Aspects of Knowing: Epistemological Essays (Perspectives on Cognitive Science)

ASPECTS OF KNOWING EPISTEMOLOGICAL ESSAYS

i

Previous books published in the Perspectives on Cognitive Science series:

Creativity, Cognition and Knowledge — An Interaction Terry Dartnall Language Universals and Variation Mengistu Amberber and Peter Collins Perspectives on Cognitive Science, Vol. 2 — Theories, Experiments, and Foundations Janet Wiles and Terry Dartnall Perspectives on Cognitive Science, Vol. 1 — Theories, Experiments, and Foundations Peter Slezak, Terry Caelli and Richard Clark Representation in Mind: New Approaches to Mental Representation Hugh Clapin, Phillip Staines and Peter Slezak Competition and Variation in Natural Languages: The Case for Case Mengistu Amberber and Helen de Hoop

ii

ASPECTS OF KNOWING EPISTEMOLOGICAL ESSAYS

EDITED BY

STEPHEN HETHERINGTON The University of New South Wales, Sydney, Australia

Amsterdam • Boston • Heidelberg • London • New York • Oxford Paris • San Diego • San Francisco • Singapore • Sydney • Tokyo iii

Elsevier The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, UK Radarweg 29, PO Box 211, 1000 AE Amsterdam, The Netherlands First edition 2006 Copyright © 2006 Elsevier Ltd. 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 electronic, mechanical, photocopying, recording or otherwise without the prior written permission of the publisher Permissions may be sought directly from Elsevier’s Science & Technology Rights Department in Oxford, UK: phone (⫹44) (0) 1865 843830; fax (⫹44) (0) 1865 853333; email: [email protected]. Alternatively you can submit your request online by visiting the Elsevier web site at http://elsevier.com/locate/permissions, and selecting Obtaining permission to use Elsevier material Notice No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging-in-Publication Data A catalogue record for this book is available from the Library of Congress ISBN-13: 978-0-08-044979-1 ISBN-10: 0-08-044979-4 For information on all Elsevier publications visit our website at http://books.elsevier.com Printed and bound in The Netherlands 06 07 08 09 10 10 9 8 7 6 5 4 3 2 1

iv

Contents

Acknowledgements

vii

Contributors

ix

1.

Introduction: The Art of Precise Epistemology Stephen Hetherington

1

Part A Epistemology as Scientific? 2.

A Problem About Epistemic Dependence Tim Oakley

3.

Accounting for Commitments: A priori Knowledge, Ontology, and Logical Entailments Michaelis Michael

17

35

4.

Epistemic Bootstrapping Peter Forrest

53

5.

More Praise for Moore’s Proof Roger White

67

6.

Lotteries and the Close Shave Principle John Collins

83

v

vi

Contents

7.

Skepticism, Self-Knowledge, and Responsibility David Macarthur

97

8.

A Reasonable Contextualism (or, Austin Reprised) A. B. Dickerson

115

9.

Questioning Contextualism Brian Weatherson

133

Part B Understanding Knowledge? 10.

Truthmaking and the Gettier Problem Adrian Heathcote

151

11.

Is Knowing Having the Right to be Sure? André Gallois

169

12.

Knowledge by Intention? On the Possibility of Agent’s Knowledge Anne Newstead

183

13.

Gettier’s Theorem John Bigelow

203

14.

Knowledge that Works: A Tale of Two Conceptual Models Stephen Hetherington

219

Acknowledgements This book originated in a conference held at the University of New South Wales in December 2004. Being the first major epistemology conference in Australia since the mid-1980s, much expert energy and enthusiasm was on display. Early versions of some of the book’s essays were among those presented. Michael Forshaw and Soon Ng assisted me with the conference organization. Peter Slezak encouraged me to develop this volume. Fiona Barron at Elsevier provided friendly and capable editorial oversight. S. H. Sydney, December 2005

vii

This page intentionally left blank

viii

Contributors

John Bigelow

School of Philosophy and Bioethics, Monash University, Australia

John Collins

Department of Philosophy, Columbia University, New York, NY, USA

A. B. Dickerson

School of Professional Communication, University of Canberra, Australia

Peter Forrest

School of Social Science, University of New England, Armidale, Australia

André Gallois

Philosophy Department, Syracuse University, Syracuse, NY, USA

Adrian Heathcote

Department of Philosophy, University of Sydney, Australia

Stephen Hetherington

School of Philosophy, University of New South Wales, Sydney, Australia

David Macarthur

Department of Philosophy, University of Sydney, Australia

Michaelis Michael

School of Philosophy, University of New South Wales, Sydney, Australia

Anne Newstead

School of Philosophy, University of New South Wales, Sydney, Australia ix

x

Contributors

Tim Oakley

Philosophy Program, La Trobe University, Australia

Brian Weatherson

The Sage School of Philosophy, Cornell University, Ithaca, NY, USA

Roger White

Department of Philosophy, New York University, New York, NY, USA

Chapter 1

Introduction: The Art of Precise Epistemology Stephen Hetherington

1. Questions If we think philosophically about knowledge and associated phenomena, then we are engaged in epistemology. Are we thereby immersed in an autonomous mode of reflection about the world of reflection? Are we focusing with exactitude and rigour upon cognitive states and occurrences? Do we fashion falsifiable theories as to how our minds uncover, in repeatable ways, the nature of the world? Will we know what it is to know? Are we rational in what we believe about rational belief? We might well think so; yet what kind of cognitive insight would we thereby have? Will it be scientific? Should epistemology be part of cognitive science? Or is it somehow methodologically independent of cognitive science? Is it ever at odds with cognitive science, either doctrinally or methodologically? This book provides data with which to refine, ponder, and even to answer, some of these questions.

2. Quine’s Naturalization of Epistemology The picture of epistemology, as not only cognitive but scientific too, gains strength from W.V. Quine (1969), one of the twentieth century’s pre-eminent philosophers. Attentive to language’s nuances, rigorous in his applications of logic, and ever respectful towards good science, Quine was a philosopher’s

Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

1

2

Stephen Hetherington

philosopher. Yet he was also a scientist’s philosopher. His most famous advice to epistemologists, quoted repeatedly since being issued, is this: Epistemology, or something like it, simply falls into place as a chapter of psychology and hence of natural science. It studies a natural phenomenon, viz., a physical human subject. This human subject is accorded a certain experimentally controlled input … and in the fullness of time … delivers as output a description of the threedimensional world and its history. The relation between the meager input and the torrential output is a relation that we are prompted to study for somewhat the same reasons that always prompted epistemology; namely, in order to see how evidence relates to theory, and in what ways one’s theory of nature transcends any available evidence. (pp. 82–83) In other words, epistemology, done properly, is science. Quine urges us to naturalize epistemology, rendering it part of cognitive psychology, tailoring its concepts and methods accordingly. Epistemology should seek scientific integration, not philosophical separateness.

3. Quine’s Farewell to the Concept of Knowledge Which traditional epistemological concepts would survive such integration? Maybe not all will do so. Consider the historically important concept of knowledge. Many epistemologists have sought to understand exactly what knowledge is. But Quine (1987) counsels, pre-emptively, against such a focus. Rather, he advocates discarding the concept of knowledge: The notion of knowledge is beset … by … a vagueness of boundary. Knowledge connotes certainty; what shall we count as certain? Even if one holds that some things are absolutely certain, and is prepared to specify a boundary between absolute certainty and the next best thing, still one would hesitate to limit knowledge to the absolutely certain… . We do better to accept the world ‘know’ on a par with ‘big’, as a matter of degree. It applies only to true beliefs, and only to pretty firm ones, but just how firm or certain they have to be is a question, like how big something has to be to qualify as big.

Introduction: The Art of Precise Epistemology

3

… [F]or scientific or philosophical purposes the best we can do is give up the notion of knowledge as a bad job and make do rather with its separate ingredients. (pp. 108–109) This is Quine the analytic philosopher writing. Conceptual boundaries possess special significance for such philosophers, who generally take philosophy to be inadequate if not precise. Analytic philosophy strives for detailed and accurate delineations of this-as-against-that, particularly of this-concept-as-against-that-one. And there can be correlative pressures to relinquish a concept if it cannot be fully distinguished from related ones. Thus (analytic epistemologists have often asked), what exactly is the difference between knowing and not knowing? Because Quine sees no non-arbitrary answer to this question, he infers that proper epistemology, functioning in the service of science, should lose the concept of knowledge. Must we study, then, not really aspects of a real phenomenon of knowing, but at most what would be knowing’s details if it were to exist? Will a scientific epistemology, imbued with an analytic zeal for precision, tell us only about belief, truth, degrees of confidence, and the like — never about the cognitive phenomenon of knowledge as such? Or is there still much for us to learn about knowledge? This book’s essays illuminate such questions by trying (in their respective ways) to understand knowledge in itself, knowledge as it relates to its possible components, these possible components in themselves, and how best to think about these matters.

4. Drowning in Details As I noted, analytic epistemologists’ hunger for exactitude — precise accuracy. But are there different forms that this could take? Might some of these possible forms assist, while others hinder, our searches for epistemological insight? In his Nicomachean Ethics (1094b24-8), Aristotle proffered this wise observation: For it is the mark of an educated man to look for precision in each class of things just so far as the nature of the subject admits; it is evidently equally foolish to accept probable reasoning from a mathematician and to demand from a rhetorician scientific proofs. So, what level of precision is most apt for an epistemologist? In particular, does analytic epistemology ever become too precise — too much so for its own good as a potentially deep and revelatory way of thinking? This is not a possibility discussed by epistemologists; it should be.

4

Stephen Hetherington

For nowhere in philosophy, it sometimes seems, does questioning become as self-defeatingly precise as can occur within epistemology. I offer this thought only as an hypothesis, ripe for testing, open to refutation. Might some epistemological inquiries have been shaped, at least to a substantial extent, by needless precision, by a drive for details for the sake only of details? Sceptical questions, for instance, all too often attract variations on this sort of frustrated reaction: “Are you really wondering whether I know that I am not dreaming, say? Surely you aren’t serious”. And the debate centred upon those sceptical questions might be rendered as one of finding distinctions where no real differences exist — a debate based upon essentially misleading precision in one’s conceptual overlay. The same possibility might pertain to questions about the nature of knowledge, especially in the wake of Edmund Gettier’s (1963) influential challenge to what had previously been assumed by philosophers to be an adequate understanding of what it is to know a fact or truth: “Are you really wondering whether, in that imagined fanciful situation, the person’s belief is knowledge? Surely you aren’t serious”. Routinely, epistemologists experience ennui at the practically endless circling of attempted solutions to Gettier’s challenge; and this ennui could be sufficient grounds for suspecting that the questioning has been taken too far. Nevertheless, the hypothesized problem runs deeper still. Could it even be that the sort of analytic quest spawned by Gettier — a search for ever greater detail in our accounts of what knowledge is, or of how it can be attained — will of itself prevent those accounts from amounting to an accurate and deep understanding of knowledge? There are at least two reasons for taking this question seriously. (1) If precision is an end in itself, finer and finer distinctions become valued in themselves. And I venture to say that clever epistemologists can always find further conceptual distinctions — possible new forms for knowledge to take, possible new sceptical challenges to our having various kinds of knowledge. The result will be that no clearly correct and complete analysis or theory of knowledge will settle into place, accepted and used by epistemologists.1 (2) There is inductive evidence in support of (1). Epistemology has in fact struggled to achieve full and precise understandings of its designated phenomena (including various epistemic concepts). So much has been promised; far less has eventuated. The history of philosophical engagement with the Gettier problem — the sequence of epistemologists seeking to respond adequately to Gettier’s challenge — is a classic case of that. There have been so many theories, counterexamples, refinements, amendments, and so on. There has been precision galore, with distinctions begetting distinctions, more and 1

In practice, epistemologists often do feel free to make convenient assumptions to the effect of knowledge’s being thus-and-so — where they are fully aware, nonetheless, that this conception of knowledge has not been shown to survive all of Gettier’s challenges, for example.

Introduction: The Art of Precise Epistemology

5

more of them, a potentially endless process. Where does this tense history leave us as epistemologists? Do we not understand knowledge? Committed as we have been to maximal precision, are we correlatively vulnerable to lacking a satisfactory theoretical comprehension of this prima facie key part of our chosen subject matter?

5. Leaving Out Lines There is a methodological alternative. Maybe we need not chase maximal precision. Such a suggestion will sound mistaken, of course. Perplexedly, many philosophers will wonder how this could ever be good epistemology. After all (they will continue), insofar as maximal precision is not sought, it is most likely not gained — in which case, at least correlatively many possible sources of mistake will not have been eliminated. In short, we would have left open doubts (akin to standard sceptical ones) as to whether our epistemological conclusions are true. How then could we know whereof we speak and think as epistemologists? But that objection itself assumes the aptness of a particular picture of how inquiry should proceed. Can epistemology realistically hope to eliminate all possible sources of falsity? Presumably not. Can maximal precision be attained? Seemingly, no. With which acceptance, perhaps we should not aim, quixotically, for such an outcome either. But then we still need to ask ourselves (as I suggested earlier) how much precision is most apt within epistemological inquiry. And at present we should be receptive to proposals. Formulating these, and evaluating them, will require us to reflect upon the nature of epistemological inquiry in general. Here is one possible line of such reflection. Consider the possibility that epistemology is best construed as an artful depiction — akin to a fine representational drawing — of the epistemic world (rather than as a forensic dissection of all of that world’s aspects). And think of different ways of drawing well, even in a representational style. One such way is especially pertinent. Skilful drawing can involve an artist’s ignoring possible lines — including ones that would, if they had not been left out, have been accurate details in a representational portrayal. Leaving out such potential details can strengthen what does appear. Sometimes, what most matters is all that matters: further details would obscure, spoiling the whole. A shadow, say, might be falsified not only for the sake of the whole (such as the balance and power) but also to accord appropriate respect to other details.2 2

Bear in mind, too, that a picture need not really be a response to a specific moment, even though it will assume that guise. It may portray, as if these actually are gathered together at some particular time, data from many moments of observation. It can be a distillation of them.

6

Stephen Hetherington

And why may good epistemology be like that? To do epistemology is to theorize. To do this well is, in part, to choose one theory over another. And, as Quine influentially emphasized, such a procedure is holistic: varied desiderata are compared, emphasized, balanced, combined, and discarded. There is imperfection — optionality and fallibility — in this process. There is logical underdetermination. Correlatively, might there be no single best theory of knowledge, rationality, justified belief, or the like? This is a possibility we should bear in mind (and one to which Quine would have been receptive). We want accuracy, and epistemologists have been professionally trained to believe that precision is essential to gaining this. But we should realize that there are times for leaving out lines — not all, but some, of them as we build the best picture available to us.3 Theorizing is not so unlike drawing in such respects — subject to the overriding demand that something relevant be accurately revealed. Too much precise detail kills a picture, including some of its accuracy. I suspect that it can similarly impede our theoretical grasp of some phenomenon or concept.

6. Discerning Lines I have gestured at the limitative presence of fallibility within epistemological theorizing. Can any of knowledge, justification, rationality, and related phenomena be understood perfectly and precisely? A concept for each has a lower bound. How weakly rational is real rationality? How poor can substantive justification be? How flimsy a link to the world can knowledge be? Where, exactly, does each of these begin and end? Quine wanted us to approach such questions naturalistically. However, as we observed, he also thought that doing so would not help us to understand the presence of knowledge, at least not scientifically. Thus he advised us not to retain the concept of knowledge, if we aspire to formulating a science of cognition. The problem, in effect, was one of leaving out lines. We dare not leave out the right line, the precise dividing line between knowing and not knowing. Exactly how justifiedly must one believe that p if one is to know that p? Exactly how wellorganized must one’s evidence be? Exactly how non-deviant must one’s causal history or social surrounds be? We might have no non-arbitrary answer to these questions (short of espousing an unwelcome and unrealistic infallibilism about 3 Perhaps even a misleading line may be retained because it helps to support a vital line, drawing attention to the latter, strengthening its impact. Elgin (2006) might appreciate this idea’s pertinence, because she talks of understanding’s not needing to eliminate all falsity from within itself, even as an insightful scientific picture is developed.

Introduction: The Art of Precise Epistemology

7

what it takes to have a piece of knowledge). However, in that case we must acknowledge every line. Each will matter equally to the whole. And then, we might infer, the whole is unmanageable as an analysis of knowledge. Hence, we should discard the concept of knowledge: rip up that picture; begin again! Then again, let’s not. We might, alternatively, settle for limits upon the level of precision we can usefully attain in any epistemological description. Does this remove knowledge beyond our epistemological ken? Not necessarily, but the possibility is accepted that knowledge might not be understandable as precisely and simply as epistemologists have generally assumed it to be. The problem (if that is what it is) need not be wholly within us. Maybe knowledge is vague — indeterminate at its boundaries (an implication with which Quine is uncomfortable). Perhaps (as, we saw, Quine allows)4 it has degrees — substantive variabilities within its nature as knowledge, within its own boundaries. And, of course, maybe we only ever discern any of knowledge’s more subtle features with less-than-total rational certainty. Such a picture of knowledge5 coheres with (while not rendering mandatory) the suggestion that our best understanding of knowledge will be at least somewhat like an artist’s, as we draw upon only so many of the possible lines of thought. We might not even notice all of the potential lines, as we ponder our next move. But even among those options we do observe, we should choose without thinking that the result will be the only equally accurate, let alone good, representation that could have been found. Quine wanted a scientifically realistic epistemology; fallibilism would be a key component of it. Can we combine all of this with continuing respect for a concept of knowledge? This book provides some evidence that we can.

7. This Book’s Essays The contributors to this volume offer varied answers bearing upon some of the questions I have been outlining (and upon related ones, naturally). Those questions may usefully be organized as clarifying and developing these two larger ones: (A) Science. How much scientific integration is possible for epistemology? When done as, ideally, it should be done, would epistemology be a science — a science of whatever is normative in cognition? (B) Knowledge. Towards that end, should epistemology discard what has traditionally been its centrepiece — the concept of knowledge? Is 4

And as I have argued, more fully, in Hetherington (2001), but without dismissing the concept of knowledge as thereby being unmanageable. 5 It is a picture explored by Hetherington (forthcoming).

8

Stephen Hetherington knowledge not sufficiently amenable to being understood as well as is needed for scientific purposes?

And so the book’s two parts arise, corresponding in turn to (A) and to (B).6 Part A. Epistemology as Scientific? The first two essays reflect upon how, if at all, we might understand some possibly basic cognitive phenomena — namely, basing and doxastic commitment. We begin with Tim Oakley’s analysis of what it is to gain a justified belief on the basis of another. At the heart of Quine’s naturalization of epistemology is the notion of our basing beliefs upon sensory input, our basing theories upon evidence, and of our thereby making epistemic progress. How well do we understand such dependence, though? Oakley questions our conceptual grasp of how justified beliefs are to emerge from a basing relationship. Nor, for that matter, might it be so clear how a specific belief — this one, as against some other — is ever present in the first place. Michaelis Michael’s essay highlights some of the associated subtleties, asking what commitments give this (rather than that) content to a particular belief or case of knowledge. The philosophical puzzles do not end there. Wherever epistemic uncertainty or underdetermination lurks, epistemologists espy sceptical potential. Is this already enough to prevent our making epistemic progress in our cognitive efforts? The next two essays, by Peter Forrest and by Roger White, bear upon this. They examine the capacity for there to be epistemic progress within a cognitive system (incorporating probability or plausibility assignments, say) without the epistemic subject occupying an “external” Archimedean point, perfect for assessing the epistemic strengths of his or her beliefs, say. The associated metaphor upon which Quine repeatedly (e.g. 1969, p. 84) called was Otto Neurath’s (1959, p. 201) famous picture of a boat at sea, to which repairs can be made while at sea. And Forrest’s and White’s essays exemplify that approach. Forrest develops and applies a concept of epistemic bootstrapping — improving one’s warrant for a belief, even in the face of sceptical possibilities, without ever stepping away from one’s corpus of belief so as first to dispose of scepticism. White explores the justificatory potential of

6

It should be noted that a few of the book’s chapters bear upon each of Part A and Part B. For example, John Bigelow’s essay tests epistemology’s capacity to formalize its own tenets, and hence could well have been included in Part A rather than Part B. And Brian Weatherson’s essay strives to understand knowledge, insofar as it reflects upon the means by which we would do this. It could therefore have been incorporated into Part B rather than Part A. Such organizational flexibility is welcome. It reflects the myriad possibilities for conceptual interweaving between the book’s two parts.

Introduction: The Art of Precise Epistemology

9

G.E. Moore’s (1959) infamous “proof of an external world”, even while accepting that it is not an anti-sceptical proof. It does not do battle with the sceptic who is looking on at our cognitive efforts “from outside”. Can it nonetheless assist us rationally? Should it help us to improve the alignments of our beliefs and acceptances with our assessments of justification and of rationality? Many epistemologists will not believe so. They will feel that justice has not been done to sceptical doubts. Much hangs on this, because then the question arises of whether such doubts are anathema to the hope of epistemology’s ever becoming suitably scientific. Quine did not believe that they need be, so long as we keep our doubts reasonable. In particular, he regarded sceptical doubts, when reasonable, as scientific — and, when not scientific, as not reasonable. They must emerge from careful sensitivity to observable fallibility. Congruently, we find John Collins’ essay seeking to undermine what has recently been thought by many epistemologists to be a potential route to scepticism — namely, the speedy assimilation of people’s not knowing themselves to have lost a lottery to their knowing … well, much at all. If lottery ticket-holders do not satisfy a minimum standard for knowing that they will not win (in spite of their excellent evidence in favour of this), then when do people ever know something? Rarely? Never? Collins resists such sceptical inferences, by distinguishing what is far-fetched from what is merely improbable, all the while attending to actualities in this scientifically describable world. Undaunted, though, the history of philosophy confronts us with further sceptical ideas, often claimed to challenge our cognitive powers rather forcefully. But (we might wonder) has science shown us how to accommodate or transcend those sceptical ideas (originating, as they did, in less scientifically sophisticated times)? Not if David Macarthur’s essay is correct. He traces part of our ongoing susceptibility to sceptical thinking precisely to our tendency to accord a scientifically influenced naturalism a significant say in how our cognitive domain is to be epistemically evaluated. Naturalistic stances (he would say) clash with first-personal deliberative ones; and each of these can be overdone, especially the naturalistic approach. Is sceptical thinking therefore wholly inevitable for us? Must we accept it? A.B. Dickerson, for one, is not convinced of scepticism’s power. He argues that it can be undermined; and, moreover, that this can take place “from within” a relevant system — in particular, by attending to ordinary language. The mid-twentiethcentury English philosopher J.L. Austin is Dickerson’s inspiration in this project; and Austin was not someone who would have accepted that philosophy should ever be supplanted by science. Part A’s final essay, by Brian Weatherson, complements Dickerson’s without concentrating upon scepticism. Dickerson regards Austin as having provided an early, and potentially insightful, version of what is now called contextualism. Contextualists claim both to understand and to de-fang sceptical doubts, by

10

Stephen Hetherington

attending to nuances of language use within different contexts. Austin saw ordinary language not as the final word on such matters, but at least as the first word. How far, in a given case, can it then take us? Weatherson’s essay asks this same question, arguing that contemporary contextualism, in spite of the current enthusiasm for it among epistemologists, is not particularly epistemically revelatory. Where does this leave us as we seek to understand the normative dimension of cognition? Various questions will have been raised. What should our data be? (Will they be linguistic, for instance?) How scientific should we expect the resulting theories to be? Will sceptical thinking, most notably, forever impede epistemology’s becoming genuinely scientific in its reflections upon evidence, belief, knowledge, and their cognitive cousins? Part B. Understanding Knowledge? We noted two strands in Quine’s epistemological thought. From Section 2: He exhorts us to understand the epistemic domain scientifically, absorbing the results within cognitive psychology. From Section 3: He also warns us against including knowledge within our epistemic science, our epistemology. Part A has tested the former exhortation; Part B amounts to a reaction to the latter warning. Must we heed Quine’s warning? Should we accept that no adequate understanding of knowledge is possible? In fact, Quine himself did not say very much about what knowledge is. But our next two essays, by Adrian Heathcote and André Gallois, do so. Situated within epistemology’s post-Gettier enterprise (alluded to in Section 4), each of these essays strives to clarify knowledge. Each highlights a potentially pivotal aspect of knowledge — respectively, facts and rationality. Heathcote argues that contemporary epistemologists have lost sight of knowledge’s being an appropriate relationship to a fact. (Instead, they have mistakenly conceived of it as being a relationship to a true proposition.) And what kind of relationship is required? Gallois contends, possibly controversially, that it is a relationship of rationality. Knowledge that p (according to Gallois) is simply the presence of a kind of rationality: it is one’s having the right to be sure as to p. Significantly, though, Gallois does not also require the presence of any actual confidence as to p. If he is right, therefore, knowledge would not be a kind of belief (or some similar cognitive state), contrary to what is usually presumed by cognitive theorists. The metaphysical richness of those two essays continues in the next one, by Anne Newstead. She urges epistemologists to expand the range of facts or cognitive contents of which, we allow, people can have knowledge. Part of understanding knowledge in general is our understanding what kinds of knowledge there are, or might be. And one intriguing possibility, according to Newstead, is that of agent’s

Introduction: The Art of Precise Epistemology

11

knowledge — the knowledge, in advance, of one’s intended actions. This possible form of knowledge has received little attention within epistemological writing. Newstead makes explicit some potential links between epistemology and action theory. Of course, the question remains of whether, when constructing such proposals as to what knowledge is, we can understand it. Might there be underlying difficulties here, even insuperable ones? Gettier’s is probably the most substantial contemporary challenge to philosophical conceptions of knowledge; and John Bigelow’s essay reinforces the depth and possible intractability of that challenge. He describes illuminating parallels with Bertrand Russell’s momentous paradox, launched so devastatingly against Gottlob Frege’s attempt to reduce mathematics to logic and set theory. If Gettier’s proper impact is as powerful as Russell’s was, mutatis mutandis, a fully precise analysis of fallible knowledge might forever elude us. On the other hand (and as Section 4 presaged), we might also ask whether epistemologists have made that project needlessly difficult. They have traditionally been prone to seeking proofs of the presence of knowledge before being willing to attribute it. Focusing upon sceptical possibilities and the Gettier challenge as test cases, my own essay counsels against this temptation. I advocate instead a thoroughly fallibilist, maybe even pragmatist, conception of knowledge and of how we will ever understand its presence. (I also describe some historical optionality in contemporary epistemological methodology’s ways of trying to understand knowledge.) Is this fallibilism a gloomy result? Not necessarily; if epistemology is ever to be scientific, we should expect an unrelenting fallibility in its pronouncements.7 Philosophers and others might have no infallible insights into the nature of knowledge, and this in itself might affect what they should say about its nature.

8. Australian Epistemology This book’s essays comprise the first substantial published body of Australian epistemology.8 Australian analytic philosophy has a deservedly strong international 7

For more on the nature of fallibilism, see Hetherington (2005). Note that one implication opened up by my fallibilist conception is a gradualism about knowledge — accepting that there can be degrees of knowledge that p, or that there can be epistemically better or worse knowledge than p. This is a position almost never discussed, let alone endorsed, by epistemologists (Note 4 adverted to it). Quine, as quoted in Section 3, is an exception to the former trend, but not to the latter. At least he noticed the gradualist potential in how we should think about knowledge. Unfortunately (as we also saw), this made him wary of the concept of knowledge. But if I am right, such wariness was not needed. We can profitably conceive of there being degrees or grades of knowledge that p. 8 By “Australian”, I am referring to philosophers Australian and who are presently or recently attached to Australian universities.

12

Stephen Hetherington

profile, especially in the central areas of methodology, metaphysics, ethics, philosophy of language, and philosophy of science. But implicit in much of that work have been various epistemological ideas and commitments. And this book aims to uncover many of these, building upon the conceptual links between Australian epistemology and some of these other aspects of Australian philosophy. For example, attention to the origins and applications of analytic methodology underlies the essays by Bigelow, Oakley, and me. A command of aspects of the philosophy of science informs Forrest’s, White’s, and Collins’ essays. Philosophy of language is examined and applied in the papers by Dickerson, Michael, and Weatherson. And metaphysical details underlie the contributions by Gallois, Heathcote, Macarthur, and Newstead. The collective result thus takes its place as a distinctive continuation of Australia’s fine tradition in analytic philosophy, while also expanding that tradition by doing fuller justice to its epistemological connections. Moreover, the book’s appearance within a Series on cognitive science is most apt. Australian philosophers have paid noteworthy attention to questions about the mind, the brain, and associated details. They have sought systematicity and rigour, often aspiring to being as scientific as possible. Yet how scientific is it possible to be, when thinking philosophically about cognition? As philosophers realize, often frustratedly, the relevant data (of which such reflection must take account) include putatively normative aspects of how we think. That is where this book enters the conceptual fray. Even if the philosophy of mind, for instance, can become a science, will epistemology follow suit (as Quine envisioned)? How scientific can epistemology ever become, if it is to understand cognition’s normative aspects? Can it become as scientific as its paradigm analytic practitioners would like it to be? How, if at all, can it understand the nature of knowledge? The picture that will artfully and accurately answer those questions is still being drawn by theorists. This book contributes some potentially pleasing lines to that picture.

References Aristotle, Nicomachean ethics. In: R. McKeon (Ed.), The basic works of Aristotle. New York: Random House (1941). Elgin, C. Z. (2006). From knowledge to understanding. In: S. Hetherington (Ed.), Epistemology futures (pp. 199–215). Oxford: Clarendon Press. Gettier, E. L. (1963). Is justified true belief knowledge? Analysis, 23, 121–123. Hetherington, S. (2001). Good knowledge, bad knowledge: On two dogmas of epistemology. Oxford: Clarendon Press. Hetherington, S. (2005). Fallibilism. The internet encyclopedia of philosophy, http://www.iep. utm.edu/f/fallibil.htm.

Introduction: The Art of Precise Epistemology

13

Hetherington, S. (forthcoming). Knowledge’s boundary problem. Synthese. Moore, G. E. (1959). Proof of an external world. Philosophical papers (pp. 127–150). London: George Allen & Unwin. Neurath, O. (1959 [1932/33]). Protocol sentences. In: A. J. Ayer (Ed.), Logical positivism (pp. 199–208). Glencoe, IL: The Free Press. Quine, W. V. (1969). Epistemology naturalized. Ontological relativity and other essays (pp. 69–90). New York: Columbia University Press. Quine, W. V. (1987). Quiddities: An intermittently philosophical dictionary. Cambridge, MA: Harvard University Press.

This page intentionally left blank

14

PART A EPISTEMOLOGY AS SCIENTIFIC?

15

This page intentionally left blank

16

Chapter 2

A Problem About Epistemic Dependence Tim Oakley

The idea of epistemic dependence is this: in some cases, a person’s being justified (at a time, in a given situation) in one belief, p, will epistemically depend on that person’s being justified in another belief, q. For short: justifiedness in p depends on justifiedness in q; or even shorter: Jp dep Jq. The main point to be made in this paper is that it is extremely difficult, if not impossible, to give a coherent analysis of this apparently innocent and perspicuous notion. Consequential doubt is cast on the very intelligibility of some large issues about the structure of our systems of justified beliefs — in particular, the debate between foundationalists and coherentists. In Section 1 of this paper, I will outline the role of epistemic dependence in these structural debates.

1. Foundationalism, Coherence, and Epistemic Dependence Here is a natural way of briefly explaining foundationalism: foundationalism claims that most of our beliefs depend for their justifiedness on the justifiedness of other beliefs, but that there are some beliefs (referred to as “basic” or “foundational”) which are justified for a person at a time, which do not depend for their being justified on any other beliefs of that person being justified. Allowing the transitivity of dependence, foundationalism also states that all non-basic beliefs depend for their justifiedness on basic beliefs.1 1 Such an account of foundationalism of course presents it as a theory about the structure of our system of justified beliefs, and is not an analysis of the concept of justification. If it were supposed to be

Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

17

18

Tim Oakley

Such an account of foundationalism may be natural, but if we wish to think of coherence theories as rivals of foundationalism, it is completely unsatisfactory, since most coherence theories assert the existence of basic beliefs, and have other beliefs depending for justifiedness on them. So on the above account, there would be no difference between foundationalist and coherentist theories. It is true that some forms of coherence theory may deny the existence of basic beliefs, and claim that all justified beliefs depend for their justification on the justifiedness of other beliefs. However, most coherence theories (while not using the terminology of “basic beliefs”) avoid any use of the concept of a belief’s justifiedness depending on another belief’s justifiedness, and say simply that a belief is justified if it coheres with a body of other beliefs, with no presupposition that the body of beliefs is either severally or jointly justified.2 (Neither is it necessary to explain the concept of coherence in terms of dependence or any similar concept. The coherence of one belief with a set of beliefs may be accounted for in various ways — for example, in terms of support relations, or in terms of the belief in question adding to the explanatory and predictive power of the set with which it is said to cohere, and so forth.3) Just as most coherence theories in fact assert the existence of beliefs that do not depend for their justifiedness on the justifiedness of other beliefs, so too do contextualists. A contextualist will say (something to the effect that) that many beliefs are justified by virtue of being derived from currently unquestioned other beliefs via currently unquestioned principles of inference, there being no requirement that the currently unquestioned beliefs or principles be justified. These justified beliefs do not depend for their justification on any other beliefs being justified. So we had better not define foundationalism simply as a theory that asserts the existence of basic beliefs — i.e. beliefs, which do not depend for their justifiedness on the justifiedness of any other beliefs. Rather, it needs to be specified as not just claiming that there are basic beliefs, but as (a) offering particular accounts of these beliefs and conditions sufficient for their being justified, and (b) an account of how all the rest of our justified beliefs depend for their justifiedness on these basic beliefs — a matter on which I shall say more immediately below. The commonest modern form of foundationalism remains, of course, empiricism, an analysis, it would clearly fail due to circularity. In fact, the account above is compatible with a number of analyses of justification, including reliabilist and evidentialist ones. Note also that we are concerned only with justification foundationalism. Points that may be made about the structure of knowledge, as opposed to justified belief, may or may not parallel the points made here. 2 For discussion of recent proposals about coherence theories, see Bender (1989). 3 Thus, it should be noted also that coherence theories should not be characterised (or criticised) as theories that permit circularities of dependence — that is, permit justifiedness in p to depend on justifiedness in q at the same time as justifiedness in q depends, albeit via a chain of other beliefs, on justifiedness in p.

