ed. Terence Horgan, Marcelo Sabates, and David Sosa (Cambridge, Mass.: MIT Press, forthcoming).

11 Can Supervenience and “Non-Strict” Laws Save Anomalous Monism? In “Thinking Causes,” Donald Davidson proposes to defend his doctrine of “anomalous monism” (AM) against “misunderstandings and misrepresentations” of his critics, myself included, who have called attention to its epiphenomenalist tendencies.1 Although part of what I am going to say will be in direct reply to Davidson’s specific points, I believe that several points of more general interest will emerge.

I. Have the Critics of AM (or AM þ P ) Charged it with Inconsistency? AM is the claim that, although mental properties are irreducible to physical ones, mental events are in fact physical events; and P is the conjunction of “the three premisses,” as Davidson calls them, of AM: (1) mental events cause, and are caused by, physical events; (2) causally related events instantiate “strict” laws; and (3) there are no “strict” psychophysical laws. Davidson quotes me as having said that “under Davidson’s anomalous monism, mentality does no causal work,”2 and he apparently takes this remark to contradict (1) and hence AM þ P. What he says is this: 1 “Thinking Causes,” in John Heil and Alfred Mele, Mental Causation (Oxford: Oxford University Press, 1993). 2 J. Kim, “The Myth of Nonreductive Materialism,” Proceedings of the American Philosophical Association 63 (1989), p.35.

CAN ANOMALOUS MONISM BE SAVED?

235

If “mentality does no causal work” means that mental events do not enter into causal relations, the first premiss of AM is false, for it says mental events cause, and are caused by, physical events. This is not enough to prove AM itself inconsistent, but it certainly would show the three premisses of AM inconsistent with one another.3 I don’t dispute any of this. What is curious, though, is that Davidson does not defend, or even explicitly affirm, the reading of “mentality does no causal work” suggested in the first sentence of this quotation. Thus it is puzzling why he is so certain that I have characterized AM þ P as inconsistent; the paragraph in which my offending sentence occurs makes it abundantly clear, I dare say, that by “mentality” I was referring to mental properties, not individual mental events.4 In the context of AM þ P, the claim “Mental events cause physical events” only comes to the assertion, which is not disputed by his commentators, that events with some mental property or other are causes of events with some physical property or other. The difficulty that has been voiced by the many critics whose names Davidson cites, with an impressive if unsurprising unanimity, is precisely that the truth of this assertion does not ensure the causal efficacy of mental properties (compare: “These orange pills will relieve your headache”).

II. Have the Critics of AM Made an Error in Claiming that AM þ P is a Form of Epiphenomenalism? It must be admitted that Davidson’s commentators have not always been careful to distinguish between the following two claims: (1) AM þ P entails the causal inertness of mental properties, and (2) AM þ P fails to provide mental properties with a causal role. According to Davidson, (1) is false;

3 “Thinking Causes,” p. 6. 4 The sentence that immediately precedes the one in question reads as follows: “For anomalous monism entails this: the very same network of causal relations would obtain in Davidson’s world if you were to redistribute mental properties over its events any way you like; you would not disturb a single causal relation if you randomly and arbitrarily reassigned mental properties to events, or even removed mentality entirely from the world” (Kim, “The Myth of Nonreductive Materialism,” pp. 34–35).

236


and in this he is arguably right.5 However, this does not necessarily absolve AM þ P of the charge of epiphenomenalism; for if something that purports to be a theory of mental causation assigns no causal role to mental properties—if it has nothing to say about the causal powers of mental properties while saying plenty about those of physical properties—the theory can, it seems to me, reasonably be said to be epiphenomenalistic with regard to mental properties. Plainly (2) is true, and has never been seriously disputed; and the defenders of AM have focused, by and large, on extending AM by adding a positive account of the causal efficacy of mental properties.6 In fact, that is Davidson’s own approach in ‘Thinking Causes’: he wants to supplement AM þ P with supervenience (S), and perhaps also with “non-strict laws,” to restore causal efficacy to mental properties, tacitly acknowledging that within the framework of AM þ P mental properties have no causal role to play.

III. Have the Critics Tried to Turn the Causal Relation into a Multi-Termed, DescriptionDependent, Intensional Relation? Throughout “Thinking Causes,” Davidson complains that his critics have tried to turn the binary relation of causation, “c causes e,” into a multitermed (that is, more than binary), possibly non-extensional, relation by employing such expressions as “c qua P causes e qua M,” “c under description D causes e under description D*,” etc.7 He is anxious to defend causation as an extensional binary relation whose relata are concrete events (“no matter how described”). But none of this has much to do with the main issue on hand, and getting rid of these admittedly inelegant locutions will not make it

5 I believe Brian McLaughlin was the first to argue this point; see his “Type Epiphenomenalism, Type Dualism, and the Causal Priority of the Physical,” Philosophical Perspectives 3 (1989): 109–135. 6 e.g. ibid.; Ernest LePore and Barry Loewer, “Mind Matters,” Journal of Philosophy 84 (1987): 630–640. See also T. Horgan, “Mental Quausation,” Philosophical Perspectives 3 (1989): 47–76; C. and G. MacDonald, “Mental Causes and Explanation of Action,” Philosophical Quarterly 36 (1986): 145–158. The strategies that have been tried include the use of non-strict laws and certain causal counterfactuals. 7 Davidson includes me among those who have used such expressions, on the basis of my writing “on anomalous monism, events are causes only as they instantiate laws.” The culprit in Davidson’s light is the word “as”; I used it in the sense of “because” or “since,” but Davidson apparently takes it in the sense of “qua” or “in the role of,” which is a bit curious, given that “as” in my sentence functions as a grammatical construction, not a preposition.


237

go away. The issue has always been the causal efficacy of properties of events—no matter how they, the events or the properties, are described. What the critics have argued is perfectly consistent with causation itself being a two-termed extensional relation over concrete events; their point is that such a relation isn’t enough: we also need a way of talking about the causal role of properties, the role of properties of events in generating, or grounding, these twotermed causal relations between concrete events. To talk about the role of properties in causation we don’t need to introduce the “qua” locution or any other multi-termed causal relation, although I see nothing in principle objectionable about them; all that is necessary is the recognition that it makes sense to ask questions of the form “What is it about events c and e that makes it the case that c is a cause of e?” and that we are able sometimes to answer them, intelligibly and informatively, by saying something like “Because c is an event of kind F and e is one of kind G (and, you may add if you favor a nomic conception of causality, there is a law of an appropriate form connecting F-events with G-events).” This is only to acknowledge that the causal relation obtains between a pair of events because they are events of certain kinds, or have certain properties. How could anyone refuse to acknowledge this—unless, that is, he believed that causal relations were brute facts about events, having nothing to do with the kinds of events that they are? In fact, Davidson himself acknowledges in the end that it makes sense to discuss the causal relevance of properties; for, after all, he offers an account of it, based on supervenience and non-strict laws.

IV. Can You Have Psychophysical Supervenience without Psychophysical Laws? Well, that depends on what sort of supervenience you have in mind.8 Davidson says that I made an error about the logic of supervenience in closely associating supervenience with laws; according to him, “supervenience does not imply the existence of psychophysical laws,” because “although supervenience entails that any change in a mental property p of a particular event e will be accompanied by a change in the physical properties of e, it does not entail that a change in p in other events will be 8 See J. Kim, “Concepts of Supervenience,” Philosophy and Phenomenological Research 45 (1984): 153–176 and “Supervenience as a Philosophical Concept,” Metaphilosophy 21 (1990): 1–27.

238


accompanied by an identical change in the physical properties of those other events”.9 So far so good. But he goes on to add, “only the latter entailment would conflict with AM þ P.” Here, Davidson is plainly looking for the wrong kind of law; when the question of law is discussed in connection with supervenience, it almost always concerns laws from the base (or subvenient) properties to the supervenient properties (thus, physical-to-mental laws), not laws going in the opposite direction (mental-to-physical laws). Thus, assume that two systems are in the same total physical state (at the same or different times); psychophysical supervenience implies this: if the systems change in some identical physical respect Q, they must change in an identical psychological respect M. In fact, mind–body supervenience (and supervenience in general) can be explained in terms of the existence of generalizations from the subvenient to the supervenient, thus: whenever anything has mental property M there is some physical property Q such that it has Q and everything that has Q has M. On certain plausible assumptions concerning property compositions, this formulation is demonstrably equivalent to the usual definition of supervenience in terms of indiscernibility in respect of supervenient and base properties.10 There is of course a question as to whether the kind of supervenience Davidson says he has in mind, which appears to be equivalent to what I have called “weak supervenience,” can impart to these generalizations an appropriate nomic force; but that isn’t a question Davidson raises, and there is in any case a doubt as to whether weak supervenience can provide the kind of dependency relation that most philosophers want to associate with supervenience.11

V. Does AM þ P þ S (that is, Davidsonian Supervenience) Provide a Satisfactory Account of the Causal Relevance of Mental Properties? On this question, Ernest Sosa makes a number of cogent points in his reply, “Davidson’s Thinking Causes,”12 with which I am by and large in 9 “Thinking Causes,” p. 7. 10 See Kim, “Concepts of Supervenience,” and “‘Strong’ and ‘Global’ Supervenience Revisited,” Philosophy and Phenomenological Research 48 (1987): 315–326. 11 See Kim, “Concepts of Supervenience.” 12 In Heil and Mele (eds.), Mental Causation, chap. 4.


239

agreement. So I will make just one point.13 “Causal relevance” may be one thing; “causal efficacy” another. An epiphenomenalist may argue, mimicking Davidson, that on his view mental properties are indeed causally relevant, since, according to his doctrine, what mental properties an event has makes a difference to what physical properties it has, and physical properties are causally efficacious. But that doesn’t mean that he contradicts himself in refusing to allow causal efficacy to mental properties. If this is right, supervenience can at best show that mental properties are causally relevant, not that they are causally efficacious. And it would seem that to sustain the kind of position he has argued for in “Actions, Reasons, and Causes” (1963),14 Davidson may very well need causal efficacy, not just causal relevance, for mental properties. Mere causal relevance seems too weak to support the causal-explanatory “because” in rationalizing explanations. And it seems to me that most philosophers who believe in mental causation would want efficacy, not just relevance.

VI. What Then of AM þ P þ NS (‘Non-Strict’ Psychophysical Laws)? From the text of “Thinking Causes,” I am not certain that Davidson wants to embrace non-strict psychophysical laws to account for the causal efficacy of mental properties, although that is the impression one gets. In any case, I think there are some serious difficulties with this approach for anyone who accepts AM þ P. Whether NS can comfortably fit in with AM þ P is a question that has to wait until we are in possession of a clearer account of just what the “non-strictness” of non-strict laws consists in, or just what the much-bandied phrase “ceteris paribus” means when it qualifies a law. Davidson says that his position is consistent with Fodor’s defence of mental causal efficacy based on non-strict laws hedged by “ceteris paribus” clauses. 13 I have myself given an account of mental causation on the basis of supervenience; see e.g. Kim, “Causality, Identity, and Supervenience in the Mind-Body Problem,” Midwest Studies in Philosophy 4 (1979): 31–49 and “Epiphenomenal and Supervenience Causation,” Midwest Studies in Philosophy 9 (1984): 257–270. My account is based on a supervenience relation stronger than Davidson’s. I am inclined to think, however, that even this stronger supervenience relation may not be strong enough for a fully adequate account of mental causation. 14 Journal of Philosophy 60: 685–700.

240


But it would be ill-advised for the anomalous monist to buy into Fodor’s notion of “ceteris paribus law.” For, according to Fodor, such a law has something like this form, “There exist conditions C1, . . . , Cm such that when they are satisfied, F-events cause G-events”; and when the Ci’s have been identified, that will give us a strict law of the form “Under C1*, . . . , Cm*, F-events cause G-events,” where each Ci* is some value of the variable Ci that satisfies the open inner sentence15. Thus, on this account, a non-strict law is simply a strict law with some of its antecedent conditions existentially quantified. But that means that where there is a non-strict psychophysical law, there must be a strict psychophysical law waiting to be discovered. I think it obvious that the anomalous monist must reject this notion of non-strict law. Moreover, it seems to me that, however the non-strictness of non-strict laws is explained, non-strict laws are laws and must carry an appropriate nomological force; given this, it isn’t obvious that Davidson’s fundamental argument against psychophysical laws can allow even non-strict laws between the mental and the physical. As I understand it,16 the gist of the argument is something like this: the mental domain and the physical domain are each governed by their own special synthetic a priori constitutive constraints (principles of rationality, in the case of the mental), and the existence of psychophysical laws with their strong nomic force would ultimately bring these two sets of constitutive principles into conflict or at least jurisdictional disputes. Hence, if each domain is to retain its own integrity, there cannot be laws connecting them. It isn’t clear why this argument, if it succeeds in banning strict psychophysical laws, doesn’t banish non-strict ones as well; at least, an explanation is called for. I have always thought that the power of the Davidsonian argument for mental anomalism is seen in the fact that, if it works at all, it should work against laws of all kinds—for example, statistical laws as well as deterministic ones (after all, the only strict laws we have may be statistical). Remember: non-strict laws, whatever they are, are supposed to be laws!

15 J. Fodor, “Making Mind Matter More,” Philosophical Topics 17 (1989): 75–76. 16 Kim, “Psychophysical Laws,” in Ernest LePore and Brian McLaughlin (eds.), Actions and Events: Perspectives on the Philosophy of Donal Davidson (Oxford: Blackwell, 1985).


241

VII. Are There Other Reasons for Being Wary of NS (‘Non-Strict’ Laws) if You Are an Anomalous Monist? Yes, there are. First, if you accept non-strict laws as nomological grounds of causal relations, you will need a convincing rationale for retaining Davidson’s strict law requirement on causation. It can be seen, in fact, that having laws of both kinds ground causal relations opens up a serious new problem, ‘the problem of exclusion.’17 Suppose a mental event, m, causes an event e (which can be either mental or physical); m, as a mental event, must have some mental property, M, and let’s assume that M, in virtue of a non-strict psychophysical law relating it to some physical property of e, is causally efficacious in m’s causation of e. But, given the strict law requirement, m must also have a certain basic physical property P which is connected, by a strict law, to some property (presumably, another basic physical property) of e, and this fact grounds the causal relation between m and e. Thus, m turns out to have two properties each of which is causally efficacious in m’s causation of e, and, on AM, M and P are irreducibly distinct. We now face this question: given that the causal relation from m to e is grounded in the basic physical properties of m and e and a strict law relating them, what causal work is there for M to do?18 M’s precise causal role in this picture— exactly what contribution M makes in the causation of e—is in need of an explanation. There are various moves one can make at this point, but the problem is there, especially for the adherents of AM. The exclusion problem is a general problem with mental causation, something most of us have to contend with. There is, however, a further specific problem with NS that Davidson and friends of AM seem not to have recognized. It is this: NS may put antireductionism in serious jeopardy. One can still hold on to Davidson’s claim that psychology is not reducible, by strict law, to some underlying physical theory. But why insist on reduction by strict laws only? What’s wrong with non-strict psychophysical

17 See Kim, “Mechanism, Purpose, and Explanatory Exclusion,” Philosophical Perspectives 3 (1989): 77–108. 18 LePore and Loewer raise a similar difficulty in “Mind Matters” with respect to Fodor’s account of mental causation in terms of ceteris paribus psychological laws. There is a brief discussion of this issue in “Thinking Causes.”

242


laws as “bridge” laws? This is not an idle question; nor is it merely a verbal issue. For there seems to be a general consensus, among those who speak of the “strictness” of laws, that there are no strict laws outside basic physics, and Davidson seems to agree.19 If this is correct, there isn’t going to be, and there has never been, any reduction anywhere in science—that is, if you insist on reduction via strict laws. You are going to find strict laws only in basic physics, and you aren’t going to reduce basic physics to basic physics! (At least, you are not going to find reductions outside basic physics.) This surely cannot be a sense of reduction that holds serious philosophical interest for us. If psychology is reducible by the same standards that apply to the best cases of theory reduction in the sciences (pick your favourite examples), why isn’t that reduction enough? There has been a tendency, among some current antireductionists, to base their arguments on an unrealistically stringent and idealized model of reduction, thereby weakening their conclusions. I think “non-strict laws” are bad news for anomalous monists. In embracing them they may end up losing anomalism from anomalous monism. 19 Davidson says, “I made it clear that what I was calling a law in this context was something that one could at best hope to find in a developed physics” (“Thinking Causes,” p. 8). Others who hold a similar view include Fodor, and LePore and Loewer. I am not sure I understand what Davidson means by “developed physics”; whatever it is, it follows that Davidson isn’t going to find any reductions outside “developed physics.”

12 Causation and Mental Causation I An epistolary event occurred in 1643 which will live in the history of the mental causation debate. In the May of that year, Princess Elisabeth of Bohemia dispatched to Descartes what must be one of the most celebrated philosophical letters, challenging him to explain: How the mind of a human being can determine the bodily spirits [i.e., the fluids in the nerves, muscles, etc.] in producing voluntary actions, being only a thinking substance. For it appears that all determination of movement is produced by the pushing of the thing being moved, by the manner in which it is pushed by that which moves it, or else by the qualification and figure of the surface of the latter. Contact is required for the first two conditions, and extension for the third. [But] you entirely exclude the latter from the notion you have of the soul, and the former seems incompatible with an immaterial thing.1 A need for explanation arises for Elisabeth because she takes contact as a necessary condition for physical causation: the cause—at least, the proximate cause—of the motion of a material body must be in spatial contact with that body, a condition that plainly cannot be met by an immaterial causal agent outside physical space. The idea that causation requires contact survives even in Hume, a philosopher who is commonly thought to have held a deflationary view of causation as consisting solely in de facto regularities. One of the conditions Hume laid down for causation is that of contiguity in “space and time” between cause and effect, either direct or

1 Elisabeth to Descartes, May 1643. This quotation is taken from D. Garber, Descartes Embodied (Cambridge: Cambridge University Press, 2001), p. 172.

244

CAUSATION AND MENTAL CAUSATION

mediated by a chain of contiguous cause-effect pairs. (We will recur to the contiguity condition below.) Elisabeth’s challenge is intelligible and surely reasonable, both in commonsensical terms as well as in light of what Descartes had written, in Meditation II, about bodies and causes of their motions: By a body I understand whatever has determinate shape and a definable location and can occupy a space in such a way as to exclude any other body; it can be perceived by touch, sight, hearing, taste or smell, and can be moved in various ways, not by itself but by whatever else comes into contact with it.2 To be sure, Descartes doesn’t say that is the only way to set bodies in motion; however, if the mind’s causation of bodily motion is an exception, we are in need of an explanation. There may well have been earlier philosophical concerns about the powers of the mind to bring about changes in the physical world,3 but, for many of us, the exchanges between Descartes and Elisabeth are our first encounter with the mental causation debate in the early modern period. Descartes continues to loom large in contemporary discussions of many central issues in the philosophy of mind, and our current concerns with mental causation are no exception. What is of interest to us here is Elisabeth’s appeal to a specific feature of causation in her challenge to Descartes, the requirement that to cause a material body to move, physical contact with the body is required. Such contact, in more modern terms, represents the imparting of energy, or transfer of momentum, from one body to another, and this fact constitutes the causal action. Elisabeth’s simple complaint, which still resonates with us, is that such a conception of causation leaves no room for mental causation within Cartesian dualism. Minds, being essentially extensionless and without location in physical space, cannot meet Descartes’ contact requirement for causation of bodily motion; in fact, the idea of contact between an immaterial mind and a material body lacks a coherent sense. It appears, therefore, that on Cartesian terms, mental causation is a metaphysical impossibility; the very idea would seem unintelligible. 2 Descartes, Meditations on First Philosophy (1641), The Philosophical Writings of Descartes, vol. ii, tr. J. Cottingham, R. Stoothoff, and D. Murdoch (Cambridge: Cambridge University Press, 1985), p. 17. 3 V. Caston, “Epiphenomenalisms, Ancient and Modern,” Philosophical Review 106 (1997), pp. 309–363.


245

II Mental causation has been a flash point of debates in the contemporary philosophy of mind for well over three decades, ever since the publication of Donald Davidson’s “Mental Events” (1970).4 In this paper Davidson put forward what was then considered a startling thesis to the effect that there are no “strict” laws about mental phenomena—neither mental-physical laws nor mental-mental laws. Strict laws, if they exist, are found only in “developed physics.”5 Further, Davidson claimed that strict laws are required to underwrite causal relations. The two claims together appear at first blush to entail the impossibility of mental causation. However, Davidson had a deft reply: All that his two principles imply is that any causal relation must instantiate a strict physical law, and that what is required for a mental event to enter into a causal relation is to have an appropriate physical description under which it instantiates a physical law. From this Davidson derived his “anomalous monism”: All individual mental events—in fact, all individual events6—that enter into any causal relations are physical events. The only events that escape Davidson’s argument are those that are both causeless and effectless—entities hardly worth worrying about. This ingenious solution was the spark that brought back the problem of mental causation to contemporary philosophy. It all began when several philosophers noticed,7 apparently independently, that although Davidson’s anomalous monism allowed individual (or “token”) mental events to be causes and effects, it failed to give any role to mental properties, or mental descriptions, of these events in determining what causal relations they enter into. The reason is simple: since, on Davidson’s view, all strict laws are physical laws and they apply to individual events solely in virtue of their

4 Reprinted in Davidson, Essays on Actions and Events (New York and Oxford: Oxford University Press, 1980). 5 D. Davidson, “Thinking Causes,” in J. Heil and A. Mele (eds.), Mental Causation (Oxford: Oxford University Press, 1993). 6 Here we assume that the mental-physical dichotomy is both exhaustive and exclusive. 7 F. Stoutland, “Oblique Causation and Reasons for Action,” Synthese 43 (1980): 351–367; T. Honderich, “The Argument of Anomalous Monism,” Analysis 42 (1982): 59–64; J. Kim, “Epiphenomenal and Supervenient Causation,” Midwest Studies in Philosophy 9 (1984): 257–270; E. Sosa, “Mind-Body Interaction and Supervenient Causation,” Midwest Studies in Philosophy 9 (1984): 271–281; B. McLaughlin, “Type Epiphenomenalism, Type Dualism, and the Causal Priority of the Physical,” Philosophical Perspectives 3 (1989): 109–136.