A Problem About Epistemic Dependence

19

which regards beliefs about observations as basic, and which offers either experientialist or reliabilist accounts of the justification of basic beliefs. Call a chain of beliefs in which one depends for its justifiedness on the justifiedness of the next, which depends for its justifiedness in turn on the justifiedness of a third, and so forth, a “dependence chain”. Foundationalism of course envisages a tree structure with branching dependence chains. Justifiedness in any given belief, say the belief that p, if it depends on the justifiedness in another belief, say, that q, is likely to depend on justifiedness in a number of other beliefs as well, and each of these on a number more. It may also depend in part on other conditions being fulfilled as well, for instance the believer making a certain observation, or something else again. (Thus, the making of an observation or the having of a certain experience may, on some foundationalist theories, be sufficient for justifiedness in basic beliefs, but that is not to say that such observations or experiences may not also be necessary for the justification of some non-basic beliefs.) So, when we speak of justifiedness in p depending on justifiedness in q, we are not of course speaking of q justifying p. The former, like any other sort of dependence — logical dependence, causal dependence, dependence in virtue of a convention, etc. — is a notion related to necessity, the latter to sufficiency. If x depends on y in any sense whatever, then the idea is that y is necessary, in that sense, for x. The familiar “infinite regress of reasons” argument is standardly presented as an argument for foundationalism. This is wrong.4 What the argument establishes, if it establishes anything, is simply that there are basic beliefs, which claim we have noted will gain the assent of contextualists and most coherentists. This is not an unimportant conclusion however: if we indeed have some justified beliefs, and if we wish to avoid the conclusions either that dependence chains can be (indeed, are) infinitely long, or that dependence chains can be circular, the dependence chain must end (so there is a basic belief). My main concern in this paper is with the concept of dependence as it is found in the context of foundationalist theories, and in the more general context of the infinite regress argument. But it shows up in other areas as well. Many think (rather plausibly) of beliefs of one type being “justified in terms of” beliefs of another type: beliefs about the future in terms of beliefs about the present; those about other minds in terms of those about behaviour; those about the physical world in terms of those about sensory experiences. It is plainly being assumed, and is sometimes stated explicitly, that justifiedness in beliefs of the first sort depends on justifiedness in beliefs of the second sort. Epistemologists seek arguments with premises entirely of the second sort (statements about the present, about

4

This is a mistake I once made myself. See Oakley (1976).

20

Tim Oakley

behaviour, and so forth) and conclusions of the first sort (statements about the future, other minds, and so forth) to try to establish that we are indeed justified in our beliefs of the first sort. All this only makes sense on the above assumption: that justifiedness in the one sort of belief depends on justifiedness in the other sort.

2. Why an Account of Dependence is Needed I will argue that no satisfactory account of epistemic dependence has been given, and probably none is capable of being given. Accounts of dependence, however framed, all have defects, such as the consequence that there are no basic beliefs. (Perhaps in fact, there are indeed no basic beliefs, but this conclusion is surely a substantive one, and had better not flow straight out of the conceptual machinery used to set up the issue.) As a result of this lack of an account of dependence, sense cannot actually be made of foundationalism and the dispute between foundationalists and their opponents, and serious doubt is cast on the other epistemological enterprises that have been making use of the notion of dependence. But do we need an analysis of dependence? In this section, I will consider two reasons that may be proposed for there being no need for one. First, it may be said that the notion is perfectly clear as it stands, and we all know what it means, so it can be used without any further clarification. Second, it may be said that even if there are found to be problems with the concept of dependence, there are other terms in which the regress argument can be (and indeed frequently is) framed. I take these matters up in the remainder of this section. First, must we have an analysis of every concept we use? Of course there is no universal rule that dictates that we must abandon a concept if we find that we cannot provide an analysis of it. It is notoriously difficult to provide an account of the concept of propositional knowledge, but that on its own is not accepted as a reason for rejecting the concept as bogus, or saying that our attributions of knowledge are nonsensical. However, there is a rule (not always obeyed, but that is another matter) that those using a term of art, introduced in the context of setting up a theoretical problem or position, are under an obligation to give a precise account of its meaning. A definition of “dependence” is needed because the notion involved is a technical rather than an ordinary language one, as in the case of “knowledge”, where an unanalysed intuitive understanding might have been sufficient. It will be acceptable to say in the case of “knowledge” that despite the difficulty in coming to an analysis of the concept, it is nonetheless clear that we have an intuitive grasp of the notion, and that there is no doubt that there is a coherent concept (or perhaps a systematically connected set of concepts). In the case of dependence, the notion is introduced by philosophers to set up a certain theoretical issue: simply

A Problem About Epistemic Dependence

21

put, the notion of dependence is introduced in order to assist the neat articulation of the infinite regress argument for basic beliefs. There is in fact another rule about when we need analyses of concepts (again not always observed). We need analyses of concepts that are introduced to do an explanatory job. Dependence features in the foundationalist explanation of how we get to be justified in our beliefs. The explanation of our being justified in most of our beliefs is that the justification depends on justification in a particular subset. Lack of an exact account of the concept of dependence involves a corresponding lack of explanatory power in the theory.5 I conclude that we do need an analysis of the concept of epistemic dependence. Let me turn, then, to the second proposed reason for dismissing the need for an analysis. It will be objected that discussions of foundationalism, and of justification of various categories of beliefs in terms of other categories, need not be — and often are not — set out in terms of dependence. So even if there are, as I claim, problems about that concept, these theories are not threatened. Most commonly, in fact, foundationalism is introduced via a distinction between inferred and uninferred beliefs.6 So it is tempting to see the concept of dependence as being either analysed in terms of, or replaced by, the concept of inference, which is, after all, a good common-sense, ordinary language notion, which we all understand whether or not we have a precise analysis of it. Can we not simply say that a subject’s justification in p depends on her justification in q if and only if p has been inferred from q, in accordance with a sound

5

Compare one strand of the Quinean attack on the concept of analyticity, which was to the effect that no adequate account of the notion had ever been given. Quine was happy enough to agree that we could learn to classify together certain statements, which intuitively seem to belong together, and which we could learn to dub “analytic” or “necessary” in contrast to other statements, which similarly seem to belong together, and which we could learn to call “synthetic” or “contingent”. His concern was not with these intuitive classes, but with the fact that the notion of analyticity had been pressed into explanatory service — (allegedly) explaining, inter alia, our possession of certain sorts of knowledge, and our capacity to understand certain parts of language. He thought that the lack of an adequate account of the notion was serious in view of the use to which the concept was being put in philosophy. He was right (see, e.g. Quine, 1967). 6 See, for example, Everitt and Fisher (1995, pp. 74ff.); Pojman (1995, pp. 94–95); Post (1992); Feldman (2003, pp. 50ff.); Bonjour (2003); and Fumerton (2002, pp. 210ff.). Some mix their use of the terminology of inferential and non-inferential justification with other terms such as “based on”, “derived from”, “premise beliefs”, “conditionally justified beliefs”, etc. Interestingly, Russell (1967, p. 64) is more cautious than many others, qualifying his use of the terms “inference” and “reason” as follows: “Almost all our common beliefs are either inferred, or [my emphasis] capable of being inferred, from other beliefs which may be regarded as giving the reason for them. As a rule, the reason has been forgotten, or has even never been consciously present to our minds”. As will emerge, I think he was right to put matters this way.

22

Tim Oakley

principle of inference, and the subject is justified in q? More briefly, assuming we have one subject in mind throughout, and, a trifle casually, using the abbreviation “Jp” for both the noun phrase, “justifiedness in p”, and for the sentence, “p is justified”: Jp depends on Jq iff (1) p has been inferred from q in accordance with a sound principle of inference, and (2) Jq. Such an account admits to variation of detail: for instance, we might ask whether the principle of inference has to be sound, or whether the subject must merely be justified in believing that it is, or perhaps even just believe, justifiedly or not, that it is. We may skirt such matters here. The basic problem for any account in terms of inference is that, on any normal understanding of the term “infer”, it is quite clear that there are such things as uninferred beliefs. There is no need for an argument that says that there have to be such beliefs on pain of infinite regresses or circularities. And these uninferred beliefs are not the observation-based beliefs normally thought of as basic by foundationalists. Beliefs of all sorts, including quite theoretical ones, may pop into one’s mind spontaneously, and not as the result of anything describable as an inferential process. And they may be justified, as Russell pointed out (see footnote 6), not by having been inferred, but by being inferable from other propositions in which we are justified. We may accept these other propositions, or (once again following Russell’s suggestion) they may be propositions that we do not accept, but that we would be justified in accepting, were we to accept them. So our justifiedness in these uninferred beliefs will often depend on justifiedness in other propositions that we may not actually accept, and from which we certainly did not infer the proposition in question. There are other reasons why the dependence relation has nothing to do with inference. Accounts of an inferential link between beliefs are accounts of the genesis of beliefs. But often the genesis of a belief has nothing to do with its justifiedness. Belief p, long held unjustified since it was inferred from bad reasons, may come to be justified on the strength of our acquiring, late in the piece, good reason q. Here justification in p depends on justification in q, but p was not inferred from q. We might be tempted to explain inference in terms which make inference not a matter of genesis of beliefs, but a matter of something which sustains belief, but it will be seen later why this will not do either. Similarly, we might also be tempted to make use of a notion of unconscious inference, but then we will face the problem of accounting for just when one belief is supposed to be unconsciously inferred from another, a problem not soluble by appeal to common intuitive

A Problem About Epistemic Dependence

23

understanding, and a problem essentially similar to that of accounting for the idea of one belief’s justifiedness depending on that of another.

3. With What Kind of Justification Are We Concerned? To make my case concerning the problems of accounting for epistemic dependence, we need to attend to exactly what sort of justification we are talking about here, since much work in the last 20 years or so has pointed to distinctions between different kinds. In particular, a case has been made for a distinction between a subject, S, being justified in believing p, and the belief in p being justified for S.7 Before discussing this distinction, a clarification can usefully be made between a belief and a proposition. The word “belief ” can refer to a psychological state of a subject, a state that has causal and psychological relations to other states of the subject and of the world. On the other hand, it can refer to the content, the proposition believed. The proposition has logical relations to other propositions, but it does not, as an abstract object, have causal relations with anything else. Let us refrain from using the word “belief” for propositions believed, and reserve it for those psychological states that are assentings to the truth, or acceptances, of propositions. Sometimes we speak of a proposition being justified, making no reference to anyone believing it to be true. For example, we say that it is well justified that the Earth is (roughly) spherical. We might even say the same of a proposition without any presupposition that anyone actually does accept it. (For instance, we may have evidence for the truth of a mathematical claim that is not believed by anyone, simply because it is literally too long and complex to formulate, and where we refer to it indirectly in some way.) A claim that it is well justified that the world is spherical presumably means that there is much evidence available for this claim. But availability is always availability to someone. A proposition will thus be justified relative to the evidence held by an individual (at a time, in a given situation) or to the evidence held by a particular group, community, civilisation, or whatever. The evidence may not be accepted by any particular people, collectively or jointly, and may simply be the evidence to be found in a certain body of texts. In any case, there is always the possibility that there will be others for

7

For discussions of this and other distinctions among types of justification, see, for example, Oakley (1988); Engel (1992) and the criticism of Engel in Reiter (1998); Fumerton (2002); and for a view of “justification” as covering a considerable range of notions, see Alston (2005). Note that the point here is not of diversity among analyses of justification; it is rather that there are more than one separate analysandum requiring attention.

24

Tim Oakley

whom the proposition is not justified. The justifiedness of a proposition is always in the end a matter of some person or persons, real or hypothetical, in real or supposed possession of certain evidence, being such that they are or would be justified in accepting the proposition. Russell’s reference, quoted in footnote 6, to “other beliefs”, any one of which constitutes our reason for a given belief, but which possibly “has been forgotten or has never consciously been present to our minds”, can presumably be seen to be speaking of propositions which support the one for which they are the reason, and which are such that, were we to accept them, we would be justified in doing so. I turn now to the distinction between a belief’s being justified (which we may call doxastic justification) and the person’s being justified (personal justification).8 The fundamental idea is that a belief may be justified because of the support available for it, or perhaps because it has been formed through the use of a reliable belief-forming process, while the judgment that a person is justified may be a claim much more specifically about the person — usually about the satisfactory nature of their epistemic performance. (This in turn may be explained in terms of behaving in an epistemically responsible fashion, or fulfilling epistemic duties, or being attentive to and aware of the evidence which supports the belief in question, or in some other way again.) The basic distinction is best illustrated by the (relatively rare) sort of case where doxastic and personal justification come apart. Someone believes he faces financial ruin on the basis of an unpleasant dream he has just had. In fact, he also has available to him much evidence that points to the same conclusion, and he believes this evidence. However, prior to the dream, he has refused to accept the truth of the conclusion that (we might say) he ought to have drawn from the evidence in his possession. In such a case, the answer to the question, “Is the person justified?” is likely to be “No”. But the answer to the question, “Was the belief justified?” might well be “Yes”. Generalising, we are prepared to call the belief p justified if good grounds can be found for it from among the corpus of justified beliefs held by the subject, S, but will be inclined to call S unjustified in believing p if S neglects those good grounds, and believes p on other grounds which are not good, or on no grounds at all. Putting the two points together in the same sentence jars, of course, but we usually do not make them in the same breath, but raise different concerns in different contexts. Our motivations for asking about justifiedness vary. Sometimes we are interested in a person’s epistemic performance; we are marking a mathematics examination and want to know if the student reached a correct answer by a

8

Others have different terminology for much the same distinction. See, e.g. Fumerton (2002, p. 206) for the distinction between having justification for a belief, and having a justified belief.

A Problem About Epistemic Dependence

25

good process of reasoning or by a lucky guess. We are not interested in the justification of the belief — we already know it to be true, and have more evidence for its truth than we know what to do with. On the other hand, we may be very keen to know whether some belief that has been communicated to us by a person holding it is justified. We may want to know whether we should adopt it for ourselves. As long as there exists evidence for its truth, we may not care at all about whether the person informing us used that evidence, or merely based his belief foolishly on a dream. It is true that the notion of a satisfactory epistemic performance is hard to specify, but we have a fair idea of what we are prepared to call a bad performance, and I take it that the intuitive notion will suffice here.9 The justification of the belief may be granted in other cases where we would deny justification to the person: e.g. where the subject is in possession of good reasons for a belief, but is unable to formulate the reasons coherently without a lot of prompting, or perhaps is still unable to articulate her argument even after prompting. Or where a person’s justification was “gappy”, that is, where she would, on the basis of other justified beliefs she holds, be justified in accepting proposition q, and where q would support p, but she fails to accept q. (Once more, see the quote from Russell in footnote 6.) Similarly of course, S may be justified while the belief is not justified, as where she justifiedly but falsely believes herself to have good evidence, or to be using a reliable belief-forming process, etc. Assuming the acceptability of the doxastic/personal distinction, then, the issue obviously arises as to which sort of justification — doxastic or personal — has been the concern in the discussions between foundationalists and their critics. It seems clear that these theorists have standardly taken themselves to be discussing beliefs and not believers. This is because they have been interested in the shape and structure of belief systems. These systems are, granted, belief systems of individuals, though the concern with individuals seems to fade to the background rather,

9

Some may be inclined to object that there is really no such thing as doxastic justification separate from personal justification. They may attempt to reinterpret the “dream of ruin” case as one where the person (S) who believes that p (that he faces ruin) is unjustified (for the reasons outlined) but where we, the assessors of the situation, are justified in believing p, because we, like S, are in possession of evidence (q, r, and s) which supports the belief, but unlike S, we recognise the evidential connection, and believe p because we believe q, r, and s. This would be wrong. It could be the case that q, r, and s are false and that we know they are false. We might also know that S’s belief in them was very well justified. (Perhaps for some odd reasons, S had been cleverly and systematically deceived and led by an elaborate scheme to hold various false beliefs about his financial situation. To the chagrin of the plotters, he failed to draw the conclusion that he faced ruin. Finally, however, a dream did the trick.) In such a situation, we are not justified in believing p, but may still want to speak of the belief as justified.

26

Tim Oakley

and they could as well be dealing with systems of communally shared beliefs. Indeed, the interest has largely been with propositions and the evidence for them. Belief systems were in that sense idealised: they were sets of propositions linked by support relations. In asking about a belief system, the interest has been in whether or not any particular belief(s) could be shown to be justified by reference to other beliefs, and ultimately to beliefs justified without reference to other beliefs. The whole distinction between “derivative” or “inferred” or “non-basic” beliefs that owe their justifiedness to other justified beliefs on the one hand, and “intuitive” or “uninferred” or “basic” beliefs that do thus derive their justification on the other, is one between beliefs rather than believings. The philosophers interested in this distinction were concerned with such matters as ultimate sources of knowledge, and whether or not there were touchstones against which doubts or disagreements could finally be settled, and not with the presence or absence of flaws in individuals’ actual epistemic performances. They were concerned especially with reassuring themselves as to the rejectability of scepticism, where scepticism is a theory to the effect that nobody, not even people performing epistemically as flawlessly as people are able to, not even people fulfilling their epistemic responsibilities to the highest degree possible, knows (or is justified in) what we ordinarily take ourselves to know. As a result, foundationalists and their opponents, and those who discuss the regress or reasons argument, are interested in beliefs rather than believers being justified. Before leaving the matter of the personal/doxastic distinction, it is worth remarking that recent discussions of the basing relation have been concerned with the personal rather than the doxastic. These discussions have seen the basing relation as the relation that has to obtain between a reason r and a belief that p in order for p to be justified. The assumption is that if a person holds r, and r supports p, that is not enough for the person to be justified in p. Something else is required.10 (One thing that might well be seen to be required is that the subject be justified in r, a condition almost never even mentioned by those discussing the basing relation. However, let us neglect this point.) What sorts of suggestions have been made about this basing relation? Some have wanted to insist that the belief that p must be caused or at least causally sustained,

10

It would be natural to think of basing as a relationship between beliefs, rather than between the justifiedness of beliefs, just as inference is, and as being entirely independent of matters of justification. One may infer one belief from another, and both may be completely unjustified. Similarly, it would be reasonable to expect, one unjustified belief may be based on another. However, as the basing relation is now widely understood, it is a relation between one belief being (justifiedly?) held and another belief being justified. So it looks, at first glance at least, to be relevant to our present concerns. For a review of attempts at analysis of the notion, see Korcz (1997).

A Problem About Epistemic Dependence

27

by the belief that r (for example, Koppelberg, 1999). Others have wanted a metabelief held by S, a belief about the support relation between r and p. In the financial ruin case, we plausibly say that the subject’s belief that he will be ruined is based on the dream, and not the better evidence he possesses, precisely because both the causal connection and the appropriate meta-belief connect his belief with the dream rather than with the better evidence. Thus he is not justified in his belief. It can be seen then that these discussions of the basing relation have been concerned with the justification of the believer rather than the belief, i.e. personal rather than doxastic justification. The man whose belief that he faces financial ruin is based on a dream is not personally justified in the belief, but given the (admittedly neglected) evidence in his possession, we are inclined to say that his belief itself is (doxastically) justified (by that evidence). Were there to have been a basing relation between his belief and a reason that really did support the belief, we would be able to say that he was justified — that is, that we would then have a case of personal justification. However, if we are correct in thinking that our concern is with the justification of the beliefs, as opposed to the believers, then the basing relation, however explained, is of no concern for our present purposes. I turn therefore to other attempts to provide a coherent account of the notion of dependence, where the type of justification we are concerned with when we speak of one belief’s being justified depending on another belief’s being justified is doxastic justification.

4. An Account in Terms of Necessary Conditions We have already remarked on the close relation between the concept of dependence and the concept of a necessary condition. So it is very tempting to try something along the lines of: Jp dep Jq iff Jq is necessary, given the subject’s other justified beliefs, for Jp. However, this account has the consequence that justifiedness in any belief, p, will always depend on justifiedness in any other belief which p immediately and obviously entails. Let q be (p or r) for any arbitrary r. If someone were not justified in ( p or r), he/she would normally not be justified in p. The consequence that Jp depends on itself disjoined with any arbitrary belief is, it should hardly need to be pointed out, wildly counter-intuitive. An example: you cannot be justified in p*

Peter broke the window

28

Tim Oakley

unless you are justified in p* or r* Either Peter broke the window or Robert broke the window or for that matter, unless you are justified in p* or r** Either Peter broke the window or the moon is made of green cheese. So justifiedness in ( p* or r*), and in ( p* or r**), is necessary for justifiedness in p*. But in very few circumstances would it be the case that justifiedness in p* depended epistemically on justifiedness in either ( p* or r*), or in ( p* or r**). Aside from counter-intuitive consequences, the analysis proposed also results in there being no basic beliefs, since every belief immediately and obviously entails a disjunction of itself disjoined with some other arbitrary belief. As remarked, it may be true that there are no basic beliefs, but we need a concept of dependence that enables that issue to be argued out, rather than a concept that settles it in advance of any other considerations. The same troubles that make the account of dependence in terms of necessary conditions untenable is equally bad for any account in terms of whether a person will remain justified in p if she is no longer justified in q. A somewhat more elaborate account along these lines was long ago proposed by Pastin (1977, 1978) and criticised in similar terms to those above by Feldman (1977, 1978). Various ways out of the problem may suggest themselves. You could try ruling out cases of justifiedness in p depending on justifiedness in q where p entailed q, thus: Jp dep Jq iff (1) Jq is necessary, given the subject’s other justified beliefs, for Jp; (2) It is not the case that p entails q. This fails for two reasons. First, it is too weak in that it fails to include cases that ought to count as dependence. In some cases it is clear intuitively that justifiedness in p does indeed depend epistemically on justifiedness in something that p entails. Thus, we may be justified in t All persons lose membership rights immediately on becoming unfinancial.

A Problem About Epistemic Dependence

29

and this will depend on our justifiedness in u

This person has lost membership rights immediately upon becoming unfinancial.

plus v Either all persons lose membership rights immediately on becoming unfinancial or no members do. This will be so even though t entails u. Indeed, justifiedness in p may depend on not only a disjunction that p entails, but on a disjunction in which p is disjoined with some completely unconnected further proposition (as in the case above of (p* or r**)). Our evidence for p might be: the computer has just issued the disjunctive sentence (p or s), and whenever the computer issues a disjunction, the disjunction is true, and the second disjunct is false. But such a case of dependence would be ruled out by this account. But the account is also too strong and would, like its predecessor, rule in as cases of dependence instances that we should not want ruled in. Consider: w All members of a very large and varied set of As have all been observed to be Bs. z The next A will be a B. Consider the situation where S is justified in z on the basis of justification in w. (S’s evidence for z is w. If you think we need additional beliefs in order for z to be justified, such as beliefs about absence of defeaters, then add them as additional clauses in w. S has no independent evidence for or against z.) Here we would say that Jz depended on Jw. And indeed, in the context of S’s other beliefs — with no independent evidence for or against z — Jw seems indeed necessary for Jz. However, in exactly the same circumstances, is not Jz necessary for Jw? If S were not justified in z, could S be justified in w? Surely not. If normal epistemic principles say that justification in w is sufficient in the context for justification in z, then justification in z in the same context is necessary for justification in w. (It could happen, of course, that w was justified and z not, if there were indeed independent evidence against z, but that is precisely what we do not have in the case as described.) So on the account of dependence just given above, we would have to say that Jp depended on Jq, but also that Jq depended on Jp, a wildly counter-intuitive conclusion. Clearly, this is not the concept of epistemic dependence we are trying to capture.

30

Tim Oakley

5. An Account in Terms of Support Relations Let us turn then to an attempt to explain dependence in terms of support. By “support”, I have in mind the logical relations that we find in cogent arguments (of any sort — deductive or any form of inductive argument) between premises and conclusions. Premises or reasons severally support conclusions. An account of epistemic dependence in terms of support looks as though it ought to work. Aside from controversy on the manner of justification of basic beliefs, where reliabilism has looked promising to some, the whole debate on foundationalism has traditionally been conducted in evidentialist terms, so it seems natural to think of dependence chains in terms of evidential or support relations. Further, the interests that lie behind much of the discussion of whether one class of beliefs depends on another for justification — beliefs about other minds on beliefs about other people’s behaviour, for example — is largely a matter of whether the members of the first class are ultimately supported in principle, by members of the second. So consider this attempt: Jp dep Jq iff (i) Jp; (ii) Jq; (iii) q is part of an argument supporting p, all premises in which are justified for the subject; (iv) without q, there would be no such argument supporting p in which all the premises were beliefs justified for the subject. Clause (iv) is needed because if there were some other argument that supported, and thus provided justification for p, then p would not depend for its justifiedness on the justifiedness of q. There is however an immediate objection to this account, viz., that condition (iv) will never be fulfilled, as there will always be other arguments frameable from the corpus of the subject’s justified beliefs. And if this is so, no belief will ever depend on any other, which is presumably sufficient reason for dismissing the analysis offered. On the assumption that the subject is justified in some arbitrary belief, r, there will be the following argument for p: p and r, so p. Or, on the assumption of the subject being justified in some arbitrary belief, not-s, there will be the following further argument for p: p or s, and not-s, so p. The examples will have the air of a cheat about them of course, since they both seem to involve circularity of argument. Both presuppose p in the course of an argument which supports p. Should we not insert the term “non-circular” before the word “argument” in each of the clauses (iii) and (iv) above? We had in fact better not do this, for two reasons. First, it would be worrying if foundationalists, in the course of defining the fundamental concept in terms of which the theory is to be framed, ruled out the idea of a circular dependence chain.

A Problem About Epistemic Dependence

31

As remarked, most coherentists do not speak in terms of requiring the notion of epistemic dependence at all, but at least some non-foundationalists would probably want to say that we are sometimes justified in a belief in virtue of precisely such a circular chain of support. (Some still explicitly identify coherence theories with circular justification — for example Pojman (1995, pp. 96–97).) If we ban circularity in dependence, we rule out any such claims trivially, by the stroke of our definition, just as surely as the definition of dependence in terms of necessity ruled out basic beliefs. Since the point of a definition of dependence is to permit a framing of the debate about foundations, which may then be settled as a substantive matter, and not a verbal one, these two accounts are both unsatisfactory in the extreme. A second reason is that if one explains dependence in part by reference to the concept of circularity, it is extremely hard to see how one is to avoid a different sort of circularity — circularity of definition. For it is very plausible to say that we will have to explain the notion of circularity in justification by saying that it is the situation where both justification in p depends on justification in q, and justification in q depends on that in p.

6. Personal Justification and the Basing Relation Reconsidered We have seen reasons, in Section 3, for treating discussions of foundationalism in terms of relations between doxastically justified beliefs. In the light of our present difficulties, it may seem that a reconsideration of personal justification is appropriate, as the demand for a basing relation of the kind discussed would at least permit the avoidance of the problems noted above in connection with accounts in terms of necessary conditions, and in terms of the support relation. We could then say that in general, no one’s justifiedness in the belief that Peter broke the window depends on his/her being justified in the absurd disjunction, “Either Peter broke the window or the moon is made of green cheese”, simply because there is no basing relation: there is no causal connection, and there is no meta-belief. Similarly, the counter examples raised above against the account of dependence in terms of support may be dismissed for the same reason. It is possible then to conceive a foundational structure among the beliefs in which a subject is personally justified. (The additional question of which beliefs are personally basic for a subject will be answered in different ways, depending on which general theory of justification is adopted.) Will this do? The first thing to note is that there is very little structure to be found in a subject’s personally justified belief system. Most of any given person’s beliefs have probably not been based on beliefs that have been acquired in the course of a fully

32

Tim Oakley

responsible epistemic performance, or been derived in an epistemically responsible way from other beliefs which were so acquired. Most humans are thoroughly fallible in their thought processes, if not stupid, lazy, careless, pigheaded, and prone to wishful thinking in the greater part of their cognitive activities. But, secondly, even with such a theory applied to personalistic justification, doxastic justification still exists as well, and we are still interested in it, and may well want to know if there is a structure there too, in the system of justified beliefs as opposed to a justified person. This is certainly true of those like coherentists and foundationalists concerned with the overall shape of our belief systems. As stressed in Section 3, the reason for this is probably that the foundationalist’s concern was with the health of our whole epistemic enterprise. Was our knowledge — conceived generally and “in the abstract” — soundly based in some way? Was there some definitive way, in principle at least, for settling disputes about the truth? And if there was, was it by reference to beliefs gained through the exercise of reason or of the senses? Given these concerns, which in fact most epistemologists still share, why should we be concerned with the vagaries of the thought processes of individual humans? I conclude, then, that admirable as the aims of traditional foundationalists may be, and central to the setting up of many epistemological issues as the regress of reasons argument may be, there are serious problems with the conceptual machinery here. I have of course not said enough to show that the position is completely beyond rescue. But at the very least, some serious repairs are necessary.

Acknowledgements I am grateful to many people for help with the issues dealt with in this paper, and in particular to audiences at the University of New South Wales Epistemology Conference in December 2004, at La Trobe University in May 2005, and at the Conference of the Australasian Association of Philosophy held at Sydney University in July 2005. Very early thoughts on this topic were developed an embarrassingly long time ago, and at that time I had invaluable assistance from discussions with Frank Jackson.

References Alston, W. (2005). Beyond ‘justification’: Dimensions of epistemic evaluation. Ithaca, NY: Cornell University Press. Bender, J. (Ed.) (1989). The current state of the coherence theory: Critical essays on the epistemic theories of Keith Lehrer and Laurence BonJour, with replies. Dordrecht: Kluwer.

A Problem About Epistemic Dependence

33

BonJour, L. (2003). A version of internalist foundationalism. In: L. BonJour and E. Sosa (Eds), Epistemic justification. Internalism vs. externalism, foundations vs. virtues. (pp. 3–96). Malden, MA: Blackwell. BonJour, L., & Sosa, E. (2003). Epistemic justification. Internalism vs. externalism, foundations vs. virtues. Malden, MA: Blackwell. Dancy, J., & Sosa, E. (Eds) (1992). A companion to epistemology. Oxford: Blackwell. Engel, M. (1992). Personal and doxastic justification in epistemology. Philosophical Studies, 67, 133–150. Everitt, N., & Fisher, A. (1995). Modern epistemology: A new introduction. New York: McGraw-Hill. Feldman, F. (1977). On the analysis of warranting. Synthese, 34, 497–512. Feldman, F. (1978). Final comments on the analysis of warranting. Synthese, 37, 465–469. Feldman, R. (2003). Epistemology. Upper Saddle River, NJ: Prentice-Hall. Fumerton, R. (2002). Theories of justification. In: P. K. Moser (Ed.), The Oxford handbook of epistemology (pp. 204–233). New York: Oxford University Press. Koppelberg, D. (1999). Justification and causation. Erkenntnis, 50, 447–462. Korcz, K. A. (1997). Recent work on the basing relation. American Philosophical Quarterly, 34, 171–193. Moser, P. K. (Ed.) (2002). The Oxford handbook of epistemology. New York: Oxford University Press. Oakley, I. T. (1976). An argument for scepticism concerning justified belief. American Philosophical Quarterly, 13, 221–228. Oakley, I. T. (1988). Scepticism and the diversity of epistemic justification. The Philosophical Quarterly, 38, 263–279. Pastin, M. (1977). Counterfactuals in epistemology. Synthese, 34, 474–495. Pastin, M. (1978). Warranting reconsidered. Synthese, 37, 459–464. Pojman, L. (1995). What can we know? An introduction to the theory of knowledge. Belmont, CA: Wadsworth. Post, J. F. (1992). The infinite regress argument. In: J. Dancy & E. Sosa (Eds), A companion to epistemology (pp. 209–212). Oxford: Blackwell. Quine, W. V. (1967). On a suggestion of Katz. The Journal of Philosophy, 64, 52–54. Reiter, D. (1998). Engel on internalism and externalism in epistemology. Erkenntnis, 49, 175–184. Russell, B. (1967 [1912]). The problems of philosophy. London: Oxford University Press.

This page intentionally left blank

34

Chapter 3

Accounting for Commitments: A priori Knowledge, Ontology, and Logical Entailments1 Michaelis Michael

In giving an account of our cognitive nature, modern epistemology has focused, understandably, on the question of what it is to know something. There are other aspects of that nature which are not settled by that focus. My focus here is not that perfectly respectable question but a further question: in taking on an explicit commitment which may or may not be knowledge, what other commitments have I taken on? This further question is part and parcel of our understanding each other, and ourselves. It requires an account of what we are implicitly committed to by dint of our explicit commitments. Were such an account forthcoming, we would easily account for what one knows in knowing something, or again, what one comes to believe in believing something. One very popular answer to that issue is to focus on the entailments of our explicit commitments. I am broadly in favour of this approach but find that beyond that agreement in generality there is room enough for major disagreement. I shall approach the matter by outlining the views of Frank Jackson who represents a particular view I find mistaken. Jackson introduces what I call the

1

This paper derives from work I did while supervised by David Lewis. I benefited immensely from his advice. I also benefited from remarks made on the work by Frank Jackson at the time. More recently I have benefited from comments by Aislinn Batstone, Stephen Hetherington, and Ralph Kennedy.

Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

35

36

Michaelis Michael

necessitation theory of entailment and defends it successfully against one interesting objection. I shall develop another kind of objection to the necessitation theory of entailment which I maintain is a more serious objection to this theory. I take a number of lessons from the problems with the necessitation account and offer an alternative which focuses on the notion of a priori consequence and uses a theory of ideal rationality to spell that difficult notion out. One approach to the problem of accounting for the ontological commitments carried by a theory is that advocated by Frank Jackson in his (1989) paper entitled “On a Puzzle about Ontological Commitment”. According to Jackson, the ontological commitments of a theory are those existential generalizations such that necessarily, they are true if the theory is true. This is in concert with his views, argued for elsewhere, about the nature of material conditionals and, by natural extension, entailment.2 These views constitute the standard picture about the nature of entailment. In general, we say that a set of sentences entails a sentence just in case the latter is true whenever all the sentences in the set are true. The entailments of a theory, T, are given by the superset relation on the set of metaphysical possibilities determined by T. Any theory, T* which determines a set of metaphysical possibilities which includes the set determined by T is entailed by T. Let us call this relation among theories “necessitation”. So Jackson’s view is that the ontological commitments of a theory are the necessitations of that theory which are of the form of existential generalizations. While I agree that Jackson is correct in finding a key role for the notion of entailment in an account of ontological commitment, and as I would suggest a general theory of implicit commitment, I shall suggest that his theory goes wrong in the details. In particular, even though the relation of necessitation is important, it is not a plausible candidate to be the entailment relation of which we are seeking an explication.

A Puzzle and Jackson’s Solution Frank Jackson considers a problem for his necessitation analysis of commitment. He observes that it is a consequence of his view that if something exists necessarily then any theory is committed to it. So, in particular, if pure sets or nonactual possible worlds exist and exist necessarily, then every theory is ontologically

2

See for example his (1987) in which he puts forward a view of “if … then”. His (1982) paper with Lloyd Humberstone is also a defence of an extensionalist conception of consequence against certain intensionalist objections raised by Anderson and Belnap (1975). My criticisms of his extensionalist account of consequence do not constitute a criticism of these arguments which he and Humberstone rightly deploy.

Accounting for Commitments

37

committed to them. But now it seems that when we develop a theory of some seemingly unrelated subject matter, say the law of Torts, we commit ourselves to the existence of these entities. In fact if pure sets exist necessarily, even nominalists about sets, in every utterance they make, commit themselves to the existence of pure sets. This is so even when their utterances are to the effect that sets do not exist. On the face of it, this looks bad for the necessitation account. It does not enable us to make sense of the way in which, as it might naively be put, some theories are committed to sets in a way that other theories are not. Jackson uses the example of David Lewis, who is both a materialist and modal realist. Lewis must (if he were to accept Jackson’s characterization of commitment) regard both of the following arguments as valid and sound. B1 Materialism is true. B2 Materialism is ontologically committed to non-actual possible worlds. BC Non-actual possible worlds exist. C1 Modal realism is true. C2 Modal realism is ontologically committed to non-actual possible worlds. CC Non-actual possible worlds exist. Yet is it not obvious that there is an intuitive difference here between the way Lewis’s materialism is committed to non-actual possible worlds and the way his modal realism is? Jackson feels the force of this intuition and offers us an account of the difference between the two arguments. The difference is located in the epistemology of the two cases. Doubt about the common conclusion of the two arguments does not reflect symmetrically on them. Such doubt reflects on C1 in the one case and on B2 in the other. This is in turn explicable, since the grounds for holding B2 to be true are just the grounds for thinking that BC is true and necessarily so. Any doubt about BC being necessarily true sheds doubt directly on the holding of the necessitation relation between materialism and the existence of non-actual possible worlds. As Jackson (1989, p. 198) puts it, anyone who accepts his account of entailment and ontological commitment cannot say that one theory is, and the other is not, ontologically committed to non-actual possible worlds. Both are committed. What they can say though, is that doubts about non-actual possible worlds have the potential to reflect back on modal realism but not on materialism. The argument can, at least potentially, be reversed to cast doubt on modal realism but not on materialism.

38

Michaelis Michael

Frank Jackson’s account (1989) of our intuition of a real and significant difference between the two cases identifies and exploits a real phenomenon, and one which is required to explain phenomena like the paradoxes of entailment and the location of the question-begging aspect of arguments. This phenomenon ought to be acknowledged by everyone. Jackson has identified and solved the puzzle he presents. In so doing he has shown us how, if we are suitably sensitive to the dynamics of belief systems, with a rather crude notion of entailment we can find distinctions where there seemed to be only unsupported intuition. This is a good thing for his account of entailment since, on the whole, in cases of conflict such as this, it is better to opt for unsupported intuition over technical theory.

A More General Puzzle for the Necessitation Account There is a phenomenon which like the puzzle that Jackson has noted, might give us reason to pause over whether necessitation is the best explication of the intuitive notion of entailment. Unlike Jackson’s phenomenon this is a problem just for the necessitation account. Necessitation is supposed to be an account of entailment. Naively, we think that the entailments of theories to which we are committed are also things to which we are in some sense committed. This would make sense of our conviction that failing to see that various things are among our commitments is a rational failing; something for which we can be held responsible. It is consistent with this picture of rationality that we can conceive of ideal rational agents who do not have to engage in reasoning to see whether something is a consequence of something they believe. Such rational agents need not be omniscient. Since there may be many matters of fact that do not follow as a matter of reason from what they believe, not all ideally rational beings are omniscient. However, the difference between the notion of an ideally rational being and that of an omniscient being is not that the omniscient being is even more rational than the ideally rational being. Rather, an omniscient being differs (if at all) in knowing more than the ideally rational being. As far as rationality is concerned, an omniscient being and an ideally rational being are equally rational. What is ruled out for ideally rational beings is that they should deny anything which is a consequence of something they accept, or that they should ever be in a position of having inconsistent beliefs. Should this threaten they will revise back to a consistent kernel of beliefs. These agents are ideal. Failing to meet these criteria is no reason to give up philosophy. You can be as smart as the cleverest human and still fail to be ideal by these criteria. However this does not commit one to thinking that rationality and irrationality come as all-or-none properties. People can be ranked as being more or less

Accounting for Commitments

39

irrational by these criteria.3 Some contradictions are more obvious than others. We might think a person who should have seen an obvious contradiction should be criticized in the way we may not think that a person holding onto the naive theory of sets should be criticized.4 According to the necessitation account of entailment, one sentence entails another just when the right inclusion relation holds between the propositions determined by the sentences in question. However, and intuitively, whether two theories which determine precisely the same set of worlds entail some other theory may depend on the way those theories are presented. If Fort Denison is simply identical to Pinchgut, then Fort Denison’s being an island just is Pinchgut’s being an island.5 However, someone who believed that both Fort Denison and Pinchgut were on Sydney Harbour would not be irrational if he thought that one was an island and the other was not an island but a strategic point. Although there are no possibilities determined by this person’s beliefs, since none are consistent with his beliefs, we feel reluctant to ascribe any irrationality to him. This person’s problem is not a matter of logical failure and no amount of inferential power could show him the error in his belief system.6 His failure is due to a lack of information. This makes perfect sense. Contrast

3

In fact, these criteria are likely to lead to paradox if we are not very careful. Montague and Kaplan (1974) showed how, in certain contexts, three simple assumptions lead to a result analogous to the liar’s paradox which they call the “knower’s” paradox. 4 What is to be made of this obviousness of contradictions? Is there an independent scale on which this can be weighed against someone’s rationality? At this point we may run into a holism. There may be no independent scale on which the obviousness of contradictions is to be ranked except the scale of how they tell against the rationality of someone who accepts them. Note that in general there will not be a simple value for this obviousness. Someone who has done a bit of logic, forgets the simple proof that naive set theory is inconsistent, and later scratches down the axioms of naive set theory and finds them all very obvious is more proscribed than someone who wrote before Russell discovered his paradox. 5 Pinchgut was so-called by the convicts who faced starvation diets when sent there. It is located on a very small island in Sydney Harbour. It is indeed identical to Fort Denison. 6 Making the distinctions we have made allows us to see a way of understanding Kripke’s (1988) enigmatic puzzle of Pierre. Pierre, raised a Frenchman, has learned French the native way and has been told and sincerely assents to “Londres est jolie”. He moves to London and an ugly bit of London at that. He learns English not by translation into French but by learning it directly from the people about him. He comes to assert “London is not beautiful”. Kripke suggests (ibid., p. 114) that we hold that de dicto attributions are subject to the following principle “If a sentence of one language expresses a truth in that language, then any translation of it into any other language also expresses a truth (in that other language)”. Of course I am presenting the situation by discussing the attributions de dicto to Pierre and in particular using dicta in a number of languages. The real puzzle arises when we insist on asking what we should say about his attitude to the dicta in just one language. On my way of presenting this puzzle it is due to the fact that when we seek to assign sentences of our language to Pierre’s beliefs, as expressed in his two languages, we meet the problem that there is no way of doing

40

Michaelis Michael

this situation with the case of someone who believes the axioms of naive set theory. This person certainly has displayed his irrationality. There is a proof that what he believes cannot be true. The proof, of course, is Russell’s and is a priori. These two cases are intuitively very different. In the second case what was wrong was something to do with the ability of the agent to see what follows from what he is committed to. This is not so in the first case. The necessitation account of entailment, at least as we have used it so far, does not enable us to draw this important distinction. We are used to describing entailment to students in the following ways: (a) B entails C iff it is necessary that if B then C. (b) B entails C iff any way that the world could be in which B is true is a way the world is in which C is true. (c) B entails C iff C follows from B a priori. (d) B entails C iff any interpretation of the non-logical vocabulary in B and C which makes B true also makes C true. These are all familiar ways of speaking and it ought to be familiar also that the first two are not equivalent to either of the latter two. The first two couch entailment in terms of necessity; the latter two couch it in terms of a priori consequence and in terms of strictly formal consequence, respectively. I want to make the old-fashioned claim that implicit commitment has more to do with a priori consequences than it has to do with necessary consequences. This is something philosophers have underemphasized because we have worried too much about the algebra of propositions determined by sentences and not enough about the epistemic relations among beliefs and sentences. How can this big claim be justified? The points made against the necessitation theory are a partial justification but in what follows I want to strengthen the case against the necessitation theory. It ought to be noted that whenever we have a necessitation which has a necessary truth as a premiss we can simply omit that premiss and still have a necessitation for the same conclusion. Look again at the examples that Jackson has given us.

this and respecting the deducibility relations among his beliefs without importing new names into our language. Even if we try to introduce new names to try to mirror the relevant deducibilities, we cannot do that in the appropriate manner, since we do know that the two names are co-referential. Kripke is right that the puzzle does not turn on the fact that the names are directly referential. It turns on the fact that the two names (directly referential or not) are each assigned the same name in the translation. This assignment would depict a relation among the original beliefs which was originally merely that of possessing contradictory metaphysical contents, and which does not on its own impugn the rationality of the believer, as one which is the possessing of contradictory epistemic contents, a relation which does impugn the rationality of the believer.

Accounting for Commitments

41

B1 Materialism is true. B2 Materialism is ontologically committed to non-actual possible worlds. BC Non-actual possible worlds exist. C1 Modal realism is true. C2 Modal realism is ontologically committed to non-actual possible worlds. CC Non-actual possible worlds exist. It ought to be noted that the second premiss in each of these arguments is, if true, necessary and so is redundant. In fact the conclusions are supposed to be metaphysically necessary which means that any premiss is redundant. The point about the redundancy of any necessary premiss does not depend on the necessity of the conclusions in these examples and is quite general on the necessitation account. However the first argument looks bad without the second premiss. How can that be so on the necessitation view of entailment? What is correct here is that if one of our premisses is entailed by everything it can be omitted from the set of premisses that an argument needs. This is a reply that Achilles could have made to the Tortoise in Lewis Carroll’s little gem (1895). But there are many necessary truths which need to be mentioned in the premisses of arguments if they are to be entailments. Therefore there are necessary truths which are not entailed by everything. An obvious example is that a ⫽ b entails a ⫽ b. Or again that a ⫽ b together with b ⫽ c entails that a ⫽ c. These are good arguments, arguments which are intuitively valid. They are each both necessitations and a priori. If the premiss a ⫽ b is dropped from either argument the resulting argument is intuitively invalid and they are no longer a priori. However they remain necessitations, if the conclusions are true. This is one good reason to doubt that the intuitive notion of validity is best understood in terms of necessitation. There are criteria apart from validity and a priority which are relevant to whether an argument is a good one to show that a certain conclusion follows. Even when a premiss is a priori and so can be omitted from the premiss set, it is often better not to do so since this affects the obviousness of the conclusion from the premisses. A perfect reasoner would not require the additional premiss, and indeed many less than perfect reasoners will not require it. However, such a premiss can facilitate the movement of the mind from the premisses to the conclusion. Again, just how the obviousness of an argument is to be accounted for is not a real problem for my project. A good account of this would nicely supplement what I have to say. However all I need is the claim that failure to see something which is obvious reflects on the rationality of an agent. I shall develop another example of how the necessitation account of entailment yields the wrong sorts of commitment in general and ontological commitment in particular. An example which trades on the idea that there are a posteriori necessities.

42

Michaelis Michael

One type of response to these sorts of cases is to insist there are no a posteriori necessities. Whether or not there are such cases, the crucial thing is that validity marches in step with the epistemic notion of the a priori and not with the metaphysical notion of necessity. Anyone who rests their notion of necessity on the a priori will doubt the significance of these claims simply because they already endorse the priority of the epistemic. That is not forced on us and when we are tempted to pull these important notions apart, we need to distinguish which notion fits with the intuitive notion of consequence. Let us suppose that we humans are necessarily made of biological cells. Suppose that this is one of those a posteriori necessities Kripke mentions. Let us also suppose that the law of Torts entails the existence of adult humans. Then on the necessitation account it seems to follow that anyone who accepts the law of Torts is committed to biological cells. This is so regardless of whether they believe that persons are necessarily made of biological cells. In fact a legal theorist like Grotius, writing well before anyone had thought of biological cells, is committed to them. Moreover, on further reflection it seems he is also committed to atoms, electrons, and maybe even quarks, as well. This is irrespective of the fact that neither cells, atoms nor electrons exist necessarily. All that needs to be necessary is that humans (and perhaps ordinary things in general) are made of the stuff they are actually made of. According to this view, the fact that they are in fact made of cells means that they are necessarily made of cells. The cells themselves turn out to be necessarily made of atoms which are in turn necessarily made of electrons and other subatomic particles. All these claims could turn out to be necessary and yet there should be no inclination to regard Grotius as failing to recognize his commitments in this respect. This ought to be regarded as a fairly damning result. The case of Grotius’ commitments to biological cells is different from the cases that Jackson gives. If there are sets or non-actual possible worlds then presumably the reason for accepting them is an a priori argument. The failure of an agent to discover that a priori argument displays the irrationality of that agent. There is no question of this in the case of Grotius and biological cells — hence the point of noting that these are a posteriori necessities. There is a rejoinder to this objection which is available to a necessitarian. The rejoinder is that the argument which has just been given against the necessitarian misses its mark because we know that if the theory we endorse is true then all its consequences are true. So whether or not there are consequences about which we cannot make a judgement, if we can decide whether our theory is true (which is our only real concern) then we will be in a position to know that all its consequences are true. This is a serious rejoinder. The phenomenon pointed to is quite accurately rendered — a true theory has only true necessitations. However, discerning that a theory is true just is discerning that it has no false consequences. So not knowing what the consequences are is tantamount to not knowing whether the theory is true. An

Accounting for Commitments

43

account of how terms get their meaning might show that if all of a certain restricted class of consequences of a theory is true then the theory is true. For example, if the account of the origin of terms is a verificationist one, and the restriction is related to our epistemic abilities, it might be that the guarantee is in place. However, some accounts of the origin of meanings of terms might not ensure this guarantee. An account in terms of natural properties, for example, will not guarantee that a theory which is true as far as we can tell is really true. But even if the theory in question is true and so all of its consequences are true, problems remain for the necessitarian. Suppose Grotius were to jump to the conclusion that there are biological cells, out of the blue as we might say. Can he justify his newly acquired belief by pointing out that it is a logical consequence of the law of Torts? Not plausibly. We might feel that he had managed to fluke a true belief, but not that this belief follows, in any reasonable sense, from his other beliefs, especially not his reasonable belief in the law of Torts. Since (we can suppose) there are a priori arguments for the existence of sets, Jackson’s solution to his puzzle must account for the intuitions we all have that not all theories are committed to sets. Is there a worry left over though? When Goodman makes the mistake (on our supposition) of asserting that there are no sets, he is making a mistake that is a priori determinable. Still, it may not be an obvious mistake. This would be similar to the case of someone who asserts the axioms of naive set theory. And in both cases the obviousness of the mistake may depend on what else is known to be the case. We can distinguish, in the way Jackson does, among the different theories which each have ontological commitment to sets, by distinguishing among them by virtue of the reasonableness of the agent who accepts the theory and yet denies the existence of sets. What would we think if there were necessary existents whose existence was utterly unestablishable a priori? Even if we come to accept, on the ground of empirical evidence, that these things exist, does this mean that, before the evidence has come to hand, everyone — positivists, Quineans, atheists, agnostics, and sceptics no less — is committed to the existence of these things? It does so on Jackson’s account. But I maintain that coming to accept that these things exist is not a matter of recognizing your commitments, it is a matter of extending them. The main problem with the necessitation view of entailment is that necessitation is a relation between the premisses and conclusion which is quite often not determinable by us in any manner. In that case necessitation does not play a role in the practice of appraising the extent to which we have recognized our commitments. But the commitments of a theory are relevant to determining whether believing the theory brings in unpalatable ontological commitments. That is, if a theory carries implicit commitment to S then it must be a priori knowable that this is so. What do I mean that this be so knowable? Perhaps the happiest answer

44

Michaelis Michael

to this question is to look to the discipline of logic. One thing logic does is partially to explicate the notion of a priori consequence. Indeed, when we prove completeness theorems in formal logic we do not prove that a certain interpreted theory in a certain formal language has as its necessitations exactly the same set of sentences as its deducibilities. This remains so even though we may speak loosely to students in ways that suggest that we have done just that. What we do show is that when we take the intersection of truths of all the possible interpretations of the theory, they coincide with the proof-theoretic deducibilities of the theory. When we define the relation of semantic consequence for a formal language we appeal to what is true in all models. Notice that a consequence of the completeness theorem for first-order logic is that, for any necessitation which is not a priori, there are truths that can be added to the premiss set that will then make the relation between the premisses and conclusion a priori. At this point I want to raise the substantive character of the idealization involved in the notion of the a priori. Previously I said that rationally ideal agents do not need to reason to see whether something is a consequence of their beliefs. I might as well have said that they can immediately see whether any given sentence is a consequence or not. But this would leave open the possibility that the belief set of an ideally rational agent is inconsistent. Perhaps it is a characteristic of this ideal that such an agent checks to see if the negation of an arbitrary belief is a consequence of his beliefs.7 When it comes to the procedures followed by such an ideally rational agent for the acquisition and testing of beliefs, how wide are we allowed to cast our net for abilities? Are these abilities limited to moves allowed in the language of the theory? Are they limited to proofs of a certain sort, say of finite size which are in some sense surveyable? I suggest that the answers to these questions need not be determined ahead of time. Further, there is no reason to think that we should be able to develop a complete account of these abilities, nor that they will apply similarly in all cases. Most people would not think that a proof becomes easier to find just because someone somewhere in the world has just found it.8 I want to suggest that attributions of commitment depend on the notion of criticizability of rationality. Criticizability is a matter of degree. Someone might be more criticizable than someone else. Now if there are necessities which are a posteriori,

7 Notice that at this point I am assuming that a non-paraconsistent logic captures the formal skeleton of the entailment relation which goes with implicit commitment. The point made does not depend especially on that relation being classical logic, though I believe it will be. The procedure would still work even if the base logic were intuitionistic. 8 The case in mind is not one in which the prior discovery of the proof shows how difficult the problem was but rather where that discovery changes the difficulty of the proof.

Accounting for Commitments

45

then the denial of these is not criticizable in the way that denials of a priori truth are.9 A person who denies that the heavenly body the Romans actually called “Lucifer” is either a planet or a comet, while asserting repeatedly that Venus is a planet, is someone who is denying something which is a necessitation of something he believes. This is so since Lucifer just is Venus. However, this does not impugn his rationality. One serious problem for the theory I have developed is the possibility that contradictions may knowingly and self-consciously be asserted as true. According to the account of commitment developed here this means that theorists such as Graham Priest are committed to everything. But Priest does not think that all sentences are both true and false. Only some sentences are dialethic. For example, he denies that drop-bears exist. Moreover, Priest refuses to acknowledge that his beliefs entail the existence of drop-bears. It is not that he is unaware of C. I. Lewis’s argument for ex falso quod libet. Being a talented logician, he is aware of the argument but denies that it is truth-preserving; for according to Priest, there are some true contradictions but not all sentences are true. The problem of interpreting Priest is not unlike that of interpreting Kripke’s puzzling Pierre. But unlike Pierre, Priest’s contradictions are epistemically accessible to himself on the natural assumption that he is using our language. The prospect of treating Priest as though he were really talking another language altogether which merely seems superficially like ours, is not easy to work out in detail. The problem would be that it is not just English that works this way in Priest’s mouth, any other language he seems to speak would be subject to just the same problem. My preferred solution is to treat his implicit commitments in accordance with the notion of a priori consequence, but to treat the attributions of commitment to Priest as going with what is, or ought to have been, evident to him. In discussions with him it can happen that one becomes more open-minded about even logical truths. This is not to say that we let our confidence in them drift too far from the maximum. However, it is to say that insofar as we treat logic as an area of significant research, in which disagreement and knowledge are possible, we do lessen our conviction that we have discovered the logical truths, even if this lessening of our conviction is a context-relative phenomenon. How are the notions of the a priori and rational criticizability related? On my view the a priori consequences of a theory just are those for which a believer is 9 I think that there are necessary truths which are a posteriori and which are more criticizable than many a priori truths. Many truths of mathematics are such that doubting them or even denying them is far less criticizable in our current state of knowledge than denying that Socrates is mortal. But my point is that while denials of necessary a posteriori truths can be no reason to doubt that an agent has ideal reason, doubting a priori truths always impugns your reason.

46

Michaelis Michael

to be held rationally responsible in the limit. This is not to say that we do not use weaker criteria for the attribution of commitments on occasion or even usually. But the ontological commitments properly ascribable to agents depend on our judgements concerning what those agents should have seen as being among the consequences of their beliefs. Mere necessitations when they are not a priori are not among the consequences they should have seen. The ideal case is determined by what would follow from the belief set by means of rationally sanctioned inference (which may involve inductive inferences as well as deductive, but let us avoid complications here) or, as we might say, as determined by the a priori closure of the belief set. This discussion raises other issues. Do we include only what follows from a finite set of premisses by a finite number of steps or may we idealize beyond those limitations and include infinitely long proofs? In the context of a noncompact logic (which is required by any language containing locutions like “many” and “most”) and in particular if defeasible (i.e. non-deductive) inferences are allowable, this question is pressing. But it is only pressing insofar as we are trying to characterize the notion of the a priori. The a priori enters the story as one way of fleshing out the notion of a rational ideal. Insofar as we are interested in different ways of fleshing out the notion of a rational ideal, there is no need to choose between the different notions indicated above. In different contexts different notions will be apposite. The question to be addressed is whether certain sorts of failure should count against the rationality of the agent in question. This will be criterion-dependent. The pressing question is which of these notions is the basic one. I am tempted to think, tentatively, that if one of these is the basic notion it is that of rational criticizability, understood in the correct way. Rational criticizability is basic in that it plays a constitutive role in determining the language the agent uses which in turn is constituted by a series of normative criteria holding. That is, an agent uses a language L when he shows the appropriate sensitivity to the features of the world to which he should be sensitive. However, I do not want to pursue this issue at this point, although it will have to be dealt with by an adequate account of the a priori. The salient issue here is that I have provided an account of commitment that aligns an agent’s commitments with the entailments of that agent’s beliefs. Entailment is cashed out as the relation of a priori truth preservation. It was then noted that if something follows a priori from a belief set then an ideal agent will be in a position to realize this. This means that insofar as we fail to see that something is an entailment of what we accept, we show that we are not ideally rational. This is as it should be. The necessitation account cannot yield this result since on that account not even the most ideally rational agents will be able to recognize all of their commitments.

Accounting for Commitments

47

Is Jackson Really a Necessitarian? In this section, I shall re-examine Jackson’s paper in light of the previous discussion to see whether Jackson is really a thorough-going necessitarian or whether he has not after all got a foot in the a prioristic camp. Of course I would be happy to welcome him into my camp. I have argued that the commitments of a theory properly include the ontological commitments of that theory, and also that these commitments are the consequences that are precisely what we should look at to see whether the agent’s epistemic state is criticizable or not. If these conclusions were accepted it would follow that any purported consequence which the agent is not to any degree criticizable for failing to recognize is only tendentiously to be called a commitment. As I read him, Jackson holds that the commitments of a theory are the necessitations of that theory; however, he suggests that not all such necessitations are relevant to the acceptability of the theory. To be relevant, according to Jackson, a commitment has to be such that doubt about it could reflect back on the theory. In the first instance, I am rather uncomfortable with the modality involved here. Surely what is of relevance is not whether a commitment could reflect back on the theory, since any commitment of a theory (or indeed any non-commitment of a theory) could so reflect, but rather whether it should so reflect back. Doubting anything can cause the doubting of anything else. However, many things that could happen in an agent’s beliefs reflect badly not on the theory but on the rationality of the agent. Incorporating that observation, the idea that Jackson is arguing for is that, according to the right semantics, a statement like (1) Theory T is committed to sets is to be judged for truth along the lines of the simple necessitation account, but that the pragmatic story about (1) is a good deal more complicated. Whether (1) is assertible in a context depends on whether doubt about the existence of sets could (rationally) upset a commitment to T. But when does this sort of situation arise? When will doubt about sets have the potential (rationally) to upset a commitment to a theory T? When Jackson says that “Possible doubts about sets bear on the reasonableness of the full theory [of classes] but are irrelevant to the reasonableness of the virtual theory”, he aligns the notion of commitment with reasonableness.10 I agree. He says this is a purely pragmatic matter. I disagree. In fact, in general, I am sceptical of semantic accounts of truth conditions and pragmatic 10

Jackson (1989, p. 198). The emphasis in the quote is mine.

48

Michaelis Michael

assertability conditions that are too divergent, if both can be given compositionally and if the recursive clauses for the assertability conditions do not require the output of the recursive clauses for truth. This scepticism is especially marked when the divergence is not due to the fact that if a sentence is asserted it will change the context of utterance so that the sentence changes its truth-value.11 When the problem is that the meaning of the terms involves what has been called, following Paul Grice, conventional (as opposed to conversational) implicature, an implicature which cannot be cancelled by explicit denial without self-contradiction, we have an element in meaning which affects the reasonable inferences a listener can legitimately draw without allegedly affecting the truth-conditions of the sentence. This sort of thing is allegedly ubiquitous. But when, as I am tempted to do, we characterize the content of a sentence by the role the sentence plays in deliberation and in action, the idea that there is an aspect to the meaning which plays a role in determining what follows from the sentence and which is yet not part of the truth conditions, and so not a part of the content of the sentence, is rather dubious. Ideally there is a confluence of two approaches to meaning. On the one hand there is the pragmatic account I have just sketched, and on the other there is the recognition that there is a compositional story to be told, of how the parts of sentences contribute to the meaning of the whole. Often there is a divergence between the theory of truth and the theory of (conventional) assertability in terms of subjective conditions. This is not to say that when some sentence is subjectively assertible the metalinguistic sentence saying that that sentence is true is not also assertible. That is always the case; subjective assertability aims at truth and not another thing. On the other hand, it is not the case that whenever a sentence is subjectively assertible, it is true. Truth and subjective assertability are related one to another, but are distinct. The explanation of the divergence must be given in terms of how the subjective assertability conditions are attempts to get at the truth of the sentence in question. When the divergence cannot be so understood, 11

Compare the often-seen disclaimer on pamphlets, “This page has been left empty intentionally”. If only they had not written it, it would be true! But then what is the “it”? Not the token, since the token is false. Not the type, since the type itself takes no truth-value, depending as it does on the specification of semantic value for the indexical “this page”. Can we find something between the two, something to take a truth-value without tokening? What about the proposition expressed by the false token? That proposition can be taken as true prior to the tokening of the sentence on the page, and false thereafter. Whether propositions are treated as capable of changing truth-value over time in this way depends on whether they are treated as analogous to sets of worlds or to sets of world-moments. In this footnote I have chosen the former option, though not too much rests on this choice. The time index can be built into the content of the proposition, in which case the propositions take their truth-value timelessly, or the truth of the propositions can be indexed to times, in which case the truth-values can change over time. As long as we are clear about what we are doing we should find no problems adopting either option.

Accounting for Commitments

49

then we ought to be sceptical of the posited account of truth-conditions. In the case in hand, we find a case can be made that the requisite relations between truth and subjective assertability conditions do obtain. According to both Jackson’s account and my own, the sentence saying that an agent B is committed to objects of kind F is assertible when and only when it is reasonable to think that doubt about Fs can reflect back on B’s theory. But Jackson’s account masks another problem. I have argued that whether B is committed to Fs depends only on B’s beliefs. Jackson, on the other hand, would have the assertability of “B is committed to Fs” depend on whether we think that B holds onto beliefs which necessitate the existence of Fs. So we now should regard “Grotius is committed to biological cells” as true. On my account what matters in determining commitments is what follows from B’s theory in an a priori manner. If it does not so follow from Grotius’s theory that humans are made of biological cells, then Grotius is not committed to biological cells. So in my view the more disambiguated account of subjective assertability of commitment is: A sentence saying that an agent B is committed to objects of kind F is assertible for a speaker C when and only when it is reasonable for C to think that if B were to entertain doubts about Fs then B’s doubts should reflect back on B’s theory. Consider the case of “A but B” and “A and B”. These are alleged to be cases of sentences with the same truth conditions though with different conventional implicatures, so that they will behave differently in suggesting to the linguistic agent different classes of inferences.12 For example, asserting “A but B” licences the inference to “There is (or there is commonly thought to be) a contrast between A and B”. Now how does this occur? On my way of describing the situation, the sentences embed alike in the metaphysical dimension, while diverging in the epistemic dimension. This mere possibility is perfectly coherent in the abstract, but it is an important question how such a situation could have come about. On one important conception of the relationship between the metaphysical and the epistemic, such a situation could not have come about.13 On this picture, it is the epistemic dimension of content that determines metaphysical content. Metaphysical content is derived from the mere fact that beliefs with such and such epistemic content are held by beings located thus and so in the world. That is, what underpins the fact that the beliefs we hold about water have a metaphysical content relating to H2O and not XYZ?14 Nothing introspectible could discern the difference, for there is no introspectible difference. The relevant difference has to do with the situation of believers; how they and their linguistic fellows are and have been situated in the world. 12

This example is developed in Jackson (1987). See for example Sellars (1974), Harman (1988), and other papers emphasizing semantic externalism. 14 See Putnam (1975). 13

50

Michaelis Michael

In the case in hand, if saying that someone is committed to biological cells were a purely pragmatic matter, then we would have little purchase on the importance of commitments to reasonableness. We would need to explain why agents need not accept all their commitments. Indeed, we would need to explain why these commitments are theirs at all. The fact that sensitivity to whether or not something exists is utterly irrelevant to the “reasonableness” of your belief state is a reason for thinking it is not among your commitments. Of course, this is not going to be compelling to Jackson. He will simply reiterate his account of the assertion conditions of attributions of ontological commitments. Still, theoretical simplicity is a virtue. If an account of truth-conditions of such attributions which does not make them diverge greatly from the (conventional) assertability conditions can be provided, this will arguably be an account with greater theoretical simplicity. The account I have argued for, in terms of inferences which are a priori, has that virtue. There would be a stronger reason to accept that the notion of a priori consequence is the basic one if there are a priori truth-preserving consequences which are not necessitations. Conversely, there would be a reason to think that necessitation is basic if it were to be crucial for determining the reasonable commitments of a theory. That is, necessitation would be basic if the commitments for which an agent is responsible are always a subset of the necessitations of the agent’s theory, a subset which would not be determinable without the prior determination of the set of necessitations. Arguably, this would be evidence but not a conclusive reason for thinking that the notion of necessitation is basic, since order of epistemic priority need not reflect order of conceptual priority.15 If it were accepted that anything which follows a priori from a theory is a commitment of that theory and so an entailment of that theory, it would be equivalent to the claim that all a priori inferences are necessarily truth preserving. Assertability would then seek to align itself with necessitation. This seems to be Jackson’s view and to this extent he is really a necessitarian and thus subject to the criticisms we have seen.