246


physical properties, the mental properties they may have—or the mental kinds (e.g., pain, desire, thought) under which they may fall—are rendered irrelevant to the events’ causal relations—or so it seemed to his critics. Though Davidson’s anomalous monism is not a form of “token” epiphenomenalism, it was generally perceived as a form of “type” epiphenomenalism (a distinction due to McLaughlin8), the view that psychological characteristics and features contribute nothing to the causal powers of objects and events that have them. The position has the consequence that if we were to redistribute psychological, and other nonphysical, properties over the events and objects of this world—even if these properties were entirely removed—that would not change a single causal relation as long as the distribution of physical properties (and relations) remains the same. Davidson tried to defend anomalous monism against the epiphenomenalist charges;9 however, few seem to have found his efforts persuasive. It did not escape philosophers’ attention that Davidson’s troubles with mental causation crucially depended on his conception of causation—in particular, the condition that causally related events must instantiate “strict” laws. There has been some controversy about how to understand the strictness of strict laws, or what Davidson intended with the term. Strict laws of course must be laws—that is, as Davidson explicitly notes,10 must be capable of supporting counterfactuals and subjunctives and also be capable of confirmation by observation of positive instances (that is, inductively projectible). There seem two further features that make for strictness: first, strict laws are totally exceptionless (in this regard, they contrast with laws or generalizations hedged with ceteris paribus clauses), and, second, they are often (always?) found as part of a theory that is in some sense “complete” and gives comprehensive coverage over its domain. Paradigm cases of such theories will include physical theories, like classical mechanics, electromagnetic theory, and quantum mechanics. It is not easy to spell out this second condition in precise terms, something that Davidson himself seems never to have done. In practice, however, exceptionlessness is what does most of the work, and for most purposes this has seemed sufficient. In any case, a natural question to raise, when we are faced with Davidson’s epiphenomenalist 8 McLaughlin, “Type Epiphenomenalism.” 9 Davidson, “Thinking Causes.” 10 Davidson, “Mental Events.”


247

predicament, is why we should tie causation to strict laws. Why can’t there be causation where there are no strict laws in Davidson’s sense? This question is especially appropriate given the fact that Davidson never stated a clear reason, much less a detailed argument, for his requirement of strict laws for causal relations.

III One strategy that will naturally occur to many is to abandon Davidson’s strict law requirement, or relax it by allowing nonstrict laws, or laws with “ceteris paribus” hedges, to ground causal relations. Jerry Fodor is one such philosopher. He admits that “even the best psychological laws are very likely to be hedged,”11 and then continues “it [is] no longer clear why hedged psychological laws can’t ground mental causes; and, presumably, if hedged psychological laws can, then strict physical laws needn’t.”12 But how can a ceteribus paribus law, a law whose antecedent, say F-events, does not necessitate its consequent, G-events, ground a causal relation between F-events and G-events? Given the law, it is amply possible for an F-event to occur without being followed by a G-event. Being qualified by a ceteris paribus clause, the law is immune to falsification by such counter-instances; that is exactly the point of “ceteris paribus” hedges. Fodor’s argument is based on a special construal of “ceteris paribus” clauses that he proposes: The first—and crucial—step in getting what a robust construal of the causal responsibility of mental requires is to square the idea that Ms [mental events of kind M] are nomologically sufficient for Bs [bodily events of kind B] with the fact that psychological laws are hedged . . . [If] it’s a law that M ! B ceteris paribus, then it follows that you get Bs whenever you get Ms and the ceteris paribus conditions are satisfied. This shows us how ceteris paribus laws can do serious scientific business since it captures the difference between the (substantive) claim that Fs cause Gs ceteris paribus and the (empty) claim that Fs cause Gs except when they don’t.13 11 J. A. Fodor, “Making Mind Matter More,” Philosophical Topics 17 (1989): 72. 12 Ibid. 13 Ibid. 73.

248


The heart of Fodor’s strategy, then, appears to be the thought that whenever we have a serious ceteris paribus law “Ms cause (or are followed by) Bs, ceteris paribus,” there is a set C of conditions (as yet not fully identified) such that “Whenever C obtains, Ms cause (or are followed by) Bs” is a strict, exceptionless law. The reader will have noticed the alternate formulations, “cause (or are followed by),” in the preceding sentence. This was to reflect Fodor’s unexplained move, in the quoted passage, from “M ! B ceteris paribus,” which presumably only states that M is (nomologically) sufficient for B ceteris paribus, to “Ms cause Bs ceteris paribus.” This slide between nomological sufficiency and causation occurs throughout “Making Mind Matter More,” and it is indicative of the fact that Fodor’s operative conception of causality is straightforwardly based on nomological regularity.14 In any case, on Fodor’s view, “ceteris paribus” hedges represent only our ignorance of the details of causal/nomological regularities, and nothing metaphysical about the regularities themselves. So, for him, wherever there is a ceteris paribus law, there is a strict law waiting to be discovered. A ceteris paribus law connecting M-events with B-events can support the hedged causal claim “This M-event caused this B-event ceteris paribus,” and if the unspecified ceteris paribus conditions are in fact satisfied (whether or not we know it), we have the true unhedged causal judgment “This Mevent caused this B-event.”15 But this, to my mind, is to mislocate the real problem. What one should worry about in this context is not ceteris paribus clauses but a more fundamental question about causation and regularities. This question concerns whether or not we can get causation out of regularities, whether these are strict or hedged, or whether they are mere de facto regularities (Humean “constant conjunctions”) or given a suitable modal force (“physical/nomological necessity,” “lawlikeness,” “projectibility,” and the like). As early as 1925, C.D. Broad made the following simple observation: Again, if causation be nothing but regular sequence and concomitance, as some philosophers have held, it is ridiculous to regard psycho-neural 14 Fodor briefly considers the possibility of noncausal laws, that is, nomological regularities that do not constitute causal relations (ibid. 65) but waves it off in his typically lighthearted way. 15 It’s a bit of a mystery how we can ever know these “unknown” conditions are satisfied and hence how on Fodor’s account we can ever know a causal relation exists on the basis of a “ceteris paribus” law. Maybe we could infer it from the fact that the supposed effect actually occurred? But this event might have been caused by another event.


249

parallelism and interaction to be mutually exclusive alternatives. For interaction will mean no more than parallelism, and parallelism will mean no less than interaction.16 Actually, the situation is not quite as simple as Broad describes; for causation, or causal interaction, has directionality whereas psychoneural correlations under a strict parallelism are entirely symmetric and it would be difficult to determine which of the two symmetrically correlated events, one mental and one physical, is the cause and which the effect. But that is the least of the problems faced by the regularist, or nomological, approach to causation. First, there is the much discussed situation in which two phenomena are correlated, with nomological necessity, because they are collateral effects of a single cause.17 One of the two effects may always occur a little earlier than the other so that we may mistakenly think that the first is the cause of the second. There must be many such cases in medicine where a single underlying pathological state gives rise to two distinct symptoms, one occurring earlier than the other. The regularity connecting the two symptoms may be projectible and lawlike, and there seems no reason to deny that it holds with nomological necessity (we could even suppose it strict and exceptionless, though this is unlikely). Regularities of this kind, though they arise from underlying causal processes, do not constitute causation—a relation in which one event brings about another. Situations with the following structure present an analogous difficulty. We observe a regular connection between two events, A and B, with A preceding B, and we may be tempted to postulate a causal connection between them, with A as cause and B as effect. A believer in a purely regularist-nomological conception of causation would be committed to this conclusion. However, it may well turn out that the observed correlation between A and B is due to A’s regular correlation with event C and B’s regular correlation with event D, where C causes D. The correlation from A to B is only a superficial manifestation of an underlying causal process involving C and D. It will be easy enough to find instances exemplifying 16 Broad, The Mind and Its Place in Nature (London: Routledge & Kegan Paul, 1925), p. 96. Thanks to Brian McLaughlin for bringing this paragraph to my attention. 17 For extensive dicussions of “causal forks,” see Wesley C. Salmon, Scientific Explanation and the Causal Structure of the World (Princeton: Princeton University Press, 1984). Broad notes such cases in The Mind and Its Place in Nature, esp. pp. 115–117.

250


this situation in medicine in which an underlying pathological process gives rise to regular connections between symptoms caused by the various stages in the progression of the pathology. Closer to home, where mental causation is concerned, regular sequences of mental events may well be cases of this kind. If so, the impression that we are observing mental-to-mental causation would only be an impression. The fact would be that the observed sequence of mental events is grounded in, and is explained by, an underlying causal process between the neural substrates of the mental events. There is no direct causal relationship between the successive mental events in the sequence; the only genuine causal relations present are those between the neural substrates underlying the mental sequence. Thus, the relationship between the two successive mental events is like that between a series of shadows cast by a moving car at successive instants. The moving shadows do not constitute a causal sequence; nor would a sequence of mental events grounded in (or supervenient upon or realized by) a series of causally connected neural substrates.18 It should be clear that the issues about ceteris paribus laws do not touch these difficulties with the regularist-nomological conception of causality. Trying to soothe our fear of epiphenomenalism (or “epiphobia” as he calls it), Fodor tells us: According to the present view, the properties projected in the laws of basic science are causally responsible, and so too are the properties projected in the laws of the special sciences. Notice, in particular, that even if the properties that the special sciences talk about are supervenient upon the properties that the basic sciences talk about, that does not argue that the properties that the special sciences talk about are epiphenomenal.19 But this is no cure of epiphobia. Our discussion shows that though there may be projectible special-science properties and there may be specialscience laws, that does not guarantee that there is causation in the special sciences. Fodor continues as follows:

18 This is the gist of what I have called the “supervenience argument”: it is often called the “exclusion argument”; see J. Kim, Mind in a Physical World (Cambridge, Mass.: MIT Press, 1998) and Physicalism, Or Something Near Enough (Princeton: Princeton University Press, 2005). 19 Fodor, “Making Mind Matter More,” p. 66.


251

Not, at least, if there are causal laws of the special sciences. The causal laws of the special sciences and causal laws of basic sciences have in common that they both license ascriptions of causal responsibility.20 To be sure, if there are causal laws in psychology, they will license ascription of causal responsibility to psychological properties and ground psychological causal relations. The crucial question unaddressed by Fodor is whether psychological laws are causal laws—that is, whether the regularities we observe in the psychological domain are causal regularities, or mere reflections of the causal regularities at a more fundamental level. Fodor’s neglect of this question is evident in his seemingly unconscious slide between regularities and causal regularities, and between laws and causal laws. All this seems to be an outcome of his unreflective assumption of a regularistnomological conception of causation.

IV So the issue of mental causation cannot be resolved by simply invoking the regularist or nomological approach to causation. In saying this, I don’t mean to say that the nomological model of causation doesn’t work anywhere; it is possible that it is the right account of causation at the fundamental physical level. Precisely because it is the bottom level with nothing below, regularities, or “constant conjunctions,” may be all we can get; it makes no sense to speak of “underlying” mechanisms, or “real” causal processes at a lower level. Or we can perhaps take this to mean that, although only “constant conjunctions,” but no causation, exist at the fundamental level (the textbook Hume was right about this level), causal relations can, and do, exist (or “emerge”) at higher levels. These are interesting and intriguing issues but we must set them aside and move on. The nomological conception, which once was the reigning approach to causation and scientific explanation, seems to have lost favor with a significant number of philosophers. The tide now seems to have turned in favor of a broadly sine qua non conception of causation whose most influential modern version is due to David Lewis’s seminal account of causation in 20 Ibid.

252


terms of counterfactual dependence.21 This approach has recently attracted a lot of attention from an active and energetic group of philosophers, and there appear to be numerous ongoing research projects attempting to develop a satisfactory version based on Lewis’s basic insights. The idea of counterfactual dependence is this: e is counterfactually dependent on c just in case if c had not occurred, e would not have occurred. Since counterfactual dependence as defined is not transitive whereas causation is, Lewis explained “c causes e” as the ancestral of counterfactual dependence, that is, in terms of there being a series of events linking c with e such that any event is this series is counterfactually dependent on its predecessor. There are numerous outstanding difficulties with the counterfactual approach, among them the problems of overdetermination and preemption—problems that seem highly resistant to solution. The current literature is rife with increasingly complex and clever counterexamples and equally complex and ingenious remedies to evade them.22 The impression one gets from looking in from outside is that we are still very far from achieving the desired end, and that a reasonably simple and intuitively well-motivated counterfactual account of causation is not yet in sight. The increasing number of epicycles being piled on top of the epicycles already there reminds one of the ultimately fruitless searches for the “fourth” condition of knowledge prompted by the Gettier Problem. I don’t want to say that the ongoing research on the counterfactual analysis of causation is without value; far from it, it may yield—I believe it has already yielded—some valuable insights into our causal talk, just as the Gettier-inspired work in epistemology has contributed much to our understanding of knowledge and justification. Our present concern is not with the ultimate viability of the counterfactual approach to causation; it is a more restricted one of assessing the prospect of explaining mental causation in terms of counterfactuals, although in doing this our discussion will inevitably involve some general issues about causation and counterfactuals. One such general issue concerns the apparent dependence of counterfactuals, at least those involved in causal attributions, on laws and regularities, and if this is the case, embracing the counterfactual approach to 21 Lewis, “Causation,” Journal of Philosophy 70 (1973): 556–567; repr. with postscripts in D. Lewis, Philosophical Papers, vol. ii (Oxford: Oxford University Press, 1986). 22 A good place to sample some of this is J. Collins, N. Hall, and L. A. Paul (eds.), Causation and Counterfactuals (Cambridge, Mass.: MIT Press, 2004).


253

causation will have no advantages over the regularist-nomological approach we considered earlier. Consider the causal claim: The striking of the match caused it to light. On a simple counterfactual analysis, this amounts to the assertion of the following counterfactual: (C) If this match had not been struck, it would not have lighted. Almost all current counterfactual theorists of causation use the semantics of counterfactuals developed by Robert Stalnaker and David Lewis,23 based on comparative similarity among possible worlds. According to this scheme, (C) is true just in case the consequent of the conditional “the match lighted” is true at the world that, apart from the fact that the match was not struck in that world, is the closest to the actual world in all other respects (we use this somewhat simplified formulation; this will make no difference). Assume (C) is true. That means that in the closest world in which the match was not struck, it did not light. How do we know that this world is closer to the actual world than is the closest world in which the match was not struck but it nonetheless lighted? The obvious, and the only possible, answer seems to be that, in the actual world, dry matches struck in the presence of oxygen usually and reliably ignite, and that it is our knowledge of this regularity, or law, combined with knowledge of the actual circumstances in which the match was struck (e.g., it was dry, oxygen was present, etc.), that accounts for the judgment that the world in which the match that was not struck does not light is closer to the actual world than is the world in which the unstruck match lights. Perhaps the laws involved might be more theoretical and concern the chemical composition of the match head, its combustibility, the characteristics of the surface against which the match was struck, and so forth. In any case, one crucial respect in which the comparative similarity of worlds is to be determined evidently involves the similarity of laws holding in them. It is difficult to see how evaluations of conditionals like (C) could avoid adverting to laws and regularities. Let us see how this affects the use of counterfactuals to account for mental causation. Consider the claim that a sudden attack of migraine headache caused Susan a frightful sense of anxiety. For the counterfactualist, this amounts to the truth of:

23 Stalnaker, “A Theory of Conditionals,” in Nicholas Rescher (ed.), Studies in Logical Theory (Oxford: Blackwell, 1968); Lewis, Counterfactuals (Cambridge, Mass.: Harvard University Press, 1973).

254


(D) If Susan had not had the sudden migraine headache, she would not have experienced frightful anxiety. We can concede that our commonsense “knowledge,” or assumption, that counterfactuals like (D) are often true grounds our belief in the reality of mental causation. Our job as philosophers is to see what makes the likes of (D) true and whether this justifies the claim that Susan’s migraine headache caused her anxiety. If our observations relating to (C) are correct, the truth of (D) must depend on the regularity connecting sudden attacks of migraine headaches and feelings of anxiety. This regularity could be limited to Susan and others like her in relevant (presumably neurophysiological) respects, or it could be a (ceteris paribus) law for all people with migraine headaches. It seems to me that even an epiphenomenalist like T.H. Huxley can, with consistency, accept a regularity of this kind and acknowledge it to be lawlike (surely, the connection isn’t accidental or coincidental—not even for an epiphenomenalist). However, the epiphenomenalist will deny that (D) warrants the causal claim that the attack of migraine headache caused the sense of fearful anxiety. The observed regularity arises out of a genuine causal process connecting two neural substrates on which the headache and the anxiety respectively causally depend. The situation is fundamentally the same if you believe that the dependence relation between mental states and their neural substrates is better described in terms of supervenience or realization than causation.24 And these observed psychological regularities may very well underwrite their corresponding counterfactuals. As we saw with Fodor, however, there is no guarantee that they are causal regularities, and if they are not, there is no reason to think that the counterfactual dependencies they ground yield genuine causal relations.

V In spite of these and possibly other difficulties, the counterfactual approach to causality remains popular—among philosophers working on issues about causation (especially, the analysis of the concept) and among philosophers who aim to defend mental causation against various epiphenomenalist 24 I doubt that Huxley would have cared one way or the other. Obviously the idioms of supervenience and realization were not available to him.


255

threats.25 The intuition that supports the counterfactual approach, I believe, is the close association we form between a cause of an event and a sine qua non condition of its occurrence. A cause is the condition but for which the effect would not have occurred. We can grant the legitimacy of this intuition, without necessarily wedding it to any particular way of making it precise and exact—without, that is, necessarily explicating it in terms of counterfactuals with a Stalnaker/Lewis-style semantics, or any other special semantics of conditionals. But there is another strong intuitive conception of causation that contrasts sharply with the conception tied to counterfactual dependency, or the sine qua non condition. It is a productive or generative conception of what causing consists in. On this conception, a cause is something that produces, or generates, or brings about its effects, something from which the effects derive their existence or occurrence. This idea was given its classic expression when Elizabeth Anscombe wrote: There is something to observe here, that lies under our noses. It is little attended to, and yet still so obvious as to seem trite. It is this: causality consists in the derivativeness of an effect from its cause. This is the core, the common feature, of causality in its various kinds. Effects derive from, arise out of, come of, their causes. For example, everyone will grant that physical parenthood is a causal relation.26 Indeed, in a recent article, Ned Hall makes a plausible case for the thesis that there are two fundamentally distinct notions of causation: Causation, understood as a relation between events, comes in at least two basic and fundamentally different varieties. One of these, which I call “dependence,” is simply that: counterfactual dependence between wholly distinct events. In this sense, event c is a cause of (distinct) event e just in case e depends on c, that is, just in case, had c not occurred, e would not have occurred. The second variety is rather more difficult 25 For example, E. LePore and B. Loewer, “Mind Matters,” Journal of Philosophy 93 (1987): 630–642; T. Horgan, “Mental Quausation,” Philosophical Perspectives 3 (1989): 47–76; B. Loewer, “Comments on Jaegwon Kim’s Mind in a Physical World,” Philosophy and Phenomenological Research 65 (2002): 655–662 and “Mental Causation, Or Something Near Enough,” in Brian P. McLaughlin and Jonathan Cohen (eds.), Contemporary Debates in Philosophy of Mind (Oxford: Blackwell, 2007). 26 Anscombe, Causality and Determination (Cambridge: Cambridge University Press, 1971); page refs. are to the reprint in M. Tooley and E. Sosa (eds.), Causation (Oxford: Oxford University Press, 1993). The quotation here is from pp. 91–92.

256


to characterize, but we evoke it when we say of event c that it helps to generate or bring about or produce another event e, and for that reason I call it “production.”27 According to Hall, three characteristics are central to productive/generative causation: transitivity, locality, and intrinsicness. Of these what is relevant to our present concerns is locality, which Hall states as follows: “Causes are connected to their effects via spatiotemporally continuous sequences of causal intermediaries.”28 This seems equivalent to Hume’s contiguity condition alluded to earlier. Hume’s own statement is this: I find in the first place, that whatever objects are consider’d as causes or effects, are contiguous; and that nothing can operate in a time or place, which is ever so little remov’d from those of its existence. Tho’ distant objects may sometimes seem productive of each other, they are commonly found upon examination to be link’d by a chain of causes, which are contiguous among themselves, and to the distant objects; and when in any particular instance we cannot discover this connexion, we still presume it to exist.29 As Hall notes, causal relations conforming to the dependence idea need not meet the locality condition; we will see some examples below. What we have seen in earlier sections is that neither the nomological nor the dependency conception of causation can properly ground mental causation. I argued that nomological relationships do not deliver the kind of causal efficacy, or productivity, we want for mentality, and that the counterfactual approach seems to presuppose, or collapse to, the nomological conception and thereby inherit the latter’s shortcomings. Many counterfactualists will dispute this claim. We need not concern ourselves with this general issue about causation. In this section, I will try to argue that the relation of causation as dependence, or counterfactual dependence, even if it is a proper and useful causal relation, is not the source of our worries about mental causation. That is, even if we succeed in showing that mental causation, with causation construed as dependence, is real, that would not suffice to vindicate mental causal efficacy and thereby dissipate our 27 Hall, “Two Concepts of Causation,” in Collins, Hall, and Paul (eds.), Causation and Counterfactuals, 225. 28 Ibid. 29 Hume, A Treatise of Human Nature (London, 1739); page refs. are to the L. A. Selby Bigge edn. (Oxford: Clarendon Press, 1888). The quotation here is from p. 75.


257

epiphenomenalist worries. Fundamentally these worries arise, I believe, from the question whether mentality has the power to bring about its effects in a continuous process of generation and production—or the question whether we can show that this is so. Why should we resort to this “thick” variety of causation in thinking about mental causation? My answer is pretty simple: We care about mental causation because we care about human agency, and agency requires the productive/generative conception of causation.30 I don’t have a knockdown argument to prove that agency requires productive causation; I hope what I will say here makes my claim at least plausible. It seems to me that mere counterfactual dependence is not enough to sustain the causal relation involved in our idea of acting upon the natural course of events and bringing about changes so as to actualize what we desire and intend. An agent is someone who, on account of her beliefs, desires, emotions, intentions, and the like, has the capacity to perform actions in the physical world—that is, to cause her limbs and other bodily parts (e.g., the vocal cords) to move in appropriate ways so as to bring about changes in the arrangement of objects and events around her—open a door, pick up the morning paper, and make a cup of coffee. It seems to me that without productive causation, which respects the locality/contiguity condition, such causal processes are not possible. These causal processes all involve real connectedness between cause and effect, and the connection is constituted by phenomena like energy flow and momentum transfer, an actual movement of some (conserved) physical quantity.31 In saying this we need not impugn the dependency conception of causation; all we need is the point that agency requires productive causation.32 Note, further, that we need not 30 In correspondence, Barry Loewer has challenged this claim. According to him, “thin” causation, or dependence, is sufficient to ground agency. See Loewer, “Mental Causation, Or Something Near Enough.” 31 I am of course referring to the so-called conserved quantity approach to causation. See P. Dowe, “Wesley Salmon’s Process Theory of Causality and the Conserved Quantity Theory,” Philosophy of Science 59 (1992): 195–216 and Physical Causation (Cambridge: Cambridge University Press, 2000); W.C. Salmon, “Causality Without Counterfactuals,” Philosophy of Science 61 (1993): 297–312; for an early statement see D. Fair, “Causation and the Flow of Energy,” Erkenntnis 14 (1979): 219–250. See also the exchange between Dowe, “Causes Are Physically Connected to Their Effects,” and the dependence theorist J. Schaffer, “Causes Need Not Be Physically Connected to Their Effects: The Case for Negative Causation,” both in C. Hitchcock (ed.), Contemporary Debates in Philosophy of Science (Oxford: Blackwell Publishing, 2004). 32 For our purposes we need not claim that all cases of action involve productive causation; perhaps we are willing to regard certain cases of omissions as actions and consider omissions as eligible as causes. See below for further discussion of omissions.

258


claim that dependency is not involved in actions; it may well be that the dependency involved, say between a limb movement and a desire, has an explanation in terms of the productive/generative relations between them.33 Consider the component of mental-to-physical causation involved in action, namely the causation of bodily movements by our desires, beliefs, intentions, and the like. To endow our mental states with causal powers to move our limbs (or, more proximately, the powers to bring about changes in our neural states), would it be enough to show that counterfactuals like the following are true—and that we have reason to believe them true? If Susan had not wanted to open the window, neural state N would not have occurred in her brain (where we suppose N triggers an appropriate sequence of bodily movements). If Susan had not experienced the sudden migraine headache, neural state N* would not have occurred (where we assume N* to be the neural substrate of anxiety attacks). We earlier argued that counterfactuals like these ultimately involve reference to psychological or psychophysical regularities, and that their significance for mental causation depends on the question whether these regularities are causal regularities. Apart from this issue, there are reasons to be suspicious about relying on counterfactuals alone to defend mental causation—what such a strategy could show. Friends of the counterfactual approach often tout its ability to handle omissions and absences as causes and note the productive/generative approach’s inability to account for them. We are inclined to take the truth of a counterfactual like: If Mary had watered my plants, the plants would not have died as showing that Mary’s not watering, an omission, caused the plants’ death and take that as a basis for blaming Mary for killing the plants. But obviously there was no flow of energy from Mary to the plants during my absence (that exactly was the problem!); nor was there any other physical

33 Sometimes in terms of the absence of such relations, e.g., in cases of causation by omissions and absences; see below. Can there be any causation—any dependency causation—without there being at least some productive causation? I don’t know the answer.


259

connection, or any spatio-temporally contiguous chain of causally connected events. One issue with regarding Mary’s omission as a cause of the plants’ death on the strength of the foregoing counterfactual is that there are indefinitely many other counterfactuals like it inappropriately certifying an indefinitely large number of other omissions as causes of the plants’ death:34 If George W. Bush had watered my plants, the plants would not have died. If Laura Bush had watered my plants, the plants would not have died. If Hillary Clinton had watered my plants, the plants would not have died. Well, you get the idea. I blame Mary for not watering the plants; we may blame agents for their omissions. But we don’t have to say that the omissions are causes. We need not say that I am blaming Mary for killing my plants (she would have killed them if she had sprayed them with a herbicide); the fact is that she didn’t do anything to them while I was away, and that was the problem. I am blaming her for breaking a promise—her promise to water the plants. If omissions should count as actions, something we do, then by staying in my room “doing nothing” (I could be taking a long nap), I would be performing countless actions, like not watering my plants, not writing an email to my niece, not doing the MS walk, not space-walking out of the shuttle Discovery, . . . Of course these are not intentional omissions (at least they don’t have to be), but it is difficult to see how intentional omissions and mere omissions could differ metaphysically, specifically with respect to their causal powers. At any rate, it is by no means clear that its apparent ability to handle omissions as causes is something that the dependency theorist should celebrate. Not thinking, not believing, not desiring, and so on are mental omissions. If causation by mental omissions counts as mental causation, that would make mental causation easy—too easy. My not believing (or disbelieving) that a chest of treasures is buried in my backyard is a cause of my not 34 I saw examples like these for the first time in B. Abbott, “Some Problems in Giving an Adequate Model-Theoretical Account of CAUSE,” in C. Filmore, G. Lakoff, and R. Lakoff (eds.), Berkeley Studies in Syntax and Semantics, vol. i (1974). This paper is recommended to those interested in the counterfactual approach to causation. See also S. McGrath, “Causation by Omission: A Dilemma,” Philosophical Studies 123 (2005): 125–148.

260


digging in my backyard; my not believing treasures are buried in your backyard causes my not digging in your backyard, and so on ad infinitum. This doesn’t look like causation worth having.

VI But if we understand causation in mental causation in the productive/ generative sense, wouldn’t that rule out mental causation—in particular mental-physical causation—too quickly, without any need for an argument? Especially if we require that causation requires energy flow or momentum transfer, how could there be such a process from a mental entity to a physical entity, or in the converse direction? Remember Elisabeth’s challenge to Descartes: the causation of physical motion requires spatial contact, but how could an immaterial mind outside physical space be in such contact with a material body? Notice that this problem is not special to Cartesian physics; it arises even under Hume’s concept of causation, which, as we saw, requires a spatially contiguous chain of causally connected events. Don’t all such conceptions of causation, conceptions that require some “real” connections between cause and effect, automatically rule out mental-physical causation (and hence human agency)? Further, what could “contiguity” mean unless it meant spatial contiguity? Further, what “real” connection can there be between two immaterial and nonspatial substances? Wouldn’t the productive/generative conception of causation preclude, without much ado, mind-to-mind causation as well as mind-to-body causation—that is, all mental causation?35 An answer—the right answer, in my opinion—is contained in a followup letter Elisabeth sent to Descartes. In June 1643, she wrote to Descartes: And I admit that it would be easier for me to concede matter and extension to the mind than it would be for me to concede the capacity to move a body and be moved by one to an immaterial thing.36 This, I believe, is a remarkable statement attesting to the philosophical astuteness of Elisabeth. She is saying what some of us have been saying for 35 For an argument for an affirmative answer to this question based on spatial considerations, see Kim, Physicalism, chap. 3. 36 Garber, Descartes Embodied, p. 172.


261

the past two decades—namely that to make sense of mental causation, she would rather physicalize the mind (“concede matter and extension to the mind”) than acquiesce in the unintelligible idea that immaterial minds can be in causal commerce with material bodies. Elisabeth is voicing the thought that we would now express by saying that mental causation is possible only if mentality is physically reducible. Her avowal may well have been the first causal argument ever for physicalism. The dominant strain of physicalism on the contemporary scene has been the nonreductive kind. Nonreductive physicalists, while rejecting Cartesian immaterial minds or any nonphysical object, nonetheless resist the idea that mental properties are reducible to physical properties. Beliefs, desires, intentions, pains, visual images, and the rest, though they may be supervenient on neural/biological processes, are irreducible to them; nonetheless, these mental states are claimed to be causally efficacious. But how can the idea of productive/generative causation be applied to them in relation to neural/physical states? How can there be energy flow or momentum transfer from a desire, as an irreducible mental state, to the firing of a group of neurons? In his characterization of nonreductive physicalism, Terence Horgan, a leading proponent of the position, writes: First, mental properties and facts are determined by, or supervenient upon, physical properties and facts. Second (and contrary to emergentism), physics is a causally complete science; the only fundamental force-generating properties are physical properties. More specifically, the human body does not instantiate any fundamental force-generating properties other than physical ones. Third, mental properties nonetheless have genuine causal/explanatory efficacy, via the physical properties that “realize” mental properties on particular occasions of instantiation.37 So, on Horgan’s antireductionist view, a desire (as an individual “token” event) has “genuine” causal efficacy, say the power to move my arm to reach for a glass of water, in virtue of the fact that its neural realizer, an instance of a neural property on which the desire supervenes, has this causal power. Horgan’s suggestion, I believe, is fundamentally right: mental events and states have the causal efficacy that they have because their neural/physical

37 Horgan, “Reduction, Reductionism,” in Donald M. Borchert (ed.), The Encyclopedia of Philosophy: Supplement (New York: Simon & Schuster Macmillan, 1996), 498.

262


realizers have causal efficacy. In fact, a mental state, occurring on a given occasion, in virtue of being realized by a certain neural/physical state, has exactly the causal powers of that physical state.38 Where I differ from Horgan is that once we are prepared to say what we have just said, the next natural step to take—in my view, a step we are compelled to take—is to reductively identify this particular mental state with its neural/physical realizer.39 This of course is to jettison the “nonreductive” part of nonreductive physicalism. To say what Horgan says, namely that the belief is a distinct state from its neural realizer, and go on to take each as a sufficient cause of the arm rising, is to walk smack into the problem of overdetermination. And to say that the mental state has causal efficacy “via” the causal efficacy of its neural realizer carries an apparent epiphenomenal implication: Given that the neural realizer is a full cause of the arm rising, what causal work is there for the mental state to contribute? Or, to put the question another way, what could “via” mean here? What is it for an event to cause something “via” another event that presumably does the real causing? And what is it about the realization relation that grounds this “via” connection? To resist the reductive move of identification is to recognize the existence of something whose causal work is at best superfluous, and nonexistence at worst. So the idea of causation as production and generation, or causation as requiring a “real” connection between cause and effect, can be applied to mentality as long as, and presumably only so long as, mental states have physical realizers. Whether an approach of this kind leads to reductionism, as I just claimed, or it is compatible with a nonreductive view of mentality is a further, currently much debated, issue.40 38 Kim, “Multiple Realization and the Metaphysics of Reduction,” Philosophy and Phenomenological Research 52 (1992): 1–26. 39 Kim, “Postscripts on Mental Causation,” in Kim, Supervenience and Mind (Cambridge: Cambridge University Press, 1993); and Kim, Mind in a Physical World. 40 Kim, Physicalism.

13 Two Concepts of Realization, Mental Causation, and Physicalism I The concept of realization, unlike most philosophical concepts currently in use, is relatively new. In this regard, it’s like supervenience. Yet there is a difference. Arguably, the idea of supervenience, though perhaps not the term, had a much earlier origin, going back by several centuries,1 before R.M. Hare introduced the term in 1952.2 Realization is different; so far as we know, there was no such concept in philosophical use before the second half of the 20th century. I believe that we owe the term, and the concept, to Hilary Putnam who wrote in his 1960 paper “Minds and Machines”: In particular the ‘logical description’ of a Turing machine does not include any specification of the physical nature of those ‘states’ [a Turing machine’s ‘internal states’] . . . In other words, a given ‘Turing machine’ is an abstract machine which may be physically realized in an almost infinite number of different ways.3 1 Leibniz’s use of the Latin supervenire seems consistent with the current philosophical use of “supervene” when he writes: “Relation is an accident which is in multiple subjects; it is what results without any change made in the subjects but supervenes from them; it is the thinkability of objects together when we think of multiple things simultaneously” (Die Leibniz-Handschriften der ko¨nighlichen o¨ffentlichen Bibliothek zu Hannover, ed. E. Bodemann (Hanover, 1895), VII, c, p. 74; quoted in Hide´ Ishiguro, Leibniz’s Philosophy of Logic and Language (Ithaca, NY: Cornell University Press, 1972), p. 71). Paul Hoffman has informed me that Aquinas used variants of supervenire numerous times in his works but in a sense similar to the colloquial meaning of “supervene.” According to Hoffman, there is a Greek equivalent of supervenire that Aristotle is thought to have used to describe the soul’s relation to the body. See Robert Heinaman, “Aristotle and the Mind-Body Problem,” Phronesis 35 (1990): 83–102. 2 In The Language of Morals (Oxford: Clarendon Press, 1952). 3 Hilary Putnam, “Minds and Machines,” reprinted in Putnam, Mind, Language, and Reality (Cambridge: Cambridge University Press, 1975); originally published in Dimensions of Mind, ed. Sydney Hook (New York: New York University Press, 1960). The quotation is from p. 371 of the Putnam volume (with added emphasis).

264

TWO CONCEPTS OF REALIZATION

It is interesting to note that the idea of multiple realization was born at the very same time as the idea of realization—in fact, in the same sentence. In any event, this is the first occurrence of the term “physical realization” that I know of. Apparently, Putnam himself thought so, too, as he writes a bit later in the same paper: The functional organization (problem solving, thinking) of the human being or machine can be described in terms of the sequences of mental or logical states respectively . . . without reference to the nature of the ‘physical realization’ of these states.4 It seems plausible to conjecture that Putnam put quotes around “physical realization” thinking that he was introducing a technical neologism. It is somewhat remarkable that he offers no explanation of just what he had in mind, apparently counting on the reader’s intuitive understanding of the term; it is even more remarkable that no one raised a serious question about what “realization” meant, or should mean, while the term was gaining quick currency in philosophical literature through the rest of the century, figuring in some of the most important claims and arguments in philosophy of mind and philosophy of science. As everyone knows, the so-called multiple realizability argument was what was primarily responsible for bringing down the mind-brain identity theory, and, more generally, reductionist physicalism in philosophy of science as well as philosophy of mind. The demise of reductionist physicalism, which began in the late 1960s, and the concurrent rise of nonreductive physicalism to a dominant position during the ’70s was one of the most important philosophical events in the second half of the last century. It brought about a sea change in the way philosophers viewed the status of “higher-level” properties and phenomena, in the domains of the special sciences, in relation to the “lower-level” physical properties and processes, and thus the relationship between higherlevel sciences and basic physics. What is remarkable in retrospect is the fact that all this happened because of some arguments crucially based on an unanalyzed and undiscussed concept of realization, a concept that philosophers at the time apparently accepted, and used, as clear and unproblematic with a robust shared meaning.

4 Putnam, Mind, Language, and Reality, p. 373.


265

As century turned, however, the situation began to change; realization has now become a target of philosophical analysis and debate. A new generation of philosophers, including Lawrence Shapiro, Thomas Polger, Ronald Endicott, Lenny Clapp, and Carl Gillett among others, have begun asking serious and probing questions about the concept of realization and the thesis, which had achieved the status of accepted wisdom, that specialscience properties, including psychological properties, are multiply realizable and are often so realized.5 And we now have a substantial book, Physical Realization,6 devoted to the topic, by Sydney Shoemaker, an esteemed metaphysician and philosopher of mind. It is fair to say, I believe, that Shoemaker’s account of what it is to realize a property is the only genuinely new proposal we have seen in this area since Putnam first introduced the concept half a century ago. Shoemaker argues that his conception of realization is superior to the standard conception in that it leads to better and more satisfying resolutions of certain vexing philosophical problems about the mind, in particular the mental causation problem and the issue of reductive type physicalism versus nonreductive physicalism. In this paper, I want to examine Shoemaker’s concept of realization, especially in regard to his claims about its advantages over the standard notion.

II As Shoemaker says, the dictionary meaning of “realize” is something like “make real.” This seems to fit Putnam’s original usage as well: a physical computing machine “makes real” or “brings into concrete reality” an abstractly characterized Turing machine, a mathematical entity. In this sense, what is realized are things like plans, designs, blueprints, and the like, and realization is a relation between concrete objects in the world and abstract items like Turing machines and designs. But, as we know, this has changed: most philosophers who now speak of realization think of 5 For example, see Lawrence Shapiro, “Multiple Realizations,” Journal of Philosophy 97 (2000): 635–654; Shapiro, The Mind Incarnate (Cambridge, Mass.: MIT Press, 2004); Lenny Clapp, “Disjunctive Properties: Multiple Realizations,” Journal of Philosophy 98 (2001): 111–136; Carl Gillett, “The Metaphysics of Realization, Multiple Realizability, and the Special Sciences,” Journal of Philosophy 100 (2003): 591–603; Thomas Polger, Natural Minds (Cambridge, Mass.: MIT Press, 2004). 6 (Oxford: Oxford University Press, 2007).

266


it primarily as a relation between properties; we talk about neural/ physical properties realizing mental properties, physicochemical properties realizing biological properties, microphysical properties realizing observable macroproperties of physical objects, and so on. (Shoemaker extends this to include “microphysical states of affairs” as things that can realize properties; however, we won’t be concerned with this aspect of Shoemaker realization.) As Shoemaker observes,7 realization as currently used by philosophers is a term of art. Although its provenance in Putnam and its earlier uses by philosophers like Jerry Fodor and Ned Block deserve due consideration, we are free to define it pretty much any way we like. What speaks in favor of a proposal is the philosophical usefulness of the concept introduced. In the present case, the usefulness consists in its capacity to illuminate, elucidate, and possibly resolve a cluster of well-known issues concerning the mind-body problem, reductionism, emergence, mental causation, and the like. What drives Shoemaker’s search for his own concept of realization is his belief that the standard account of realization has undesirable consequences, consequences that ought to be avoided, regarding many of these issues. He contrasts his “subset” account of realization with the “standard” concept according to which a property that is realized is a “second-order” functional property defined in terms of a causal role and its realizers are the first-order properties that fill, or satisfy, the specified causal role.8 Let us look at this standard conception in a bit more detail. As an example, suppose the property of being in pain could be given the following functional characterization (this is for illustration only; I don’t advocate a functional analysis of pain, or any other qualitative state of consciousness): x is in pain ¼ def. x has some property P such that P is apt to be caused to instantiate in x (and systems like x) by tissue damage, and in x (and systems like x) an instance of P is apt for causing winces and groans.

7 As has Brian McLauglin in “Mental Causation and Shoemaker-Realization,” Erkenntnis 67 (2007): 149–172. 8 The idea of functional properties as “second-order” properties is also due to Putnam. See his “On Properties,” in Essays in Honor of Carl G. Hempel, ed. Nicholas Rescher et al. (Dordrecht: D. Reidel, 1970); reprinted in Putnam, Mathematics, Matter and Method (Cambridge: Cambridge University Press, 1975).


267

For humans, let’s say it is the familiar (if harmlessly fictional) C-fiber stimulation (Cfs) that plays the causal role as specified; that is, where x is a human, Cfs is the property apt to be instantiated when x suffers tissue damage and which is apt to cause x to groan and wince. So Cfs is a realizer of pain in humans. And for octopuses (you may recall they were the original example used by Putnam to illustrate pain’s multiple realizability), it may be “O-fiber” stimulation that fits the causal specification, and this makes O-fiber stimulation pain’s realizer for octopuses; for Martians “XYZ-fiber” stimulation might be pain’s realizer; and so on. All this is pretty familiar. Let us refer to this concept of realization as the “second-order” account. However, Shoemaker thinks that this view of realization is philosophically flawed and needs to be replaced, and this is what motivates him to propose his own account, the “subset” view. Why would anyone find the second-order view unsatisfactory? There has been a feeling that this view of realization leads to epiphenomenalist consequences for mental properties. Nearly two decades ago, Ned Block expressed the fear that mental properties construed as functional properties face the threat of their causal powers being preempted by those of their realizers—realizers on the second-order view.9 Shoemaker shares Block’s fears; he thinks that the standard view exposes mental properties to the threats posed by the exclusion-style epiphenomenalist argument. He writes: A prima facie objection to this [second-order account of realization] is that it seems to make it true, by stipulation, that any causal role we might want to assign to the realized property is preempted by its realizers. So any effects—e.g. wincing—we attribute to someone’s being in pain are really due to whatever neural property realized pain on that occasion (11).10 Shoemaker’s language here, in particular his “seems,” is a bit noncommittal as to whether the causal exclusion actually follows—or whether he believes it follows—from the application of the second-order view to mental properties. In a footnote, he says that there would be no preemption if a mental property can be identified with its neural realizer, but that the phenomenon of multiple realization precludes this. He is also against the view that 9 Ned Block, “Can the Mind Change the World?” in Meaning and Method, ed. George Boolos (Cambridge: Cambridge University Press, 1990). 10 Numerals embedded in the main text refer to page numbers in Shoemaker’s Physical Realization.