Conclusion The idea of logical entailment playing a key role in explicating our commitments is an attractive one. Once upon a time logic was seen as the science of reasoning. 15

For an example where the order of conceptual or semantic priority differs from the order of epistemic access, consider David Lewis’s account of counterfactuals and the notion of similarity there invoked. The notion of similarity is conceptually prior to the notion of what would have happened, but often we need to look to our judgements about what would have happened to determine the issue of similarity. Lewis (1986) supports this view.

Accounting for Commitments

51

Aristotle denied that any proposition logically implied itself because in moving from premise to conclusion there was no movement of mind. Such a conception of logical entailment pitches entailment as an account of cognitive dynamics. At the very least, this picture misses the fact that logical inference is a subset of rationally guided changes of our explicit commitments. Many of our rationally guided changes of explicit commitment are not deductively valid, whether they are inductive, whether they use default assumptions, probabilistic reasoning, or some other modes. They cannot one and all be assumed to be deductively valid. More importantly and more tellingly, the existence of a logically valid argument does not force us to accept the conclusion of that argument even when we have explicitly accepted the premises of the argument. We may choose to use the existence and recognition of that argument as a reason to revise our commitment to the premises. In that way a better picture of logical entailment is as an account of cognitive statics. Logic on this understanding gives an account of consistent sets of commitments. The recognition that we have an inconsistent set of commitments does not force us to update our commitments in any particular way. As the saying goes one person’s modus tollens is another’s modus ponens. Entailments spell out our implicit commitments, and failing to be sensitive to our implicit commitments is a way of failing to be ideally rational. Thus we see that since there are necessitations which it is not a rational failing not to recognize, these necessitations cannot be implicit commitments of every theory. Thus necessitation as a relation does not have a central role in spelling out the implicit commitments of cognitive agents.

References Anderson, A. R., & Belnap, N. D. (1975). Entailment: The logic of relevance and necessity (Vol. I). Princeton: Princeton University Press. Carroll, L. (1895). What the tortoise said to Achilles. Mind, 4, 278–280. Harman, G. (1988). Wide functionalism. In: S. Schiffer & S. Steele (Eds), Cognition and representation (pp. 11–20). Boulder, CO: Westview Press. Jackson, F. (1987). Conditionals. Oxford: Blackwell. Jackson, F. (1989). A puzzle about ontological commitment. In: J. Heil (Ed.), Cause, mind, and reality: Essays honoring C. B. Martin (pp. 191–199). Dordrecht: Kluwer. Jackson, F., & Humberstone, L. (1982). On a challenge by Anderson and Belnap. Analysis, 42, 179–181. Kripke, S. A. (1988 [1979]). A puzzle about belief. In: N. Salmon & S. Soames (Eds), Propositions and attitudes (pp. 102–148). Oxford: Oxford University Press. Lewis, D. (Ed.) (1986). Counterfactual dependence and time’s arrow. In: Philosophical papers, Volume II (pp. 32–66). New York: Oxford University Press.

52

Michaelis Michael

Montague, R. L., & Kaplan, D. (1974). A paradox regained. In: R. H. Thomason (Ed.), Formal philosophy: Selected papers of Richard Montague (pp. 271–285). New Haven, CT: Yale University Press. Putnam, H. (Ed.) (1975). The meaning of ‘meaning’. In: Mind, language and reality: Philosophical papers (Vol. 2, pp. 215–271). Cambridge: Cambridge University Press. Sellars, W. (1974). Meaning as a functional classification. Synthese, 27, 417–438.

Chapter 4

Epistemic Bootstrapping1 Peter Forrest

This paper concerns the connection between internalist justification and externalist warrant, with the aim of defending what might seem a blatantly circular way of reasoning. Because herds of thirsty epistemologists have muddied the waters I shall not, however, assume any fixed meanings for terms like “justification” and “warrant”. I shall not even assume that William Alston (2005) is being unduly defeatist in giving up on the concept of justification, as in his recent work. Instead I shall examine two examples, leaving it to readers to generalise as they see fit. Some preliminary remarks are in order, however, concerning what I mean by the phrases “internalist justification” and “externalist warrant”. Here I shall be more specific than my argument requires, so as to give readers some sense of the context of the examples. I shall assume a foundationalist, internalist account of justification. A belief is justified, I say, if it is, or can easily be, shown to be probable given various evident judgements concerning premises and concerning the probability of rules of inference. Evidence here includes both that which is self-evident, that is obvious when you think about it, and that which is evident to the senses, that is obvious because it is perceived. It may include other sorts of evidence such as what is morally or aesthetically evident but that is not my present concern. To say something is evident is to be committed to its truth, but on occasion what is evident to

1

Many thanks to Stephen Hetherington for his insightful comments on earlier drafts.

Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

53

54

Peter Forrest

someone may in fact be false. Thus, the axioms of naive set theory may well have been self-evident to Frege — before he was corrected by Russell. Warrant, on the other hand, I take to be a reliabilist, externalist concept. A belief is warranted if it is arrived at by some truth-seeking process that is reliable in its normal context and that is in fact operating in its normal context. On an externalist theory of knowledge, warrant in this sense is conceptually tied to knowledge in that if I believe (or know) a belief to be warranted I must believe (or know, respectively) that I know it. For further discussion see, for example, Alvin Plantinga’s works on epistemology (1993a, b). Again, largely to provide readers with some context, let me state my position in epistemology. It is that a good case has been made for a reliabilist account of knowledge, and hence of warrant, but that this is a Pyrrhic victory if it carries the consequence that our knowledge-claims are, although sometimes true, never justified in a way accessible to the reasoner. Fortunately, there is nothing to prevent a shell-back (i.e. foundationalist, internalist) theory of justification from incorporating an externalist account of warrant, by tackling the question of when we are justified in believing that a truth-seeking process is reliable. Epistemic bootstrapping may be seen as part of this incorporation, whereby metaphysical theses justified on internalist grounds may be used in turn to justify otherwise problematic claims that some truth-seeking process is reliable. As I have already said the above remarks are merely to provide readers with some context. The examples stand alone.

Plantinga on Christianity I begin with the example that got me interested in this topic. Plantinga (2000) has argued that Christianity is warranted if it is true. He has also argued (1993b, Chapter 12) that Evolutionary Naturalism is unwarranted (even) if it is true, and so we should not believe it to be the case.2 Both these claims are controversial. My aim, however, is not to examine the arguments for them but to see what might follow from them. It might seem blatantly circular to say that because Christianity is warranted if true we are justified in being Christians, and as far as I know Plantinga has not argued in that way. Nonetheless if he is right to conclude that

2 With a slight reservation, however, Plantinga considers the probability of our reasoning on such matters conditional upon evolutionary naturalism being correct, and argues that this probability is either low or “inscrutable”. For a recent statement of his position see Plantinga (2003).

Epistemic Bootstrapping

55

Evolutionary Naturalism should not be believed, because it is not warranted (even) if true, then we should ask whether similarly the warranted-if-true character of Christianity does provide some support. I shall answer with a qualified affirmative, although it turns out there is not perfect symmetry between the warranted-if-true and the unwarranted-if-true cases. This might explain why we are quick to say that if Evolutionary Naturalism is unwarranted if true then it is probably false, while more cautious about the way Christianity is supported by its warranted-iftrue character.

With One Bound the Brain Leapt Out of its Vat Let us use the verb “envat” to cover being today’s plaything of the deceitful demon, dreaming an atypically coherent and waking-like dream, being a brain in a vat, experimenting with Salvia divinorum, etc. Envattment is a condition characterised by saying that if you were envatted you would not notice anything unusual and yet your beliefs about the world would be systematically in error. If you are not envatted I shall say you are vat-free. The standard sceptical argument has the premise that we do not know we are not envatted and the conclusion that we know very little. That is my concern only in so far as it prompts the reply that even if, strictly speaking, we know very little we “know” a great deal that is warranted without complete certainty, but, say, with 99% confidence. Or, there is the reply that the probability that we know a great deal is at least 99% (even if it is not 100% probable). Presumably the sceptical rejoinder to these replies is to say that although we have good reason to behave as if vat-free we should seriously doubt that we are vat-free, where a serious doubt would consist in having, say, less than 99% confidence. Now there are many highly probable arguments that show that you are vat-free. From what little we know about demons if there are any, playing epistemic games is not high on their agenda compared, say, to corrupting once-honest politicians. Likewise there are various tests we can apply to decide if we are dreaming. Does the light intensity stay the same when you switch a light on or off? Can you go through walls or do you have to open the door? And so on. As for brains in vats, we have to estimate what proportion of the gross domestic product is devoted to envattment, and hence what the proportion of envatted to vat-free brains is likely to be. The trouble with all these arguments is not that they are highly probable rather than conclusive, so much as that they have premises drawn from our experience, which experience would be delusional if you are envatted. So, it could be said, all we have shown is that the belief in your vat-free status is warranted (in a way that involves at least 99% probability) if true.

56

Peter Forrest

An Informal Analysis Informally, we may reason thus: We start with the, not especially high, degree of support
that a proposition p derives from metaphysical considerations alone. This constrains the degrees of confidence, or as I shall call them subjective probabilities, so that Prob(p) 
and then we notice that p is warranted if p is true. This then increases the lower bound for Prob(p), so that Prob(p) 
, which in turn increases the lower bound for Prob (p is warranted), which increases the lower bound for Prob(p), and so on. Presumably these iterations converge to some limit , which might be significantly greater than
, such that eventually Prob(p)  . By epistemic bootstrapping I mean this increase from
to , arising from the judgement that p is warranted if true. Notice that it would often be foolish to put Prob(p) 
, if
is the degree of support the metaphysical arguments lend to p. For — Shock! Horror! — it may well be that metaphysical arguments have rather little to contribute to the discussion, in which case the degree of support given to not-p could be significantly less than 1 
, yet it is standard to take Prob(p)  Prob(not-p)  1. Because I am not saying that Prob(p) 
, but rather that Prob(p) 
, there might well be no increase in Prob(p) as a result of bootstrapping. What has increased is the lower bound that the all-things-considered philosophical arguments put on Prob(p). This increase is due to taking into consideration externalist warrant as well as the internalist justification provided by metaphysical argument. And it may be interpreted in either of two ways. One is to suppose there is an ideal subjective probability, equal to the logical probability conditional upon total evidence, and the inequalities concern this, only partially accessible, ideal. The other is to deny that there is a unique ideal probability but to take philosophical argument to be about putting additional constraints on subjective probabilities, which are taken to be those of a rational person who is not, however, aware of all the relevant arguments and so can be subject to alterations in subjective probabilities.

The Bayesian Analysis How should we formalise this analysis? Let us start with the epistemic innocent, who has beliefs expressible in a language containing no epistemic terms. We may idealise by supposing the epistemic innocent has a subjective probability assigned to every proposition in this restricted language, which subjective probabilities must obey the usual Bayesian principles. To say they are subjective is not, however, to say that they are uninfluenced by metaphysical arguments concerning such matters as the existence of a non-deceitful God or the question of whether we are vat-free or envatted.

Epistemic Bootstrapping

57

There is subsequently a loss of epistemic innocence, and an accompanying change of language so as to incorporate beliefs about the reliability of various belief-forming processes, not just beliefs about their past success. Now consider a metaphysical thesis m, which supports r, a statement of the reliability of a process that in fact leads to m. There is a subjective probability of m prior to the loss of innocence. Call this Prob(m). After the loss of innocence there is a new subjective probability Prob(m). Bootstrapping depends critically on the conditional probabilities Prob(m/r) and Prob(r/m). That these are high may be expressed by the conditionals, “If the process that leads to m is reliable then probably m is correct”, and “If m is correct then the process that leads to m is reliable”. It is the second of these that explicates the claim that m is warranted if true. The first follows from the concept of reliability, and there is an implicit “probability” qualification in the reliability of the process. For reliability implies a high probability of success rather than 100% probability. Bootstrapping would fail if we interpreted these conditionals as material ones. But I take it that even those such as Frank Jackson (1987) who interpret indicative conditionals as material ones grant that we would not ordinarily assert the indicative conditional unless the corresponding conditional probability were high. I am entitled, therefore, to take the bootstrapping scenario as one in which both Prob(m/r) and Prob(r/m) are high. Initially Prob(m) 
, and we aim to find  such that Prob(m)  . Let Prob(m/r)  1  and Prob(r/m)  1  . The judgement that m is warranted if true may now be unpacked as saying that is small. A small is part of what we mean by reliability. Clearly, however, there is a trade-off: if we require too small a value for then we cannot have small . Bootstrapping occurs when a fairly small and show that  is significantly greater than
. I now make a crucial inference from Prob(m) 
to Prob(m/not-r) 
*, where
*/
is fairly high, say over 80%. This asserts that the support that the metaphysical argument provides is to a great extent independent of the reliability of the belief-forming process being considered in the example. The crucial inference is not proposed as always correct, for it fails if the process is itself a metaphysical argument. But it does hold in the examples being discussed. I now make some straightforward calculations, using nothing more controversial than the multiplication principle that Prob(p/q)  Prob(q)  Prob(p & q). This multiplication principle makes precise a version of modus ponens: If q then probably p But probably q So (somewhat less) probably p & q.

58

Peter Forrest

Here are the calculations. Prob(m/not-r) 
*

(1)

Hence, Prob(m & not-r)  Prob(m/not-r)  Prob(not-r) 
* Prob(not-r)

(2)

Prob(m/r)  1 

(3)

Prob(m & r)  Prob(m & r)  Prob(r)  (1  ) Prob(r)

(4)

Prob(r/m)  (1  )

(5)

Hence,

Hence, Prob(m & r)  Prob(r & m)  Prob(r/m)  Prob(m)  (1  ) Prob(m)

(6)

Let x  Prob(r), y  Prob(m), z  Prob(m & not-r). So y  z  Prob(m & r). Then from (2), (4), and (6) we have, respectively, the inequalities: z 
*(1  x)

(7)

y  z  (1  )x

(8)

y  z  (1  )y

(9)

Substituting (7) into (8) and (9) we have: y 
*(1  x)  (1  )x

(10)

y  (
*/ )(1  x)

(11)

To solve the pair of inequalities, (10) and (11), we need no more probability theory, but only a little algebra. We first solve the simultaneous equations t 
*(1  x)  (1  )x and t  (
*/ )(1  x), obtaining: t  t  (1  )(1  t/
*)

(12)

Epistemic Bootstrapping

59

Hence,


*(1  )  (1  )t 
*(1  )

(13)

Thus we have the fairly simple formula: 1/t  (1  )/(1  )  /
*

(14)

We may then say that Prob(m)  , where  is whichever is the smallest of t,
*/ , and (1  ). It is easy to see that t
*/ , so  is whichever is the smaller of t and (1  ). The bootstrapping scenario is one in which is small, so if  were (1  ) we would have a most excellent result. Typical values will in fact be such that t (1  ). So we may suppose that: 1/  (1  )/(1  )  /
*

(15)

Because and are both small we have the approximation: 1/
1  /
*

(16)


1/(1  /
*)

(17)

So

(17) shows that we may bootstrap from quite a low
to a fairly high , provided
* is several times . For example if   5%, and if
 25% then
*  20%, and   80%. So 1   1   95%. That is, we are asserting with 95% confidence that if m, the metaphysical theory, is correct then r is the process leading to belief in m is 95% reliable. Taking that to be the explication of the claim that m is warranted if true it then follows that we may bootstrap from an initial judgement that m is at least 25% probable to a final one that m is at least 80% probable. Next, suppose we thought that 95% was too high a level of confidence and reduced this to 90%, so   10%. In that case, starting with an initial estimate for the probability for m of at least 25%, then the final probability is greater than 2/3. We would get the same final result from a very low-initial estimate of at least 2.5% provided   1%, that is if we are 99% confident that if m is correct the process leading to m is itself 99% reliable. If  
*, then   50% which would be interesting if , , and
were all very small. Spectacular bootstrapping occurs if both and are very small

60

Peter Forrest

and
* is N times , where N is fairly large. For then 
1  1/N, For instance if   1% (i.e. 99% confidence) and
 12.5%, then
*  10%, so   90%. Or, as might well be the case when worrying about envattment, if

  0.1% (99.9% confidence) and
 12.5% then   99%.

Unwarranted Even if True Suppose a metaphysical hypothesis m is in the unhappy position of being unwarranted if true. In this case we suppose Prob(m) , that is the initial estimate of the probability of m is that it is less than . And we aim to find  such that Prob(m) , that is the final estimate is that m has a significantly lower probability. Let Prob(not-m/r) 1   and Prob(r/m)  1  . The judgement that m is unwarranted even if true is asserted with confidence 1  , where  is small, and r is a process that reliably results in the negation of m, where the reliability in question is given by the probability 1  . Self-refutation occurs, to some extent at least, if a small  and  show that  is significantly less than . We have calculations analogous to the previous ones. Prob(m/not-r) 1

(18)

Prob(m & not-r) Prob(not-r)

(19)

Prob(m/r) 

(20)

Prob(m & r)  Prob(r)

(21)

Prob(r/m)  (1  )

(22)

Prob(m & r)  (1  ) Prob(m)

(23)

Hence,

Hence,

Hence,

Let x  Prob(r), y  Prob(m), z  Prob(m & not-r). So y  z  Prob(m & r). Then from (19), (21) and (23) we have respectively: z (1  x)

(24)

y  z x

(25)

y  z  (1  )y

(26)

Epistemic Bootstrapping

61

Substituting (24) into (25) we have: y (1  x)  x

(27)

y  (1  )x 1

(28)

(1  )y  x

(29)

Hence,

From (25) and (26):

As before it now requires just a little algebra to solve the pair of inequalities (28) and (29). Their solution is y , where y   is part of the solution of the simultaneous equations: y  (1  )x  1 and (1  )y  x. We have, therefore, 1/  (1  )/  

(30)

An interesting case occurs when  and  are equal and small. So suppose that     1/N. Then from (30), we have, 1/  (N  1)    N  1,

(31)

 1/(N  1)

(32)

Hence

Consider for instance the case of     10%. Then, by (32),  1/9. If     1% then, by (32),  1/99. The extent of self-refutation in the two cases is then the difference between  and the final figure of 10% or 1%, respectively. Notice that in this case there was no analogue of the “crucial inference”. This might help to explain why self-refutation is more intuitive than bootstrapping, the latter being applicable only where the crucial inference may be made correctly. It is important, however, that, apart from the crucial inference, the reasoning in the two cases is analogous. If, therefore, self-refutation is acceptable so should be bootstrapping.

Back to the Example of Christianity Some care is required in applying the Bayesian analysis of bootstrapping, so as to justify the crucial inference from Prob(m) 
to Prob(m/not-r) 
*, where

62

Peter Forrest


*/
is high. In the Christianity example, we might take for m the hypothesis that there is a non-deceiving God.3 Being a non-deceiving God makes it probable that the history of apparent revelation to individuals or communities is in fact a history of unsteady progress towards true beliefs about God. It makes this probable because if there is a non-deceiving God it is rather unlikely that this history of apparent progress is just a temporary state that God chooses not to interfere with, perhaps because later stages of human history will see this as a preliminary floundering around, standing to genuine religion the way alchemy does to chemistry. What I mean by a history of unsteady progress towards true beliefs may be understood by comparison with the history of scientific discovery. That God is non-deceiving does not make it antecedently probable that there should be such a history but does make it probable that if there is a history of apparent scientific progress it does reveal more and more about the nature of things. We may put this more succinctly by saying that the hypothesis of a non-deceiving God supports Scientific Realism. Likewise, the hypothesis of a non-deceiving God might be said to support Religious Realism, the thesis that a careful and critical acceptance of apparent revelations and apparent religious experiences leads on the whole to true beliefs. We may now let m be the hypothesis that there is a non-deceiving God and let r be the hypothesis that if there is history of apparent revelations and apparent religious experiences then it is on the whole reliable. We may then assign a high probability, without near certainty, to m conditional on r and to r conditional on m. These probabilities are themselves conditional upon background evidence that is not in dispute, notably that there is a history of apparent revelation and apparent religious experience. We are supposing there are metaphysical arguments that make no mention of religious experience or revelation and that establishes a moderate, but not high probability for m. We shall then obtain a fairly significant, but not mind-crushing, increase in the probability of m using bootstrapping, provided the crucial inference is justified. In passing it should be noted that a further discussion within the scope of religious realism would be required to argue from m (or more accurately from r) to Christianity. In this case, the crucial inference is justified because the metaphysical case would not be affected much by the supposition of not-r, which would merely tell us that we were mistaken in thinking that God intended the apparent progressive history of revelation and instead merely chose not to interfere with it.

3

My use of the phrase “non-deceiving” could be taken as stipulative, but in fact it is based on the idea that deception can be either active or passive, like euthanasia, and that both should be distinguished from not interfering (as in not officiously preserving life).

Epistemic Bootstrapping

63

The Refutation of Scepticism Some readers may think me perverse for being more interested in God than the deceitful demon. So let us now apply bootstrapping to envattment. Here m is the common sense world view that there are re-identifiable objects, among which are human beings who are persons, together with various beliefs about the behaviour of objects and persons (see Strawson, 1959; Joske, 1967). This hypothesis m makes it highly probable that r is the reliability of memory, perception and testimony, provided they are cross-checked in various ways. Moreover m is itself highly probable on r. In this case Prob(m) might be a probability based upon Descartes’ argument in the Meditations to the conclusion that there is a non-deceiving God.4 We do not need this, however, Prob(m) might simply be a priori subjective probability, that is, one assigned independently of observational evidence. But assigned by whom? Either we are considering inequalities based on our estimates of the subjective probabilities of an ideal thinker, or we are considering constraints more directly on our own subjective probabilities. But in either case I am considering a sceptic who, thinking about the hypothesis of envattment, begins with a genuine suspension of judgement. Imagine, then a first year student who wants to defend the sceptical position, saying, “Now you mention it, that I am envatted one way or another is more likely than this crazy tutorial being real”. How far will the student go? As far as saying the following? “I find all this so-called common sense totally weird, or implausible, or just too complicated to be taken seriously”. I think not. As I have said, I think I am entitled to concentrate on a sceptic who begins by suspending judgement. And that implies that there is a moderate a priori probability
of m, say
 10%. But let us be conservative. The sceptic, at least the one I am prepared to take seriously, does not claim that she or he should believe in envattment, even if we are adopting relaxed standards in which a 99% probability is taken as sufficient for belief. Instead the sceptic is inviting us to suspend judgement. So we may assume that
 1%. In this case we may take and to be significantly smaller than in the previous example, perhaps as small as 0.008%. So even if we were quite modest with the assessment of the a priori confidence in m bootstrapping would result in a high probability for m, about 99% with the above figures, refuting the sceptic. In this case the crucial inference is justified not merely with the safety factor built in so that
*/
 0.8 but even with
*/
 1. For the sceptic would agree that our envattment makes no difference to the a priori probability.

4

This is not the place for Descartes scholarship. It suffices to point out that for present purposes the argument from having the idea of a perfect being to the existence of such a being would have to have some not too low probability.

64

Peter Forrest

Replies to Two Objections Idealisation One objection might be that there are various idealisations built into the Bayesian theory of subjective probabilities. Among these is the rather peculiar assumption that degrees of confidence, even of an ideal thinker, may be expressed as precise betting quotients. Surely the actual human situation is messier than that. My reply is quite general. Bayesian calculations are far worthier of trust when they concern inequalities than when they concern equalities. Suppose first we thought that subjective probabilities were vague. That would be no obstacle to a claim such as Prob(m)  25%, however strange it might be to say that Prob(m)  26%. Likewise if we adopted Henry Kyburg’s account in which subjective probabilities were indeterminate and represented by intervals of reals we might have Prob(m)  {x: 25% x 40%}, but it is still the case that Prob(m)  25% (Kyburg, 1974, chapter 10). Yet again Isaac Levi (1980) suggests there is indeterminacy between credence functions, that is the assignment of subjective probabilities to propositions rather than individual subjective probabilities. But once again, provided for each credence function d in the state of epistemic innocence we have d(m)  25%, we may say that Prob(m)  25%. Neither vagueness nor indeterminacy threatens the inequalities. Reliability I seem to be assuming a reliability theory of warrant, and that is controversial. I reply that I am merely assuming that people have beliefs about the reliability of various truth-generating processes, such as perception, memory, testimony, religious experience and so on. Because there are such beliefs there can also be subjective probabilities concerning the reliability of truth-generating processes. It is the interaction between the subjective probabilities and the reliability that generates bootstrapping or self-refutation as the case may be. So it is reliability not the concept of warrant that is being considered here. It could be retorted that I should still be embarrassed by the way the reliability of a belief-forming process depends on the class to which it is taken to belong.5 For instance, suppose we are considering the reliability of purported revelations. We might ask in what context the revelation is being considered, and what are the criteria which have to be met by the purported revelations. The result is that we shall consider revelations belonging to class A, revelations belonging to class B,

5 For a recent discussion of the problems in characterising the reliability of a belief-forming process see Alston (2005, pp. 114–132).

Epistemic Bootstrapping

65

etc. and they will have different reliabilities depending on the classes. So for some function  of classes and percentages, there might be subjective probability greater than (X, ) that revelation of class X is reliable to a degree at least . The probability-theoretic calculations used in my discussion of bootstrapping and self-refutation would be accurate if the classes were mutually exclusive. For in that case we would just select the class to which the revelation being considered belongs. The objection, though, is that any class X of revelations may be divided into subclasses with differing reliabilities, with the very finest division being into singletons, for which the reliability is, it could be said, trivially either 0% or 100% depending on whether the particular process in question in fact leads to a false or a true belief. So which is the relevant division into classes? I submit that the relevant class X to which belief-forming process z belongs is one such that no further information about it is relevant. Hence if Y is any sub-class to which we are able to assign z then for all , (X,  )  (Y,  ), even if Y is just the singleton {z}. Hence, because we are considering the subjective probabilities about these reliabilities, the supposed relativity of reliability to the class being considered is not a threat to the probability-theoretic analysis — even if in fact the reliability varies with the class, becoming either 0% or 100% in the case of the singleton.

Conclusion That beliefs are warranted if true may result in increased support for the beliefs in question, although considerable care is required in assessing just how much is the increase. Such bootstrapping is not the exact opposite of self-refutation, for the former but not the latter requires the crucial inference based on the judgement that the metaphysical considerations and considerations of reliability are largely independent.

References Alston, W. (2005). Beyond “Justification”: Dimensions of epistemic evaluation. Ithaca, NY: Cornell University Press. Jackson, F. (1987). Conditionals. Oxford: Blackwell. Joske, W. (1967). Material objects. New York: St. Martin’s Press. Kyburg, H. (1974). The logical foundations of statistical inference. Dordrecht: Reidel. Levi, I. (1980). The enterprise of knowledge: An essay on knowledge, credal probability, and chance. Cambridge, MA: MIT Press.

66

Peter Forrest

Plantinga, A. (1993a). Warrant: The current debate. New York: Oxford University Press. Plantinga, A. (1993b). Warrant and proper function. New York: Oxford University Press. Plantinga, A. (2000). Warranted Christian belief. New York: Oxford University Press. Plantinga, A. (2003). Probability and defeaters. Pacific Philosophical Quarterly, 84, 291–298. Strawson, P. (1959). Individuals, an essay in descriptive metaphysics. London: Methuen.

Chapter 5

More Praise for Moore’s Proof Roger White Love it or hate it, G. E. Moore’s (1939) “Proof of an External World” continues to fascinate us. Recent appraisals range from William Lycan’s (2001, p. 43) insistence that “there is nothing deeper, in all philosophy, than Moore’s response to the skeptic”, to Crispin Wright’s (2002, p. 333) dismissal of it as “the episode of simple-minded question-begging which it has always seemed”. How could such a simple argument so divide us? There is much that is intriguing about Moore’s argument. On the one hand, it seems to exemplify Moore’s notorious adherence to commonsense, not allowing that philosophizing might rationally upset our firmest convictions. On the other, it appears to be a radical and provocative challenge to a kind of philosophical commonsense. Not many of us have seriously doubted that there is an external world, or that we know there is. But proving the existence of an external world has struck so many philosophers as a profoundly challenging or perhaps impossible task. Along comes Moore with a wave of his hands claiming to have provided the elusive proof. I will be joining the chorus of praise for Moore’s argument, but in ways that have not been recognized by others. I will suggest first of all that even some of Moore’s recent defenders have underestimated the dialectical utility of his proof. And secondly, even if we reject the crucial assumptions that many contemporary defenses of Moore appeal to, his argument can still have epistemological lessons for us. Let me state up front some assumptions I will be making in considering the merits of Moore’s argument. I will assume there is indeed an external world which is more or less how it appears to us, including the existence of hands and the like. Furthermore, we are typically justified in believing, and even know, that there are Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

67

68

Roger White

external objects before us when there appear to be. Some will say that these matters should be up for grabs too. But I don’t think that making these assumptions stacks the deck too far in favor of Moore to make the assessment of his argument interesting. Most who have thought that Moore’s proof is deeply unsatisfactory do not think that its merits hinge upon the truth or epistemic status of its conclusion. Even if we do have hands and we know it, it is often thought, there is something obtuse about Moore’s appeal to such knowledge in his argument for an external world. My purpose is to consider what value we might find in Moore’s argument if we already take for granted that skepticism is false.1

Moore’s Argument and Meta-Argument Moore claimed that he could prove that two human hands exist by raising his hands saying “Here is one hand, and here is another”. He had earlier argued that hands, if there are such, are “things outside of us”, and hence that in proving that there are hands he will have ipso facto proved that there exist external objects. It seems fair enough to render Moore’s overall argument as follows: P1 Here are two hands. P2 If there are two hands then there is an external world. C So, there’s an external world. Ironically, while Moore spends most of his energy defending P2, there is so little interest in disputing it today that only the last five pages of the paper are typically reprinted in anthologies. Of course, probably even fewer would wish to dispute P1. Most interest focuses on the claims that Moore makes about his argument. For in addition to this argument, Moore presents a meta-argument: an argument that his first argument constitutes a proof of its conclusion. This is what is crucially at issue, for Moore begins his discussion by quoting Kant’s concern that without such a proof the existence of external objects “must be accepted merely on faith”. The meta-argument could be rendered thus: MP1 Three conditions suffice for an argument to constitute a rigorous proof: (i) The premise must be different from the conclusion, (ii) the premise is known to be true, and (iii) the argument is valid. MP2 My argument meets each of these conditions. MC So my argument is a rigorous proof. 1

For a different defense of Moore working from some similar assumptions see Hetherington (2001).

More Praise for Moore’s Proof

69

This is actually a slight overstatement, as Moore admits that there may be some further condition required for a proof that he hasn’t thought of. But he backs up his case by noting that there are plenty of arguments of this general form that we have no trouble recognizing as conclusive proofs. We might, Moore suggests, establish that a certain book contains some misprints by pointing and saying, “There’s a misprint, and there’s another”. And if we can be sure that these are misprints (as indeed we might be) then a fortiori we can be sure that Moore just raised his hands. The challenge then is to say what is lacking in Moore’s argument if it is not a similarly conclusive proof.

What’s Wrong with Moore’s Proof ? It is hard to deny that Moore’s argument meets his suggested criteria for proofhood. If we balk at granting that Moore has proved the existence of external objects, what crucial feature might his argument lack? A typical first complaint is that his argument is question-begging. Unfortunately, it is rather obscure just what this means or what is wrong with it. It won’t do to say that he is merely “assuming what he is trying to prove”, for as Moore rightly notes, his premise does not simply state that there are external objects but rather provides some specific examples. Perhaps in some sense his conclusion contains information already contained in the premise — it is after all a valid argument — but this can hardly be one of its flaws. Sometimes the accusation that an argument begs the question appears to amount to the complaint that it is just too obvious. “Of course if I were to accept that then I would have to accept the conclusion on pain of inconsistency! But I’m not stupid, so my doubts about the conclusion naturally extend to your premises as well”. Moore’s argument may well invite this kind of response from a skeptic, and even non-skeptics might sympathize with the reaction. But it is not clear what this complaint amounts to beyond accusing Moore of bad manners by insulting the skeptic’s intelligence. Any valid argument for the existence of external objects will be such that an external world skeptic cannot coherently believe the conjunction of the premises. It is a contingent psychological matter that some arguments are more obviously valid than others, and hence that we are more likely to find skeptics believing the premises of some arguments than others. But it is far from clear how this could be a complaint of epistemological significance. And I will argue at any rate that Moore’s argument needn’t even be psychologically ineffective. A more subtle diagnosis (Wright, 2002) of Moore’s argument notes that if we are justified in believing Moore’s premise P1, then it is by virtue of having a certain visual experience: it appears to us that there is a pair of hands before us. But certain

70

Roger White

conditions must be met in order for an experience as of some hands to justify us in believing that there are indeed some hands. In particular, it would appear that we must be able to rule out, or at least to put very little credence in, alternative explanations of my experience such as that I am merely dreaming that I have hands. But this is just what the skeptic typically claims to have trouble doing (he insists that we ought to have the same trouble too). The skeptic doubts that there is anything external to his mind because he suspects that he may be subject to a massive illusion. An argument that can justify a skeptic in concluding that there is an external world will have to give him justification for denying that all his visual experiences are part of a dream and that there is really nothing out there. But a justified denial of this dreaming hypothesis is required if the appearance of two hands is to give the skeptic any reason to believe that some hands exist. So it appears that one can’t gain justification for believing in the external world by way of Moore’s argument. If we are already justified in believing in the external world then we don’t need Moore’s argument. If we are not, then Moore can’t help us.