268


any given instance of the realized property can be, or should be, identified with the particular instance of the realizer property involved on the occasion. Shoemaker goes on to say: “I favor an account that is designed to avoid this consequence” (11). We will have a chance later to see what Shoemaker’s alternative account of realization can, or cannot, do for mental causation. First, though, there is an important question of philosophical strategy to consider here. Suppose that on Shoemaker’s account of realization the exclusion argument cannot be carried through; it may even be that this argument cannot even be coherently formulated. What does that show about the second-order view of realization, or the exclusion argument formulated on the basis of it? The (supposed) fact that the second-order account leads to an epiphenomenalist consequence is hardly a reason to think the account is in some way defective. Maybe we have to swallow some unpalatable implications of our assumptions and presumptions. Definitions are just definitions—remember that here we are dealing with a stipulative definition. They fix meanings of our terms and concepts, and they can’t have substantive consequences all on their own. What needs to be done to defeat the exclusion argument, based on the second-order account of realization, is to show that mental properties are not realized by neural/physical properties in the sense provided for by that account. If they are so realized, then the exclusion argument goes through, whether or not we are particularly fond of this concept of realization, and we will be stuck with the epiphenomenalist consequences (if the exclusion argument is otherwise unproblematic). Whether or not we can fashion another definition—or a dozen definitions— of realization under which the exclusion argument cannot be run seems to me entirely irrelevant; it does not make the need to deal with the epiphenomenalist implications of the second-order account go away. The point is quite general. We cannot hope to escape the exclusion problem by developing new and novel concepts of realization, or by positing other relations between mental properties and their underlying neural/physical properties. This becomes especially clear when we remember that realization in the standard sense is not the only mind-body relation that encourages the exclusion-style reasoning. As is well known, versions of the exclusion argument have been formulated in terms of mind-body


269

supervenience rather than mind-body realization.11 The crucial characteristic of any mind-body relation that gets an exclusion-style argument going is dependence/determination—the idea that mental properties depend for their instantiation and character on the instantiation of appropriate physical/ neural properties. Given a dependence thesis of this sort and the thesis of mindbody antireductionism as premises, it is not difficult to develop a causal exclusion argument that apparently renders the mental epiphenomenal.12 This is why, in my view, even the classic emergentist conception of the mind-body relation falls prey to exclusion arguments.13 This means that if you are looking for an account of realization that will insulate the mental from causal exclusion arguments, you had better come up with a realization relation that does not entail a dependency/determination relation. In addition, you must be able to show that your realization relation is the most appropriate, and strongest, way to characterize the mind-body relation—more apt than supervenience, standard realization, emergence, and the like. To me, this looks like an impossible task—especially if you have any physicalist sympathies. We must conclude, then, that whatever advantages the new subset account of realization might enjoy over the second-order account, the relief it might provide us from exclusion arguments isn’t one of them.

III Let us now turn to the alternative account of realization, the “subset” view, offered by Shoemaker. According to him, properties are individuated by causal profiles. What are “causal profiles”? To quote: The causal profile of a property consists of two sorts of causal features— forward-looking causal features, having to do with how the instantiation of the property contributes to producing various sorts of effects 11 See my Mind in a Physical World (Cambridge Mass.: MIT Press, 1998), chap. 2, and Physicalism, or Something Near Enough (Princeton: Princeton University Press, 2005), chap. 2. At first, I called this argument “the supervenience argument”; I now believe that it is clarifying to divide the argument into two stages, one to be called “the exclusion argument” and the other “the supervenience argument.” See my Philosophy of Mind, 2nd edn. (Boulder, Clo.: Westview Press, 2006), chap. 7. 12 On this point, see Marcelo Sabates’ helpful discussion in his “Varieties of Exclusion,” Theoria 16 (2001): 13–42. 13 For details see my “Making Sense of Emergence,” Philosophical Studies 95 (1999): 3–36. Essay 1 of this volume.

270


(and contributes to bestowing causal powers on its possessors), and backward-looking causal features, having to do with what sorts of states of affairs can cause the instantiation of the property (12). What does it mean to say that properties are “individuated” by their causal profiles? Shoemaker says that for the purposes of his project on realization, a full causal theory of properties, under which properties have their causal profiles essentially, is not needed, and that he can work with a weaker assumption that in the actual world, no two properties have identical causal profiles.14 As a “first approximation,” Shoemaker offers the following definition of realization: . . . property P has property Q as a realizer just in case (1) the forwardlooking causal features of P is a subset of the forward-looking causal features of property Q, and (2) the backward-looking causal features of P have as a subset the backward-looking features of Q (12). Shoemaker’s “causal features” are not exactly what are usually called causal powers. For him, corresponding to each forward-looking causal feature of a property there is a “conditional power” bestowed by that property upon an object that has it. An object with a conditional causal power (I am adding “causal”) can have a certain causal power simpliciter if the object also has certain other properties. To use Shoemaker’s example, being knife-shaped is a conditional causal power; when it is combined with “made of wood,” the object having these properties has the power to cut butter. When the object is made of steel, it would have the power to cut wood. And so on. From here on, I propose to focus on forward-looking causal features only and also to speak of causal powers of properties. And I will think of the causal profile of a property as the set of its causal powers. I do not believe that this simplification will make a difference in what follows. So I will be working with a simplified definition of the subset account, in the following form: Q is a realizer of P just in case P’s causal powers are a subset of Q’s causal powers—or P’s causal profile is a subset (part?) of Q’s causal profile.15 14 Whether the weaker assumption suffices for Shoemaker’s purposes seems an open question. Brian McLaughlin argues that Shoemaker’s account of realization requires a full-fledged causal theory; see his review of Physical Realization in the Notre Dame Review of Philosophy. 15 I am also simplifying the idea of a causal profile accordingly. I believe this more closely resembles the earlier version of the subset account in Shoemaker’s “Realization and Mental Causation,” in Physicalism and Its Discontents, ed. Carl Gillett and Barry Loewer (Cambridge: Cambridge University Press, 2001). At


271

How does this conception of realization elude the exclusion-style argument? Here is what Shoemaker says: The subset account obviously avoids the threat that the causal role of the realized property will be preempted by its realizers. It starts with the assumption that the realized property has a causal profile, and nothing in the account takes this assumption back (13). Let us return to pain and its (supposed) realizer Cfs. According to the subset account, this means that pain’s causal powers are a subset of Cfs’s causal powers. Pain’s causal profile is the set of its causal powers (or it is constituted by its causal powers). So when Shoemaker says his account “starts with the assumption that the realized property has a causal profile,” implying that this is a sufficient response to the epiphenomenalist threat, he must mean that the causal profile is nonempty. Similarly, the assertion that pain’s causal powers are a subset of the causal powers of its realizers must be understood to mean that they are a nonempty subset. There is no need to be pedantic here, but the empty set is a subset of any and every set, and the claim that the causal powers of mental properties are subsets of the causal powers of neural/physical properties is consistent with epiphenomenalism—that is, the claim that none of these mental properties have any causal powers. Here I just want to highlight Shoemaker’s “starting assumption,” namely this: Mental properties have causal powers—indeed, they have physical causal powers, at least causal powers of the sort that physical properties have. I find this procedure—the structure of Shoemaker’s argument—troubling and worrisome. Suppose you are impressed by one or another of various considerations that apparently argue for mental epiphenomenalism. The exclusion-style argument is by no means the only source of epiphenomenalist pressures. As far as I can tell, T.H. Huxley’s reasons for advocating epiphenomenalism had nothing obviously to do with the exclusion-style considerations. There are also more recent arguments directly based on the

a book symposium on his Physical Realization at the meeting of the American Philosophical Association, Pacific Division, April 2009, Shoemaker stated that he no longer wished to include references to backward-looking causal features, and seemed willing to revert to his earlier, and simpler, formulation. For some possible complications arising from including them, see McLaughlin’s review of Shoemaker’s book in the Notre Dame Review of Philosophy.

272


causal closure of the physical domain, for example.16 And we don’t need a second-order functionalist view of mental properties to motivate an epiphenomenalist argument. I think a simple line of consideration like the following, for example, is sufficient to stimulate doubts and anxieties about mental causal efficacy. Suppose I experience a sharp pain in my hand and jerkily withdraw the hand. We say: the pain caused the hand movement. That gives the mental event its causal due. But suppose we trace back the sequence of physiological events that culminated in my hand movement. I believe we understand the physiology of such limb movements pretty well. There is every reason to think that if we traced this causal chain backward, we will end up in a neuronal event somewhere in our nervous system as its initiating cause. It is highly unlikely that neuroscientific research will identify and recognize pain (qua a mental event) as a causal source in this causal chain; it is likely that the causal chain will entirely bypass the pain— unless of course we are prepared to say that the pain is the very same event as some event in the neural chain. If the pain causes my hand movement, it is difficult to imagine its performing the causal work by some telekinetic causal action; we are apt to think it must somehow ride piggyback on the neural/ physical causal chain, if that makes any sense. In any case, it seems clear that this line of thinking, which makes no use of realization or supervenience, can give rise to a doubt about mental causal efficacy. It seems to me that Shoemaker’s starting assumption that mental events have causal powers will do little to allay these worries. Shoemaker’s line of argument only gives us what seems only like a verbal transformation of the problem: the question whether mental properties have physical causal efficacy has been turned into another, but essentially equivalent, question, the question whether mental properties (nontrivially) have physical realizers. I think T.H. Huxley might respond to Shoemaker’s account thus: Now, you show me that mental properties have physical realizers! If the subset account is your favored notion of realization, then to avoid begging the question on mental causation, I suggest you take the subset definition of realization literally—as saying that P has Q as a realizer just in case P’s causal powers are a subset, possibly an empty one, of Q’s causal 16 See, e.g., David Papineau, Thinking about Consciousness (Oxford: Oxford University Press, 2002), chap. 1. However, Shoemaker’s mental causation is probably consistent with physical causal closure; this is so because, as I will argue, Shoemaker’s overall picture is strongly physicalist.


273

powers—and then try to show, by independent considerations, that the supposed causal powers constitutive of a mental property are indeed a nonempty subset of the causal powers of some certifiably physical property. That would have quieted my worries about Shoemaker’s attempt at a vindication of mental causal efficacy. What is clear is that the adoption of the subset definition of realization is not enough in itself to do the job. If “subset” in the definition means “nonempty subset,” the claim that mental properties have physical realizers directly begs the question. If “subset” means what it usually means, namely “possibly empty subset,” then we are back to square one; the hard part is to show that mental properties have a nonempty set of causal powers.

IV And I have a specific problem with mental causation within Shoemaker’s scheme. Consider the following kind of situation: Suppose M has P as one of its physical realizers, and on a particular occasion, an instance of M occurs in virtue of its realization by P. So there is a P-instance. Suppose further that M (think, e.g., pain) is constituted by causal powers C1 and C2, and that P’s causal profile consists of causal powers C1, C2, C3, and C4. Assume that the P-instance causes another physical property P* to be instantiated, and that for this causation to hold, C3 and C4 are also needed; that is, no property whose causal profile consists merely of C1 and C2 can cause a P*-instance. It so happens, we suppose, that P* is a realizer of another mental property M* (say, a sense of distress). So on this particular occasion, an instance of M (pain) is followed by an instance of M* (distress). So we have the following picture: M (C1, C2)

M*

Realized by

Realized by

P (C1, C2, C3, C4)

— causes —›

P*

274


Should we say in this situation that the M-instance caused the M*instance? I am strongly inclined to say yes—that that particular occurrence of pain brought about the feeling of distress. But on Shoemaker’s account we apparently must say no, since, ex hypothesi, the possession of causal powers C1 and C2 does not suffice to cause P*, the realizer of M* on this occasion. I believe that the fact that P, M’s realizer on this occasion, caused P*, M*’s realizer on this occasion, is sufficient to warrant the causal attribution, namely that M caused M* on this occasion. Does this lead to causal overdetermination? No. On the view I advocate, there are not two property instances on this occasion; if M is realized by P on this occasion, the M-instance is identical with the P-instance (and the two events have the same causal powers). This probably is an empirical scientific question, but I am inclined to believe that the diagram presented above must hold in typical situations involving mind-brain realization and causation. Appeals to “intuitions” perhaps don’t count for much. But consider the following possibility: it so happens that all physical realizers of M, not just P, have causal profiles such that each of them causes P*. As before M’s causal profile doesn’t suffice to cause P*. In such a case, M-instances will invariably be followed by M*-instances, and this holds as a matter of causal law. Even so, it seems that Shoemaker’s account doesn’t allow us to say that the Minstances cause their associated M*-instances. And that seems to me wrong. My recommendation has been that we identify a given M-instance with the P-instance that realizes it on that occasion. This means that the realized property instance has the same causal powers as its realizing property instance, and the kind of situation sketched above does not arise.17

V For these reasons, I am inclined to believe that a possible epiphenomenalist threat is not what distinguishes Shoemaker’s subset view from the standard 17 Shoemaker objects to this token identity on the ground that property instances can be identical only if the properties instantiated are identical. On his view, M and P are distinct properties (one realizes the other). I am not sure that this principle is plausible when P is a realizer of M; I would be willing to waive it in this case. If that looks ad hoc, my overall view of this area takes a deflationary attitude toward the propertyhood of functional properties; I am happy with functional predicates/concepts, and willing to recognize mental concepts and predicates, while taking a skeptical view of mental properties.


275

second-order view of realization. As far as I can see, the two views are pretty much in the same boat in regard to the problem of mental causation. I will now turn to what appears to me like a more significant difference between the two, a difference that I believe speaks in favor of the second-order view. Briefly, there is reason to believe that second-order realization is an explanatory relation. That is, to know that Cfs is a realizer of pain, in the second-order sense, enables us to explain why pain is experienced when Cfs occurs—or why pain and Cfs are correlated as they are. As you see, this is the so-called explanatory gap problem; so the claim is that if Cfs is a physical realizer of pain, we can explain why pain correlates with the neural state with which it correlates,18 and this bridges the explanatory gap between Cfs and pain. Note that the metaphor of “bridging” a gap is given a concrete sense—it is accomplished by deducing a pain fact from neural/physical facts. Consider the following array of statements: (1) Jones is undergoing Cfs at t. (2) In systems like Jones, tissue damage is apt to cause Cfs and Cfs in turn is apt to cause winces, groans, and escape behavior. (3) For something to be in pain is for it to be in a state that is apt to be caused by tissue damage and that is apt for causing winces, groans, and escape behavior. (4) Therefore, Jones is in pain at t. This seems to me like a good explanation of why Jones is in pain on the basis of the fact that she is suffering Cfs at the time. Notice that the deduction involves a law, in premise (2), and that it qualifies as a Hempelian deductivenomological explanation. Actually, I think whether, and in what sense, we consider this an “explanation” is less important than the fact that it shows us how pains can be deduced from Cfs and other facts at the physical/neural level alone. That is what could close the explanatory gap. You might ask: What about premise (3), which speaks about pain? The answer is that (3) is a definition, and that definitions don’t count as premises in a deduction; it is not a report of facts about pain—if it is a report of anything, it is a report on the concept of pain, or the meaning of the term “pain.” At least, we can say this: the foregoing deduction shows that on the second-order view of realization it is

18 I believe that Ned Block has formulated the explanatory gap problem as follows: Why do conscious states correlate with the neural states with which they correlate?

276


possible to deduce pain statements (or facts) from neural-physical statements (facts) plus conceptual truths.19 I have argued elsewhere20 that in contrast to the functional reduction of pain, the psychoneural identity reduction, which identifies pain with Cfs, claims that the so-called explanatory gap does not exist, rather than that it helps close the gap. If pain ¼ Cfs, there is just one thing and you need at least two things to create a gap. I think this is a perfectly valid way of dealing with the explanatory gap. When we are faced with an explanatory demand “Why p?”, there are two ways of meeting it: first, we can provide a correct answer to this why-question; second, we can show that p is false and that there is nothing to be explained. The explanatory request “Why p?” is improper and cannot be met if p is false. It seems to me that Shoemaker’s subset view is somewhere near the identity view on this explanatory issue. On the subset view, pain’s causal profile is a proper subset, proper part, of Cfs’s causal profile. If only the two profiles completely coincided, we could have the identity “pain ¼ Cfs” (assuming a full causal theory of properties), but on Shoemaker’s view we are not there yet and won’t ever get there, on account of the multiple realizability of pain. So where are we with the subset account of realization in regard to the explanatory gap issue? It obviously doesn’t allow the kind of reductive explanation of pain in terms of Cfs that I sketched above. On the other hand, since it eschews the identification of pain with Cfs, it prima facie seems that it cannot adopt the identity theorist’s strategy and deny that there is such an issue. So does this put the subset view in an untenable situation? Unlike the identity theory, it must acknowledge the explanatory gap as a problem, but unlike the second-order view, it is not in a position to provide an explanation to close the gap. But things are almost never as neat as that. Talk of a “gap” is a metaphor. Perhaps, the subset theorist could argue that, like the identity theory, the subset view doesn’t allow a gap between a mental property and its realizers. The causal profile of a mental property is a part of those of its realizers, and

19 This accords with the idea of reductive explanation promoted by David Chalmers and Frank Jackson, in their “Conceptual Analysis and Reductive Explanation,” Philosophical Review 110 (2001): 315–360. 20 In “Reduction and Reductive Explanation: Is One Possible Without the Other?”, Essay 10 of this volume.


277

there cannot be a “gap” between a thing and a part of that thing, any more than between something and the very same thing. In any event, I believe there are issues to think about in this area.

VI To my knowledge, Shoemaker doesn’t say outright that he embraces physicalism. At the very end of his “Introduction,” he writes: Nevertheless, much of this work will be concerned with the physical realization of mental properties, and this does require the truth of physicalism. I will not undertake to establish the truth of physicalism, or to defend it against standard objections; my concern will be with what must be true of mental properties and their instances if physicalism is true (9, emphasis added). His initial characterization of physicalism is contained in this sentence: But if physicalism is true, all of these [ostensibly nonphysical] properties must in some sense be determined, constitutively rather than just causally, by physical properties or physical states of affairs (1). With the notion of realization in hand, Shoemaker offers a more exact statement of physicalism: . . . it is arguable that [the notion of realization] provides the most revealing characterization of physicalism itself: physicalism, we can say, is the view that all states and properties of things, of whatever kind, are physical or physically realized (1). It would seem fair, though, to take Shoemaker of this book to be a physicalist. In chapter 6, he defends the claim that qualia have physical realizers. Although I won’t take that up for discussion, this defense seems unconditional; it isn’t merely that the truth of physicalism requires it (obviously it does); it seems that Shoemaker unconditionally advocates the position that qualia are physically realized. What more can we expect from a physicalist? And yet Shoemaker rejects reductive physicalism, or type physicalism, the position that mental properties are physical properties. On pp. 10–11, he describes the familiar multiple realization argument against type physicalism

278


in a way that I can only take to imply his endorsement. In a footnote on p. 11, he says that the exclusion argument could be stopped if pain could be identified with a neural property, but that the multiple realization of pain does not allow it. And his rejection of the view that an instance of a realized property can be identified with an instance of its realizer (on that particular occasion) leaves no doubt as to his negative stance on type physicalism. But the large picture Shoemaker presents to us about the relationship between the mental and the physical strikes me as a strong form of physicalism, in fact, reductionist type physicalism. I think there are some subtle differences between reductionist physicalism and type physicalism, but I won’t go into that, except to note that while type physicalism is a form of reductionist physicalism, the converse need not be the case.21 So then, why does Shoemaker’s overall scheme impress me as reductionist physicalism, perhaps even type physicalism? On an intuitive level, the picture I have of Shoemaker’s scheme is something like this: The fundamental ontological items of this world are physical causal powers—or Shoemaker’s “causal features.” These are packaged, or bundled, into properties, or Shoemaker’s “causal profiles.” Of course, not every set of causal powers need to turn out to be, or coincide with, a property; to constitute a property, the causal powers of the set must, at least, be able to be co-instantiated in an object. I don’t know whether this co-instantiability should be taken as a purely metaphysical notion, or understood in relation to fundamental laws of physics. In any event, the picture I have is a world of properties which are, or are constituted by, bundles of causal powers, all of them physical causal powers. I won’t be surprised if this misses many of the subtleties of Shoemaker’s constructions and proposals, but, as I said, that is the intuitive picture I come away with. And that picture is a starkly physicalist one. In this picture, where do mental properties find a place? If all properties are bundles of physical causal powers, or physical causal features, what makes some properties mental and others physical? Can there be a principled distinction between them? As far as I know, Shoemaker does not address, on any level, the question what makes mental properties mental, or what makes physical properties physical. On p. 20, he distinguishes between “mental causal features” and “physical causal features.” He writes: 21 Briefly, because reductionist physicalism could be a token identity thesis claiming that mental property instances are identical with physical property instances.


279

Let’s say that a forward-looking causal feature is a mental causal feature if the properties referred to in specifying it are mental properties, and that it is a physical causal feature if the properties referred to in specifying it are physical properties. Following Shoemaker, let Br be the property of having the belief that it is raining. We can take this property to include, again following Shoemaker, the causal feature of “being such that if it is instantiated together with the desire to keep dry and the belief that umbrellas keep off rain, this results in the subject’s taking an umbrella” (19). I am not sure whether this causal feature counts as mental or physical as it is specified by reference to both mental and physical properties (e.g., desire, umbrella). Shoemaker points out that, on the subset view, if a physical property, P1, is a realizer of Br, the aforementioned causal feature must be part of P1’s causal profile—it is one of the causal powers that constitute P1. But, as Shoemaker goes on to say, those other mental states mentioned in the specification of this causal feature, a belief and desire, must themselves be physically realized. What Shoemaker doesn’t say is whether or not this and other such mentalistic causal features—“mentalistic” because their specifications refer to mental properties even though not exclusively— are in some sense ultimate, or they will in the end be cashed out in purely physical terms. That is, whether the references to mental properties in the specifications of these causal features are ultimate and ineliminable, or they will at some point be eliminated in favor of physical properties. It is possible that Shoemaker says something about this in his book; if he does, I have missed it. I raise this question because it is relevant to the correctness, or approximate correctness, of my overall picture of the Shoemaker world, the world Shoemaker envisages under physicalism. The reason is that one might claim my picture is wrong because, on Shoemaker’s view, a physical property can have nonphysical causal powers (or Shoemaker’s mental causal features); for example, property P1 above, which realizes believing that it is raining, has in its causal profile the power to cause someone to carry an umbrella when that person also has the desire not to get wet, and so forth. So it is not the case that all physical properties are constituted by purely physical causal powers; many of them (including those that realize mental properties) will include nonphysical causal powers, or causal powers that are not purely physical, in their causal profiles.

280


Well and good. But the question this raises for me is the following. If these nonphysical causal powers are ineliminably constitutive of physical properties, that is, if they cannot somehow be eliminated so that the causal profiles definitive of physical properties are constituted exclusively by purely physical causal powers, then that may render Shoemaker’s characterization of physicalism inadequate. As you will recall, physicalism, for him, is the thesis that “all states and properties of things, of whatever kind, are physical or physically realized” (1). If mental properties are real and if they are realized by physical properties that essentially include nonphysical causal features or powers, then physicalism as characterized by Shoemaker cannot count as a form of physicalism. For the Shoemaker world as pictured would include in its ontology mentalistic causal features as ineliminable fundamental entities. On the other hand, if physical properties are constituted by purely physical causal powers and mental properties are physically realized in Shoemaker’s sense, there seems no good reason not to consider these supposedly mental properties to be physical properties, pure and simple. They, like any other antecedently certified physical properties, are constituted by purely physical causal powers and individuated by their purely physical causal profiles. What I am suggesting, then, is a dilemma for Shoemaker: Either his physicalism does not qualify as physicalism, or his physicalism is a version of reductionist type physicalism.

VII I have argued that Shoemaker’s subset account of realization is not able to resolve the problem of mental causation all by itself, and that it must at least be supplemented with an independent argument showing that mental properties indeed have nonempty sets of causal powers. I have also argued that Shoemaker’s overall picture is a strongly physicalist one, and that his account of mental properties is best viewed as a form of type physicalism, not nonreductive physicalism. But I have not argued that subset realization is not an interesting idea, or that it does not have significant applications to philosophical issues. It may well have a potential for illuminating applica-


281

tions beyond the mind-body problem; Shoemaker’s Physical Realization, in which the idea is put to work on a number of issues in general metaphysics, makes a positive case for the fruitfulness of the concept. I don’t believe it’s productive, or even meaningful, to ask which account of realization is the correct one, the more accurate one, and so on. I don’t think there are facts of the matter to settle such questions. What we have before us are two relationships defined for properties. The subset account introduces a relation between properties defined by one property’s causal powers including those of another. The second-order account introduces a relation that holds for two properties just in case one of them occupies the causal role that defines the other. There is no need to think of these two relations as competitors in any sense that matters. Whether mental and physical properties are related by these relations is a question that has to be answered on further considerations, probably both conceptual and empirical. It is possible that the psychophysical relation is an instance of both of these relations, or one but not the other, or perhaps neither. If it is a case of, say, subset realization, we could explore its implications; similarly if it is an instance of second-order realization. And the two concepts may have applications outside the mind-body problem, in metaphysics and philosophy of science. Exploring the relationship between the two realization relations can itself be an instructive philosophical problem. So I say let a thousand realization concepts bloom! There is of course no guarantee that each of the thousand, in fact any of them, will turn out to be an interesting and useful concept.

14 Why There Are No Laws in the Special Sciences: Three Arguments

So, then, why is there anything except physics? . . . Well, I admit that I don’t know why. I don’t even know how to think about why. I expect to figure out why there is anything except physics the day before I figure out why there is anything at all . . . 1 Jerry Fodor There is physics. The rest is engineering.2 J.J.C. Smart In science, there is only physics. All the rest is stamp collecting.3 Sir Ernest Rutherford

Physicalism, Physics, and the Special Sciences The central claim of physicalism is the proposition that the physical domain is all-encompassing. All things in space-time—all bits of matter and their For helpful comments I am indebted to Ronald Endicott, Marc Lange, Barry Loewer, Brian McLaughlin, Alyssa Ney, and Jeff Poland. 1 Jerry Fodor, “Special Sciences: Still Autonomous After All These Years,” Philosophical Perspectives 11 (1997): 149–163; the quote is from p. 161. In 2004, Barry Loewer gave a colloquium paper “Why There Is Anything Except Physics” at Brown University, and this talk has stimulated me to think about the issues discussed in this paper. Loewer’s paper was later published in Being Reduced, ed. Jakob Hohwy and Jesper Kallestrup (Oxford: Oxford University Press, 2008). 2 J.J.C. Smart, Philosophy and Scientific Realism (London: Routledge & Kegan Paul, 1963). This is not a verbatim quotation. Smart’s view will be discussed in detail later. 3 This is one version of the famous remark attributed to the noted physicist in the early 20th century.

NO LAWS IN THE SPECIAL SCIENCES

283

aggregates, however complex—belong in this domain and behave in accordance with the prevailing laws of physics. This has an immediate corollary, the thesis that this domain is comprehensive and closed, from the causal and explanatory point of view, in the following sense: If any physical event has a cause, or an explanation, it has a physical cause, or a physical explanation.4 We also expect fundamental physical laws to be reasonably simple and few in number, and formulated in terms of a manageable number of fundamental properties, magnitudes, and forces. Sir Ernest Rutherford and J.J.C. Smart notwithstanding, few of us today would be inclined to think that physics is the only science. There are also the “special sciences,” like biology, geology, astronomy, psychology, cognitive science, and the rest. Most of us believe that these disciplines are no less scientific than physics, even though their scope and aspirations may be more limited and modest. We expect these sciences to aim at discovering lawful regularities and causal connections in their respective domains and often succeed in this effort, thereby generating law-based explanations and predictions and helping us control events and phenomena of interest to us. For most nonscientists, the special sciences, such as biology, geology, and psychology, enjoy greater salience and accessibility, and command a wider interest, than basic physics which, at its deeper levels, can be baffling and intimidating if not outright incomprehensible. It is fair to say that sciences like biology, psychology, and economics have a more prominent and direct impact on our lives, from our daily copings with objects and people around us to the control of our physical and biological environments and the management of a nation’s finances and economy. A need to understand how the special sciences are related to physics and, in particular, how laws, explanations, and causal claims of the special sciences stand vis-a`-vis those of physics arises from two sources. First, as noted, the domain of physics is thought to be all-encompassing; it includes all objects and events in the space-time world. In contrast, each special science is “special” just because its domain is a specially demarcated subregion of the physical domain. This means that all entities and phenomena in the domain of any special science also belong in the wider domain of physics and come under the jurisdiction of physical laws. Further, as earlier noted, 4 This is the way the closure principle is often formulated but as stated it is slightly defective. It allows the following causal sequence “ . . . p1 causes m causes p2 . . . ,” where p1 and p2 are physical events and m is a nonphysical event. Note: by transitivity, p1 causes p2, and p2, therefore, has a physical cause.

284


there is a strong sense in which physics is, or is thought to be, causally closed, in that it does not tolerate injections of nonphysical causal influences into its domain; nor does it allow the possibility of physical phenomena receiving explanations in terms of nonphysical causal agents. This prompts us to wonder how the laws and explanations of a special science are related to those of physics. If the phenomena investigated by the special sciences are part of the all-encompassing physical domain, how can there be specialscience laws and explanations in addition to physical laws and explanations? That is, how are special sciences possible? And even if they are possible, do we need them? Why shouldn’t developed physics meet all our needs? Or, to repeat Fodor, why is there anything except physics? Moreover, unless we take a radically irrealist or instrumentalist stance toward the special sciences, we must grant that they, like physics, are about this one world that we inhabit. All our sciences aim at telling us something true about what goes on in the same world—a true story about some aspects of this one reality—and we think that at least sometimes they succeed. So it is perfectly natural for us to want to know how the stories told by these sciences hang together, among themselves and with the story that physics tells us. We want to be able to piece together these stories, stories about genes and photosynthesis, about cognition and behavior, about consciousness and the emotions, about the birth and death of planetary systems, about inflations and employment rates, and the rest, into one coherent picture that makes sense to us. Something like this presumably was the goal of the philosophers, like the early positivists, who promoted the idea of “unified science”—that all the sciences form a “unity” in some clear and substantive sense. As everyone knows, few philosophers now subscribe to such an ambitious monolithic view—either that science is unified in its language or subject matter, or that some identical methodology regulates all the sciences. The idea of a unified science has been out of fashion well over half a century. But do disunity and disorder prevail everywhere as many have claimed?5 (If they do, should they be allowed to?) What is a realistic picture of the relationship that the special sciences bear to basic physics? In this paper, I hope to make a start on exploring these issues. But it will be only a start because I am going to focus on a single issue here, the 5 See, e.g., John Dupre´, The Disorder of Things (Cambridge, Mass.: Harvard University Press, 1993); The Disunity of Science, ed. Peter Galison and David J. Stump (Stanford: Stanford University Press, 1996).


285

question whether there are laws, or “strict” laws, in the special sciences. (However, I will not explicitly take up the question whether there are strict laws in basic physics.) Obviously this issue is directly relevant to the question in what ways the special sciences differ from, and are related to, basic physics. It is also relevant to questions about the nature of explanations offered by the special sciences and the status of causal claims made in these sciences. These further issues are important and, I believe, deserve serious attention. But my present concern is with the question of special-science laws. More specifically, I will present three arguments, each with substantial presumptive plausibility, to the effect that there are no laws in the special sciences. The first of these begins with a close look at Donald Davidson’s famous, and famously opaque, argument for the claim that psychology is “anomalous”—that is, psychology has no laws. Although the argument I will formulate goes considerably beyond Davidson, I hope that the reasoning leading up to it will be of interest as a Davidson exegesis to those who know his work in this area. The second argument will take as its starting point J.J.C. Smart’s claim that biology, unlike physics, is not in the business of discovering and formulating its own laws, and that biology is akin to engineering in that it borrows, and makes use of, laws from physics and chemistry. The final argument to be considered will be based on considerations on the nature of special—science properties—in particular, the widely accepted view that these properties are “multiply realizable” in diverse physical structures.6

Davidson Against “Strict” Laws in Psychology In a series of influential papers published in the 1970s, Donald Davidson advanced the claim, which many found surprising and not credible at the time, that there can be no laws, or no “strict” laws at any rate, about psychological phenomena. This is the thesis he called “the Anomalism of

6 To those who are inclined to retort at this point “Whoever would have thought that there are strict laws in the special sciences anyway!”, all I can say is “Perhaps, but it surely wouldn’t hurt to know why.”

286


the Mental.”7 According to him, there are neither psychophysical laws, laws connecting psychological with physical phenomena, nor purely psychological laws, laws connecting psychological phenomena with other psychological phenomena. By “psychological” phenomena, he means content-carrying intentional/cognitive events and states, like beliefs, desires, and intentions (we may include their analogues in cognitive science and psychology), and he explicitly excluded from his consideration sensations and other sensory/ qualitative states and events. As the title of one of his papers on the topic, “Psychology as Philosophy,”8 suggests, Davidson held, at least at the time, that psychology is not, and cannot be made into, a science, and that it is more like philosophy (perhaps hermeneutics) than a science. He apparently thought that the discovery of laws and the use of laws to formulate explanations and predictions is a defining characteristic of science. Since, according to his mental anomalism, no laws are to be found about psychological phenomena, psychology does not qualify as a science. Second, the reason why there are no laws in the mental domain is important. On Davidson’s view, it is of the essence of intentional psychological states like belief and desire that they are regulated by normative principles of rationality and coherence, and this precludes a causal-predictive investigation and nomological systematization of these phenomena.9 Here we will set aside Davidson’s claim that psychology is not a science and his emphasis on the normative and rational aspects of mentality. What I am interested in, because this is directly relevant to our present concerns, is Davidson’s argument for mental anomalism, the thesis that there are no “strict” laws about psychological phenomena. The final phase of Davidson’s argument is contained in the following short, and notoriously cryptic, paragraph from “Mental Events”: It is not plausible that mental concepts alone can provide [a comprehensive and closed framework like that of physics], simply because the mental does not, by our first principle, constitute a closed system. Too much happens to affect the mental that is not itself a systematic part of the mental. But if we combine this observation with the conclusion 7 The most important paper in which this is argued is “Mental Events,” reprinted in Davidson, Essays on Actions and Events (New York and Oxford: Oxford University Press, 1980); first published in 1970. 8 Reprinted in Davidson, Essays on Actions and Events; first published in 1974. 9 See “Mental Events.” For Davidson’s later take on these issues see “Could There Be a Science of Rationality?”, in his Problems of Rationality (Oxford: Clarendon Press, 2004); first published in 1995.


287

that no psychophysical statement is, or can be built into, a strict law, we have the Principle of the Anomalism of the Mental: there are no strict laws at all on the basis of which we can predict and explain mental phenomena.10 What is a “strict” law? There is some interpretive uncertainty about exactly what Davidson meant by strict laws. One condition on which most are agreed is that, unlike “mere” or “rough” generalizations or “ceteris paribus” laws, strict laws allow absolutely no exceptions.11 There is the further idea that strict laws occur only as part of “comprehensive” and “closed” theories over a domain, but it is not an idea that is easily made precise in this context, and in any case it isn’t clear whether this is to be taken only as a fact about strict laws or it is constitutive of the very idea of strictness. Fortunately, we need not settle this issue, for the only property of strict laws we will be using is their exceptionlessness; for our purposes, then, strict laws are laws that hold without exceptions and tolerate none. In any case, our interest in the quoted paragraph above stems largely from its potential generalizability beyond the psychological domain—that is, we are interested in the question whether an argument along similar lines can be constructed for the conclusion that the special sciences in general are “anomalous.” Such an argument would show that there are no “strict” explanatory/predictive laws in biology, geology, and other special sciences any more than in psychology. If an argument to this effect can be constructed, as I believe it can be, that would be a matter of great interest. But what is Davidson’s argument? Return to the quoted paragraph above: Davidson, at this point in “Mental Events,” is assuming that he has already established psychophysical anomalism, the thesis that there are no strict laws connecting psychological and physical phenomena. To derive the full anomalism of the mental, the claim that there are no laws at all about psychological phenomena, what he needs is psychological anomalism, the proposition that there are no purely psychological laws, laws connecting psychological phenomena with other psychological phenomena. His reasoning in the paragraph appears to go like this: (1) The mental domain is not causally closed—that is, some mental events are caused by nonmental events. 10 “Mental Events,” p. 224. 11 Statistical laws can be strict; if there are basic physical laws that are statistical, they do not allow “exceptions”—any more than nonstatistical, deterministic laws do.

288


(2) Therefore, there are no purely psychological laws (psychological anomalism). (3) There are no psychophysical laws (psychophysical anomalism). (4) Hence, there are no laws about psychological phenomena. For present purposes, we consider (3) as given; that is, we assume that Davidson has adequately established psychophysical anomalism. Given (2) and (3), the anomalist conclusion (4) immediately follows. The only significant question for us, therefore, concerns the transition from (1) to (2)—how from the failure of causal closure of the psychological domain, Davidson is able to derive the nonexistence of psychological laws. More than a few philosophers have pondered this question, trying to develop a line of reasoning that would show how the derivation might work, or reveal what Davidson might have had in mind. To my knowledge, none have succeeded. One intractable obstacle is Davidson’s appeal to his “first principle” as the source of the failure of causal closure for the mental domain. This principle is the following: Principle of causal interaction. “At least some mental events interact causally with physical events.”12 The chief exegetical difficulty lies in the fact that this statement is completely symmetric as between the mental and the physical. The statement that “at least some mental events interact causally with physical events” is equivalent to the statement that “at least some physical events interact causally with mental events.” So if this principle entails, or warrants, the claim that causal closure fails for the psychological domain, it becomes unclear, in fact wholly mysterious, why it does not also entail a parallel conclusion, namely that causal closure fails for the physical domain as well. But it is one of Davidson’s bedrock beliefs that the physical domain differs from the mental domain—in fact, from all other domains—precisely in that it alone is governed by strict laws, and that it alone is a “comprehensive and closed system” of the kind that allows strict laws. It would seem that if we could derive the nonexistence of psychological laws from his “first principle,” as Davidson is apparently suggesting, we should also be able to derive the nonexistence of strict physical laws from the very same premise! So clearly something is amiss with Davidson’s argument, or our reading of it. 12 “Mental Events,” p. 208.


289

I have come to believe that Davidson’s invocation of his “first principle” was a crucially misleading factor, a stumbling block for his readers. Notice that the events that the principle speaks of are most naturally taken as Davidsonian token events, that is, concrete dated occurrences, not kinds or properties of events (or events construed as property instantiations13). So when Davidson says some mental events causally interact with physical events, what this means is: Token causal symmetry. Some token events falling under mental event kinds cause, and are caused by, some token events that fall under physical event kinds. As earlier noted, the first principle understood this way is symmetric for mental and physical events. Possible causal asymmetries between the physical and the mental become visible when—and, I believe, only when—we consider event types or kinds. For example, consider the following possible mental-physical causal asymmetry, something that isn’t implausible and seems consistent with Davidson’s views in this area: Type causal asymmetry. Whenever a mental event, m, causes a physical event, m is also a physical event, that is, m falls under some physical event kind P, whereas it is not the case that when a physical event, p, causes a mental event, p must also be a mental event by falling under some mental event kind M. In the latter case, p may be a “purely” physical event, an event that falls under no nonphysical kinds. This asymmetry is entirely consistent with the symmetry of Davidson’s Principle of Causal Interaction. But reading causal asymmetry into Davidson this way is something un-Davidsonian because the basic idea here is that when a mental event does any causal work (passive as well as active), it does so in virtue of falling under a physical kind—that is, the causal efficacy it has is due to the fact that it is an event of a certain physical kind. Of course, when a physical event is implicated in a causal relation, it is only the fact that it is an event of a certain physical kind that is causally relevant. Remember that on Davidson’s event ontology, an event is mental or physical according as it falls under a mental kind or under a physical kind. The “or” here is 13 See my “Events as Property Exemplifications,” reprinted in Supervenience and Mind (Cambridge: Cambridge University Press, 1993); originally published in 1976.

290


nonexclusive: an event can fall under both a physical kind and a mental kind, and therefore it can be both a physical event and a mental event. The mental/physical asymmetry that I am suggesting, therefore, depends on an assumption that Davidson, at least “officially,” rejects, namely the thesis that properties of events, or kinds under which events fall, are causally relevant, and that it is in virtue of falling under the event kinds under which they fall that events have the causal efficacy that they have, and enter into the causal relations into which they enter. Davidson has rejected this kind of talk as confused and misguided, insisting that causation is an extensional two-place relation over token events, and that talk of properties “in virtue of which” events cause, or are caused by, other events is incoherent.14 In any case, what these reflections show is that causal closure must be formulated in terms of event kinds and types, or properties, not solely in terms of Davidsonian token events. Standardly, physical causal closure is stated like this: (C) If a physical event has a cause (occurring) at t, it has a physical cause at t. This will not work as intended if “event,” or “physical event,” is understood within a Davidsonian framework of token events. Something like (C) works only because of the tacit assumption that a physical event is the instancing of a physical property. If an object x’s instantiating, or having, a physical property P has a cause—that is, if an event causes this object to have P—then there must be a physical event, namely an event consisting in an object’s instantiating some physical property P*, that causes x to instantiate P. Perhaps it is more perspicuous to put this in terms of explanation: If there is a causal explanation of why an object has physical property P, there must be a causal explanation of this fact in terms of some object’s having a certain physical property P*. I propose that we consider the following two causal closure theses formulated in terms of properties/kinds:

14 Davidson has not been entirely consistent on this point. See his “Thinking Causes” (in Mental Causation, ed. John Heil and Alfred Mele (Oxford: Clarendon Press, 1993)) in which he begins by strongly affirming this point and then goes on to give his own account of the “causal relevance” of mental properties in terms of supervenience.