The Dogmatist Defense We should, however, proceed more carefully here, for the worry sketched above glossed over an important distinction. It may be true that one is justified in believing that there are two hands only if one is justified in denying that one is falsely dreaming that there are two hands. This simply follows from a kind of closure principle on justification. Obviously, it can’t both be the case that there really are two hands and that one is falsely dreaming that there are. One can’t coherently be sure that there are hands while being entirely unsure whether one is only dreaming that there are. So, if one is justified in believing the former then one is justified in denying the latter. But by itself, this fact does not bar one from gaining justification via Moore’s argument. For all we have said, one might see Moore raise his hands, come to thereby know that there are two hands and as a consequence that one is not just falsely dreaming that there are hands. Since we are justified in this last conclusion, the condition on having empirical justification for Moore’s premise will indeed have been met. What is required to block this kind of move is a claim to the effect that in order for the experience of Moore’s hands to give one justification for believing that there are two hands, one must have justification for denying the dreaming hypothesis that does not depend on this very experience. It might seem plausible to suppose that one can legitimately conclude that P when it appears that way only if one is already justified in denying that one is merely dreaming that P. If this is so, then it is clear that I can’t, as a result of watching Moore wave his hands, gain

More Praise for Moore’s Proof

71

justification for denying that I am merely dreaming about it and hence that there is an external world after all. Following Wright (2002) and Davies (2000), we can say that while we are justified in believing Moore’s premises and hence also its conclusion — this much follows from the closure of justification under known entailment — our justification for the premise does not transmit to the conclusion. It is not by virtue of our justification for the premises that we are justified in drawing the conclusion. Rather, an independent justification for believing the conclusion is required in order to gain justification for the premise in the first place. If this is right, it reveals one kind of failure that Moore’s argument suffers: the argument is no help whatever in giving one justification for a belief in the external world. However, it is just this point at which many of Moore’s defenders get off the boat. According to the view that Pryor (2000) calls dogmatism, one needn’t have independent justification for denying that one is merely dreaming that there are some hands, in order for the experience as of some hands to give you justification for supposing that there are. What is required for perceptual justification on this view is only the weaker condition that one lack reasons to suspect that one is dreaming or otherwise deceived.2 In this way, Pryor (2004), Peacocke (2004), and Davies (2004) suggest that by watching Moore raise his hands we may gain justification for the belief that there are hands and hence in turn that we are not merely dreaming and there is indeed an external world.

The Dialectical Critique As even Moore’s defenders recognize, there seems to be something wrong with his alleged proof. It has generally been thought that a full defense of Moore requires a diagnosis of why it appears so inadequate. The popular answer here is that his argument is hopelessly inadequate for certain purposes, most obviously for the purpose of rationally persuading someone of its conclusion. Pryor (2004) argues that Moore’s argument has this limitation even though it might provide one with some justification for its conclusion.3 To support this line, we need to draw a distinction between having justification to believe that P, and being rational in drawing the conclusion that P. To the extent that I suspect that I am merely dreaming of some hands, it does not seem appropriate for me to conclude that there are two hands on the basis of my experience. 2

Versions of dogmatism have been proposed by Alston (1986), Burge (1993, 2003), Chisholm (1989), Ginet (1975), Peacocke (2004), Pollock and Cruz (1999), and Pryor (2000, 2004). 3 For a similar critique without the dogmatist commitment see Jackson (1987).

72

Roger White

My suspicion that I am dreaming rationally obstructs me from concluding that there are some hands before me. This appears to be so regardless of whether my suspicion is justified. A justified suspicion that I am dreaming would serve as an undercutting defeater for my perceptual justification for believing that Moore is raising his hands. If my suspicion that I am dreaming is entirely unjustified, then according to Pryor my visual experience may well provide me with justification for a belief that there are hands. But given my suspicion that I’m dreaming of hands it would be inappropriate reasoning to judge by appearances that there are hands before me. On this way of looking at it, while Moore’s proof may provide us with some justification for its conclusion, it cannot rationally persuade us to accept it. If we doubt that there is an external world, or at least if this doubt is due to the suspicion that we are subject to a visual hallucination or the like, then we cannot rationally conclude that there are hands when there appear to be. On the other hand, as long as our doubts are not justified, the appearance of two hands will give us some justification for believing that there are hands.

Might Moore’s Argument Rationally Persuade? I argue elsewhere against the dogmatist position on perceptual justification.4 So, I can’t accept that when we see Moore’s hands we thereby gain some justification for denying that we are merely dreaming of hands.5 Nevertheless, I think that Moore’s defenders may have underestimated the utility of his argument for rational persuasion. Let’s suppose that Skeptic Sue has come to doubt that there is an external world, and gives a fair portion of her credence to the possibility that all her experiences are illusory. So, she travels to Cambridge to watch Moore rehearse his proof and see if she is persuaded. The first thing to note is that it is a psychological fact about us that it is very hard not to believe that there is a hand in front of us when there vividly appears to be. When we consider the matter very abstractly from the armchair, it can be tempting to think that the competing hypotheses that I see some hands and that I am merely dreaming that there are hands are epistemically on a par. Such reflections might cause someone to give a

4

See White (forthcoming). For other objections to dogmatism see Cohen (2002) and Schiffer (2004). It is not so clear that an appearance of hands couldn’t possibly give me any evidence in support of the existence of an external world, even though it cannot support the thesis that I am not merely dreaming of hands. Suppose that somehow I first became conscious with my conceptual capacities already developed (perhaps that’s not possible). If my first experience was as of some hands this might give me more reason to suppose that there is an external world than if I enjoyed no experience at all. 5

More Praise for Moore’s Proof

73

large measure of credence to the dreaming hypothesis. But it is very difficult to maintain this doubt when you look at a hand up close, turning it around to view at all angles. While we attend to such a vivid experience and consider whether there is a hand before us, we typically can’t help but be struck by the conviction that it is indeed a hand. Most likely then, when Sue sees Moore’s hands she will find herself believing that there are hands. So far this is just a psychological fact about the impact of perceptual experience. While it may be irresistible, to draw this conclusion given her suspicions about possible illusions does not seem fully reasonable. Nevertheless, the question remains how she ought to respond now that she does believe Moore’s first premise. She cannot coherently maintain her conviction that there are hands while giving a large share of her credence to the possibility that she is merely dreaming that there are some hands. Other things being equal, rationality requires some adjustment in her attitudes. The question is whether she ought to abandon her conviction that there are two hands, or conclude instead that the appearance of hands is no illusion, and that there is indeed an external world. What can be said for or against either of these possible readjustments of belief? On the one hand, it may be that it was not rational for Sue to conclude that there are two hands when there appeared to be, given her doubts about the veracity of her perceptual experience. But on the other, she is not justified in the credence she gives to the possibility of visual deception, and she does have adequate justification for believing that there are two hands. She has such justification because it appears to her that there are two hands, and she is not justified in believing or giving more than negligible credence to any alternative hypothesis — such as that she is only dreaming — that could potentially undermine her perceptual justification for supposing that there are two hands. In the light of this, it seems only appropriate that she continues to believe that there are two hands. Rather than abandon this conviction, she should cease to give more than the slightest credence to the possibility that her senses are deceiving her and hence conclude that there are external objects. This conclusion is entirely compatible with the anti-dogmatist position that Skeptic Sue gains no justification for Moore’s conclusion by witnessing his proof. She already had sufficient justification for supposing that the appearance of a world outside was no illusion; it is just that her convictions were not in accord with her justification. The experience as of two hands merely prompts her to believe as she ought. In doing so, she may make an irrational transition to the justified belief that two hands exist. But perhaps this is the best that can be expected from someone who is already being irrational in doubting that there is an external world. In this way we see that not only might Moore’s proof be psychologically very effective in persuading an external world skeptic, it can in a most

74

Roger White

important sense be epistemically effective. Delivering Moore’s proof is like a kind of invasive epistemic surgery. We need to force the skeptic through a brief episode of irrationality (judging that there are hands while strongly suspecting that she is merely dreaming of hands) in order to help her form a belief for which she already had adequate justification (that there is an external world). I have been following Pryor in supposing that it is not rational to conclude that there are hands on the basis of experience if you think (even without justification) that your experience may well be just a dream. But even this might be questioned. The source of a skeptic’s doubt about the veridicality of her experiences is usually prior doubts about the epistemic status of competing hypotheses. Sue might first think that she would be no more justified in supposing that there are hands than that she is merely dreaming that there are. Even if she is wrong about this, it is appropriate that she bring her convictions in line with what she takes to be justified, and hence to give considerable credence to the dreaming hypothesis. But when she looks carefully at Moore’s hands she may be struck not only with the conviction that there are two hands but that this is the reasonable thing to believe, and that by contrast the dreaming hypothesis is not. It is not that her experience as of two hands provides evidence for the epistemological thesis that she is justified in giving very little credence to the possibility that she is merely dreaming of hands. Quite apart from this experience, ideally she ought to have been able to judge that she ought not to think that she is dreaming. But the experience might prompt her to make a correct judgment concerning the epistemic status of the dreaming hypothesis. If it does, then she should drop her credence in the dreaming hypothesis to bring it into accord with what she correctly judges it ought to be. If together with this change of attitude she comes to think that there are two hands in response to her evidence, then it seems that she can’t even be charged with any irrationality in her initial belief transition.

Moore’s Proof Versus Philosophical Proofs It might be worth comparing Moore’s proof with the kind of proof that philosophers have usually hoped for. It has been thought that the only adequate proof of an external world would be one that makes no appeal to perceptual experience and hence whose premises are assessable a priori. The hope is to find some premise (or conjunction of premises) P such that careful armchair reflection reveals that P is true, and it can be shown to entail the existence of external objects. Needless to say, the prospects for finding such a premise have seemed dubious. But let us suppose that we have one. How should a skeptic respond to such an argument? The hope is that our friend Sue will consider the proposition P and

More Praise for Moore’s Proof

75

come to see that it is true by reason alone. Then upon realizing that P is true only if there is an external world, she will come to believe the latter also. But now given that Sue doubts that there is an external world, she is rationally committed to having corresponding doubts about P. That is, given that P entails there is an external world, her credence in P should not exceed her credence that there is an external world. In persuading Sue that P is true, we are inducing an (hopefully temporary) incoherence in her degrees of belief. And this seems to be less than fully rational. Of course, if she was fully rational in the first place there would be no need for this, for she would already be confident that there is an external world. But as with Moore’s proof, if Sue is struck by the intrinsic plausibility of premise P, she may come to believe it despite her doubts about the external world. And then once she believes P, it seems quite appropriate that she increases her confidence in the external world to cohere with her confidence in P, since she is justified in believing in the external world anyway. This a priori proof from P to the external world appears to have no epistemic advantage over Moore’s proof. Indeed, it appears that Moore’s proof is in crucial respects stronger than any such purely philosophical proof. In order for the proof from premise P to be psychologically effective, P will have to be something not very obviously connected with the existence of an external world. Suppose instead that we let P be the proposition that 2 ⫹ 2 ⫽ 4 and that there is an external world. Or perhaps that either there are external objects or there are round squares, but there are no round squares. Even though we are justified in believing these (since we are justified in believing that there is an external world), we can hardly expect an external world skeptic to find them very plausible. What we need for P is some proposition that can seem very plausible independently of whether there is an external world. Of course there is no such premise P that ought to seem plausible independently of whether there is an external world. For since if the argument is valid the absence of an external world entails not-P, a fully rational subject’s judgment as to whether P is true should not float free of her attitude to the existence of an external world. But given our contingent cognitive limitations, we might judge that P is true and then be surprised as we follow the subtle logical steps that this commits us to a belief in the external world. But now if the entailment from P to the existence of the external world is far from obvious, this weakens the epistemic force of the argument. For one might very well raise doubts as to whether the subtle chain of reasoning from P to the external world has led us astray. By contrast, the entailment from Moore’s premises to his conclusion is as obvious as it gets. There is no room for even a minimally competent thinker to doubt that his conclusion follows. Nevertheless, Moore’s proof does not necessarily suffer from the defect that it is unlikely that a

76

Roger White

skeptic would believe its premise. For as I have suggested, focusing on the vivid experience as of some hands has a powerful psychological effect regardless of one’s initial doubts. All things considered then, Moore’s proof appears to be better than the best that a purely philosophical proof of its conclusion could be.

Why Does Moore’s Argument Seem So Bad? Nevertheless, it does have its limitations. I must admit that I have had students who claim at least to be ignorant of anything beyond the nature of their immediate experience. And I have not always managed to pull them out of it by showing my hands. Indeed they have often thought that I must be joking. The suspicion might remain even among non-skeptics that my students’ reaction is somehow appropriate. I think this may be partly so, but that we can accommodate it without insisting that Moore’s proof is dialectically worthless. The students’ reaction is along the lines of “Obviously this hand-waving business can’t rationally persuade me!” We must distinguish between this reaction’s being understandable and its being correct. I admit the former but not the latter. The student presumably doubts that there is an external world because she thinks that she is not justified in giving more credence to the hypothesis that there are external objects than that there vividly appear to be such objects but there are not. It might be my fault that she thinks this, as I have just finished making the prima facie case that one’s total empirical evidence can have no bearing on which alternative is true, and that there can be no non-empirical justification for deciding such a matter. However, I have led her astray. As we are assuming for the purposes of this paper, we are justified in believing in the external world, and hence in giving very little credence to the skeptical alternative. If my student were justified in giving a fair share of her credence to the possibility that her life’s experience is nothing but a dream, then she would not be justified in believing in an external world, and would not be justified in believing that I have hands when I display them. So given her mistaken belief that she is justified in suspecting that she is merely dreaming, it is entirely understandable that she would mistakenly judge that Moore’s proof should not rationally persuade her. In this way, we can diagnose what seems inadequate about Moore’s proof even if it is not.

Epistemological Lessons from Moore Most of us when we encounter skeptical arguments do not go as far as seriously doubting the existence of the external world. But we may start to wonder about the epistemic status of our belief in the external world. We can, temporarily at least, fall into a kind of cognitive dissonance. While we can’t help but think that external things

More Praise for Moore’s Proof

77

surround us, we wonder whether we really ought to be so sure about this. There is a tension here, as ideally we ought to believe as we believe we ought to believe. The worry can arise as follows. Suppose we let E capture everything about the qualitative character of any experience I have ever had. Let External World be the proposition that there is an external world, and Illusion be the hypothesis that while E is true, there is no external world. Since the existence of external objects is obviously incompatible with Illusion, if I’m justified in believing in an external world then I must surely be justified in disbelieving Illusion. Yet, it is very hard to see how my justification for denying Illusion could possibly have anything to do with experience. Since E is just the course of experience that we can predict that I will have if Illusion is true, then if anything, E supports Illusion, and hence can hardly be the basis of my justification for denying it. Now it is clearly a contingent matter whether Illusion is true. But can armchair reflection reveal the truth of any contingent proposition? To suppose so can seem to attribute mysterious powers to human minds. It is tempting to think that it is only with the aid of some sensory input that one can draw conclusions on contingent matters. Hence there seems to be no way that we could be justified in denying Illusion, and hence no way of being justified in believing in an external world. I want to suggest that reflection on Moore’s argument can be instructive concerning the skeptical worry even if we reject the dogmatist position of Moore’s supporters according to which one can gain some justification for the hypothesis that there is an external world by looking at our hands. What Moore’s argument shows us is that we are indeed justified in believing that there is an external world. For when we look at our hands we clearly know that we have two hands. As Moore (1959, p. 146) says, “How absurd it would be to suggest that I did not know it, but only believed it, and that perhaps it was not the case!” But it straightforwardly follows that there is an external world, and further that Illusion is false. Hence, surely we are justified in believing these also. The thesis that there are external objects is much more modest than that there are two hands, and notIllusion is much weaker still. Since we are obviously justified in believing that there are two hands when there appear to be, a fortiori we are justified in believing in these weaker claims. While observing two hands may not give us any justification for supposing that there is an external world, reflection on the fact that we are justified in believing that there are two hands before us can aid us in rationally concluding that we are justified in believing that there are external objects. What we have is a variation on Moore’s proof in terms of justified belief. P1* We are justified in believing that here are two hands P2* If we are justified in believing that here are two hands, then we are justified in believing that there is an external world C* So, we are justified in believing that there is an external world.

78

Roger White

No doubt some will complain that this argument is question-begging in the sense of being just too obvious to be of any use. Won’t any sane person who doubts the conclusion also doubt that we are justified in believing that these are hands, especially when the source of his doubt about the conclusion is the suspicion that we are no more justified in believing External World than we are in believing Illusion? I don’t think so. Of course, as with any valid argument, anyone who has thought the matter through thoroughly enough will already either believe both premise and conclusion or deny both, or at least put no more confidence in the premise than in the conclusion. But one needn’t be stupid to find P1* more obviously true than C*, at least when they are considered independently. The reason is that we tend to do a better job at evaluating the epistemic status of a proposition for which we have a clear body of empirical evidence than a proposition for which we have little or no evidence, even if the latter is a much weaker proposition. Perhaps empirical evidence has some bearing on whether there is an external world. If your entire existence involved no experience at all, perhaps you would have a hard time telling if there was anything outside of your mind. But it does not take us very far. In particular, my total evidence E does not favor External World over Illusion. I have no empirical evidence against Illusion at all. Reflecting on whether I am justified in believing either Illusion or External World can be difficult if we consider them in isolation from our justification for anything else. However, when we turn to consider the far more specific question of whether there are two hands before us, we have no trouble judging that we are justified in believing that there are. For we have ample evidence — a clear visual appearance as of hands right before us — to conclude that there are hands. The epistemic version of Moore’s argument aids us in seeing that we are justified in believing External World, and denying Illusion by drawing our attention to their entailment by a proposition that we are clearly justified in believing.

Moore Meets Bayes We can set out the matter in Bayesian terms as follows. P(E/Illusion) ⫽ 1

(1)

For Illusion entails that E is true and External World is false. And, P(E) ⬍ 1

(2)

For it is unlikely that I should have the very experiences that I have had. From (1) and (2) we get P(E/Illusion) ⬎ P(E)

(3)

More Praise for Moore’s Proof

79

from which it follows by Bayes’ Theorem that P(Illusion/E) ⬎ P(Illusion)

(4)

P(not-Illusion/E) ⬍ P(not-Illusion)

(5)

and further that

That is, far from supporting my denial of Illusion, my total evidence E actually lowers its probability to some degree. But now External World is inconsistent with Illusion, from which with (5) it follows that P(External World/E) ⫽ P(not-Illusion/E) ⬍ P(not-Illusion)

(6)

That is, the probability that the external world exists given my total evidence is no greater than the probability that Illusion is false given no evidence at all. But what value should we assign to P(not-Illusion)? It is perhaps not easy to say by just considering it in isolation. But if we do not give it a high probability, then we cannot coherently judge it likely that there is an external world. But now let Hands be the proposition that there are two hands before us. P(Hands/E) is very high

(7)

This is obvious upon reflection. My total evidence — which includes in particular my current vivid experience as of two hands before me — renders it very likely that there are two hands. But Hands entails not-Illusion. And so P(Hands/E) ⱕ P(not-Illusion/E) ⬍ P(not-Illusion)

(8)

And hence from (7) and (8) we get P(not-Illusion) is very high

(9)

So we have shown that we ought to be quite confident that Illusion is false quite apart from empirical evidence. This can seem a little surprising, as Illusion is a contingent proposition about matters external to us. We have arrived at this conclusion not by considering Illusion in isolation, and trying to judge directly what degree of confidence we should put in it with no evidence to go on. Rather we have followed Moore in focusing on the specific claim that there are two hands before us when there appear to be.

80

Roger White

There is perhaps a general lesson here concerning the application of probability theory to reasoning and the assessment of evidence. A popular simplistic Bayesian picture has it that we begin with certain prior probabilities and likelihoods and move to new probabilities by the method of conditionalization. On this approach, in evaluating P(Illusion/E) we begin with a prior judgment concerning P(Illusion) and calculate P(Illusion/E) by Bayes’ Theorem as P(Illusion) ⫻ P(E/Illusion)/P(E). The discussion above suggests that this is not always the best way to approach the matter. Perhaps an ideally rational subject can make perfect judgments concerning prior probabilities and when conditionalizing on these everything will go fine. But the rest of us are often better off working backwards from posterior probabilities to priors as we did above. For we are often better at judging the appropriate values of certain posterior probabilities such as P(Hands/E) directly than we are at assessing prior probabilities.

Conclusion I hope to have shown that Moore’s proof of an external world has more going for it than has generally been acknowledged. It may well be rationally persuasive for some people. And even for those of us who don’t need persuading, reflection on it can still be fruitful.

References Alston, W. P. (1986). Epistemic circularity. Philosophy and Phenomenological Research, 47, 1–30. Burge, T. (1993). Content preservation. The Philosophical Review, 102, 457–488. Burge, T. (2003). Perceptual entitlement. Philosophy and Phenomenological Research, 67, 503–548. Chisholm, R. (1989). Theory of Knowledge (3rd ed.). Englewood Cliffs, NJ: Prentice-Hall. Cohen, S. (2002). Basic knowledge and the problem of easy knowledge. Philosophy and Phenomenological Research, 65, 309–328. Davies, M. (2000). Externalism and armchair knowledge. In: P. Boghossian & C. Peacocke (Eds), New Essays on the a Priori. New York: Oxford University Press, 384–414. Davies, M. (2004). Epistemic entitlement, warrant transmission and easy knowledge. Supplement to the Proceedings of the Aristotelian Society, 78, 213–245. Ginet, C. (1975). Knowledge, Perception, and Memory. Dordrecht: D. Reidel. Hetherington, S. (2001). Good Knowledge, Bad Knowledge: On Two Dogmas of Epistemology. Oxford: Clarendon Press. Jackson, F. (1987). Conditionals. Oxford: Blackwell. Lycan, W. (2001). Moore against the new skeptics. Philosophical Studies, 103, 35–53.

More Praise for Moore’s Proof

81

Moore, G. E. (1939). Proof of an external world. Proceedings of the British Academy, 25, 273–300. Reprinted in Moore (1959), Philosophical Papers. London: Allen and Unwin, 127–150. Peacocke, C. (2004). The Realm of Reason. Oxford: Clarendon Press. Pollock, J., & Cruz, J. (1999). Contemporary Theories of Knowledge (2nd ed.). Totowa, NJ: Rowman & Littlefield. Pryor, J. (2000). The skeptic and the dogmatist. Noûs, 34, 517–549. Pryor, J. (2004). What’s wrong with Moore’s argument? Philosophical Issues 14, 349–378. Schiffer, S. (2004). Skepticism and the vagaries of justified belief. Philosophical Studies 119, 161–184. White, R. (forthcoming). Problems for dogmatism. Philosophical Studies. Wright, C. (2002). (Anti-)Sceptics simple and subtle: G. E. Moore and John McDowell. Philosophy and Phenomenological Research, 65, 330–348.

This page intentionally left blank

82

Chapter 6

Lotteries and the Close Shave Principle John Collins

1. The Lottery Problem When I hold a single ticket in a fair lottery with many tickets, then, no matter how probable it is that my ticket will lose, I cannot correctly be said to know that it will lose. Call this fact the lottery observation. If the moral to be drawn from this observation is that mere probability of truth, no matter how great, is never sufficient for knowledge, then the threat of a new style of skeptical argument looms. For it appears that many of the things I take myself to know are things that I can know only if I can rule out certain quite mundane counter-possibilities, against which my evidence is merely probabilistic. In recent years this lottery problem has assumed a place alongside the Gettier problem and more traditional skeptical considerations (of the Evil Demon or brainin-a-vat variety) as one of the major challenges facing a satisfactory philosophical account of the concept of knowledge. In this paper I will argue that the importance of the lottery observation has been exaggerated in the recent literature. The lottery poses no special skeptical threat. For while the lottery observation is quite correct as far as it goes, and illustrates something important about the concept of knowledge, the lottery phenomenon is also a fairly isolated one that fails to generalize in the way that has sometimes been claimed. I will outline, in Section 2, some of the cases that have been offered as generalizations of the lottery situation. Section 3 presents a characterization of the original lottery case that distinguishes it from the alleged generalizations. I will propose Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

83

84

John Collins

a rather weak necessary condition on knowledge that fails to be satisfied in the lottery example. I claim that a lottery ticket holder fails to know that her ticket will lose precisely because of the failure of this condition. And since this necessary condition on knowing is satisfied in each of the supposedly generalized cases, I say that they are not really analogous to the lottery example at all. For all that lottery considerations have shown, it is quite possible that in each of these other examples the subject does possess knowledge. Lotteries pose no special skeptical threat. There is no lottery problem. This attempt to isolate the lottery example is defended, in Section 4, against an objection stemming from cases of belief based on a certain kind of statistical reasoning. The aims of the present paper are quite modest. In arguing that there is no special skeptical threat from lottery considerations, I am certainly not suggesting that skepticism is false. And in saying that lottery considerations do not force us to deny knowledge in the allegedly analogous cases, I am not claiming that in all or some of those cases the subject really does know.

2. The Lottery Generalized It is the belief that the lottery observation readily generalizes that has impressed the lottery problem upon many contemporary epistemologists. Jonathan Vogel was one of the first to suggest that the phenomenon is a widespread one with potentially devastating skeptical consequences. The following example is paraphrased from Vogel (1990). Car Theft: Suppose that several hours ago Smith left his car parked on a side street in a major metropolitan area. Since Smith clearly remembers where he parked the car, we may be inclined to say he knows where his car is. But does he know that his car has not been stolen in the last couple of hours and driven away from where he parked it? Many people would say that he does not. Now knowledge is commonly held to satisfy the following closure principle. Closure Principle: If S knows that p, and S knows that p entails q, then S knows that q. Given the Closure Principle, there is an obvious tension between the pair of responses Vogel says are standardly elicited by the previous example. That Smith’s

Lotteries and the Close Shave Principle

85

car is still where he left it entails that it has not been stolen in the last couple of hours and driven away. So from his failure to know that his car has not been stolen and the Closure Principle, we may draw the skeptical conclusion that Smith does not in fact know where his car is. Our primary interest here, however, is with Vogel’s reason for thinking that there is an analogy between the Car Theft example and the lottery case that explains why Smith does not know that his car has not been stolen. Here is part of what Vogel has to say on the matter (ibid., p. 16): In effect, when you park your car in an area with an appreciable rate of auto theft, you enter a lottery in which cars are picked, essentially at random, to be stolen and driven away. Having your car stolen is the unfortunate counterpart to winning the lottery. And, just as one doesn’t know that one will not have one’s number come up in the lottery, it seems one doesn’t know that one’s number won’t come up, so to speak, for car theft. Vogel goes on to offer a range of further examples, which he suggests are also analogous to the lottery case. Here are two of them presented as conversational exchanges (ibid., pp. 20–21): Luncheonette: Q. Do you know where I can get a good hamburger? A. Yes, there’s a luncheonette several blocks from here. Q. Do you know that a fire hasn’t just broken out there? A. No. Meteorite: Q. Do you know what stands at the mouth of San Francisco Bay? A. Yes, the Bay is spanned by the Golden Gate Bridge. Q. Do you know that the Bridge wasn’t just demolished by a falling meteorite? A. No. In a similar vein, John Hawthorne devotes the opening pages of his recent book Knowledge and Lotteries to persuading the reader that “the problem posed by lotteries is not an isolated oddity, but is actually widespread” (2004, p. 3, n. 5). The first of Hawthorne’s examples approaches the point directly, by drawing attention to the problem of claiming to know anything which entails that a particular person fails to win a major prize in a lottery. It should be noted that an example of this sort was already discussed in Harman (1986, pp. 71–72).

86

John Collins Safari: Suppose someone of modest means announces that he knows that he will not have enough money to go on an African safari this year. We are inclined to treat such a judgment as true, notwithstanding various far-fetched possibilities in which that person suddenly acquires a great deal of money. We are at some level aware that people of modest means buy lottery tickets from time to time, and very occasionally win. And we are aware that there have been occasions when a person of modest means suddenly inherits a great deal of money from a relative from whom he had no reason to expect a large inheritance. (Hawthorne, 2004, p. 1)

Hawthorne then proceeds to give several other cases that he claims are also analogous to the lottery case, despite the fact that there is no actual lottery involved. I will give only one of these here, to conclude the present section: Heart Attack: I am [Hawthorne writes] inclined to think that I know that I will be living in Syracuse for part of this summer. But once the question arises, I am not inclined to think that I know whether or not I will be one of the unlucky people who, despite being apparently healthy, suffer a fatal heart attack in the next week. (If only medical self-examination were so easy!) Indeed I am just as unwilling to count myself as knowing about the heart attack as I am to count myself as knowing about the lottery. The analogy continues. Just as I have excellent statistical grounds for supposing that any given lottery ticket will lose, I have excellent statistical grounds for supposing that a given apparently healthy person will not have a heart attack very soon. … And just as many of our ordinary commitments entail that this or that person will lose a lottery, many of our ordinary commitments entail that this or that person will not soon suffer a fatal heart attack. (Ibid., p. 3)

3. The Far-Fetched and the Merely Improbable Now I think that in fact there is a stark disanalogy between the original lottery case and the attempted generalizations of the previous section. I suggest that this disanalogy is only overlooked when we conflate the notion of a state of affairs being far-fetched with the quite distinct notion of its being (merely) improbable; when we confuse what is outlandish or far from the truth with what is just unlikely. Let me introduce the point by examining a case (adapted from Dretske, 1970) that is clearly not analogous to the lottery example.

Lotteries and the Close Shave Principle

87

Zebra: Smith is at the zoo with his two sons. They are looking at a zebra standing in an enclosure marked “Zebra”. Smith’s younger son asks, pointing, “Daddy, do you know what kind of animal that is?” Smith replies: “Yes, it’s a zebra”. Smith’s teenage son then turns to his father and says: “But Dad, how do you know that it’s not just a mule cleverly painted to look like a zebra?” My primary concern here is not with whether the teenager’s challenge is appropriate, or whether Smith does in fact fail to know that the animal they are viewing is not a painted mule, and hence fails to know that it is a zebra. Neither is the point here to evaluate the status of the Closure Principle, or to ask whether, perhaps, an epistemologically relevant shift of context might have taken place in the course of the conversation. I would simply like to ask the following question: is the Zebra case plausibly construed as a lottery example? In other words: do lottery considerations show that Smith fails to know that the animal is not a painted mule? I think the answer to that question is, quite clearly, no. I know that fraud and deception do take place, even at certain apparently respectable public institutions such as (we may imagine) the zoo in question. But I would deny that, when Smith purchases his admission ticket to the zoo, he is in any sense, in effect purchasing a ticket in a lottery in which some ticket-holders will be picked, essentially at random, to be defrauded and deceived. There is simply no reason to suppose that any such mechanism is in place in the situation described in the example. The skeptical possibility raised by the teenage son’s question is not just improbable, and Smith’s evidence against it is not “merely probabilistic” in the same sense in which my evidence that my ticket will lose the lottery is merely probabilistic. Given Smith’s background knowledge about the trustworthiness of zoos and of public institutions in general, and of this zoo in particular, the hypothesis is not just very probably false. It is very probably a far-fetched possibility. That is to say, it is a possibility which, very probably, is not at all similar to the way things actually are. Smith does not just have probabilistic reasons for thinking that the animal is not a painted mule. He also has probabilistic reasons for believing that the closest alternatives to actuality in which it is a painted mule are fairly remote. Vogel seems to acknowledge this point about the Zebra case. He writes (1990, p. 14): The reason you know that an animal in the pen is not a disguised mule (if you do know it’s a zebra) is that you have a true belief to that effect backed up by good evidence. That evidence includes background information about the nature and function of zoos. You know

88

John Collins that zoos generally exhibit genuine specimens, and that it would be a great deal of trouble to disguise a mule and to substitute it for a zebra. Only under the most unlikely and bizarre circumstances, if at all, would such a substitution be made, and there is no reason whatsoever to think that any such circumstances obtain.

But now contrast this with the Lottery case. While the possibility that my single ticket will be the winner is very improbable, it is also a possibility that I know to be very close to actuality (if not actual!) and not at all far-fetched. At least, that is the case if I know that the lottery is a fair one. I propose that the reason I fail to know that my ticket will lose is that I know that, no matter what happens, I will either win, or, at worst, come very close to having won. In other words, I am suggesting that the reason I do not know that my ticket will lose, despite the overwhelming probability I assign to that prospect, is that the following necessary condition on knowing is not satisfied: The Close Shave Principle: If S knows that p, then there is no possibility that is very close to actuality at which p is false and to which S assigns non-zero probability. Equivalently: If there is some possibility that is very close to actuality at which p is false, and to which S assigns non-zero probability, then no matter how subjectively improbable this possibility is, S doesn’t know that p. Now let us return to the first of the supposed analogies to the Lottery: Vogel’s Car Theft case. If my suggestion is correct, then much will depend on how we fill in the details of the example. First of all, suppose that, in the neighborhood in which Smith parked his vehicle, there is a gang that, quite literally, steals cars at random. Each evening they commence by drawing a ball from an urn to determine which side street they will hit. Then they toss a coin to decide whether to take a car from the north or south side of the street, then another ball is drawn to determine which car on that side of that street will be taken. If this is the case, then, yes, if Smith parked in that neighborhood, he certainly does not know that his car has not been stolen, and for the same reason that I do not know that my ticket will not win the lottery.