291

Physical causal closure. If x’s having a physical property P has a cause that occurs at t, there is a physical event, an object y’s having a physical property P* at t, such that this event, in virtue of being an instancing of P*, causes x to instantiate P. Mental causal closure. If x’s having a mental property M has a cause at t, there is a mental event, an object y’s having a mental property M* at t, such that this event, in virtue of being an instancing of M*, causes x to instantiate M. I believe these two principles could be stated for the Davidsonian ontology of token events; however, these restatements would likely be more complex and cumbersome, more so than the already convoluted formulations given here. In any case, physical causal closure as stated is arguably true; if you believe, as with almost all physicalists, in the comprehensiveness and closedness of physics, our closure principle seems to capture this idea quite nicely—in the way a bare statement like (C) taken to be about token events does not. In contrast, mental causal closure as stated seems plainly false: as Davidson says, “too much happens to affect the mental that is not a systematic part of the mental.” Cases in which mental properties are caused to instantiate not by instantiations of other mental properties but by instantiations of nonmental, physical properties are familiar and numerous; we all know that retinal stimulation causes visual experience, that tissue damage causes pain, and all the rest. The mental/physical asymmetry represented by physical causal closure and the failure of mental causal closure is a good start. But it is not enough to yield the asymmetry we seek, which is the impossibility of any strict laws in the mental domain in spite of the presumed existence of such laws on the physical side. The reason is simple: psychological anomalism asserts that there are no strict psychological laws at all whereas the failure of mental causal closure only means that there are some mental events—one, at least— whose causation is not covered by a strict psychological law. Here is where the failure of mental causal closure gives us an idea: What if causal closure failed for every putative case of mental causation—that is, everywhere in the mental realm? If that should happen, that must be because no strict mental law could cover any mental event in a causal relation. I suggest then that we consider the following proposition:

292


Massive failure of causal closure for the mental.15 Let M and M* be mental event kinds and consider a putative causal relation from an M-event to an M*-event. Then there always is a physical event kind P such that if a P-event were to occur along with the M-event, that would prevent the M*-event from occurring. That is, every mentalto-mental causal relation is liable to disruption by the occurrence of a physical event. This, I believe, is not an implausible thesis. We can see that if mind-body supervenience holds, the claim is likely true. For M* to occur, one of its physical supervenience bases must occur. It seems that after the putative cause M occurs we can, perhaps in all cases, intervene and make sure that none of the physical bases of M* occur. (We can always have the brain destroyed—if it must come to that!) Davidson has said that “too much” that is nonmental happens to causally affect the occurrence of mental events; the thesis of massive failure upgrades “too much” to “always” (but weakens “happens” to “could happen”). Every instance of mental-to-mental causation is vulnerable to disruption by the occurrence of a nonmental event. Given the massive closure failure for the mental, we can quickly show that there are no strict laws in the mental domain. Suppose M ! M* is a strict psychological law. This law suffices to ground a causal relation between an M-event and an M*-event.16 Since the M ! M* law is strict, it admits of no exceptions and this means that there can be no interfering physical event to disrupt the M-to-M* causation. This contradicts the thesis of massive failure of causal closure for the mental. It follows that M ! M* is not a strict psychological law, and that there are no strict laws on the mental side.17 To wrap it up: a strengthened form of the failure of mental causal closure gives us psychological anomalism. To complete the argument for the full anomalism of the mental, we need only to observe that this thesis is a conjunction of psychological anomalism and psychophysical anomalism. 15 It may be a bit misleading to call this a failure of “causal closure,” whether massive or otherwise. If this bothers you, just pay attention to what it says (and call it anything you like!) 16 Davidson requires that causal relations be subsumed by strict laws; see “Mental Events.” 17 It should be clear that the foregoing considerations (including Davidson’s own) apply to what may be called “transition laws,” laws that connect psychological events diachronically. They do not affect the possible existence of synchronic laws involving psychological properties, a possibility that Ron Endicott pointed out to me. However, it will be seen that the second argument below, against special-science laws, affects laws of that sort as well.


293

For the present purposes, as you may recall, we have conceded the latter to Davidson. Even without this concession, what has been shown is not insignificant: there are no strict laws connecting psychological kinds with other psychological kinds.

The Fragility of Higher-Level Causation I believe the reasoning is plausibly generalized to other special sciences. If there are no strict psychological laws and psychological causal relations are in principle vulnerable to disruption, the same must hold for the higher special sciences that psychology underlies, such as sociology and economics. So let us consider biology, a lower-level science in relation to psychology: Can there be strict biological laws and undisruptable biological causation? Not likely: nonbiological physical events (exposure to high levels of radiation, the unavailability of necessary nutrients, ecological changes, natural disasters, and so on) can always intervene to break up biological causal processes. Biological entities are complex aggregates of physical elements subject to physical laws, and something can always, and often does, go wrong with the physical underpinnings of biological processes to interrupt their “normal” progress. Normal developmental processes at the biological level can be affected by all sorts of lower-level nonbiological occurrences. The progress of a pathological condition can be interrupted, or be steered in another direction, by intervention at the physicochemical level, such as administration of drugs or surgical procedures. It is plausible, then, that a generalized analogue of the massive closure failure for the mental domain holds: General failure of special-science causal closure. For any putative causal relation from an S1-event to an S2-event, involving special-science kinds S1 and S2, there always is a lower-level condition C such that if C were to occur along with the S1-event, that would disrupt the causal process and prevent the S2-event from occurring.18 18 Wouldn’t this mean that there is the following strict law: “If S1 occurs and C occurs, then S2 does not occur”? Not necessarily. Even when C occurs, if another lower-level event C* also occurs, it may counteract C and S1 may still lead to S2; and so on. Notice also that the consequent of the law is to the effect that a certain psychological phenomenon—to pick an arbitrary special science—does not occur. The nonoccurrence of a psychological phenomenon is not itself a psychological phenomenon. For this reason, it would be problematic to consider a law of this form a “psychological” law.

294


From this, by the argument sketched above for the mental domain, it follows: The nonexistence of strict special-science laws: For any special-science kinds S1 and S2, there is no strict law of the form S1 ! S2. From this perspective, Davidson’s psychological anomalism, the thesis that there are no strict laws about mental phenomena, is only a special case. A more general thesis is that there are no laws in any of the special sciences. Strict laws can be found only in fundamental physics. The basic argument for this claim is the same as the argument for mental anomalism: causal/ nomological relations at higher levels are always susceptible to disruption from below—that is, to interference from events occurring at a lower level.19 The physical level being the bottom level, physical causal relations at the fundamental level suffer no similar vulnerability, and this allows for the possibility of strict physical laws.20 Davidson is not alone in advancing the thesis that there are no strict laws in some special-science domain. Philip Kitcher, for example, has argued against the reducibility of classical genetics to the lower-level sciences (in particular, molecular biology) in part on the ground that there are no “laws” in classical genetics. On the Nagelian model of bridge-law reduction,21

19 To make the argument fully parallel to Davidson’s argument, we will need to establish auxiliary premises that correspond to Davidson’s psychophysical anomalism—that is, the claim that there are no bio-physical laws, that there are no geological-physical laws, etc. (as Abe Roth has mentioned to me). I believe a general argument to this effect could be constructed on the basis of the same considerations invoked here. There is also the possibility of constructing the needed arguments on the basis of the second and third arguments against special-science laws to follow. For an argument based on considerations of multiple realization to show the nonexistence of laws in the social sciences, see John T. Roberts, “There are no Laws of the Social Sciences,” in Contemporary Debates in Philosophy of Science, ed. Christopher Hitchcock (Oxford: Blackwell, 2004). Also of interest is Lee McIntyre, “Davidson and Social Scientific Laws,” Synthese 120 (1999): 375–394. 20 What if there is no bottom level, there being a lower level for each microphysical level? Would that show that there are no strict laws anywhere? The answer is not clear; whether or not the considerations put forward in the paper (in particular, the idea of “interference from below”) will generally apply to microphysical levels seems like an empirical physical question, and I have no idea whether there is received wisdom regarding this question. On the possibility of there being no bottom level, see Ned Block, “Do Causal Powers Drain Away?”, Philosophy and Phenomenological Research 67 (2003): 133–150; Jonathan Schaffer, “Is There a Fundamental Level?”, Nouˆs 37 (2003): 498–517. 21 Ernest Nagel, The Structure of Science (New York: Harcourt, Brace and World, 1961). I have elsewhere argued that this is not an adequate model of reduction; see, e.g., “Reduction and Reductive Explanation: Is One Possible Without the Other?”, Essay 10 of this volume.


295

assumed by Kitcher, the reduction of one theory to another is accomplished by deriving the laws of the former from those of the latter (with “bridge laws” as auxiliary premises). But if there are no laws of genetics, there is nothing to reduce. Why are there no laws in genetics? According to Kitcher, if there are laws in classical Mendelian genetics, they must be about the transmission of genes. But he can’t find any such laws: . . . when we read the major papers of the great classical geneticists or when we read the textbooks in which their work is summarized, we find it hard to pick out any laws about genes. These documents are full of informative statements. Together, they tell us an enormous amount about the chromosomal arrangements of particular genes in particular organisms, about the effect on the phenotype of various mutations, about frequencies of recombination, and so forth.22 But Kitcher doesn’t see any laws among these otherwise informative statements. Take, for example, Mendel’s “second law,” roughly to the effect that the probabilities of a gamete receiving any of the possible genetic combinations are equal. Kitcher points out that this “law” does not hold in general. Once we understand that genes are chromosomal segments, we see that alleles that are on the same chromosome will tend to be transmitted together. Kitcher goes on to argue, plausibly, that various possible restrictions or emendations to Mendel’s second law will not make it completely precise and exceptionless.23 Let us accept Kitcher’s claim that genetic laws are not found in the scientific literature of genetics. But is there a principled argument that would show that, given the nature of genetics, or genetic properties, there could not be laws about gene transmission? Perhaps we find no laws of genetics in the scientific literature, but how do we deal with the simple retort that if we would only wait long enough, or allocate generous enough research funds, someone might come up with genetic laws, strict laws about genetic phenomena? When we look through Kitcher’s discussion with this issue in mind, we can discern a pattern of considerations that fits well with

22 Philip Kitcher, “1953 and All That: A Tale of Two Sciences,” Philosophical Review 93 (1984): 335– 373. The quote is from pp. 340–341. 23 Kitcher also argues against possible “bridge” laws connecting biological predicates (e.g., “is a gene”) with molecular physical predicates. His argument here appears to be an instance of the classic multiple realization argument.

296


the quasi-Davidsonian argument just presented. As you will recall, the argument was that for any putative special-science law, there are “lowerlevel disruptors”—that is, conditions at lower levels such that were they to occur, the law would not hold. Return to the reason Kitcher offers for the failure of Mendel’s second law: once we see what lower-level phenomena are involved in genetic transmissions, we realize that Mendel’s law cannot hold in general. Or see what Kitcher says in explaining why it will not do to restrict Mendel’s second law to genes on nonhomologous chromosomes. He writes: “Unfortunately, this will not quite do. There can be interference with normal cytological processes so that segregation of nonhomologous chromosomes need not be independent.”24 Again, the idea is that lowerlevel interference can, and often will, disrupt higher-level regularities. All we need to add is that this is something that cannot be avoided, that it will bedevil all higher-level special-science properties everywhere and all the time. It is an instance of what we called the general failure of special-science causal closure: Higher-level regularities and causal relations are fragile— they are liable to breakdowns due to interference from below.

Smart on Biology as Engineering About two decades before Kitcher argued that there are no laws in genetics, J.J.C. Smart advanced, in his Philosophy and Scientific Realism (1963), the broader claim that there are no laws in biology or psychology, and, by implication, none in any other special sciences. Physics, according to him, is the only science whose business is to discover and formulate laws. He wrote: Not only do I deny the existence of emergent laws and properties, but I even deny that in biology and psychology there are laws in the strict sense at all. There are, of course, empirical generalizations. There are not any biological laws for the very same reason that there are not any laws of engineering. Writers who have tried to axiomatise biological and psychological theories seem to me to be barking up the same gum tree as would a man who tried to produce the first, second, and third 24 “1953 and All That: A Tale of Two Sciences,” p. 342.


297

laws of electronics, or of bridge building. We are not puzzled that there are no laws of electronics or of bridge building, though we recognise that the electronic engineer or bridge designer must use laws, namely laws of physics . . . I shall try to show that the important analogy is not between biology and the physical sciences but between biology and the technologies, such as electronics.25 I believe this is an insightful perspective—not only about biology but about other sciences such as geology and astronomy. What does Smart mean by “laws in the strict sense”? He writes: Physics and chemistry have their laws. For example, there are the laws of motion in classical mechanics, the laws of electrodynamics, and the equations of quantum mechanics. . . . These laws are universal in that it is supposed that they apply everywhere in space and time, and they can be expressed in perfectly general terms without making use of proper names or of tacit reference to proper names. Such laws I call laws in the strict sense.26 Smart’s “laws in the strict sense,” therefore, seem akin to, if not identical with, Davidson’s “strict laws.” For both, strict laws are wholly exceptionless and completely general, applying everywhere in the space-time world, and, according to Davidson, they are only found in theories that give a comprehensive and closed coverage of their domains. Both Smart and Davidson claim that there are strict laws only in basic physics.27 Smart conceives of the relation between physics and biology in analogy with that between physics and electronic engineering: From a logical point of view biology is related to physics and chemistry in the way in which radio-engineering is related to the theory of electromagnetism, etc. . . . Just as the radio-engineer uses physics to explain why a circuit with a certain wiring diagram behaves as it does, so the biologist uses physics and chemistry to explain why organisms or parts of organisms (e.g., cell nuclei), with a certain natural-history description, behave as they do.28 25 J.J.C. Smart, Philosophy and Scientific Realism (London: Routledge & Kegan Paul, 1963), p. 52. 26 Ibid. 53. 27 Or, in Davidson’s term, “developed physics.” See Davidson, “Thinking Causes,” in Mental Causation, ed. John Heil and Alfred Mele (Oxford: Clarendon, 1993). 28 Philosophy and Scientific Realism, p. 57.

298


But why does Smart think that there can’t be laws in biology—laws about genes, mutation, photosynthesis, and numerous other biological entities and processes? Prima facie, the existence of biological laws seems consistent with the fact, which we may concede to Smart, that “the biologist uses physics and chemistry to explain” biological phenomena. This could be true not because there are no biological laws but because biological laws are reducible to, or derivable from, laws of physics and chemistry. So we need to ask: what is it about biological entities and processes that make laws about them unavailable? What are the pertinent differences between them and the entities dealt with in physics that explain the nomological differences between them? Smart says that “there are no real laws of biology for the very same reason that there are no special ‘laws of engineering’.” For Smart, then, biological organisms are like radio receivers, internal combustion engines, and suspension bridges. In designing and building them, and in understanding their behavior (their malfunctions as well as their proper, designed-in behavior), we use laws of physics and chemistry. In engineering there may be useful rough and ready empirical generalizations and rules of thumb; but there seem no laws, or strict laws, about designing or manufacturing radios or internal combustion engines. The analogy with engineering seems apt and is thought-provoking, but it does not answer the question why there are no laws in biology—or, for that matter, the question why there no laws of engineering. That is, what do biology and engineering have in common that is responsible for there being no laws in either? It is not easy to give a clear interpretation or reconstruction of Smart’s thoughts on this question (it is to his credit that he does raise the question, however). After a meandering discussion, he settles for the claim that the structures of interest in biology, for example, cells, genes, and organs, are vastly more complex than those studied in physics. In his final paragraph on this issue, Smart writes: However, though there is not a sharp division in nature between the objects of the physical sciences and those of the biological sciences, there is, of course, a non-sharp division, which is one of complexity of structure. The methodological division does reflect this non-sharp division in reality.29

29 Philosophy and Scientific Realism, p. 61.


299

Smart appears to be saying that on account of the (“non-sharp”) differences in the complexity of their respective subject matters, we make a methodological decision, to put physics in the business of coming up with laws and give biology the lesser role of applying the laws discovered by physics. It is clear that this decision makes sense only if structural complexity has a decisive negative impact on the existence of laws. Such a simple connection between complexity and laws, however, is implausible, for, as we will soon see, structural complexity alone cannot preclude laws. What is of greater relevance is a point that Smart himself refers to when he calls biological entities “complicated and idiosyncratic structures” (emphasis added).30 He continues on to say “No one expects even all motor cars of a certain make and year to behave exactly alike,” and points out that a living cell is vastly more complex than an automobile. Here, what does the work is not the complexity of cells itself but rather what must be a contingent byproduct of their complexity, namely their idiosyncrasy—that is, individual variability among cells. No two cells are exactly alike—not even those serving identical biological functions in the same organism; hence no two cells behave exactly the same way; hence, we cannot expect to find exceptionless, strict laws about them—or any significant subclasses of them. On this line of reasoning, complexity itself does not have a fundamental role. Cells could be as complex as you please, but if they—or those in some theoretically significant subclass of cells—had an identical microstructure, to the last molecule or basic particle, there would be no prima facie reason, on Smart’s general view, why there could not be strict, exceptionless laws about them. They would behave identically in identical situations—just as electrons do. This means that what really accounts for the absence of laws in biology is not structural complexity of biological entities per se but rather its contingent byproduct, as I said, namely the individual variability (or, following Smart, “idiosyncrasy”) among the entities grouped together under a single biological kind.31

30 Ibid. 55. 31 Actually, the divergence between entities grouped under a biological kind arguably goes much deeper and wider. Biological kinds, like mental kinds, are standardly considered to be “multiply realizable”; think of how diversely, say, the heart or the visual system, is realized in diverse biological species. The structural differences between two human hearts are very slight indeed when compared with differences between a human heart and a reptilian heart. Issues involving multiple realizability will be taken up in our “third” argument below.

300


I believe these considerations can be generalized to other special sciences along the following lines: the entities of any special science are complex aggregate structures of the entities dealt with in physics, and there inevitably will be structural/compositional differences among entities falling under a single kind in any special-science taxonomic system. This is important because we expect the behavior, or behavioral dispositions, of a complex aggregate system to depend on, and be determined by, its structural/compositional detail. In general, entities that are similar in their microstructure behave in similar ways when placed in similar situations; in fact, the greater the structural similarity the greater the behavioral similarity. The converse of this principle also holds: structurally dissimilar systems are apt to exhibit dissimilar behavior, and, as a general rule, the greater the dissimilarity, the greater the behavioral dissimilarity. Consider all actual and possible samples that fall under a biological kind. We can expect the kind to be idiosyncratic in the sense that these samples show a substantial range of individual variability in their microstructural composition, and given the dependence we just noted of the behavior of wholes on their microstructure, it is highly unlikely that there will be strict laws that apply to this biological kind. It does not strictly follow that there are, or can be, no strict, exceptionless correlations between biological kinds or properties. Let B1 and B2 be two biological kinds/properties: it is conceivable that in spite of the high degree of idiosyncrasy of both B1 and B2, the range of individual variation allowed by B1 is matched by that allowed by B2, generating a precise, strict correlation between them. This is logically possible but highly unlikely. It is true of course that whether or not a generalization holds over all instances of a kind depends on the predicate, or property, projected over them, and that when the projected predicate is extremely general (e.g., “has a mass,” “is made of cells,” etc.), we can easily secure truth (e.g., “Every heart has a mass,” “Kidneys are made up of cells”). But we can safely say that these are not the kind of laws, whether strict or not, biological researchers look for. With these caveats in mind, we can say that the considerations based on the idiosyncrasy of special-science kinds and properties show, or at least make it highly likely, that we will not find strict, exceptionless, laws in the special sciences. Moreover, the same considerations also explain why we can expect broad empirical generations of interest and usefulness (you might call them “nonstrict” laws or “ceteris paribus” laws) in the special sciences. This is because in spite of the idiosyncrasy, there are also broad points of


301

compositional similarity and resemblance among samples falling under a given kind. These considerations invoke two substantive metaphysical premises. One is the thesis, already mentioned, to the effect that the behavior of an aggregate structure of physical entities (elementary particles, quarks, atoms, molecules, or what have you) is determined by, or supervenes on, their basic structural/compositional details. This must be so because the intrinsic properties, and hence the causal potentials, of complex systems supervene on their microstructure. But the premise that plays a more direct role in our argument is the converse of this supervenience principle, namely that structurally dissimilar systems will tend to behave in dissimilar ways under identical conditions. A deep discussion of the grounds for accepting these principles, or of their precise formulations, is beyond the scope of this paper. But it seems to me that principles like these are operative, for example, in our thinking about quality control in industrial manufacturing: In order to maximize a uniform level of performance and functionality across samples of a product, we try to maximize uniformity over the parts to be assembled into the product—that is, to minimize differences (or “idiosyncrasies”) among them—and to ensure the uniformity of the mode of assembly as much as possible. Automobiles rolling off an assembly line are not exactly identical with one another and don’t behave in exactly identical ways; the “lemons” would be among the extreme cases—and if there are lemons there must also be “anti-lemons,” those samples markedly superior to their cohorts. So there are no exceptionless laws about, say, all samples of the 2006 Honda Accord LX sedan, and it would be silly to look for them. However, they are similar enough to yield many useful empirical generalizations, or rules of thumb, about how they behave and function. That’s what makes it possible to write service manuals and owners’ handbooks. The same goes for biology: although there are no biological laws, conspecifics are structurally similar enough so that there are informative and useful empirical generalizations (about their longevity, size and weight, developmental phases, susceptibility to diseases, responsiveness to medical interventions, and so on). Smart seems to implicitly endorse the claim that these generalizations are not explanatory; that they are like useful rules of thumb about auto repair and maintenance, and not capable of generating genuine scientific understanding of why organisms function and behave as they do. Smart’s position on biology

302


encourages the view that genuine explanatory understanding in biology comes only from applying physicochemical laws to explain why, given its physical structure, an organism functions and behaves as it does. Given the idiosyncrasy of biological kinds, such explanations will not be generalizable even across different samples falling under one biological kind. That is, biological explanations are likely to show as much idiosyncrasy as biological entities do. These reflections suggest a possible account of why there are, or can be, strict laws in physics. Let us make this assumption: the entities studied in microphysics, like electrons and photons, are metaphysical simples with no internal structure, or, if they are complex, those falling under a basic kind in fundamental physics have an identical structure. On this assumption, the entities belonging to a basic physical kind have exactly the same set of intrinsic properties and hence exactly identical causal potentials (one electron is exactly like the next one, in all intrinsic properties). In consequence, they behave the same way under the same conditions. This is why the basic physical taxonomy makes exceptionless physical laws possible. I am not suggesting that this is the only way in which strict laws can be seen to be possible in basic physics—in particular, I am not saying that this sort of structural consideration applies to all basic laws of physics; nor do I know whether the assumption with which we began the present paragraph is strictly true or will be accepted by the working physicists. So I am leaving this as a speculative conjecture.32

Implications: A New Argument for the Anomalism of the Mental? The Smart-derived argument of the preceding section suggests a new, and simpler, argument for psychological anomalism. Psychological systems, like human beings and other higher organisms,33 are complex physical systems, and organisms belonging to the same species can, and do, exhibit a fairly 32 As the reader will have noticed, this explanation of why strict laws are possible in physics differs from the explanation I earlier suggested in connection with the “interference from below” argument. I believe there may be interesting connections between the two explanations; however, they have to be set aside for another occasion. 33 If you like, you could include sophisticated computing machines, or robots, as well.


303

large degree of idiosyncrasy from the point of view of underlying physical/ biological structure and mechanism, though of course as members of the same species they are bound to show a good amount of similarities as well. Moreover, token psychological states, or events, that fall under a single kind, in the scientific or vernacular psychological taxonomy, are certain to show a significant range of individual variability at the underlying biological/physicochemical level, even within the same species—in fact, even for the same organism at different times. As in the case of biology, we should expect that the greater the structural differences are at the biological/ physical level, the greater the psychological/behavioral differences will be. Conversely, the greater structural similarity, the greater the psychological similarity. It follows then, first, that strict, exceptionless laws connecting psychological kinds to psychological kinds will likely be unavailable, and, second, that, in spite of that, we can expect to find useful empirical generalizations, or ceteris paribus laws. It seems that the same line of consideration could also help establish psychophysical anomalism—the claim that there are no strict laws connecting psychological with physical phenomena. Briefly, given the range of individual variability among the token physical/biological states underlying psychological states falling under a psychological kind, it is highly unlikely that the “same” psychological states, that is, states belonging to some single psychological kind, will be followed by, or correlate with, identical physical states. Human animals, as conspecifics, are biologically/physically quite alike but not exactly so. As a result, they will not behave exactly the same way in psychological contexts any more than they do in purely physical contexts. So there will be no strict laws connecting psychological events with physical events. (There is one exception: exactly identical physical conditions in psychological systems will lead to identical psychological behavior.34 But such laws will be of little practical or scientific interest.) As in the case of laws connecting psychological phenomena with other psychological phenomena, there are enough structural similarities among psychological systems and their properties to yield rough but stable and

34 If physical identity is understood in terms of identity in intrinsic physical properties, that would make this statement highly dubious, at least for many philosophers, in view of the externalist considerations regarding content-carrying states, such as belief, desire, and intention.

304


useful empirical generalizations, and this is part of what makes the psychological and cognitive sciences possible. If these considerations are not entirely off the mark, it means that we have a new argument for Davidson’s Anomalism of the Mental, the claim that there are no strict laws about psychological phenomena. As most everyone who has worked through “Mental Events” will agree, Davidson’s arguments are complex and difficult to grasp as they involve a web of subtle considerations concerning rationality, normativity, holism, radical interpretation, and the rest. In contrast, our Smart-derived arguments stay clear of such deep considerations, resting instead on what seem to me to be relatively simple and transparent metaphysical theses about structure and behavior. If this is right, the anomalousness of the mental need have nothing specifically to do with the special character of minds; it is merely one instance of a general and pervasive natural phenomenon, namely the structural idiosyncrasy of psychological systems, including human persons. Beyond the realm of the sciences, the metaphysical considerations motivating the arguments of this section also explain why there are no strict, exceptionless laws about macroobjects and their observable macroproperties—about rocks, rivers and mountains, rains and winds, trees and shrubs, tables and chairs, and so on. These are the familiar sundry things that make up the world in which we pass our lives. Again, our considerations explain why, even though there are no strict laws, there often are useful generalizations about them that are reliable for most of our purposes and capable of grounding counterfactuals in appropriate cases. Needless to say, without access to such generalizations, it would hardly be possible for us to cope with our constantly changing surroundings and navigate our way through the natural world. These rough and ready generalizations, not the strict laws (if there are any), serve as an indispensable guide for us in our quotidian dealings with the world.

A Note on “Ceteris Paribus” Laws It is widely thought that laws in the special sciences—in fact, all laws outside physics—are “hedged” by “ceteris paribus” conditions, conditions that exempt these laws from direct falsification by counterexamples, instances


305

in which the antecedent event of the law is not followed by the specified consequent event. These “other things being equal” hedges are often also parsed as “under normal conditions,” “under ideal conditions,” and the like. But what exactly do these hedges mean? There is one popular understanding of the “ceteris paribus” hedge that our considerations show to be inadequate as an explanation of the nonstrictness of biological and other special-science lawlike generalizations. Against Davidson’s requirement that all singular causal statements be underwritten by strict laws, Jerry Fodor argues: It seems to me that we are now lacking any convincing argument for accepting [Davidson’s condition that causal relations instantiate strict laws]. Suppose it’s true that causes need to be covered by laws that necessitate their consequents; it doesn’t follow that they need to be covered by strict laws. Hedged laws necessitate their consequents in worlds where their ceteris paribus conditions are satisfied. Why, then, should mental causes that are covered by hedged intentional laws with satisfied antecedents and satisfied ceteris paribus conditions require further covering by a strict laws of physics? . . . Strict laws and hedged laws with satisfied ceteris paribus conditions operate alike in respect of their roles in covering causal relations and in respect of their roles in covering law explanations. Surely this is as it should be: Strict laws are just the special case of hedged laws where the ceteris paribus clauses are discharged vacuously; they’re the hedged laws for which “all else” is always equal.35 On Fodor’s conception, then, the nonstrict law “Ceteris paribus, F-events cause G-events” comes to an existentially quantified statement “There are conditions C1, . . . , Cn (unknown—yet to be identified) such that when they are satisfied, F-events cause (or necessitate) G-events.” When these C’s have been identified and we are in a position to affirm “When conditions C*1, . . . , C*n are present, F-events cause G-events,” we are in possession of a strict law. This means that for each hedged ceteris paribus law, there is a strict law out there waiting to be discovered. The ceteris paribus hedges are only a reflection of our epistemological limitations, not something objectively defective in the regularities described by the special sciences. 35 Jerry A. Fodor, “Making Mind Matter More,” Philosophical Topics 17 (1989): 59–79 (the quotation is from p. 75).

306


It should be clear that this conception of ceteris paribus hedges has nothing to do with the nonstrictness of biological and other special-science laws uncovered in our argument deriving from Smart’s observations on the nature of biology. The nonstrictness of biological generalization is due to the objective structural differences among the entities falling under the same biological kind. It has everything to do with the absence of uniformity, from a microphysical point of view, of biological items classified in the same categories; it has nothing to do with the limitations of our knowledge about the biological and their underlying physicochemical facts. There is nothing out there in the biological domain such that if we knew it we could turn the nonstrict biological generalizations into strict laws. Nonstrictness here is metaphysical, not epistemological. Fodor speaks of intentional psychological laws as being nonstrict in his epistemological sense. However, if we are right, the nonstrictness of such laws is not some correctable defect that could be removed by doing more diligent psychological research. It is an objective metaphysical fact about psychological phenomena.36

Multiple Realization and Special-Science Laws The claim that the properties and kinds investigated in the special sciences are “multiply realizable” and in fact are so realized has become an entrenched philosophical platitude, so often repeated that it has become more than slightly tedious to hear it mentioned. As many will recall, the example that Fodor used to excellent effects, in his 1974 paper “Special Sciences—or the Disunity of Science as a Working Hypothesis,” an example that I believe helped to win over many philosophers to the antireductionist side, concerned monetary exchanges and the extremely diverse kinds of physical objects that could serve as a medium of economic transactions. Fodor wrote: Some monetary exchanges involve strings of wampum. Some involve dollar bills, and some involve signing one’s name to a check. What are the chances that a disjunction of physical predicates which cover all 36 This is not to dispute the validity of Fodor’s challenge to Davidson’s requirement that causal relations must be covered by strict laws. Perhaps, causal relations do not need to be covered by laws at all, whether strict or hedged.


307

these events . . . expresses a physical natural kind? In particular, what are the chances that such a predicate forms the antecedent or consequent of some proper law of physics?37 As Fodor puts it, such a physical predicate would be “wildly disjunctive.” The presumption is that there is something flawed and defective about such disjunctive predicates, making them unfit to serve in proper scientific laws. I think Fodor is right in his claim about disjunctive predicates, except that I would put the point by saying that such predicates are not inductively projectible, and that since projectibility is standardly taken to be the hallmark of scientific laws, these predicates are nonnomic and the concepts/ properties they express cannot serve the scientific purposes of explanation and prediction. Since I have argued these points in detail elsewhere,38 I will give only a brief sketch of my considerations. Suppose P is a special science property (we are taking an ontological turn and switch to talk of properties from talk of predicates) that is realized in three ways, Q1, Q2, and Q3, where the Q’s are properties (or mechanisms) at a lower level. Now if the “multiplicity” in multiple realization is going to mean anything, it must mean nomological/ causal multiplicity: presumably it is because the Q’s are nomologically and causally diverse from each other (“wildly diverse,” as Fodor would say) that they count as “different” realizers. In general, we may take the multiply realized higher-level property P to be a second-order functional property defined in terms of a causal role, in some such fashion as this: Having P is, by definition, having some property (or mechanism) that meets causal specification C (e.g., it is caused to instantiate by input of kind J and its instantiation causes output of kind K). Any property that meets the causal specification C counts as a realizer (or realization) of P: If an object has a realizer of P, it instantiates P, and it can 37 Jerry A. Fodor, “Special Sciences—or the Disunity of Science as a Working Hypothesis” (1974), reprinted in Philosophy of Mind: Classical and Contemporary Readings, ed. David J. Chalmers (New York and Oxford: Oxford University Press, 2002), p. 129. I don’t think this example is a particularly apt one, though it has served Fodor’s purpose well. It would be absurd to think that monetary exchange, or any other economic entity or property, might be directly reduced to physics (or “folk” physics about wampum or whatever). If there is going to be a reduction of economics, it would first have to be reduced to individual and social psychology; one could then consider the question whether these sciences are reducible to biology; and so on. 38 In “Multiple Realization and the Metaphysics of Reduction” (1992), reprinted in Supervenience and Mind (Cambridge: Cambridge University Press, 1993).

308


instantiate P only if it has one or another realizer of P. It is clear that this conception of a functional property allows multiple diverse realizers for any functional properties. On the assumption that whether or not a property meets causal specification C depends on the (causal) laws operative at a world, it is evident that the relation of being a realizer of is a metaphysically contingent (though nomologically necessary) relation. If we piece together these considerations, a certain picture emerges concerning the possibility of laws in the special sciences. Consider the instances of P, a multiply realized higher-level property: each instance of P is either a Q1-instance or Q2-instance or Q3-instance. That is, all instances of P fall into three disjoint classes, those that are Q1-instances, those that are Q2-instances, and so on. Any two Qi-instances will be nomologically/ causally similar, since we are assuming that each of the Q’s is a nomic/ causal kind. But the P-instances that are not instances of the same realizer will differ in their nomic/causal properties. What this means is that the higher-order property P will itself be nomologically and causally heterogeneous—as heterogeneous as its “wildly” diverse realizers. And as such P will be ineligible to enter into projectible generalizations, that is, laws, thereby failing to represent a homogenous causal kind. To put it another way, there will be no scientific theory of P, or P-instances, only scientific theories of the Q’s—that is, one theory for each of the three Q’s. Some recent writers have tended to stress that the diverse realizers of a higher-level properties can be more similar to each other than indicated by the foregoing considerations.39 The Q’s that realize P must each be capable of performing the causal task specified by C, the defining causal role of P, and this can place important constraints on the physical composition and structure on each Q (as Block points out, teacups don’t have complex enough structure to realize language-speakers, Disney cartoons notwithstanding). Where biological organisms are concerned, different species evolving in similar environments and coping with similar natural forces may have developed largely similar mechanisms to solve various evolutionary “design problems,” for example, developing a tissue damage detector, or a device that stores information gained through perceptual systems. These seem correct observations, but not really to the point. One reason is that here we are talking about the terrestrial biological organisms and species, 39 See Ned Block, “Anti-Reductionism Slaps Back,” Philosophical Perspectives 11 (1997): 107–132.


309

sharing a tiny corner of the universe as their habitat. What are the chances of these limited generalizations turning out to be “laws in the strict sense,” with unlimited applicability throughout the space-time world? In this context, we must consider all nomologically possible realizers of functionally defined special-science properties, not just the actual realizers in a limited space-time region. I think the diversity among the possible realizers of any functional property is so dizzyingly large40 that the possibility of exact, exceptionless laws connecting such properties has to be discounted. The conclusion concerning special-science laws, therefore, is quite deflationary: If special-science kinds are multiply realizable in diverse causal/ nomological kinds, as the standard view in this area has it, the prospects of finding significant special-science laws that apply across the diverse realizers will be slim, indeed. When we find such laws, their generality must derive from the causal/nomological similarities among the realizers, which weakens “multiple” in the claim of “multiple realizability.” Such similarities could exist in certain cases; they could be matters of pure coincidence, or derive from spatio-temporally restricted contingent facts like parallel evolutions in similar natural environments. Real scientific laws must be sought at the level of the realizers, and these laws will suffice to explain and predict all phenomena at the higher level.

Concluding Remarks There are various pending issues not addressed here which I believe deserve further exploration. One such issue concerns the question how the three arguments presented here are related to each other. Although I have made some comments relevant to this question, there is more to think about and undoubtedly more to say. I am inclined to think that the second, Smartinspired, argument is the fundamental one, and that the other two may very well depend on the considerations that motivate and ground it. Second, and more importantly, there is the question what implications our reflections on the laws, or the absence thereof, in the special sciences have on the nature of explanation and causation in these sciences. Does the absence of “strict” 40 This used to be the point emphasized and widely promoted by those who invoke multiple realizability as their principal argument against reductionism.

310


laws in the special sciences, assuming this to be the case, mean that specialscience explanations are in some ways flawed and imperfect, in comparison with explanations in fundamental physics? What of the causal claims we encounter in the special sciences? In the absence of strict laws how should we understand such claims?41 What grounds causal relations between special-science phenomena? The problem of mental causation—the problem of accounting for the possibility of causation involving psychological phenomena—has been extensively debated for well over two decades. The problem of special-science causation and explanation will likely turn out to be only a generalized version of the same problem. If so, what we have learned from the mental causation debate may carry important lessons for the general problem. On the other hand, it could be a smart philosophical strategy to tackle the general issues concerning the special sciences first, for the light it may shed on the special case concerning psychological phenomena. Either way we seem to have a rich set of philosophical problems in this area that should engage us for some time to come. 41 It is not implausible to hold that causal claims are most at home in the special sciences, while physics itself is free of causal concepts. The idea of mechanism seems to be gaining increasing popularity among philosophers in giving an account of explanations in the higher sciences; see, e.g., William Bechtel, Mental Mechanisms: Philosophical Perspectives on Cognitive Neuroscience (New York: Lawrence Erlbaum, 2007); Carl F. Craver, Explaining the Brain: Mechanisms and the Mosaic Unity of Neuroscience (Oxford: Oxford University Press, 2007).

Index Abbott, Barbara 259 n. 34 Achinstein, Peter 150 n. 2, 157 n. 20, 168 action: appropriateness of 133, 135–9, 143–5 deliberative 132 principle of 133, 144–6, 147 rational 123, 129–30 action explanation, see rational (rationalizing) explanation agency 3, 107, 127–8, 141 and mental causation 257–60 Alexander, Samuel 1, 8, 9, 11, 25–6, 28, 32, 38, 46 n. 7, 67, 77, 79, 81, 85 n, 87 Alston, William P. 123 n. 17 anomalism of the mental 112, 214, 285–94, 302–4 psychological anomalism 288, 291–2, 294, 302 psychophysical anomalism 288, 292, 303 anomalous monism 234–42, 245–7 Anscombe, Elizabeth 255 Antony, Louise 22 n. 19 Aquinas, Thomas 263 n. 1 Aristotle 44 n. 4, 167, 263 n. 1 Armstrong, David 16 n. 15 Beauchamp, Tom L. 154 n. 13 Bechtel, William 310 n. 41 Beckermann, Ansgar 8 n. 1, 95 n. 26 Bedau, Mark 81 n. 15, 87, 88 n. 5 Bellwoar, John 105 Benfield, David 148 Block, Ned 52, 53, n. 15, 60 n. 22, 64 n. 24, 103 n. 33, 209, 215 n. 15, 216 n. 18, 217 nn. 19–20, 218, 229 n. 30, 230 n. 32, 231, 233 n. 39, 266, 267, 275 n. 18, 294 n. 20, 308 Bodemann, Eduard 263 n. 1 Boghossian, Paul 148 Boyle, Robert 47 bridge law 18, 20, 73–4, 76, 99, 101, 210, 212–16, 232, 242, 295 see also reduction, Nagel (bridge-law)

Broad, C. D. 1, 2, 8, 38, 67, 69 n. 4, 71, 72–3, 77, 80, 86, 87, 88–101, 248–9 Bromberger, Sylvain 168, 196 Brown, S. C. 123 n. 17 Bunzl, Martin 165 Butts, Robert 202 n. 27 Carnap, Rudolf 57, 189, 192 Caston, Victor 8 n. 3, 244 n. 3 causal closure of the physical 272, 283–4, 288, 290–1 see also comprehensiveness of physics causal closure of the mental 288, 291 massive failure of 292 see also special-science causal closure, general failure of causal efficacy: distinguished from causal relevance 238–9 of the emergent 28 of the mental 235–7, 239, 256–7, 261–2, 272–3, 289–90 see also causal power; mental causation causal explanation 125, 153, 208–9 and causation 197–8 Hempel’s position on 164, 172 n. 6, 194–5 Lewis’s view on 174, 175 singular 202–4 see also causation, as an explanatory relation Causal Inheritance Principle 21, 23, 77 n. 11, 224 Causal Interaction, Principle of 288–9 causal law 195, 197, 204–5, 248–51 causal overdetermination 83–4, 162, 165–6, 230, 252, 262, 274 see also explanatory overdetermination causal power: of disjunctive property 227–8 downward 25, 41, 90, 104 of emergent property 11, 28, 31–2, 81–4, 87 of functional property 229–30, 267

312

INDEX

causal power: (cont.) of mental property 235–6, 271–3 of realized property-instances 21–3, 77, 224, 261–2, 273–4 causal-power actuality principle 35–7 causal preemption 165, 252 causal realism 154–9, 161 causation: and contiguity 154, 243–4, 256–7, 260 and control 83 counterfactual account of 83, 251–3 as counterfactual dependence 255–60 as explanatory relation 150–3, 172–4, 203 as extensional relation 203–4, 236–7, 290 mental causation 243–62 nomic-regularity view of 83, 152–3, 174, 195, 237, 248–51 by omission 258–60 as production (generation) 255–6, 262 strict-law requirement on 241, 245, 247 thick/thin 257 Causey, Robert 216 n. 17 ceteris paribus law 239–42, 247–51, 300, 304–6 see also special-science law; strict law Chalmers, David 15 n. 14, 16 n. 15, 24 nn. 23–4, 75 n. 10, 87, 88 n. 5, 89, 92, 95, 98, 99 n. 28, n. 30, 101, 103 n. 33, 210–11, 276 n. 19, 307 n. 37 Churchland, Paul 232 Clapp, Lenny 265 Clayton, Philip 87 n. 4, 88 n. 5 Cohen, Jonathan 6 n. 2, 103 n. 33, 255 n. 25 Collins, John 252 n. 22, 256 n. 27 comprehensiveness of physics 60, 246, 286–7, 291 contiguity 154, 243–4, 256–7, 260 Corbi, Josep 105 Cottingham, John 94 n. 21, 244 n. 2 covering-law model of explanation 112, 129, 168, 171–5, 178, 192–205 see also deductive-nomological argument; nomic expectability Craver, Carl F 310 n. 41 Cummins, Robert 150 n. 2 Davidson, Donald 77, 152 n. 7, 155 n. 18, 165 n. 35, 190 nn. 9–10, 204 n. 31, 305, 306 n. 36 on action explanation 3, 106, 111–13, 125–6, 134, 139, 146

on mental anomalism 6, 7, 214, 234–42, 245–7, 285–94, 297, 304 Davies, Martin 16 n.15 Davies, Paul 87 n. 4, 88 n. 5 deductive-nomological argument 135, 162–4, 171–5, 177, 180, 197–200 deliberation 116–17, 132–3 and self-understanding 116–19, 138–41, 147 and third-person understanding 144–5, 147 see also predictive stance Dennett, Daniel C. 113 n. 6, 120 dependence, dependency 50, 78–80, 183–4, 194, 269 causal 183, 194, 197–8 mereological 183, 184 see also supervenience Descartes, Rene´ 18, 94, 96, 243–4, 260 Desire-Belief-Action Principle 111–13, 120 diversity, causal/nomological 23, 230, 307–9 see also idiosyncrasy, of special-science kinds; individual variability double-aspect theory 18, 79 Dowe, Phil 257 n. 31 downward causation 25–40, 41, 67, 81–4, 90 diachronic 36–8 nonreflexive 33 principle of 30, 31 reflexive 33–8 synchronic 35, 37, 38 Dray, William 3, 128–9, 133–8, 143–5, 147 dualism: Cartesian (substance) 18, 43, 79, 244 naturalistic 98 Ducasse, C. J. 155 n. 16 Dupre´, John 284 n. 5 Eccles, John C. 9 n. 4 eliminativism 231–2 see also functional property conceptualism Elisabeth of Bohemia, Princess 243–4, 260–1 emergence, emergent property 8–40, 66–84, 85–104 emergent/resultant distinction 11–13, 27–8, 32, 70 epistemological 86–7, 88 n. 6, 92, 104 and functional reduction 15–24, 27–8, 70–8