Lotteries and the Close Shave Principle

89

Alternatively, let us suppose that there is indeed a car thief at work in the neighborhood, that she operates opportunistically, if not completely randomly, and that as a matter of fact she comes very close indeed to stealing Smith’s car on the night in question. Walking along the street, trying car door handles, ignoring those vehicles with obvious anti-burglary devices, she comes upon Smith’s car. It is locked but unprotected. She sees that the coast is clear, but then, just as she is about to force the lock, a police patrol car swings around the next corner. With the police car’s headlights upon her, she casually resumes her stroll down the block. By the time danger has passed, she has happened upon another vehicle just as attractive as Smith’s. She steals that car instead. In this last scenario, does Smith, sitting blithely at a nearby restaurant, know that his car has not been stolen? I am strongly inclined to say: “No, he doesn’t”. This example provides both illustration and confirmation of the Close Shave Principle, I think. Were an observer of the whole episode later to describe to Smith exactly what had transpired, Smith might well respond as follows: “Whew! That was a close thing. I thought I knew where my car was that night when I left the restaurant, but I guess I did not. I was just lucky that it was still there where I parked it”. Absent such details, what should we say about the example though? I suppose I am inclined to agree that Smith does not know that his car has not been stolen, but the situation is really not at all clear to me, and neither is it my present concern to settle the matter. What is clear to me, however, is that lottery considerations alone do not force us to say that Smith lacks knowledge in the unadorned Car Theft case. If we do not know that Smith’s car comes close to being stolen, then we cannot appeal to an alleged analogy with the Lottery case in order to deny Smith knowledge of his car’s whereabouts. What of the more direct kind of example that Hawthorne adapts from Harman, in which a claim has been made to know something that entails that a particular person will not be a lottery winner? Consider the Safari example. Suppose that Smith, being of modest means, announces that he knows he will not be able to afford an African safari this year. Should we concur with this judgment? Note that in order to cast skeptical doubt on Smith’s claim via lottery-style considerations it is not sufficient simply to remind ourselves (as Hawthorne does) that “people of modest means buy lottery tickets from time to time, and very occasionally win”. We would need to know in addition, for example, that Smith is in the habit of buying the occasional lottery ticket, or that someone will give (or very nearly might have given) Smith a lottery ticket in the near future. After all, if I know that someone does not hold a ticket in a certain lottery, then there is no apparent problem with my knowing that he will not win that lottery.

90

John Collins

In short, if I know that Smith does have a lottery ticket, then we are back in the original lottery case. We should then simply agree that Smith does not know that he will not win a major prize, and hence that he does not really know that he will be unable to afford the safari. If, on the other hand, the story leaves it unclear whether or not Smith will have, or very nearly might have had, a lottery ticket, then Smith’s claim to knowledge is as unclear as in some of the previous cases. But then the example is clearly no longer one to which the lottery observation straightforwardly applies. Similar comments apply to the Heart Attack example. Heart attacks are not the result of any process that remotely resembles an actual lottery. If I am generally in good health then I consider it not only very probable that I will not suffer a heart attack in the next few months, but also very probable that I will not even come close to suffering a heart attack. If I later learn than I did come close, perhaps, say, because a crucial artery was very nearly blocked by a blood clot, then I will certainly retract any claim to have known that I would not. Once again, this serves to confirm the significance of the Close Shave Principle. Of course, this is not to say that in the absence of any particular medical evidence it is ever correct to say that I know that I will not suffer a heart attack in the near future. All I am claiming here is that lottery considerations are not relevant to the example. I suspect that our reluctance to ascribe knowledge in this sort of case, and to seek a medical check-up even when in apparent health, has more to do with the high cost of being wrong, than to any analogy with the case of the lottery.

4. Inductive Knowledge Suppose that on the basis of empirical observation we have come to believe that a certain generalization is both true and law-like. It appears that such a belief, acquired by inductive inference from a limited sample, can straightforwardly count as an item of knowledge, despite the fact that what we know about the sample fails to entail the truth of the generalization. And in many cases the possibility of acquiring such knowledge is quite unproblematic for the view being defended here. If, for example, we have come to believe on the basis of what has been observed, that, as a matter of physical law, metals are good conductors of electricity, we will not only assign a high probability to the proposition that this (untested) piece of metal will prove to be a good conductor; we will also consider it highly probable that the closest alternatives to actuality at which the metal strip is not a good conductor will be rather remote possibilities. An apparent problem arises, however, in the case of law-like statements arrived at inductively, that turn out on closer examination to be statistical in nature.

Lotteries and the Close Shave Principle

91

Consider for example the following two cases: Ice Cube: Several hours ago, you left an ice cube in a thermos half-filled with lukewarm water. It occurs to you that the ice cube you left in the lukewarm water must have melted by now. Despite the fact that you are not presently looking at the contents of the thermos, you know that the ice cube has melted. (cf. Vogel, 1987, p. 206) Uranium: Roger places a piece of uranium on a photographic plate, and discovers that the plate has become fogged. He repeats the experiment many times. After numerous trials, he puts a piece of uranium on a plate, goes away from his laboratory, and returns some time later. Roger believes that the plate is fogged. Moreover he knows, by induction, that the plate is fogged, even before he inspects it. (see Vogel’s manuscript) The problem here is that physicists tell us that in each of these examples, the process that underwrites the inductive generalization (ice melts in the sun on hot days, uranium fogs photographic plates) is a merely statistical one. Does it follow from the Close Shave Principle that neither is a case of knowledge? If so, then the Close Shave Principle stands refuted. Now in order to see whether the Principle really applies to cases such as these, we will have to look more closely at the particular details of the examples. Let us take the uranium example first. Suppose that Roger placed a one-gram lump of uranium on a photographic plate several hours ago. The dominant isotope that makes up more than 99 percent of the element as it occurs naturally is Uranium 238, which has a half-life of about four and a half billion years. The “half-life” is a period of time in which half of the atoms present may be expected to have undergone radioactive decay. Because the half-life in this instance is so long, the probability that any single particular atom in the lump will have decayed in the few hours that Roger has left it unattended is extremely small. And, since the decay process is an essentially indeterministic one, this is the kind of situation to which the Close Shave Principle has application. Suppose Roger is somehow able to form the belief, de re, about some single particular atom in the lump, that in the several hours since he left the piece of uranium unattended in his lab, that atom has not decayed. Then this is indeed belief in a “lottery proposition” (in Vogel’s terminology). Roger’s belief is very probably true, yet, even supposing that the belief actually is true, Roger does not know that the atom has not decayed in the last several hours. Why? Because to suppose that, contrary-to-fact, the atom in question had decayed, requires us only to envisage

92

John Collins

a possible state of affairs that differs from actuality with respect to one tiny particular matter of fact. This is, then, a possibility that is very close to actuality, to which Roger assigns a (very small yet) positive probability, and in which Roger’s belief about the atom is false. The Close Shave Principle tells us that no matter how improbable Roger considers this possibility to be, he does not know that it does not obtain. But Vogel’s example is quite different in structure to that. It involves the indeterministic behavior over several hours, not of a single atom, but of a vast quantity of atoms. There are so many atoms in a one-gram lump of uranium that, even given the very small probability that any particular atom will undergo decay, one may confidently expect about 25,000 of these individually improbable events to occur each second that passes. Each of these decay events involves the emission of an alpha particle accompanied by weak gamma radiation. Over the course of several hours there will be hundreds of millions of individual decay events within the lump, and this suffices to expose the photographic film. So, in order for the photographic plate not to have become fogged since Roger left it, some hundreds of millions of independent chance events that actually did take place must be supposed not to have taken place. But this means that there is no possibility that is close to actuality in which the photographic plate is not fogged, since any possibility that differs from actuality with respect to the occurrence of some hundreds of millions of independent events is not at all close to the way things actually are. The proposition that the plate has not become fogged is not a “lottery proposition” at all; the Close Shave Principle simply does not apply here. Consistently with our observations about the subject’s lack of knowledge in the original lottery case, we can say the following things about this example: Roger knows that it is physically possible, though overwhelmingly improbable, that the photographic plate is unfogged. Roger knows, moreover, that it is overwhelmingly probable that the plate is not even close to being unfogged, i.e. that there is no possible world at all close to the actual world at which the plate is not fogged. So, consistently with the Close Shave Principle, we say in this case that Roger does know that the plate has been fogged by the uranium. Similar considerations apply to the case of the ice cube. It is true that there are physically possible initial microstates of the system consisting of the water and the ice that are (i) macroscopically indistinguishable from the actual initial microstate of the system, and which (ii) evolve according to the laws of physics over the next several hours into a state in which the ice remains unmelted in the water. (By “microstate” here I mean a complete specification of the initial position and momentum of each and every one of the water molecules in the thermos.) Do I know that the system was not initially in one of those unusual microstates?

Lotteries and the Close Shave Principle

93

It should be clear by now that lottery considerations do not force a negative answer to that question. For it is not just that I consider the possibility that the ice remains unmelted overwhelmingly improbable. In addition I think it overwhelmingly improbable that this possibility is even remotely like the way things actually are in the thermos. There is no small adjustment to the actual initial state of the system that would transform it into one in which the ice fails to have melted after floating for several hours in the lukewarm water. That would require a quite extraordinary coordination of the trajectories of the individual particles in the system. Such extraordinary coordination is physically possible, but this possibility differs from the way things actually are with respect to the positions and velocities of a vast number of molecules. Once again: any possibility that differs from actuality with respect to a vast number of independent matters of particular fact is not a possibility that is even remotely close to the way things actually are. The proposition that the ice remains unmelted might be assigned some tiny positive probability, but it is not a “lottery proposition”. I have argued that lottery considerations pose no threat to this sort of inductive knowledge. But is the discussion of the preceding paragraphs consistent with my concession that in the original, genuine, lottery example, the ticket-holder fails to know he will lose? In other words: can the original lottery example survive the same level of scrutiny to which the Uranium and the Ice Cube examples have just been subjected? Consider a typical lottery mechanism. In the New York State Lotto, for example, ping pong balls bearing numerals from “1” to “59” are released into a transparent chamber. The balls are mixed by rotating paddles in the chamber. A valve is then opened, through which six balls are allowed to pass into a clear tube that leads to the display area. The ball-mixing paddles are of crucial importance. The resulting jostling of balls is a dynamic process, which though quite deterministic, is also highly modally sensitive. Very slight changes to the positions or velocities of just a few balls will be amplified by the ensuing collisions to produce completely different results. This ensures that every possible combination of numbers will, at worst, have come very close to having won, no matter what actually transpires when the valve is opened. One consequence of this is that predicting the result of the draw in advance is a practical impossibility. Another consequence is that, by the Close Shave Principle, no ticket holder can know in advance of the draw that her ticket is going to lose. As Hawthorne (2004, p. 8) reminds us, the official slogan of the New York State Lottery is: “Hey, you never know”. Note that it would be not at all appropriate here to argue that making, e.g. a slight change to the position of single ball would really require changing a vast

94

John Collins

number of particular matters of fact, since each ball is itself made up of a vast quantity of particles. The ball is a rigid body whose parts move together in a quite predictable way. The particular matters of fact about the precise locations of the micro-particles that make up the ball are not independent matters of particular fact. Another example discussed by Vogel (manuscript) seems to be poised delicately between the genuine Lottery case and examples of the statistical-inductive kind just considered: Hole-in-One: Sixty golfers are entered in the Wealth and Privilege Invitational Tournament. The course has a short but difficult hole, known as the “Heartbreaker”. Before the round begins, you think to yourself that, surely, not all sixty players will get a hole-in-one on the “Heartbreaker”. Do you know that not all the players will get a hole-in-one? Vogel thinks that you do. I am inclined to agree with Vogel here, but for what I suspect are rather different reasons. In fact I believe that the example is under-described. I think that judgment as to the correctness of an ascription of knowledge might well go either way in this case, depending on how various missing details of the story are filled in. Let us suppose that the players in the tournament are professionals. When a skilled professional golfer aims to land his ball on the green on a par three hole, then whether or not he succeeds is typically not at all a matter of chance. In fact a professional golfer may well be able to land the ball reliably not only on the green, but on a certain part of the green. Often a player will aim his shot right at the pin; that is, attempt to “hole out”, or, failing that, to leave the ball as close to the hole as possible. Now of course not even the best golfers have the level of control that would be required to hole out reliably at a distance of, say, 100 m. So we might well take it to be a matter of chance whether or not a certain skilled golfer who is attempting to make a hole-in-one on the Heartbreaker will be successful or not. The proposition that this player makes a hole-in-one is indeed then a lottery proposition, and I do not know in advance that she will not succeed. I think she will succeed in landing the ball on the green, perhaps quite close to the hole. I doubt very much that the ball will actually land in the hole, but I do not know that it will not. That is because, although I doubt very much that the shot will result in a hole-inone, I am also fairly sure that there is a possible state of affairs that is very close to the way things actually are, one perhaps in which the trajectory of her swing is ever-so-slightly different, in which the player does make a hole-in-one.

Lotteries and the Close Shave Principle

95

Matters are complicated, however, by the fact that, not infrequently, a good golfer will not even be attempting to make a hole-in-one when she steps up to the tee at a par three hole. Suppose, for example, that the pin is positioned near an edge of the green that is bordered by a deep and treacherous sand trap. Then it may make more sense strategically to aim for a safer region of the green, from which position the player can be confident of completing the hole in at most two putts, rather than to run the risk of disaster. If that is what our golfer has in mind when she plays the tee shot, then I think our epistemic judgment about the example changes. She is aiming the ball at an area of the green some distance from the pin, and, since she is a skilled player, it is very probable that she will succeed in landing the ball somewhere in that region. We can then be confident that she will not make a hole-in-one, because doing so would now require her to mis-hit the ball, something that she will probably not even come close to doing. In this situation, the proposition that the player does not make a hole-in-one is no longer a lottery proposition at all, and lottery considerations give us no reason at all to think that one cannot know that she will not make a hole-in-one. Returning to Vogel’s own version of the example: the reason I think you do know that not all 60 players will make a hole-in-one on the “Heartbreaker” is that it is highly probable that among those 60 players in the tournament, several (at least) will, for one reason or another, be trying not to make a hole-in-one, and can be reliably counted on not to make a hole-in-one when attempting not to do so. These players might be attempting, rather, to place the ball in a position from which the hole can be completed safely in par. The particular details of the examplescheme might be filled out in many ways, of course, but it seems to me that for many of the most natural looking scenarios, Vogel is mistaken in thinking that the case is “very much like the lottery”.

5. Conclusion In the preceding sections various examples have been examined in considerably more detail than is customary, but if the central thesis of this paper is correct, then it is precisely those details that are going to determine, in any particular case, whether or not the subject can be correctly said to know. The suggestion that lottery considerations provide an alternative route to skepticism depends on the claim that lottery propositions are widespread. I doubt very much that that is so. Such propositions will be found only where there is an underlying mechanism that ensures the existence of very close alternatives to actuality in which the lottery proposition is false. In many of the ordinary examples that have been offered as generalizations of a genuine lottery situation, it is quite

96

John Collins

implausible to imagine that there is any such underlying mechanism. While it is true that a ticket-holder in a fair lottery does not know that his ticket will lose, the epistemological significance of this observation is, it seems to me, rather slight.

References Dretske, F. (1970). Epistemic operators. The Journal of Philosophy, 67, 1007–1023. Harman, G. (1986). Change in view: Principles of reasoning. Cambridge, MA: MIT Press. Hawthorne, J. (2004). Knowledge and lotteries. Oxford: Clarendon Press. Vogel, J. (1987). Tracking, closure, and inductive knowledge. In: S. Luper-Foy (Ed.), The possibility of knowledge: Nozick and his critics (pp. 197–215). Totowa, NJ: Rowman & Littlefield. Vogel, J. (1990). Are there counterexamples to the closure principle? In: M. Roth & G. Ross (Eds), Doubting: Contemporary perspectives on skepticism (pp. 13–27). Dordrecht: Kluwer. Vogel, J. (manuscript). Subjunctivitis.

Chapter 7

Skepticism, Self-Knowledge, and Responsibility David Macarthur Modern skepticism can be usefully divided into two camps: the Cartesian and the Humean.1 Cartesian skepticism is a matter of a theoretical doubt that has little or no practical import in our everyday lives. Its employment concerns whether or not we can achieve a special kind of certain knowledge — something Descartes calls “scientia”2 — that is far removed from our everyday aims or standards of epistemic appraisal. Alternatively, Humean skepticism engages the ancient skeptical concern with whether we have good reason, or any reason at all, for our beliefs, including the common or garden beliefs that are presupposed in our ordinary practical affairs. On this traditional conception, philosophical doubt is a projection of everyday doubt and the lessons of the study are potentially lessons for the street. In this paper, I shall focus on the Humean strain of skepticism whose focus concerns whether we have adequate reasons for our beliefs. Henceforth, when I speak of skepticism it is this variety of skepticism that I am primarily referring to.3 I want to relate skepticism, so understood, to two kinds of self-knowledge. I shall argue that the failure of past solutions and dissolutions of skepticism to

1

Conant (2004), alternatively, takes the important distinction to be that between Cartesian and Kantian skepticism on the ground that Descartes raises a doubt about whether we actually have knowledge of the external world, whereas Kant raises a doubt about whether such knowledge is possible at all. But this is not a robust distinction: Descartes’s doubt blurs into Kant’s. 2 See, for example, Cottingham, Stoothoff, and Murdoch (1985, Vol. I, p. 10). 3 Although there is obviously some overlap with the more narrowly focused discussions of knowledge skepticism. Cf. Stroud (1984), and Unger (1975).

Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

97

98

David Macarthur

provide a satisfying response to the skeptic can be accounted for in terms of two stances that we can take toward our own beliefs.4 One is the stance from which we endorse beliefs and, more generally, make up our minds what to think on the basis of our sensitivity to reasons. I shall call this the deliberative stance. The other is the stance according to which we regard our beliefs, so far as possible, in the same way that others do, namely, as states of oneself that can be explained in purely theoretical (typically causal) terms. I shall call this the naturalistic stance.5 A lesson of Hume’s is that adopting the naturalistic stance on one’s own beliefs plays a crucial role in motivating skepticism. I want to go a step further than Hume and explain why that is so. But my main aim is to explain why skepticism remains a threat even after we have acquired a philosophical understanding of it. In order to set the stage for the discussion of these two kinds of self-knowledge we must first consider the problem of the problem of skepticism. I want to ask: what kind of problem is skepticism? We are all familiar with the fact that in the history of philosophy there has been a long line of failed refutations of skepticism — a refutation being a non-question-begging answer to the skeptical conclusion. A refutation must demonstrate that one has the requisite knowledge or justification that has been called into question on the restricted basis of only those premises that the skeptic is happy to grant us. Repeated efforts have been made to refute skepticism by appeal to alleged certainties of reason or common sense, by transcendental argument, inference to the best explanation, and appeal to externalism about content, to name only the most familiar attempts. Despite their ingenuity and interest, these refutations are uniformly disappointing since they seem to fall on one side or other of a dilemma: either they end up begging the skeptic’s question without explaining why they are entitled to do so; or their rational reconstructions put a false cast on the ordinary concepts, say, of justification or knowledge that they are trying to defend.6 More recently, a different kind of response to skepticism, termed quietism, has gained popularity and with good reason. The quietist is not in the seemingly hopeless business of even attempting to refute or answer the skeptical conclusion. The basic insight of quietism is that it is a mistake to take the skeptical argument at face value — as the refuters do. The aim, rather, is to show that the skeptical

4

My account of the dual nature of self-knowledge and its internal tensions is indebted to Richard Moran’s richly suggestive work, Authority and Estrangement: An Essay on Self-Knowledge (2001). 5 Moran (2001, p. 3) speaks of this as “the purely theoretical or spectator’s stance towards the self”. 6 Here we find the motive for Cavell’s conception of skepticism as embracing both traditional skeptical reflections and traditional counter-arguments (1979). On this conception, skepticism is an attack on our ordinary concepts and their criteria of application.

Skepticism, Self-Knowledge, and Responsibility

99

problem is itself defective on the ground: either (1) that the skeptical problem rests on disputable (or unnatural) theoretical presuppositions (e.g. Williams, 1991) or (2) that the very posing of the “problem” is subtly incoherent (e.g. Putnam, 1998). Although my sympathies lie more with the quietists, in this paper I want to suggest that both refuters and quietists share a common Cartesian presupposition about what a response to skepticism should achieve. A certain way of thinking about skepticism, the type of problem that it is, and the way we should respond to it, has gone missing in the Cartesian focus of many contemporary discussions of skepticism. It is to the thought of Hume that we must turn if we want to recover this way of thinking.

The Idea of a Once-and-for-All (Dis)Solution Our thinking about skepticism in general, and modern skepticism in particular, is very largely conditioned by our understanding of Descartes’s attitude to skepticism as exemplified by the figure of the meditator. You will recall that by the end of his Meditations, the appeal to a non-deceiving God has laid the hyperbolic skeptical doubts of the First Meditation to rest. Looking back on the achievement of the Meditations Descartes boasted “I became the first philosopher ever to overturn the doubt of the skeptics” (Cottingham et al., 1985, Vol. II, p. 367). Here Descartes expresses the philosophical desideratum that has shaped the vast majority of philosophical responses to skepticism ever since: namely, that what is wanted is that skepticism be overturned, which I take to mean that the threat posed by skepticism is to be completely neutralized. This ideal of what a philosophical response to skepticism should aim for is one shared by refuters and quietists alike, even in spite of their differences. Whether one has adopted the ambitious aim of refuting the skeptical conclusion or the more modest task of showing that the skeptical problem is ill-conceived, the almost universally shared assumption is that skepticism is the kind of problem that can be overturned once and for all. In his reply to Hobbes, Descartes likens skepticism to a disease and his philosophy to a cure. So we might say that the common Cartesian assumption shaping our thinking about skepticism is that our aim in philosophy is to provide a final cure. Once philosophy has done its job we are supposed to have achieved a perspective from which the skeptical problem will not seriously trouble us again. The longed-for goal is a reflective life completely free of skeptical torment. Yet the very fact that no past attempt to bring such closure to skepticism has won widespread approval should at least give us pause. Surveying 2000 years of

100

David Macarthur

philosophizing, one is struck by the fact that skepticism was there at the beginning of philosophy and is with us still, despite the efforts of modern philosophers in the Cartesian mould. Given the remarkable fact that skepticism returns again and again phoenix-like after every one of its supposed burials, it is at the very least worth asking whether we should accept the Cartesian assumption that a once-and-for-all solution, or dissolution, of skepticism is possible, or even desirable. Perhaps the problem of skepticism is more like the problem of self-deception or weakness of the will than a problem that promises a theoretical resolution. Instead it may be the sort of problem that we can come to understand better without thinking that our aim in approaching the problematic phenomenon philosophically is to rid the world of it once and for all. Supposing that the causes of, say, self-deception lie deep in the human psyche then in coming to better understand self-deception one’s aim need not involve any attempt to rid the world of selfdeceivers. Self-deception, we might think, is not that kind of problem. But why, then, should we follow Descartes in his thinking about the kind of problem skepticism is? Perhaps the interminability of skepticism is not something to be overcome but a fact about ourselves that we must learn to live with as best we can. That would be to see our inability to neutralize skepticism once and for all not as a failure, but as a datum requiring explanation. On this alternative Humean approach skepticism is a much deeper and more interesting phenomenon than Descartes and the philosophical tradition influenced by him have tended to suppose since, in a sense, skepticism is ineradicable. But why should that be so? To answer this question let us turn back to Hume.

The Interminability of Skepticism In a notorious passage of the Treatise Hume remarks (1978, p. 218): This sceptical doubt, both with respect to reason and the senses, is a malady, which can never be radically cur’d, but must return upon us every moment, however we may chace it away, and sometimes may seem entirely free from it. ’Tis impossible upon any system to defend either our understanding or senses; and we but expose them farther when we endeavour to justify them in that manner. As the sceptical doubt arises naturally from a profound and intense reflection on those subjects, it always increases, the farther we carry our reflections, whether in opposition or conformity to it. Carelessness and in-attention alone can afford us any remedy.

Skepticism, Self-Knowledge, and Responsibility

101

It is clear from this passage that no system of thought, including his own naturalistic science of man, is thought of by Hume as an answer to the skeptic. The naturalist tenet that there is a natural or non-rational basis for, say, our belief in the external world is not to be thought of as refuting external world skepticism or showing that the skeptical problem of how we can rationally justify this belief is illconceived or in any other way defective. Hume’s point is simply that our natural belief replaces skeptical doubt as a matter of fact once we leave the study and its intense reflections. Naturalism is at best a partial, or temporary, cure for the skeptical malady. It does not remove the motivations that lead to the skeptical predicament. Note, too, in contrast to Descartes, Hume’s curious remark that skepticism “can never be radically cur’d”. No sooner do we enter the study or return to our intense meditating upon the problem of the rational justification of this or any other belief, than we inevitably find that there is no answer to give and we once again find ourselves in a “philosophical melancholy and delirium” (ibid., p. 269). That there is no radical cure for skepticism means that after philosophy has done its best to neutralize and accommodate itself to the skeptical threat, that threat remains a live option, something that follows inevitably upon a certain kind of reflection to which we are inevitably drawn. This is the feature of skepticism I am calling its interminability. But what kind of reflection leads to the malady of skepticism? Here one might plausibly think that there is no general story to tell: different arguments lead to different versions of skepticism. For example, specific reflections about perception can lead to external world skepticism, whereas a certain way of thinking about bodily behavior can lead to other minds skepticism; and so on. On this view there is no general structure underlying the different skepticisms — or, even, the different arguments for a given kind of skepticism7 — and the best that philosophy can do is to address each separately. I want to suggest, on the contrary, that there is a general story to tell linking these different versions of skepticism. But it is not that I want to claim that there is a common theoretical assumption or argument-structure underlying all the many and various skeptical arguments.8 Rather, Hume’s reflections on skepticism suggest that there is a general story to tell about the conditions under which we take

7

What is commonly called “external world skepticism” can arise either from arguments attacking the claim to have knowledge (by appeal to the closure principle, or a certainty condition, etc.), or from reflections attacking the claim that perception is ever reliable. 8 Nonetheless, I take it that many skeptical arguments trade on appeal to considerations that suggest an unbridgeable causal gap between our data or evidence (the effects: subjective experience, behavior, etc.) and the object(s) of skeptical concern (the causes: external world, mind, etc). Skeptical scenarios, on this view, are alternative causal hypotheses designed to explain all of the “data”. There is more discussion of skeptical scenarios as alternative causal hypotheses in Macarthur (2003, 2004).

102

David Macarthur

these considerations seriously. Our capacity to take various skeptical considerations seriously depends upon certain basic and ineliminable features of the concept of belief and the equivocal character of our capacity for self-awareness. For these reasons, skepticism is a perennial possibility of the human subject. Let me explain.

Hume’s New Science of the Mind At this point it is worth recalling Hume’s general move in the history of philosophy. If we think of the natural sciences as attempting to explain phenomena according to causal explanations, ideally in terms of efficient causal laws, then Hume’s move is to apply this causal-explanatory strategy to the human mind itself. A central tenet of naturalism is the idea that human beings do not stand over against nature but are part of it and, like other natural things, are wholly explicable in “scientific” forms of understanding by way of “scientific” methods of investigation.9 In the Treatise Hume articulates what he calls a new “science of MAN” (1978, p. xv) according to which the method of studying the human mind is explicitly modeled on the scientific study of nature (ibid., p. xvi): ’Tis no astonishing reflection to consider, that the application of experimental philosophy to moral subjects should come after that to natural. Like the sciences of nature, Hume’s science of human nature “must be laid on experience and observation” (ibid.). If the aim of the natural sciences is to limn the causal structure of the universe and discover its principles then, analogously, the aim of Hume’s new science of the mind is to discover universal causal principles or laws to account for our mental life, especially our beliefs and ideas. Hume’s general idea is to apply a causal-explanatory project that has had great successes in explaining inanimate nature to the explanation of the mind itself; and his mind, in particular.

Two Perspectives on one’s Own Beliefs The clue to the interminability of skepticism is the fact that in adopting a naturalistic stance toward the contents of his own mind — especially his own beliefs — Hume is led to a radical skepticism about virtually everything, from which there 9 The scare quotes indicate that the introspective standpoint that Hume typically adopts would not now be recognized as scientific.

Skepticism, Self-Knowledge, and Responsibility

103

seems to be no rational return. The question is: why does considering our own beliefs as natural phenomena lead to skepticism? In order to approach this question it is worth recalling some central features of our concept of belief. I shall then consider the way in which these features bear on the skeptic’s first-person reflection on his own beliefs. Consider my belief that Caesar died in 44 B.C. To believe this is, at a minimum, for me to be committed to its truth. We can go further and say that my belief that Caesar died in 44 B.C. is my commitment to the fact that Caesar died at that time on the basis of relevant considerations (e.g. documentary and artifactual evidence). It is to take the proposition “Caesar died in 44 B.C.” to represent the way things are, and so, being prepared to act on the basis of the truth of this proposition. It is to hold that the weight of the evidence speaks in its favor, or at least not against it, and that, to some extent, one be prepared to rationally defend the claim if it is challenged. We can sum up this aspect of belief by saying that belief is a reasonsensitive commitment. It is because beliefs are sensitive to reasons that we can be held responsible for them. The relevant sense of responsibility crucially depends on our being epistemic agents, a form of rational agency that is reflected in cognitive activities such as weighing evidence, replying to criticism, forming conclusions and, in general, making up our minds what to think. We are open to criticism for the beliefs we hold and the way in which they are or are not supported by good reasons. We can be blamed for holding beliefs too firmly in the face of reasonable doubts such as evidence of their falsity or insufficient evidence of their truth. Sometimes we speak of individual beliefs as fanatical or dogmatic; on other occasions, we characterize believers in this way, meaning that they have a tendency to acquire fanatical or dogmatic beliefs. We can also be criticized for acquiring or giving up beliefs too readily, e.g. accepting another’s opinion more because of their personal charm than any plausibility their opinions might have. The epistemic sin here is that of gullibility. And we can be too cautious or hesitant in our believing, or a fence-sitter unable to make up one’s mind. And no doubt we can be criticized, qua believers, on other grounds as well. These criticisms of our beliefs and of our habits of believing presuppose that there is a stance from which one can avow or endorse one’s beliefs and decide what to believe in light of the available reasons. This is what I am calling the deliberative stance. To be a believer — in the full sense in which it applies to human beings — is to have the capacity to adopt the deliberative stance toward one’s own beliefs. Of course, I do not mean to imply that all beliefs are products of deliberation. Far from it. But I do want to suggest that it is on the basis of this form of self-awareness that one identifies with one’s beliefs, regarding them as expressions of one’s own commitment to the truth of their contents. This stance

104

David Macarthur

is essentially first-personal since it is not a stance I can adopt toward another’s beliefs. From this perspective, my beliefs are normative attitudes that I identify with. They are what constitute my world-view, how I take things to be, where this is understood to involve a sensitivity to the considerations in favor of them. They express my sense of what is the case in light of the available reasons for and against. In general, then, it is from the deliberative stance that one comes to be aware of doxastic commitments for which one is responsible or criticizable. However, it is important to see that in addition to the deliberative stance, we can also adopt a naturalistic stance toward our own beliefs, which is, to some extent, in tension with it. By the naturalistic stance I intend, in the first instance, a third-person stance toward oneself;10 however, in the context of skepticism, it is better thought of as a matter of treating first-person access to one’s beliefs on the model of third-person awareness. From the naturalistic stance, the goings-on in our own minds are treated as mere natural happenings akin to any other natural occurrence in the world, say, the movements of leaves in a nearby tree. Up to a point it is possible to think of facts about one’s own mental life as no different in kind than any natural facts in the natural world except that one is in a special position to witness them. Although first-personal, this way of thinking about introspection is obviously modeled on the epistemology of observation, a third-personal stance we enjoy toward other people or physical things in the environment. From this perspective the relation one takes to one’s own beliefs is supposed to be not importantly different from the relation that one stands in to another’s beliefs, or that another stands in toward one’s own. It is not an essentially first-person stance, unlike the deliberative stance from which one determines what one to believe, a relation one does not stand in to anyone else’s beliefs. It is from the naturalistic stance that there is a tendency to objectify beliefs as causally produced inner objects of a kind of inner perception.11 Hume, of course, falls victim to precisely this tendency in likening the mind to a theatre and in speaking of mental states as if they were characters in a play that one alone is in a position to watch (1978, p. 253): The mind is a kind of theatre, where several perceptions successively make their appearance; pass, re-pass, glide away, and mingle in an infinite variety of postures and situations. 10

Treating one’s own mental life from the third-person stance involves, e.g. ascribing beliefs to oneself on the basis of one’s appearance in a mirror, or under a description one does not recognize is about oneself, or on the basis of Freudian psychoanalytic or empirical psychological theory. 11 cf. Armstrong (1968).

Skepticism, Self-Knowledge, and Responsibility

105

Hume treats beliefs and self-knowledge of them on the model of the perception of physical objects. A third-person phenomenon is transposed into a mental interior. The important point is that when one adopts a naturalistic stance toward oneself one sees one’s own beliefs in a detached way as mere states of oneself to which one is a private spectator, one loses the sense of them as having any normative significance, as states of oneself with which one identifies. The naturalistic stance involves a withdrawal from one’s sense of being committed to the truth of the content of one’s beliefs. It is almost as if we are capable of reporting what our beliefs are without taking any stand on what they are about. At the limit that, of course, is incoherent. I cannot report that I believe that I am wearing a blue shirt without at the same time being committed to the truth that this is so. Ordinarily, if I consider the question whether I believe that I am wearing a blue shirt then I do not turn my attention inward and consider my state of mind; rather, I look outward at the world — in this case, at my shirt — or evidence pertaining to it. This is what philosophers call the “transparency” of belief.12 I cannot identify my belief that I am wearing a blue shirt independently of taking a stand on this worldly fact about my clothes. This reveals the priority of the deliberative relative to the naturalistic stance, namely, that one would not so much as have a belief to adopt any stance toward, unless one had already committed oneself to it, that is, had already made up one’s mind about the matter. However, from the third-personal or naturalistic stance, beliefs are attributed to me on the basis of evidence, so that identifying my belief that I am wearing a blue shirt may come to seem something independent of whether I am wearing a blue shirt or not.13 From this perspective, evidence for psychological states of a subject who happens to be me is one thing and objective facts about my clothing is another. So in adopting the naturalistic stance on my own belief I treat the belief in isolation from the state of the world on which its truth depends. It should be clear that although both stances toward one’s beliefs are available, the naturalistic stance of detachment is in tension with the deliberative stance of commitment and that adopting the naturalistic stance toward one’s beliefs has its limits. For to believe that p at all requires that one is committed to the truth of

12

Edgley (1969). The immediacy of ordinary self-knowledge, the fact that I do not require any evidence about myself to know, say, that I believe that it is raining, depends upon the fact that I determine what to believe about the proposition “It is raining” on the basis of rational considerations pertaining to the weather outside and not about the subject of belief who happens to be myself (cf. Moran, 2001). 13 Nevertheless, despite the difference in the basis and authority between the first-person case of my knowing what I believe and the third-person case of your knowing what I believe, it is the very same state of belief that I express and you ascribe to me. I can express it by my asserting, say, “I believe that the shirt is blue” and you can report it by your ascribing “David believes that the shirt is blue”.

106

David Macarthur

p, a commitment that is typically expressed in speech and action. But it is important to see that in so far as one does adopt the naturalistic perspective it will seem, to some extent at least, as if one’s beliefs no longer stand to oneself as commitments to the truth of something beyond themselves. As natural objects of our introspective gaze, beliefs become disengaged from the believer and from their claim to truth and reasonableness. They come to be assimilated to mere states of oneself akin to, say, having indigestion or a pain in one’s foot.14 Such an understanding of one’s own beliefs is unstable, however. The naturalistic stance could not be the only, nor the primary, stance we adopt toward our beliefs, because it is the deliberative stance that is essentially tied up with being a believer in the first place. It is essential to our conception of ourselves as rational agents that we can regard at least some of our beliefs as our own and as expressive of our rational freedom, our capacity to make up our minds about what to think on the basis of our sense of the relevant considerations. That is not to say that we must see all, or even most, of our beliefs as the products of deliberation. It is rather to see that there are limits in the adoption of a naturalistic stance on one’s own beliefs: to regard a belief of one’s own as nothing more than a natural object is to lose the sense of being committed to the truth of its content and hence to its being a belief at all.

The Two Stances and the Avoidance of Responsibility I want to suggest that it is in virtue of shifting between these two perspectives toward one’s own beliefs in first-person reflection that the skeptical considerations lead to a state of skepticism. In the imaginative exercise of engaging with the skeptic, at some point we find ourselves being asked to defend beliefs for which we have no adequate reasons. And then the question is why we ought to remain committed to beliefs that we cannot provide reasons for. Of course, it is characteristic of skeptical arguments that the commitments for which we seem unable to provide reasons are basic in our system of belief. Examples include: ●

●

●

14

that perception can provide at least prima facie reason for beliefs about the external world, that behavior can provide at least prima facie reason for beliefs about other minds, that memory can provide at least prima facie reason for beliefs about the past.

This feature of the naturalistic stance helps explain Hume’s remarkable claim “that belief is nothing but a peculiar feeling” (1978, p. 624).

Skepticism, Self-Knowledge, and Responsibility

107

Our ordinary conception of experience, action, and memory, simply presupposes these basic epistemic commitments.15 That is why we are unable to provide any non-question-begging, or independent, reasons for them.16 In skeptical reasoning the pursuit of justifying reasons from the deliberative stance reaches a dead-end at basic epistemic beliefs. The skeptic’s insight is that there is no independent reason to accept these beliefs. Now, on its own, this discovery need not lead us to a skeptical crisis. Surely we could go on accepting these commitments even in the knowledge that we have no satisfying reasons for them. Hume says that is, in fact, what we do once we leave the study.17 But what Hume does not explain is why this lack of reasons is unstable, always liable to become a crisis in which our commitments are threatened or lost. I want to suggest that we can answer this question by considering our capacity to vacillate between two stances to these (or any) beliefs in first-person reflection. In the skeptical search for justification, the so-called “infinite regress of justification” is a myth. As we all know, reasons inevitably come to an end. Since we can find no non-question-begging reason to support our basic epistemic beliefs from the deliberative stance we tend to shift to the naturalistic stance to discover their causal origin. But this is precisely a stance from which we lose any sense of the normative force of these beliefs, of their status as commitments of ours to things being thus and so that we avow or endorse. It is this movement of thought — this ambivalence of self-knowledge — that turns our lack of reasons into a serious problem. It is as if only reasons could now recover the missing commitment but more reasons would not solve the problem. The naturalistic perspective leads to a detachment that — given the centrality of the beliefs in question — strikes us as a calamity.18 As Hume reports, “the skeptical doubt … always increases, the farther we carry our reflections, whether in opposition or conformity to it” (1978, p. 218). Why, then, do we not stop at the realization that our epistemic practices presuppose basic epistemic presuppositions that we can provide no reason for or, at

15

Plausibly, they are among the beliefs that Wittgenstein (1969) calls “hinge propositions”. If we retreat to the position from which experiences are subjectivized or behavior is robbed of any connection with expression then it is no wonder that we sense that an epistemic gap has opened up that we cannot bridge. We cannot get back from subjective experience to the world or from mere behavior to other minds. 17 At this point Hume appeals to human nature as a non-rational source for our beliefs that also explains their persistence in the face of skeptical challenge. Another direction for an explanation might appeal to rational faith, although that seems to rename the problem rather than solve it. 18 It is important to distinguish the “bracketing” or “stepping back” from one’s commitments that is a feature of deliberation from the detachment (exploited by skepticism) that is associated with the naturalistic stance. The former bracketing is fully compatible with continued commitment whereas the latter detachment is in tension with it. 16

108

David Macarthur

least, no reason that does not already presuppose them? Or, rather, why do we tend to adopt the naturalistic stance of detachment from which retaining our commitments seems to depend upon reasons that we cannot find? Here, borrowing a suggestion from Wittgenstein, one might say that it is because we are inveterate explainers. Where we run out of explanations in one direction we’ll go searching for them in another. Such an account may well form part of the skeptic’s selfunderstanding, as if the skeptic is driven to his conclusion by scrupulous attention to the demands of theoretical inquiry. But I think there is a deeper motivation: the perennial human tendency to avoid or disown responsibility. In order to go some way toward justifying this claim, I want to re-conceive skepticism in terms of the notion of epistemic responsibility. Let me start by recalling what sense of justification is relevant to skepticism. In one sense, “justification” refers to the reliability of the process whereby one came to hold the belief. To be justified in this sense is for it to be objectively likely that one’s belief is true (relative to some standard of objective likelihood). The second sense of the term “justification” refers to one’s epistemic entitlement to one’s belief, the question whether one’s way of forming and continuing to hold a belief is beyond reproach. Being justified on this understanding is for one to be epistemically blameless in believing as one does. It is this second sense, justification as entitlement that is relevant to skepticism. The skeptic, both ancient and modern, asks whether we are entitled to our beliefs or, in other words, whether we have an epistemic right to believe them. Ancient skeptics, for example, attempted to cultivate the skills needed to show that for any given proposition p, the reasons one has in favor of p are no better or worse than the reasons one has in favor of not-p, so that we come to see that we are not entitled to the belief that p.19 Consequently, we suspend judgment about the question whether p. Ancient skepticism was a practical philosophy teaching not doctrine but a way of life: a life of tranquility, free from argument and dispute. Ancient skeptics report (and sometimes, inconsistently, dogmatically advertise) that suspension of belief leads to tranquility — something that seems hard to accept until one realizes that this suspension is not fully global but only concerns reason-based belief (“dogmas”). Beliefs “forced” upon one by one’s senses, by sensations, or by one’s upbringing (ethical, legal, professional) are not subject to skeptical suspension (see Sextus, 1994). Similarly, what is at stake in modern skepticism is entitlement to believe. On a superficial understanding, the modern skeptic is purely theoretically driven, a 19

One formal strategy they employed was to appeal to the predicament known as Agrippa’s Trilemma in which the infinite regress of reasons for reasons is only stopped by viciously circular reasoning or an ad hoc assumption.

Skepticism, Self-Knowledge, and Responsibility

109

kind of disappointed rationalist. Like the rationalist, the skeptic thinks that in order to earn one’s entitlement to certain basic epistemic beliefs that seem to be presupposed by one’s system of belief one must have appropriate reasons. Unlike the rationalist, however, the skeptic correctly sees that we do not have any independent reasons of the sort required. But the skeptical demonstration that we lack reasons for such beliefs does not show that we are not entitled to believe them. For why should we accept the rationalist presupposition that earning entitlement is always a matter of providing sufficient reason? Ordinarily, earning entitlement to a belief is sometimes a matter of having an appropriate reason and sometimes not. In our epistemic practices there is a class of basic epistemic entitlements that one is entitled to simply by virtue of one’s status as an epistemic agent, that is, by being a competent member of a community of reason-givers and -takers.20 One is simply entitled without reason to believe that, say, perception is a source of prima facie reasons to believe things about the external world, or to believe that behavior is a source of prima facie reasons to believe things about other minds. These entitlements are part and parcel of being a member of a community of rational believers. It is important to see that a condition of entitlement to our beliefs is that we take responsibility for them as commitments of ours. The relevant notion of epistemic responsibility is not a matter of free choice and implies no voluntarism about belief.21 What it requires is that one is properly criticizable for one’s beliefs and, to some extent, for being taken to know when and how to defend beliefs against reasonable criticism. If not having reasons for basic epistemic commitments is not, by itself, enough to lose one’s sense of commitment, what explains this phenomenon? What one requires, I suggest, is a shift to the naturalistic stance on one’s beliefs, a movement motivated by the desire to avoid one’s epistemic responsibilities. Skepticism is a refusal to accept the epistemic responsibilities that inevitably come with being a believer at all. The skeptical shift from the deliberative to the naturalistic stance where reasons have come to an end — which involves both a loss of commitment to belief and a loss of the responsibility that goes with it — is not required by a scrupulous attention to the demands of epistemic responsibility but is, on the contrary, an avoidance of epistemic responsibility. What is presented as a surprising discovery producing a state of melancholy and delirium, can be seen instead as a product of willful intention, a matter of avoiding responsibility for our epistemic situation as finite creatures. Here, too, modern skepticism connects with a dominant theme 20

On a fuller account, one might see these as contextually sensitive default entitlements that are defeasible under certain circumstances. Cf. Williams (2000). 21 The connection between epistemic responsibility and voluntarism is assumed by, e.g., Alston (1988).

110

David Macarthur

of ancient skepticism: as Burnyeat (1983, p. 128) puts it, “the skeptic … is withdrawing to the safety of a position not open to challenge or inquiry”. On the present view, then, skepticism is not simply the result of a demand for reasons as the source of our entitlement to believe, as many seem to think. The suggestion is that where the skeptical demand for reasons ends in beliefs for which we have no non-question-begging reasons and which are apparently crucial to our ordinary justificatory practices, then our capacity to adopt a naturalistic stance of disengagement from such beliefs is, ultimately, responsible for our sense of a skeptical crisis. Why? Because otherwise there would be the option of continuing to believe (or be committed to) what we discover we have insufficient or no reason to believe. Indeed, that attitude seems to be quite compatible with ordinary epistemic responsibility. The naturalistic stance explains the difference between commitment without reason and our sense of commitment lost. It is notorious that the skeptical loss of commitment is inherently unstable. Upon leaving the study and entering once more into one’s practical affairs, the skeptical considerations can strike one as, in Hume’s phrase, “cold, and strain’d, and ridiculous” (1978, p. 269). The skeptical state of suspension is short-lived because no sooner do we leave the study than we find ourselves returning from the suspended judgments of our doubtful frame of mind to the commitments and responsibilities implicit in our practical lives. Shifting between the two kinds of self-knowledge explains both the skeptical sense of commitment lost — which we mistakenly put down to a lack of reasons — and the temporary nature of this malady.

Nagel on Skepticism It is worth comparing this account of skepticism to that of Nagel (1979, Chapter 2, 1986, Chapter V). He, too, shares the idea of skepticism as arising from a shift between two essential and ineliminable epistemic perspectives we can adopt to our epistemic practices. There is a “nebula’s eye view”, or a “view from nowhere”, from which we regard ourselves and our whole system of justification and criticism from an external, detached standpoint. From that perspective our justifications seem arbitrary and cannot be further justified without circularity. But from the everyday perspective of practical engagement we commit ourselves, once more, to our beliefs even in spite of the doubts that we cannot answer from the God’seye view. Nagel thus speaks (1986, p. 88) of the human condition as one fated to a certain kind of double vision: Double vision is the fate of creatures with a glimpse of the view sub specie aeternitatis. When we view ourselves from outside, a

Skepticism, Self-Knowledge, and Responsibility

111

naturalistic picture of how we work seems unavoidable. It is clear that our beliefs arise from certain dispositions and experiences which, so far as we know, don’t guarantee their truth and are compatible with radical error. The trouble is that we can’t fully take on the skepticism that this entails, because we can’t cure our appetite for belief, and we can’t take on this attitude toward our own beliefs whilst we’re having them. Nagel’s view is analogous to that defended in the present paper in so far as it traces skepticism to our tendency to vacillate between two stances toward ourselves where the fundamental question is a first-personal one about what to believe. But Nagel supposes that skepticism arises simply from an awareness that our epistemic practices rest on unarguable presuppositions.22 I have wanted to go beyond this standard explanation — which implausibly assumes that all commitments must be argued for — by claiming that the skeptical conclusion is best understood in terms of the human tendency to disavow responsibility even though it operates under the guise of doing just the opposite, i.e. being epistemically scrupulous, careful, fastidious. Nevertheless, Nagel shows rare insight in seeing that an explanation of the tendency to skepticism is one thing, whereas the final eradication of skepticism is quite another. While we can have the first, there is no hope of the second. The temptation toward skepticism remains even in spite of our attaining a philosophical understanding of it.

Naturalism and Skepticism On a first reading, Nagel appears to suppose that reflection from the naturalistic stance inevitably leads to skepticism. On the contrary, it is important to see that a purely naturalistic inquiry into one’s beliefs does not engender skepticism since, in going straight to this stance, the skeptical question does not arise.23 Skepticism arises from the perspective of rational deliberation on one’s entitlement to believe. Naturalistic inquiry, consistently pursued, simply bypasses skepticism.24 22

Although he does not carefully distinguish the Cartesian and Humean forms of skepticism, I read Nagel (1986, Chapter V) as providing a distinct motivation for Cartesian skepticism, namely, an implicit realism built into our concept of (objective) knowledge. 23 Here it is taken for granted that one has beliefs, hence that one has made up one’s mind on various matters. 24 Quine provides a good example of this naturalistically based avoidance of skepticism. See, e.g. Quine (1981).

112

David Macarthur

But perhaps Nagel is better read as following Hume in holding that the discoveries and conceptions arising from the naturalistic perspective provide materials that play into the hands of the skeptic when he considers the question of entitlement from the deliberative stance. Regarded from the naturalistic stance, beliefs come to be thought of as causally produced objects of introspection to which we seem to bear no commitment.25 Skeptical scenarios exploit this conception by providing alternative causal hypotheses to explain how our beliefs are compatible with global error. So naturalism naturally gives rise to skepticism where this is understood as a matter of shifting back and forth between the two kinds of self-knowledge. What of Hume’s own “naturalistic response” to skepticism according to which there is a non-rational basis for our basic beliefs such as that there is an external world? Strawson (1985) has argued that, although this response does not answer or refute the skeptic, it does show that arguing with the skeptic is idle since we will have the relevant beliefs no matter what.26 A feature of this response is that since basic beliefs arise from, and are sustained by, natural forces quite independently of any reasons that might be brought to bear, we cannot be held responsible for having them. So naturalism implies that we are not responsible for our basic beliefs. Hume’s naturalism is, then, not just something that provides materials that give rise to skepticism, but is itself a form of skepticism. Understanding the skeptical problematic is ultimately a matter of understanding the indefinitely many ways in which we avoid or disown the responsibility that inevitably comes with being a rational agent in the world. The lesson of naturalism is that we cannot treat our own beliefs as nothing more than natural items in the world to which we bear a merely epistemic relation (say, of inner awareness), since that would leave out of account what makes my beliefs mine, something for which I am accountable. Here modern skepticism reconnects with another feature of ancient skepticism, its ethical orientation.

References Alston, W. (1988). The deontological conception of epistemic justification. Philosophical Perspectives, 2, 257–299. 25

That is, our ordinary concept of belief is apt to undergo a subtle distortion in being considered from the naturalistic point of view. Ordinarily we think of beliefs as causally efficacious states of a person that intrinsically involve a commitment to truth. (One might think of the naturalistic conception of mental states as the “first step … in the conjuring trick” that leads to the mind–body problem. See Wittgenstein (1958, section 308).) 26 Cf. Strawson (1985). A “response” to skepticism, as I am using the term, is not to be equated with an “answer” or “refutation” of skepticism.

Skepticism, Self-Knowledge, and Responsibility

113

Armstrong, D. M. (1968). A materialist theory of the mind. London: Routledge & Kegan Paul. Burnyeat, M. (1983). Can the skeptic live his skepticism? In: M. Burnyeat (Ed.), The skeptical tradition (pp. 117–148). Berkeley: University of California Press. Cavell, S. (1979). The claim of reason: Wittgenstein, skepticism, morality, and tragedy. Oxford: Oxford University Press. Conant, J. (2004). Varieties of skepticism. In: D. McManus (Ed.), Wittgenstein and scepticism (pp. 97–136). London: Routledge. Cottingham, J., Stoothoff, R., & Murdoch, D. (Trans). (1985). The philosophical writings of descartes (Vols. I–III). Cambridge: Cambridge University Press. Edgley, R. (1969). Reason in theory and practice. London: Hutchinson. Hume, D. (1978 [1739–1940]). In: P. H. Nidditch (Ed.), A treatise of human nature (2nd ed.). Oxford: Clarendon Press. Hume, D. (1975 [1777]). In: P. H. Nidditch (Ed.), Enquiries concerning human understanding and concerning the principles of morals (3rd ed.). Oxford: Clarendon Press. Macarthur, D. (2003). McDowell, skepticism and the “Veil of Perception”. Australasian Journal of Philosophy, 81, 175–190. Macarthur, D. (2004). Naturalism and skepticism. In: M. De Caro & D. Macarthur (Eds), Naturalism in question (pp. 106–124). Cambridge, MA: Harvard University Press. Moran, R. (2001). Authority and estrangement: An essay on self-knowledge. Princeton: Princeton University Press. Nagel, T. (1979). Mortal questions. Cambridge: Cambridge University Press. Nagel, T. (1986). The view from nowhere. New York: Oxford University Press. Putnam, H. (1998). Strawson and skepticism. In: L. E. Hahn (Ed.), The philosophy of P.F. Strawson (pp. 273–287). Chicago: Open Court. Quine, W. V. (1981). Theories and things, Cambridge, MA: Harvard University Press. Sextus Empiricus (1994). In: J. Barnes & J. Annas (Trans.), Outlines of scepticism. Cambridge: Cambridge University Press. Strawson, P. (1985). Skepticism and naturalism: Some varieties. London: Methuen. Stroud, B. (1984). The significance of philosophical scepticism. Oxford: Clarendon Press. Unger, P. (1975). Ignorance: A case for skepticism. Oxford: Clarendon Press. Williams, M. (1991). Unnatural doubts: Epistemological realism and the basis of scepticism. Oxford: Blackwell. Williams, M. (2000). Dretske on epistemic entitlement. Philosophy and Phenomenological Research 60, 607–611. Wittgenstein, L. (1958). Philosophical investigations (2nd ed.) Oxford: Blackwell. Wittgenstein, L. (1969). On certainty. Oxford: Blackwell.

This page intentionally left blank

114

Chapter 8

A Reasonable Contextualism (or, Austin Reprised) A. B. Dickerson

Epistemological contextualism is the claim that different uses of a sentence ascribing knowledge (say, “S knows P”) can possess different truth-conditions, even if they refer, at the same time, to the same S and the same P. To put this another way, it is the claim that the epistemic standards that a subject, S, has to meet in order to count as knowing a proposition, P, can vary depending upon the context in which the knowledge ascription is made. Contextualism has been a focus of recent discussion, with important papers by Stewart Cohen, Keith DeRose, and David Lewis (among others) exploring its application to a number of epistemological puzzles. Given this contemporary interest, it is worth examining an earlier form of contextualism: the epistemology of J. L. Austin, as presented in papers such as “Other Minds” (1946) and the lectures series published as Sense and Sensibilia (1962). How much in the way of justification does S have to be able to provide in order to count as knowing that a is a goldfinch? Austin’s answer is Enough. And “Enough is enough: It doesn’t mean everything. Enough means enough to show that (within reason, and for present intents and purposes) it ‘can’t’ be anything else … It does not mean, for example, enough to show that it isn’t a stuffed goldfinch” — an example of a demand which would be “silly (outrageous)” in the imagined circumstances (Austin, 1946, p. 84). In other words, in order to know P, S does not have to be able to meet every possible epistemic demand that could be made on her (i.e. resolve every possible doubt; defeat every possible defeater); she has only to meet Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

115

116

A. B. Dickerson

those demands that are “within reason … for present intents and purposes”. Austin is thus proposing a contextualism in which an utterance of “S knows P” is true only if S is able to meet the reasonable epistemic demands (with regard to P) — where what counts as a reasonable demand (with regard to P) is determined by the context of ascription.1 The purpose of this paper is to explore Austin’s contextualism in a little more detail, and to defend it against some of the more obvious objections it might seem to invite. The first section discusses his account of “knows” and, in particular, the appeal to “reasonableness” that it makes. The second section sketches how Austin’s epistemology relates to his more general doctrines about language and context, and argues that if these doctrines are sound, then a space is cleared for rehabilitating an old-fashioned charge against scepticism — that sceptical arguments are not false, but nonsensical.

1. Austin’s Modest Proposal Given the presence of Austin in this discussion, it seems fitting to begin with an appeal to linguistic usage, that is, “what we should say when” (Austin, 1956, p. 181). Consider the following pair of contrasting cases in which knowledge is attributed: (A) I teach high-school science and have given my students an exam on the fundamental laws of physics. After reading through a particular exam script, which demonstrates an excellent grasp of the topic, I remark to another teacher: “At least I’m getting through to some of them — Susan knows that the second law of thermodynamics holds”. (B) Professor Evans is investigating possible refutations of the second law of thermodynamics. At a meeting of eminent scientists I remark: “Prof. Evans, there really is no point in you continuing your research into the Fluctuation Theorem. Just talk to Susan — she knows that the second law of thermodynamics holds”. It seems clear that, ceteris paribus, my remark in (A) is “in order”, while my remark in (B) is “out of order”. By this I mean that my remark in (A) seems a natural or reasonable thing to say: it would (should) be accepted without murmur by an interlocutor, and it would (should) be relied upon when deliberating upon certain future courses of action (such as whether to progress Susan onto the next 1

Travis (1989, Chapter 4) and Kaplan (2006) contain related discussions of Austin’s epistemological insights.

A Reasonable Contextualism

117

stage in the science curriculum, etc.). On the other hand, my remark in (B) is just silly (outrageous), and would (should) have no impact whatsoever on the future research plans of Professor Evans. Contrasting cases like this are the bread and butter of contextualist epistemologies, for they suggest that the context of attribution can affect what we are prepared to count as knowledge. By hypothesis, the only thing that has changed between (A) and (B) is the context of attribution (i.e. the context in which “S knows P” is uttered). For in both cases the subject (Susan) is in the same “epistemic state” (i.e. possesses the same information, has the same recognitional and cognitive skills, stands in the same causal relations to the world, etc.), and it is the same proposition (namely, that the second law of thermodynamics holds) under consideration. Yet in (A) we are happy to count Susan as knowing — or, at least, happy to say of her that she knows — that the second law holds, while in (B), we would not count Susan as knowing that the second law holds (even if she got 87% in her exam and the law, as a matter of fact, holds). Of course — and I mention this, just to put it to one side — even if this is accepted as an illustration of appropriate usage, it is another, and much more contentious, question just what weight should be given to such data in our epistemological theorizing. However, one of the key attractions of a contextualist epistemology is that it promises to allow us to take these linguistic intuitions at face value — and thus, or so it might be thought, to do justice to them. If it is accepted that the linguistic data really do show such contextual variation in what we (“ordinarily”, perhaps) are prepared to call knowledge, a primary task of a contextualist epistemology is to provide a general account of that data. In what follows, it will be useful to have a schematic outline of any such account. In order to provide this I will use, as a rough and ready tool, the notion of an epistemic challenge. In a putative case of knowing P, S takes the world to be a certain way, W, in which P obtains. The fundamental sort of epistemic challenge to S’s putative knowledge of P is an alternative specification of a way the world might be, W*, in which P does not obtain. In order to meet this challenge, S must be in a position to discern whether she is in W or instead in W*. So, for example, in her current “epistemic state”, S might be able to tell W1 (in which a is a goldfinch) from W2 (in which a is a woodpecker), but not be able to tell W1 from W3 (in which a is a stuffed goldfinch). Given this, the general form any contextualist epistemology must take can be outlined as follows. As noted above, the defining idea of contextualism is that, even with a fixed subject S and proposition P, the truth-conditions of “S knows P” can vary across different contexts of use. And this is because, in the terms defined above, the range of epistemic challenges to knowing P that S must be able to meet in order to count as knowing P (in that context) vary from one context of

118

A. B. Dickerson

ascription to another. In other words, if we imagine the field of all possible epistemic challenges to S’s putative knowledge of P, then that field must be partitioned relative to the context. That is, for any given context in which “S knows P” is truth-apt, there will be, on the one hand, those challenges that S must be able to meet for “S knows P” to be true, and on the other hand, those challenges that S need not be able to meet for “S knows P” to be true. Following recent tradition, we can call the former set the relevant challenges, and the latter set the irrelevant challenges. Using this terminology, our generic contextualist epistemology is committed to an account of the truth-conditions of knowledge ascriptions (or, at least, of a large and important family of such ascriptions) that looks something like the following: “S knows P” is true iff S believes P, P is true, and S is able to meet all the relevant epistemic challenges to knowing P — in which, of course, what counts as the relevant epistemic challenges depends upon the context in which “S knows P” is used. Not all parts of this statement of the truth-conditions of “knows” are obligatory. Some forms of contextualism may wish to omit the constraint that S believes P; others may wish to omit the constraint that P be true. However, all contextualist epistemologies will hold that the core of an account of “knows” is given by the condition that S must be able to meet all the epistemic challenges that are relevant (in the context of ascription), and need only meet those challenges. What is then required from a contextualism is an account (of some sort) of how the partition of possible challenges is determined, and what kinds of contextual factors affect that partition. So, for example, a contextualist might hold that the relevant epistemic challenges are those that involve a specification of possible worlds that are less than a given “distance” from the actual world, or those that specify events the likelihood of which is higher than a given level of probability, and where the “distance” from the actual world, or the level of probability, is fixed by various contextual factors (such as the intentions of the speaker, what is “at stake” in the discussion, and so forth). As was mentioned in the introduction to this paper, Austin’s own form of contextualism involves the claim that the relevant epistemic challenges are those that are “within reason … for present intents and purposes”. Austin is thus giving us an account in which the partition between the relevant and the irrelevant epistemic challenges is fixed by what is reasonable (in the context). In other words, we have the following account of the truth-conditions of knowledge ascriptions: “S knows P” is true iff S believes P, P is true, and S is able to meet all the reasonable epistemic challenges to knowing P.

A Reasonable Contextualism

119

In this case, what counts as a reasonable epistemic challenge depends upon the context in which “S knows P” is used. That is, the epistemic challenges (to knowing P) that S must be able to meet (in order to count as knowing P) are those that, given all the relevant facts, it would be reasonable to demand that S be able to meet. No doubt the first thing that will be asked about this account of “knows” is: what determines what is reasonable? Given the philosophical context in which it is asked, this question is a request for a general, informative account of the epistemically reasonable. By “general” I mean an account that ranges over all (or, at least, a wide and central variety of ) possible contexts of knowledge ascription; by “informative” I mean an account that does more than move in a tight analytical circle (in which, e.g. the reasonable is what an informed, reasonable person would hold to be the case). However, this request for a general, informative account needs to be rejected. There are no realistic prospects for the attempt to provide a general criterion of the epistemically reasonable. Most fundamentally, this is because our judgments of the reasonable are always defeasible. In other words, our best judgment of what counts as reasonable (in the given circumstances) always remains open to the world, in that it may (quite rightly) be overturned by the future course of our experience (e.g. by our uncovering new evidence, developing new accounts of how the world works, and so forth). So, no matter how detailed a specification of the context we possess, it is always possible for additional contextual information to be added, that would modify our previous best judgement as to whether a given epistemic challenge is relevant or irrelevant (to S’s putative knowledge of P). With the right kind of “scene-setting”, an epistemic challenge could shift from being patently silly (outrageous) to being as obvious as the demand that S be able to tell “a hawk from a handsaw”. Indeed, as Fogelin (1994) points out, this defeasibility is precisely what lies behind our open-ended capacity to generate counterexamples to putative analyses of knowledge — regardless of how ingeniously complex those analyses are. For it entails that, no matter what general criteria of epistemic rationality are proposed, a version of the “open question” argument is always applicable (“X meets your criteria, but is it reasonable?”). With its appeal to reasonableness, Austin’s epistemology thus rejects the reductionist ambitions implicit in many contemporary versions of contextualism. Cohen (1999, p. 61), for example, thinks it virtually a truism that “[t]he standards [for knowing] are determined by some complicated function of speaker intentions, listener expectations, presuppositions of the conversation, salience relations, etc.” (my emphasis). And David Lewis (1996) has proposed at least a general sketch of what such a “function” might look like. However, if Austin’s account of “knows” is anything like the truth, then this demand for a general theory that will take a range of certain facts (specified without appeal to what is “within reason …

120

A. B. Dickerson

for present intents and purposes”) as input, and which will output a detailed account of the truth-conditions of particular knowledge attributions, is destined to be disappointed. If there were such a general function (applying across multiple contexts), then the correct use of knowledge ascriptions would not involve openended judgments of epistemic reasonableness, but would be a rule-governed response to a fixed set of contextual variables. However, if Austin is correct, and judgments of reasonableness are always defeasible, then, with sufficient ingenuity, counterexamples to any proposed function for “knows” will always be able to be constructed. It may now be felt that, if these considerations are correct, then Austin’s contextualism can amount to no more than a piece of “occult ad hoccery”,2 or perhaps worse. For there is a natural enough line of thought that goes somewhat as follows. If there is no general, informative account of reasonableness to be had, then our judgments of epistemic relevance float free from any sort of objective constraint — in that there is nothing about the nature of the world that (genuinely) underpins or justifies them. So, without some sort of systematic account of how epistemic challenges are partitioned into the relevant and the irrelevant, Austin’s epistemology is just another doctrine of the “inner light”, that leaves us with nothing but a naked clash of epistemic dogmatisms masquerading as “linguistic intuitions”. (A: “My intuition tells me that that challenge is silly (outrageous) — I won’t dignify it with an answer!” B: “Well, my intuition says it isn’t silly at all — so you do have to answer it before I’ll count you as a knower!” A and B then fall to hacking at each other with broadswords, before exiting stage left.) However, it would be fallacious to move from the absence of a general, informative account of the epistemically reasonable to the claim that, therefore, our judgements of reasonableness lack any genuine objectivity. This thought — that such anti-reductionism leaves us with nothing but mere polemic — is, in large part, engendered by the necessarily abstract fashion in which philosophy must approach the subject. The point that there is no non-circular way of specifying in general what is “within reason … for present intents and purposes”, does not entail that, in any particular case, we are left with nothing but appeals to “intuition”. In any concrete context in which knowledge ascriptions occur, competent actors will, usually enough, find themselves equipped with a whole mass of detailed ways of rationally considering and negotiating what is epistemically reasonable: what epistemic challenges ought to be considered by a putative knower, where the onus of showing a challenge reasonable or unreasonable should lie, and what sorts of

2

To adapt some remarks about contextualism made by Sosa (1986, p. 585).