INDEX

irreducibility of 15, 20–4, 27–8, 39–40, 70–8, 80 metaphysical (ontological) 86–101, 104 nondeducibility of 45, 73, 90–104 strong 86, 87, 88 n. 6 and supervenience 12, 27, 50, 68–70, 78–81, 88–9, 97–101, 104 unexplainability of 15, 24, 27–8, 70–6, 79–80 unpredictability of 11–15, 24, 27–28, 45, 70–6 weak 86, 87, 88 n. 6 see also emergentism emergence law 13–14, 72, 79, 99 emergent causation 37, 38 emergentism: central doctrines of 26–8 and layered model 44–54, 59–65 and mechanism (reductionism) 72–3, 91 see also emergence empathetic understanding 113, 121–3 see also re-enactment (recapitulation) Endicott, Ronald 265, 282, 292 n. 17 Engel, Pascal 41 epiphenomenalism 11, 28, 79, 235–6, 246, 250, 271–2 Epistemological Question, about explanation 170–2, 176, 180 event: Davidson’s theory of 77, 289–90 disjunctive 210 property exemplification theory of 199 exclusion argument 38–9, 82–4, 230–1, 250, 267–9, 271, 278 exclusion problem 4, 241, 268 see also exclusion argument explanans relation 149–50, 152, 153, 157, 161, 174 see also explanatory relation explanation: as argument (derivation) 159, 163, 188, 201–5 causal theory of 168, 174–7 individuation of 160, 166 inductive vs. mechanistic (reductive) 72 intentional 127–8 microreductive 62, 63 model of 5, 167–9, 175 pragmatic theory of 168 simplification-unification approach 168, 178–83, 185

313

and understanding 158, 168–9, 175–86, 198 see also covering-law model of explanation; explanatory knowledge; reductive explanation explanation-grounding relation 174, 176, 183 explanatory completeness 161–6 explanatory exclusion, principle of 159, 162, 165–6 explanatory gap 211, 215–18, 223, 232, 275–7 explanatory internalism (irrealism) 150, 153, 157–9, 160–6, 173–4, 177–8, 179–80, 185 see also explanatory realism, explanatory relation, as internal relation explanatory knowledge 148–9, 167–86 and descriptive knowledge 167–8, 177 pattern view of 156–8 and propositional knowledge 156, 170, 176–8, 180–1 propositional view of 156–8 and theory of explanation 168–9 explanatory overdetermination 162, 165 n. 35, 166 explanatory realism (externalism) 149–53, 183, 185, 203 causal explanatory realism 173–5, 176 and causal realism 153–9 and explanatory exclusion 160–6 and explanatory knowledge 156–7, 176–7 see also explanatory internalism (irrealism) explanatory relation 131–2, 172–5, 203, 275 distinguished from explanans relation 149–53, 160–1 as internal relation 171–2, 173 between reason and action 111, 114–15 see also explanation-grounding relation; explanatory realism; nomic expectability explanatory relevance, Hempel’s requirement of 131, 136, 164 Fair, David 257 n. 31 Feigl, Herbert 25 n. 25, 216 Filmore, Charles 259 n. 34 first-person perspective 122–4, 125–8, 138 first-person knowledge: of one’s own reasons 119 of one’s own mind 119, 130 n. 10

314

INDEX

Flohr, Hans 8 n. 1, 95 n. 26 Fodor, Jerry A. 18, 209–10, 213 n. 11, 214, 229, 230 n. 32, 239–40, 241 n. 18, 242 n. 19, 247–8, 250–1, 254, 266, 282, 284, 305–6, 307 Foley, Richard 167 folk (commonsense) psychology 120, 122–3 Friedman, Michael 178–83, 185, 205 n. 33 functional property conceptualism 228, 230–2 functional property realism 228–30, 232 functionalism 16 n. 15, 22, 75, 79, 228–32 Galen 8 n.2 Galison, Peter 284 n. 5 Garber, Daniel 243 n. 1, 260, n. 36 Gettier, Edmund 252 Gillett, Carl 88 n. 6, 216 n. 18, 233 n. 38, 265, 270 n. 15 Ginet, Carl 115 n. 10, 117 n. 11, 132 n. 14 Globus, Gordon G. 26 n. 26, 31 n. 31 Goldman, Alvin 161 n. 24 Gopnik, Alison 130 n. 10 Grandin, Temple 120–2 Hall, Ned 252 n. 22, 255–6 Hanson, Norwood Russell 151, 157 n. 19 Hare, R. M. 263 Harman, Gilbert 94, 96 Heil, John 6 n. 1, 126 n. 2, 207 n. 1, 234 n. 1, 238 n. 12, 245 n. 5, 290 n. 14, 297 n. 27 Heinaman, Robert 263 n. 1 Hempel, Carl G. 5, 9, 92, 162–5, 187–206 on action explanation 3, 111–13, 120, 128–36, 146, 194 on D-N explanation 129, 168, 171–5, 178, 179, 180 n. 22, 181 on syntacticalism 189–193 Henderson, David 105 heterogeneity, causal/nomological; see diversity, causal/nomological Hill, Christopher 216 n. 18, 217 n. 20, 218 n. 23 Hintikka, Jakko 202 n. 27 Hitchcock, Christopher 257 n. 31, 294 n. 19 Hoffman, Paul 263 n. 1 Honderich, Ted 245 n. 7 Hook, Sydney 263 n. 3 Horgan, Terence 89 n. 9, 95 n. 26, 231 n. 36, 233 n. 39, 236 n. 6, 255 n. 25, 261–2

Hume, David 154–5, 243, 251, 256, 260 Humphreys, Glyn W. 16 n. 15 Humphreys, Paul 13 n. 10, 81 n. 15, 89 n. 12, 104, 167, 168, 176, 197 Huxley, T. H. 254, 271, 272 identity 21–3, 216–21 explainability of 218–21 psychoneural 216–19, 233 and reduction 208, 211, 217, 233 token 21–2, 224–9 type 22–3, 226–8, 276 see also reduction; reductionism idiosyncrasy, of special-science kinds 299–304 individual variability 299–300, 303 inductive prediction, vs. theoretical prediction 13–15, 20, 71 integral unity 52, 57–8, 61 internal perspective 116, 127–8 see also first-person perspective Ishiguro, Hide 263 n. 1 Jackson, Frank 14 n. 12, 24 n. 24, 92, 93, 94, 103 n. 33, 276 n. 19 joint cause 165 justification, and explanation 126–8, 135, 147 Kaplan, David 200 n. 25 kind, see property Kitcher, Philip 168 n. 3, 179–83, 185, 205 n. 33, 294–6 Korsgaard, Christine 126, 128 Lakoff, George 259 n. 34 Lakoff, Robin 259 n. 34 Lange, Marc 282 Laplace, Pierre-Simon 95 n. 26, 101 layered model 25–6, 41–65 in emergentism 44–6, 48–54 of Oppenheim and Putnam 46–7, 54–9 top-down, local approach to 60–5 see also level Leibniz, Gottfried 263 n. 1 LePore, Ernest 236 n. 6, 240 n. 16, 241 n. 18, 242 n. 19, 255 n. 25 level 25–6, 41–65 and free molecules 56–7, 58, 61 proper 56, 58 and supervenience 50–2

INDEX

Levine, Joseph 16 n. 15, 22 n. 19, 24 n. 24, 211 n. 7 Lewes, G. H. 12, 13 n. 11 Lewis, David 16 n. 15, 150 n. 2, 161 n. 25, 168, 174, 175–7, 202, 203 n. 29, 251–3, 255 locality, of causation; see contiguity Locke, John 47 Loewer, Barry 216 n. 18, 236 n. 6, 241 n. 18, 242 n. 19, 255 n. 25, 257 n. 30, 270 n. 15, 282 logical connection, between reason and action 112, 113–14 Lovejoy, A. O. 8, 89 n. 9 MacDonald, Cynthia 236 n. 6 MacDonald, Graham 236 n. 6 Mackie, J. L. 155 n. 18 Marr, David 42 Marras, Ausonio 74 n. 8 materialism, see physicalism Maxwell, Grover 25 n. 25, 26 n. 26, 31 n. 31 McGeever, John 87, 88 n. 5 McGrath, Sarah 259 n. 34 McIntyre, Lee 294 n. 19 McLaughlin, Brian 6 n. 2, 8 n. 1, 15 n. 13, 103 n. 33, 148, 216 n. 18, 218 n. 23, 229 n. 30, 230 n. 34, 236 n. 5, 240 n. 16, 245 n. 7, 246, 249 n. 16, 255 n. 25, 266 n. 7, 270 n. 14, 271 n. 15, 282 Mele, Alfred 6 n. 1, 126 n. 2, 207 n. 1, 234 n. 1, 238 n. 12, 245 n. 5, 290 n. 14, 297 n. 27 Mendola, Joseph 148 mental causation 243–62 and anomalous monism 245–7 within Cartesian dualism 243–4 and counterfactual dependence 251–60 and nomic-regularist view of causation 247–51 as productive causation 257, 260–2 on Shoemaker’s account of realization 273–4 Metaphysical Question, about explanation 172–5 microreductionism 44, 47, 65 and microreductive strategy 54, 62–4, 183 Mill, John Stuart 8, 12, 87, 187 Mills Eugene 83 n. 20 mind-body problem 74–5, 123

315

Morgan, C. Lloyd 8, 9, 11, 12, 13, 25, 30–2, 34, 37–8, 39 n. 36, 44–65, 67, 70, 71, 77, 83 n. 19, 85–6, 89 n. 12, 90, 91 multiple realization argument 18, 209–11, 214, 217 n. 20, 264, 295 n. 23 Murdoch, Dugald 94 n. 21, 244 n. 2 Nagel, Ernest 9, 17–8, 73–4, 92, 99, 101, 187, 212–13, 216, 294 Nagel, Thomas 93, 127–28, 140 Newman, D. V. 81 n. 15 Ney, Alyssa 282 Nicod criterion 191–2 Nisbett, Richard 108 n. 3, 130 n. 10 nomic expectability 132, 133, 136, 172–3, 181–2 nomological dangler 216 nomological rigidity, see semi-rigidity nonstrict law, see ceteris paribus law normativity, in rational explanation 133, 136, 143 see also appropriateness of action; justification, and explanation O’Connor, Timothy 38 n. 35, 81 n. 15, 87, 89 n. 12, 104 Oppenheim, Paul 9 n. 4, 25 n. 25, 44, 46–7, 54–65, 92 n. 18, 192, 199, 200 n. 25 Papineau, David 272 n. 16 partial cause 161 Paul, L. A. 252 n. 22, 256 n. 27 Pepper, Stephen C. 91 Pfeifer, Karl 162 n. 26 physicalism 211, 282–3 a priori 103 nonreductive 9, 10, 87, 261–2 reductionist 27, 233, 264, 278 Shoemaker’s statement of 277–80 type 79, 277–8, 280 see also identity; reductionism Pitt, Joseph C. 178 n. 16, 179 n. 19, 181 n. 23 Poland, Jeff 282 Polger, Thomas 265 Popper, Karl 9 n. 4, 187 practical inference (reasoning) 106, 118, 138–44 predictive stance 109–10 vs. deliberative stance 117 third-person 120, 122 projectibility 192, 246, 248, 307–8

316

INDEX

proof system 96–7, 101–2 property: causal theory of 81 higher-level functional 226–32, 307–8 intrinsic 12, 24, 301–2 mental 278–9 nomic/causal 56, 62–5, 228, 230, 308–9 physical 60, 278–9 special-science 250, 296, 300, 309 psychologism 189, 198 Putnam, Hilary 18, 25 n. 25, 44, 46–7, 54–65, 155, 158 n. 21, 210, 214, 229, 232, 263–7 Pylyshyn, Zenon 42 qualia 14, 20, 24, 75, 92–3, 211, 277 Quine, W. V. 60, 62 n. 23, 155 n. 18, 191–2 radical interpretation 146, 304 Railton, Peter 161 n. 25, 167, 168 n. 3, 205 n. 32 rational (rationalizing) explanation: autonomy thesis about 123 Davidson’s causal account 111, 113, 125–6, 134, 139, 146 Dray’s account 135–8, 145, 147 Hempel’s nomological account 111–13, 129–34, 146 projection thesis 122 rational intelligibility, circle of 145 rationality: of action 123–4, 129–30 as a characteristic of the mental 240, 286 Raven Paradox 191–2 realization: Putnam’s introduction of 263–4 second-order view of 266–9, 275–6, 281 subset view of 269–73, 276–7, 281 reason: and cause 116, 118 vs. evidence 110–13 explanatory 111, 128 justifying (normative) 127, 128 practical 118, 119 primary 106–8, 110, 111, 116, 123 theoretical 118, 119 reduction: functional 15–24, 74–8, 221–32, 233 identity 216–21, 233, 276 Nagel (bridge-law) 17, 20, 73–4, 212–15

conservative 21, 23, 208 eliminative 21, 23–4, 207–8 vs. reductive explanation 207–33 reductionism 72–3, 91, 233, 262 token 224–9 see also identity; physicalism reductive explanation 15, 72, 207–33 and bridge-law reduction 18, 214–15 and functional reduction 19, 76, 222–6 and identity reduction 217–21 and multiple realizability 209–11 see also explanatory gap; reduction re-enactment (recapitulation) 137, 144–5, 147 Reichenbach, Hans 155 n. 16 Rescher, Nicholas. 77 n. 12, 164 n. 31, 191 n. 13, 253 n. 23, 266 n. 8 Roberts, John T. 294 n. 19 Rosch, Eleanor 10 n. 5 Rosenberg, Alexander 154 n. 13 Ross, Don 209 n. 4, 229 n. 30 Ruben, David-Hillel 185 n. 30 Ruddick, William 151 n. 5 Rueger, Alexander 81 n. 15 Russell, Bertrand 155 Rutherford, Ernest 282, 283 Sabates, Marcelo 167, 233 n. 39, 269 n. 12 Sacks, Oliver 120 Salmon, Wesley C. 134 n. 19, 150 n. 2, 151, 155 n. 18, 167, 168, 174 n. 8, 176–77, 179 n. 20, 185, 197, 202, 249 n. 17, 257 n. 31 Santana, Margarita 167 Savodnik, Irwin 26 n. 26, 31 n. 31 Schaffer, Jonathan 257 n. 31, 294 n. 20 Schlick, Moritz 190 Scriven, Michael 25 n. 25, 152 n. 6, 201–2 Searle, John 10 n. 5 self-deception 117–18 self-understanding 107–8, 116–18, 123–4, 138–41, 147 inadequacy of Hempel’s model for 112–13, 131–5 Sellars, Roy Wood 8 Sellars, Wilfrid 190 n. 9 semi-rigidity 22–3 Shapiro, Lawrence 265 Shoemaker, Sydney 6, 81 n. 17, 265–81 Silberstein, Michael 81 n. 15, 87, 88 n. 5, 89 n. 12, 90, 104

INDEX

simulation theory 122, 144 Sklar, Lawrence 17 n. 17, 196 n. 20, 216 n. 17 Smart, J. J. C. 7, 207, 216, 217 n. 19, 231, 282, 283, 285, 296–9, 301, 302, 304, 306, 309 Sosa, David 233 n. 39 Sosa, Ernest 238, 245 n. 7, 255 n. 26 special sciences 60, 209–10, 250–1, 282–310 see also special-science law special-science causal closure, general failure of 293, 296 special-science law: multiple realization and 304–6 quasi-Davidsonian argument against 285–96 Smart-derived argument against 296–302 see also ceteris paribus law; strict law; strict special-science law, nonexistence of Sperry, Roger W. 8, 30–9 Spurrett, David 209 n. 4, 229 n. 30 Stalnaker, Robert 103 n. 33, 215 n, 216 n. 18, 217 n. 19, 218, 233 n. 39, 253, 255 Stephan, Achim 8 n.1, 95 n. 26 Stoothoff, Robert 94 n. 21, 244 n. 2 Stoutland, Frederick 245 n. 7 strict law: Davidson’s view on 126, 134, 245–6, 285–6 exceptionlessness of 246, 287 in special sciences 282–310 see also ceteris paribus law; special-science law strict special-science law, nonexistence of 294 Stroud, Barry 154 n. 13 Stump, David J. 284 n. 5 sufficient cause 161, 165–6 supervenience: as dependence relation 50, 183–4 and emergence 68–70, 78–81, 88–9 logical (metaphysical) 98–101 mereological 12, 69, 183, 184 nomological (natural) 98–101 as a non-homogeneous relation 78–80 psychophysical (mind-body) 50, 78–9, 184, 237–8

317

strong 50 symmetric 51 n. 13 weak 238 supervenience argument 250 n. 18, 269 n. 11 see also exclusion argument syntacticalism 189, 192 Tarski, Alfred 189 theory theory 120, 122, 144 third-person perspective 117, 125–27 see also predictive stance Thompson, Evan 10 n. 5 token causal symmetry 289 see also type causal asymmetry Tooley, Michael 255 n. 26 total microstructural property 12, 13 trans-ordinal law 72, 73, 99–101 see also emergence law, bridge law Tye, Michael 14 n. 12 type causal asymmetry 289 see also token causal symmetry Unger, Peter 165 unity of science 65, 209, 284 upward causation 28–9, 38 upward determination 33–4, 89 Valins, Stuart 108 n. 3 Van Cleve, James 92 Van Fraassen, Bas 168 Van Gulick, Robert 16 n. 15, 68–70, 87, 88 n. 5 Van Inwagen, Peter 81 n. 17 Varella, Francisco 10 n. 5 Velleman, David 105 n. 1 Villanueva, Enrique 167 von Wright, Georg 113–14, 146 n. 26 White, Nicholas 41 Williamson, Timothy 132 n. 13, 139 n. 22 Wilson, George 115 n. 10 Wilson, Timothy DeCamp 130 n. 10, Wimsatt, William C. 13 n. 10, 26 n. 26 Wittgenstein, Ludwig 155 Wong, H. Y. 81 n. 15, 87, 89 n. 12 Woodward, James 202–3

Essays in the Metaphysics of Mind

New Essays in Metaphysics

Review of 'The Metaphysics of Mind'

Metaphysics and the Philosophy of Mind

Controversies and the Metaphysics of Mind

Plantinga, Alvin Essays in the Metaphysics of Modality

Foundations of Mind (Philosophical Essays)

Foundations of Mind (Philosophical Essays)

Kant's Metaphysics of Morals: Interpretative Essays

Postures of the Mind: Essays on Mind and Morals

Debating Dispositions: Issues in Metaphysics, Epistemology and Philosophy of Mind

Unsolved mysteries of the mind: tutorial essays in cognition

The Wave Function: Essays on the Metaphysics of Quantum Mechanics

Earth's Mind: Essays in Native Literature

Mind World: Essays in Phenomenology and Ontology

Baboon Metaphysics: The Evolution of a Social Mind

Personal Agency: The Metaphysics of Mind and Action

The Metaphysics of Hyperspace

The Metaphysics of Capitalism

The Possibility of Metaphysics

The Metaphysics of Meaning

The God of Metaphysics

Changing My Mind: Occasional Essays

Mind-Energy: Lectures and Essays

The Metaphysics of Knowledge

The Metaphysics of Knowledge

The Metaphysics of Concepts

The Metaphysics of Nonfactualism

The Nature of Metaphysics

The Necessity of Metaphysics

Essays in the Metaphysics of Mind

New Essays in Metaphysics

Review of 'The Metaphysics of Mind'

Metaphysics and the Philosophy of Mind

Controversies and the Metaphysics of Mind

Plantinga, Alvin Essays in the Metaphysics of Modality

Foundations of Mind (Philosophical Essays)

Foundations of Mind (Philosophical Essays)

Kant's Metaphysics of Morals: Interpretative Essays

Postures of the Mind: Essays on Mind and Morals

Debating Dispositions: Issues in Metaphysics, Epistemology and Philosophy of Mind

Unsolved mysteries of the mind: tutorial essays in cognition

The Wave Function: Essays on the Metaphysics of Quantum Mechanics

Earth's Mind: Essays in Native Literature

Mind World: Essays in Phenomenology and Ontology

Baboon Metaphysics: The Evolution of a Social Mind

Personal Agency: The Metaphysics of Mind and Action

The Metaphysics of Hyperspace

The Metaphysics of Capitalism

The Possibility of Metaphysics

The Metaphysics of Meaning

The God of Metaphysics

Changing My Mind: Occasional Essays

Mind-Energy: Lectures and Essays

The Metaphysics of Knowledge

The Metaphysics of Knowledge

The Metaphysics of Concepts

The Metaphysics of Nonfactualism

The Nature of Metaphysics

The Necessity of Metaphysics

Recommend Documents