A Reasonable Contextualism

121

considerations could be raised to show that a given challenge is reasonable or unreasonable. In any such particular context, the partition of possible epistemic challenges into the relevant (or reasonable) and the irrelevant (or unreasonable) will not be grounded upon bare “intuition” — which really would be “occult ad hoccery” — but upon various good reasons specific to the context (in other words, upon this and this and this). Hence, although the request for a general, informative account of it must be rejected, this does not mean that the operation of epistemic reasonableness is a dark mystery. On the contrary, it is perfectly ordinary. It consists in the possession and use, by competent actors, of various particular methods, heuristics, “knacks”, criteria, rules-of-thumb, practices of reasoning, and the like. In “ordinary” contexts of knowledge ascription, these tell us what is enough (to count as knowing something). They are implicit in the varied forms of training, education, and enculturation we receive; they are also laid down explicitly in such things as risk assessment guidelines, safety procedures, building and engineering codes, quality control methods, contamination tests, standards legislation, and so on and so forth. And, of course, this great detailed mass of ways of specifying the epistemically reasonable is in a constant process of criticism, evaluation, and change. In sum, in ordinary circumstances, we partition the possible epistemic challenges (to S’s putative knowledge of P) by applying the various specific criteria of epistemic reasonableness with which our culture, training, and institutions have equipped us. With this talk of “our criteria”, I might seem to be saying that Austin’s contextualism is a form of relativism. For it may now sound as if the proposed account of knowledge ascriptions is something like this: “S knows P” is true iff S believes P, P is true, and S is able to counter all the epistemic challenges to knowing P that my community considers to be reasonable. However, this is to slide fallaciously from the truism that our present criteria (for epistemic reasonableness) are our only handle, as it were, on what is epistemically reasonable, to the relativistic claim that what is epistemically reasonable reduces to, or is defined by, those criteria. This latter claim is false, for the simple reason that our (or, our community’s) present criteria for epistemic reasonableness may turn out to be just wrong — that, in fact, our criteria partition the field of possible epistemic challenges incorrectly; that, in fact, certain challenges we (all) had considered to be silly (outrageous) are not. My point, in other words, is a repetition of that made by Putnam (1983, p. 234), namely that “Reason is … both immanent (not to be found outside of concrete language games and institutions) and

122

A. B. Dickerson

transcendent (a regulative idea that we use to criticize the conduct of all activities and institutions)”. Austin’s contextualism is thus no relativism. It is certainly the case that, given the open-ended defeasibility of judgments of reasonableness, none of my knowledge ascriptions can ever be insulated from the course of our future experience. Suppose I say “S knows P” in context C; S believes P; P is true; and S is able to meet all the epistemic challenges we consider to be reasonable in C — accordingly, we judge that what I said is true. If we then find out more about the world (the extraordinary prevalence of stuffed goldfinches, cunningly painted mules, or barn façades in the local area, for example), and, quite rightly, our conception of what was a reasonable challenge in C undergoes change, then, in such a case, what I said in C might well turn out to have been false all along. However, if our criteria of the epistemically reasonable are correct then what I said is true. In other words, although it is always conceivable that we are wrong about what is epistemically reasonable, this is quite compatible with our knowing — so long as, in fact, we are not wrong. An additional factor that may encourage the thought that Austin’s contextualism must collapse into a form of relativism, is the fact that it (like all contextualist views) entails the failure of any simple disquotational schema for “knows”. This can be explained as follows. The basic thesis of contextualism is that an utterance of “S knows P” can possess different truth-conditions in different contexts of use. Let C1 be a context in which S counts as knowing P, and C2 a context in which S does not count as knowing P. A is in C1 and says “S knows P”. Suppose I am in C2 and reason roughly as follows: 1. A said “S knows P”, so A said that S knows P; 2. What A said is true (by hypothesis); 3. So, it is true that S knows P. But, of course, this contradicts the hypothesis that, in C2 (in which I perform this reasoning), S does not count as knowing P. If the general principles behind this argument are sound, we possess a means of “moving” a knowledge ascription from one context to any another, while retaining its truth value. And, if that is the case, then contextualism must be false.3 From a contextualist’s perspective, the key manoeuvre in this argument that needs to be rejected is the application in (1) of a general disquotational schema

3

I have heard this sort of argument raised in discussion on a number of occasions. I present it here in a fairly informal way; for a fuller and more rigorous version see Hawthorne (2004, pp. 98–111).

A Reasonable Contextualism

123

for “knows”. This licenses the move from the mention of a knowledge-ascribing sentence to its use in an oratio obliqua construction. However, if contextualism is true then such a general principle of disquotation cannot be correct. To use “… knows P” in C2 is to say that the subject has met a higher epistemic standard than she would be said to have met by a use of “… knows P” in C1. Generalizing, this means that in order to report correctly what was said by a given use of “S knows P” it may be necessary to paraphrase that utterance without using the construction “… knows P”. For example, one may need to say things such as “When A said ‘Susan knew’ he meant that she knew enough to proceed to the next level of the curriculum”, or “When A said ‘Samantha knew’ he meant that she’d checked it as carefully as could be expected in the circumstances”. Now, it may be felt that if such things are allowed as correct paraphrases (in C*) of a given “S knows P” (said in C), then they represent what was really said (elliptically) by that “S knows P” — in which case it was not a genuine knowledge ascription at all. However, if Austin is right, to have used the paraphrase from C* (e.g. “Samantha has checked things as carefully as could be expected”) in C, would not necessarily have done the same job as saying “Samantha knows P” — for that paraphrase would not have said (as “Samantha knows P” does), that Samantha was able to meet all the reasonable epistemic challenges to knowing P. In general, if contextualism is correct, then what counts as a correct report in oratio obliqua of a given use of “S knows P” will itself be context-dependent, and thus there is no general disquotational schema for “knows”. At this point, it may be felt that Austin’s contextualism has so far been characterized largely in terms of what it is not (a reductionism; a relativism), and it might be asked whether it leaves room for any further positive characterisation of the semantics of knowledge ascriptions. Although I have rejected the possibility of a general, informative account of epistemic reasonableness, this does not mean that nothing further can be said about the use of “knows” in general. In particular, an important part of any fuller discussion is an account of what knowledge ascriptions are used to do in “ordinary” (i.e. non-epistemological) discourse — in other words, of their pragmatics. According to some views of language, the pragmatics of knowledge ascriptions will have little to do with determining their semantics. But if Austin’s contextualism is correct, pragmatic features will in fact play an important role in fixing the truth-conditions of knowledge ascriptions. This is because the function a knowledge ascription is performing within a context will contribute substantially to determining what is epistemically reasonable in that context. Surprisingly little work has been done on the pragmatics of knowledge ascriptions, and I do not have the space here to give it anything like the treatment it deserves. In “Other Minds”, Austin himself makes only a few suggestive

124

A. B. Dickerson

remarks about the special case of first-person ascriptions (in which ascriber and subject are one and the same). However, I would suggest that one key place in which knowledge ascriptions occur is in the context of practical reasoning.4 It is clear that any such process of reasoning cannot be indefinitely flexible or open-ended; it must occur within a framework, or against the background, of the “taken-for-granted”. Knowledge ascriptions play an important role in fixing this framework. In ascribing knowledge of P to S, I not only characterize S’s epistemic state, I also endorse the move of P from the category of “contingencies-that-oughtto-be-taken-into-account” to the category of the “ought-to-be-taken-for-granted” (for present purposes of action, deliberation, inquiry, policy formation, and so forth). To accept such a knowledge ascription is to accept that it would be inappropriate (because unreasonable) for us not to rely upon P in the practical reasoning currently at hand (and thus unreasonable, e.g. to plan for the possibility that not-P, to spend time investigating P, and so forth). Given the important role played by knowledge ascriptions in setting the frameworks of practical reasoning, it is unsurprising that we do not (except perhaps in very unusual circumstances) allow locutions such as “S knows from her perspective but not from ours” or “You know by your standards but not by mine”. For if the project of practical reasoning at hand is one that we are undertaking together, then the standard of epistemic reasonableness applied must be publicly shared by us. Austin’s contextualism is thus not the view that epistemic standards are ascriber-dependent. The partition of the possible epistemic challenges into the reasonable and the unreasonable is not settled by the beliefs and intentions of the ascriber, but by the relevant facts of the context of use (which may or may not include certain beliefs and intentions of the ascriber). The close link between knowledge ascriptions and practical reasoning also helps to account for why, outside of theoretical epistemology, debates about what we do and do not know so often circle around notions such as acceptable risk, trust, and responsibility. Austin’s contextualism, by placing the notion of reasonableness at the core of our concept of knowledge, allows us to do justice to these sorts of debates. For example, the heart of the discussion over a question like “Do we know that GM crops are safe?” concerns just what risks are acceptable when we commit ourselves to some course of action. Put in the terminology used above, it concerns just where the borderline of the partition between the reasonable and the unreasonable epistemic challenges lies. And, given the close connections between the notion of reasonableness and ethics, it is unsurprising that such 4 I do not, of course, want to suggest that practical reasoning is the only context in which knowledge ascriptions occur. For example, in contexts in which there are no reasonable challenges to a given claim, then “knows” functions there simply as an emphasis (e.g. “I know I’m in pain”).

A Reasonable Contextualism

125

debates over knowledge often hang on disagreements over what is important in human life, what is worth doing, what is of value, and so forth. One of the great virtues of Austin’s account of “knows” is thus to indicate how these ethical concerns lie at the heart of our notion of knowledge — which in turn suggests that the link between epistemology and ethics perhaps deserves a good deal more attention from epistemologists than it has so far received.

2. Scepticism and Nonsense There is obviously a great deal more to be said — both in further defence of Austin’s contextualism, and in developing an account of the pragmatics of knowledge ascriptions. However, in this final section it is important to discuss something that has so far been left untouched: the relation between Austin’s views and epistemological scepticism. For along with a desire to describe the varying standards we actually use in ordinarily ascribing knowledge, contextualism is also motivated by a desire to offer a satisfying response to scepticism. In what follows, I describe what deserves to be called the “standard account” of scepticism offered by contemporary forms of contextualism; I then contrast that account with Austin’s very different approach. It should be noted that the following discussion attempts only to show how Austin’s contextualism creates the space for an alternative approach to scepticism; it does not attempt to argue that that approach is successful (which, obviously enough, would require a great deal more discussion than there is space for here). It is argued by many contemporary contextualists (such as Cohen, DeRose, and Lewis) that their epistemology allows us to hold that many of our “ordinary” knowledge ascriptions are true, while simultaneously doing justice to the supposed power of sceptical arguments. In outline, the account they offer is very simple. Sceptical arguments occur in — or generate — a context in which the standards for knowledge are set extraordinarily high, and in which therefore most (if not all) knowledge ascriptions are false. Hence, in those contexts in which they occur, sceptical arguments are sound; outside of such “sceptical” contexts, however, the epistemic standards for knowledge revert to normal levels. The standard contextualist mechanism for achieving this result is to include a “salience rule”. Put in my terminology, this is a rule which states that if an epistemic challenge is salient in a context — by being raised in that context, or being considered relevant by the interlocutors — it is thereby a relevant challenge (i.e. one that the subject of the knowledge ascription must be able to meet in order to count as knowing). To take one example, Lewis’s version of this salience mechanism is called the “Rule of Attention”, which states that “No matter how far-fetched a

126

A. B. Dickerson

certain possibility may be, no matter how properly we might have ignored it in some other context, if in this context we are not, in fact, ignoring it but attending to it, then for us now it is a relevant alternative” (Lewis, 1996, p. 559). Such a salience mechanism thus entails that merely raising a possibility in the course of a sceptical argument (e.g. the possibility that I am being deceived by an evil demon, or that I am a brain-in-a-vat) suffices to change the partition of possible epistemic challenges, and make that challenge a relevant one. This in turn means that, in the presence of a vocal sceptic, many more knowledge ascriptions are false than would have been so if that sceptic had been absent. If so, perhaps epistemological contexts are, paradoxically, the very contexts in which we lack knowledge. There is no doubt that this salience mechanism offers what looks like a simple account of the effect of sceptical arguments. On the one hand, sceptical arguments certainly appear to create wide-ranging doubts in the minds of many of those who entertain them — that is, such arguments do seem to destroy knowledge in their immediate vicinity. On the other hand, however, as soon as those arguments are no longer being entertained, for most people the doubts they had raised come to look “cold, and strain’d, and ridiculous” (Hume 1978 [1939], p. 269). In the terms offered by contemporary contextualism, this is explained as follows. First, putting forward a sceptical argument raises the epistemic standards in the context (by making a certain possibility salient and therefore relevant), so that we no longer count as knowing (in that context). Second, when that possibility is no longer salient (and no longer relevant) the epistemic standards revert back to more normal levels, and thus we once again count as knowing. The salience mechanism embedded in many contemporary forms of contextualism may thus appear to make good sense of the phenomena of scepticism, but it does render any such contextualism far less plausible as a description of nonphilosophical linguistic usage. For it is just a fact that, outside of philosophical discussions, we do not treat the mere raising of a possibility as sufficing to make that possibility a relevant epistemic challenge. As Austin points out, in “ordinary” contexts, raising certain challenges would be treated as silly (outrageous).5 If the challenger persists in her challenge — without providing independent evidence in its favour (i.e. not merely by showing that it is compatible with everything else that we know) — then this would be taken as a clear indication that she is irrational in some way. However, if the salience mechanism were correct, then we could never legitimately reject a challenge in this way. Given his focus on describing “ordinary language”, it is therefore unsurprising that Austin’s contextualism

5

The ethnomethodologists have explored this empirically. The locus classicus is Garfinkel (1963).

A Reasonable Contextualism

127

contains nothing equivalent to the salience mechanism. The reasonable epistemic challenges are those that ought to be salient to the ascriber, but the mere fact that a certain challenge is salient is neither necessary nor sufficient for it to be reasonable (and thereby relevant). Given this absence of a salience mechanism in Austin’s account of knowledge ascriptions, it must now be asked what sort of response to philosophical scepticism it allows. I have noted that Austin’s position allows for an unreasonable epistemic challenge to be legitimately rejected as irrelevant to the truth or falsity of a knowledge attribution. However, there is no doubt that such rejection does not seem warranted in the typical contexts in which epistemological scepticism is discussed. In such contexts, rejecting a sceptical challenge outright seems merely dogmatic, and not at all like the justified enforcing of the boundaries of the epistemically reasonable. After all, merely remarking “That’s just silly (outrageous)” is obviously not a philosophically satisfying response to scepticism. As remarked above, there is no space here to develop a full and satisfying Austinian response to scepticism. But by sketching how Austin’s epistemology links to his broader doctrines about the relation between language and context, it is at least possible to open a conceptual space wherein such a satisfying account might be built. Clearing that space will involve rehabilitating what may, at present, appear to be an old-fashioned and out-dated charge: namely, that epistemological scepticism is, in some way, nonsensical. This, of course, was perhaps the characteristic term of criticism used by the so-called “linguistic” or “ordinary language” philosophy of the post-war period. It has, however, dropped almost entirely out of sight in contemporary discussions of scepticism. And with good reason — or so it might be thought — for sceptical arguments certainly do not look nonsensical in any obvious way. It is thus worth asking just what sort of nonsense Austin’s views might allow for. In order to understand Austin’s conception of nonsense, it is obviously crucial to understand his conception of sense — that is, what it is for a bit of language to say something. Austin is a contextualist about a good deal more than just knowledge ascriptions. Indeed, his epistemology is a specific instance of a more general doctrine, namely, that the semantic properties of words are, in large part, fixed by the context in which those words are used. Stated in summary fashion, Austin’s view is that an unambiguous sentence can always be used to express an indefinite variety of different statements, where which particular statement is expressed by a sentence depends upon the circumstances in which the sentence is used. Put another way, this is to say that the meanings of sentences do not suffice to fix the truth-conditions of statements expressed by those sentences (as used in particular contexts). As Austin writes in Sense and Sensibilia (1962, p. 111): “the question of truth and falsehood does not turn only on what a sentence is, nor yet on what

128

A. B. Dickerson

it means, but on, speaking very broadly, the circumstances in which it is uttered”. From the rest of the context, it is clear that this does not mean simply “that the truth of an utterance depends on just two things: what the words as spoken mean, and how the world is arranged” (Davidson, 2001 [1983], p. 139). It means that if we imagine the world fixed a certain way, and take a set of impeccably formed, unambiguous “sentences — as distinct from statements made in particular circumstances” (Austin, 1962, p. 123), then it will still not be determined which sentences are true or false. If we assume otherwise we commit the “pervasive error of neglecting the circumstances in which things are said — of supposing that the words alone can be discussed, in a quite general way” (ibid., p. 118).6 If this contextualist account of language is sound, then a space is cleared for a certain conception of nonsense. For if Austin is right, a particular unambiguous sentence can be used to make a great many different possible statements (i.e. can be used to say things with a great variety of different truth-conditions). Hence, if a sentence is to express a determinate statement in a context, then that context must meet an obvious condition of adequacy. Namely, that context must be such as to fix which statement the sentence is expressing in that context (out of the many possible statements that it could express). To put this another way, the context must be up to the job of fixing the truth-conditions of what was said. Now, suppose it were possible for the context to fail this condition of adequacy. In that case, the sentence uttered would have failed to express a determinate statement, that is, something with determinate truth-conditions. If this were possible, then it would be possible to utter sentences which were composed of meaningful words in grammatical order (indeed, which were — unlike gibberish — perfectly meaningful qua sentences), but which yet failed to say anything determinate (and, ipso facto, failed to say anything either true or false). Such utterances would, in other words, be nonsense. This general line of thought can now be applied to the particular case in hand: Austin’s contextualist epistemology. As noted, according to Austin’s account of language, our words require a certain amount of contextual background, or “stagesetting”, in order to perform their functions. With respect to knowledge ascriptions, the requirement is that the context be such as to partition the field of possible epistemic challenges into the reasonable and the unreasonable. Suppose now that some contexts fail to determine any such partition — in that an agent, no matter how competent and reasonable and well informed (i.e. even if that agent possessed all the relevant facts) would be unable to fix the boundaries of the epistemically

6

The most systematic development of these ideas can be found in Austin (1950, 1953). For an excellent contemporary discussion of Austin’s views, see Travis (1996).

A Reasonable Contextualism

129

reasonable. In such contexts, utterances of sentences like “S knows P” would fail to say anything with determinate truth-conditions, and would thus be nonsensical. Austin’s account of language thus provides the conceptual space for a possible alternative critique of scepticism. Perhaps scepticism occurs in contexts in which the needed “stage-setting” is absent, and thus the apparent knowledge ascriptions that occur in such contexts are really just nonsense. Indeed, if Austin’s account of “knows” is correct, then such contexts are likely to be the norm rather than the exception — in that rather special stage-setting will be required in order to provide determinate truth-conditions for a given use of a given “S knows P”. In any context it will not make sense to say of many of S’s beliefs either that she knows them, or that she does not know them. It will not make sense to say these things, because the context will fail to determine just what the reasonable challenges are that S needs to be able to meet (in order to count as knowing). For example, suppose that — randomly and apropos of nothing at all — I remark “Susan knows that she is called Susan”. In such a case, what I am saying (i.e. its truth-conditions) will depend upon which epistemic challenges I am saying that Susan is in a position to meet. But which ones am I saying she can meet? That she is not badly concussed? That she is not going senile? That she has recently rechecked hospital rolls to ensure that she is not a changeling? If the context does not determine this (and why should it, given the randomness of the remark?) then what I said will have no determinate truthconditions. If this is so, then in such a context it will not be true either to say “Susan knows she is called Susan” or to say “Susan does not know she is called Susan”. Not because “there is no fact of the matter”, or because “the facts are indeterminate”, but because the use of either of these two sentences in such a context would fail to say anything determinate. They would, in other words, be nonsense. We now have the beginnings of a possible Austinian diagnosis of sceptical arguments and their apparent persuasiveness. They appear compelling precisely because they occur in a context in which all the relevant facts fail to fix a partition between the reasonable and the unreasonable epistemic challenges. This, along with the fact that, prima facie, we treat people and the challenges they raise as reasonable, would make it unsurprising that we imagine ourselves to be in a context where all possible epistemic challenges have become reasonable. In fact, however, in such discussions the very concept of epistemic reasonableness has dropped out altogether, and we are thus left with only the appearance of making determinate statements about knowledge. Hence — if this line of thought is correct — what we are doing in such “sceptical” contexts is playing out a mock epistemic dialogue of claim and challenge. It is a mock dialogue, because it is occurring in a context where our use of “knows” and its cognates has been severed from

130

A. B. Dickerson

the very concerns (of rational action, reasonable risk, and practical deliberation) that give those words their life. The above, of course, is no more than a sketch of how an Austinian response to scepticism might proceed, but I hope that it has nonetheless achieved some things. To begin with, the appeal to nonsensicality in the diagnosis of scepticism is worthy of further exploration (it may, for example, be able to shed light on topics such as epistemic closure). More generally, Austin’s contextualism is useful if it focuses our attention on what might be the “pervasive error of neglecting the circumstances in which things are said — of supposing that the words alone can be discussed, in a quite general way” (1962, p. 118). For if contextualism is true, then it means that the very nature of the contexts within which epistemology is done may limit what it can achieve. After all, contextualism is precisely the claim that whether S counts as knowing P is not determined simply by S and her context, but also by the ascriber’s context. If this is correct, then “S knows P” cannot be discussed as a “proposition” in abstracto — said by nobody, in a null context, as it were. Rather, we need to understand what we are saying, in speaking of knowledge. And in doing epistemology, the context of attribution is a philosophical discussion, which, by its very nature, is typically remote from the various practical concerns that may in fact be necessary for determining the truth-conditions of particular knowledge attributions. This in turn suggests that there may be deep limitations on what can be achieved — what can be said — by a “theory of knowledge”, at least if this means an abstract attempt to specify what is common to all knowledge attributions across all contexts.7

References Austin, J. L. (1946). Other minds. In: J. O. Urmson & G. J. Warnock (Eds), Philosophical papers (3rd ed., pp. 76–116). Oxford: Clarendon Press. Austin, J. L. (1950). Truth. In: J. O. Urmson & G. J. Warnock (Eds), Philosophical papers (3rd ed., pp. 117–133). Oxford: Clarendon Press. Austin, J. L. (1953). How to talk — some simple ways. In: J. O. Urmson & G. J. Warnock (Eds), Philosophical papers (3rd ed., pp. 134–153). Oxford: Clarendon Press. Austin, J. L. (1956). A plea for excuses. In: J. O. Urmson & G. J. Warnock (Eds), Philosophical papers (3rd ed., pp. 175–204). Oxford: Clarendon Press. Austin, J. L. (1962). In: G. J. Warnock (Ed.), Sense and sensibilia. Oxford: Clarendon Press. 7

The idea that the context in which epistemology is done may set limits to what it can achieve has been discussed by a number of writers, including Stroud (1989), Hetherington (1992), and Lewis (1996).

A Reasonable Contextualism

131

Austin, J. L. (1979). In: J. O. Urmson & G. J. Warnock (Eds), Philosophical papers (3rd ed.). Oxford: Clarendon Press. Cohen, S. (1999). Contextualism, skepticism, and the structure of reasons. In: J. E. Tomberlin (Ed.), Philosophical perspectives (pp. 57–89). Oxford: Blackwell. Davidson, D. (Ed.) (2001 [1983]). A coherence theory of truth and knowledge. Subjective, intersubjective, objective (pp. 137–153). Oxford: Clarendon Press. Fogelin, R. J. (1994). Pyrrhonian reflections on knowledge and justification. New York: Oxford University Press. Garfinkel, H. (1963). A conception of, and experiments with, “trust” as a condition of stable concerned actions. In: D. J. Harvey (Ed.), Motivation and social interaction (pp. 187– 238). New York: Ronald. Hawthorne, J. (2004). Knowledge and lotteries. Oxford: Clarendon Press. Hetherington, S. (1992). Epistemology’s paradox: Is a theory of knowledge possible? Savage, MD: Rowman & Littlefield. Hume, D. (1978 [1739]). In: L. A. Selby-Bigge & P. H. Nidditch (Eds), A treatise of human nature (2nd ed.). Oxford: Clarendon Press. Kaplan, M. (2006). If you know, you can’t be wrong. In: S. Hetherington (Ed.), Epistemology futures (pp. 180–198). Oxford: Clarendon Press. Lewis, D. (1996). Elusive knowledge. Australasian Journal of Philosophy, 74, 549–567. Putnam, H. (Ed.) (1983). Why reason can’t be naturalized. Realism and reason: Philosophical papers (Vol. 3, pp. 229–247). Cambridge: Cambridge University Press. Sosa, E. (1986). On knowledge and context. The Journal of Philosophy, 83, 584–585. Stroud, B. (1989). Understanding human knowledge in general. In: K. Lehrer & M. Clay (Eds), Knowledge and skepticism (pp. 31–50). Boulder, CO: Westview Press. Travis, C. (1989). The uses of sense: Wittgenstein’s philosophy of language. Oxford University Press. Travis, C. (1996). Meaning’s role in truth. Mind, 105, 451–466.

This page intentionally left blank

132

Chapter 9

Questioning Contextualism Brian Weatherson There is currently a dizzying variety of theories on the market holding that whether an utterance of the form A knows that p is true depends on pragmatic or contextual factors. Even if we allow that pragmatics matters, there are three questions to be answered. First, whose interests matter? Here there are three options: the interests of A matter, the interests of the person making the knowledge ascription matter, or the interests of the person evaluating the ascription matter. Second, which kind of pragmatic factors matter? Broadly speaking, the debate here is about whether practical interests (the stakes involved) or intellectual interests (which propositions are being considered) are most important. Third, how do pragmatic factors matter? Here there is not even a consensus about what the options are. This paper is about the first question. I’m going to present some data from the behaviour of questions about who knows what that show it is not the interests of the person making the knowledge ascription that matters. This means the view normally known as contextualism about knowledge ascriptions is false. Since that term is a little contested, and for some suggests merely the view that someone’s context matters, I’ll introduce three different terms for the three answers to the first question. Consider a token utterance by B of A knows that p. This utterance is being evaluated by C. A semantic pragmatist about knowledge ascriptions says that whether C can correctly evaluate that utterance as true or false depends on some salient party’s context or interests.1 We can produce a quick taxonomy of semantic pragmatist positions by looking at which of the three parties is the salient one. 1

Note that the focus here is on the truth or falsity of the utterance, and not on the truth or falsity of the proposition the utterance expresses. I’m indebted to John MacFarlane for stressing to me the importance of this distinction. Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

133

134 ● ● ●

Brian Weatherson

Type A pragmatism — A’s context or interests matter Type B pragmatism — B’s context or interests matter Type C pragmatism — C’s context or interests matter

Type A pragmatism is defended by Hawthorne (2004) under the name “subjectsensitive invariantism” and Stanley (2005) under the name “interest-relative invariantism”. The theory commonly known as contextualism, as defended by Cohen (1988), DeRose (1995), and Lewis (1996), is a form of Type B pragmatism. On their theory, the semantic value of “knows” depends on the context in which it is uttered, i.e. B’s context. (And the salient feature of B’s context is, broadly speaking, B’s interests.) Recently, MacFarlane (manuscript) has outlined a very different variant of Type B pragmatism that we will discuss below. Type C pragmatism is the radical view that the token utterance does not have a context-invariant truthvalue, so one and the same utterance can be properly evaluated as true by one evaluator and false by another. This position is outlined, and defended, by MacFarlane (2005). The purpose of this paper is to argue against Type B pragmatism. I’ll show that there is a striking disanalogy between the behaviour of “knows” in questions and the behaviour of terms for which a Type B-type theory is true. The best explanation for this disanalogy is that Type B theories, contextualism included, are false.2 I won’t be addressing the second or third questions here, but I’ll assume (basically for ease of exposition) that the right answer to the second has more to do with practical than intellectual interests. So a sceptical possibility becomes relevant not because anyone is actively considering it, but because it makes a difference to someone’s actions. Everything I say here should be restatable, given any answer to that question, so little beyond exposition turns on this.

2 My own view is that if any pragmatic theory is true, it is Type A pragmatism. In Weatherson (2005), I defend the view that whether an agent is sufficiently confident in p to count as believing that p depends on features of her context. In contexts where it is very important for practical deliberation as whether p is true, a degree of confidence that might ordinarily count as belief that p might no longer so count. This is an odd kind of doxastic externalism, namely of the view that whether a state amounts to a belief is environment-dependent. Now to know that p also requires having a certain level of confidence that p is true, and it is arguable (though I haven’t argued it) that this level is also dependent on the would be knowers environment. It is also arguable that the best theory of epistemic defeaters will contain pragmatic features. Though I describe Type C pragmatism as radical, I have been convinced by John MacFarlane’s work that it is a perfectly coherent doctrine. In Egan, Hawthorne, and Weatherson (2005), we defend a version of Type C pragmatism for sentences of the form A might be F, where might is understood epistemically.

Questioning Contextualism

135

1. Basic Indexicals in Questions As well as considering these three pragmatic theories about “know”, we can consider similar theories about other words. We’ll start with words that are universally considered to be indexicals. Consider the following three theories about “here”. Type A pragmatism — A token of “here” denotes the location of the subject of the sentence in which it appears. Type B pragmatism — A token of “here” denotes the location of the speaker. Type C pragmatism — A token of “here” denotes the location of the person evaluating the sentence. So when C evaluates B’s utterance A is here, the Type A pragmatist says that it gets evaluated as true iff A is where A is, the Type B pragmatist says that it gets evaluated as true iff A is where B is, and the Type C pragmatist says that it gets evaluated as true iff A is where C is. Obviously Type B pragmatism is true in general about “here”, as it is for all obvious indexicals.3 What we’re interested in for now, however, is not its truth but the kind of evidence that can be adduced for it. One very simple way of separating out these three theories involves questions, as in the following example. Watson is at a party in Liverpool, and he is worried that Moriarty is there as well. He calls Holmes, who is tracking Moriarty’s movements by satellite, and asks, (1) Is Moriarty here? This question has three properties that are distinctive of questions involving indexicals. ●

●

3

SPEAKER — How Holmes should answer the question depends on the speaker’s (i.e. Watson’s) environment, not his own and not Moriarty’s. It would be wrong to say “Yes” because Moriarty is where he is, or “No” because Moriarty is not in the lab with Holmes following the satellite movement. CLARIFICATION — It is permissible in certain contexts to ask the speaker for more information about their context before answering. If Holmes is unsure of Watson’s location, he can reply with “Where are you?” rather than answering directly.4

Whether the “in general” is strictly necessary turns on tricky considerations about non-standard usages, such as answering machines. I’m generally favourable to the view that if any qualification is needed, it is a very small one. See Weatherson (2002) for more discussion of this point. 4 It is important here that we restrict attention to clarificatory questions about the context, and in particular to features of the context that (allegedly) affect the truth-value of utterances involving the indexical. Speakers can always ask for clarification of all sorts of features of a question, but a question only has CLARIFICATION if it is appropriate to ask for clarifying information about the nature of the questioner’s context.

136 ●

Brian Weatherson

DIFFERENT ANSWERS — If two different people, in different locations, ask Holmes (1), he can answer differently. Assume that Moriarty is at the party, so Holmes says “Yes” in reply to Watson. Lestrade then calls from Scotland Yard, because he is worried that Moriarty has broken in and asks Holmes (1). Holmes says “No”, and this is consistent with his previous answer.

Questions involving indexicals have a fourth property that we’ll return to from time to time in what follows: NEGATIVE AGREEMENT — It is coherent to answer the question by saying “No”, then basically repeating the question, inverting the verb-subject order so you have an indicative sentence. So Holmes could consistently (if falsely in this story) say, “No, Moriarty is here”. The truth of Type B pragmatism about “here” explains, indeed predicts, that (1) has these properties. If a correct answer is a true answer, then Type B pragmatism directly entails that (1) has SPEAKER. That assumption (that correctness is truth) plus the fact that you can speak to someone without knowing their location implies that (1) has CLARIFICATION, and adding the fact that different people can be in different locations implies that (1) has DIFFERENT ANSWERS. Whether Holmes can say Moriarty is here depends whether he can truthfully answer his own question Is Moriarty here?, so NEGATIVE AGREEMENT is related to DIFFERENT ANSWERS. For these reasons it isn’t surprising that other terms that are agreed on all sides to be indexicals generate questions that have all four properties. For instance — because “me” is an indexical — (2) has all four properties. (2) Does Moriarty want to kill me? This suggests a hypothesis, that all terms for which Type B pragmatism is true generate questions with those four properties. This hypothesis seems to be false; some terms for which Type B pragmatism is true do not generate questions that have NEGATIVE AGREEMENT. But a weaker claim, that all terms for which Type B pragmatism is true generate questions with the first three properties, does seem to be true.5 Or so I will argue in the next section. 5

The failure of NEGATIVE AGREEMENT for questions involving quantifiers, comparative adjectives, and similar terms is interesting, and is something that a full semantic theory should account for. My feeling is that the best explanation will draw on the resources of a theory like that defended by Stanley (2000, 2002) but it would take us far away from our primary purpose to explore this here. It is certainly an argument in favour of the “semantic minimalism”, defended by Cappelen and Lepore

Questioning Contextualism

137

2. Other Type B Terms in Questions Various philosophers have argued that a version of Type B pragmatism is true for each of the following four terms: “tall”, “empty”, “ready”, and “everyone”. Various defenders of Type B pragmatism about “knows” have argued that “knows” is analogous to terms on this list. In this section I’ll argue that questions involving those four terms have the first three properties of questions listed above, though they don’t seem to have NEGATIVE AGREEMENT. In the next section I’ll argue that questions involving “know” do not have those three properties. These facts combine to form an argument against the hypothesis that Type B pragmatism is true of “knows”. (The argument here does not turn on supposing that Type B pragmatism is true of these four terms. All I want to argue for is the claim that any term for which Type B pragmatism is true generates questions with the first three properties. It is possible that other terms also generate these kinds of questions, but this possibility doesn’t threaten the argument.) The examples involving these four terms will be a little complicated, but the intuitions about them are clear. Moriarty has hired a new lackey, a 12-year-old jockey. She is tall for a 12-yearold girl, and tall for a jockey, but not tall for a person. Moriarty is most interested in her abilities as a jockey, so he worries that she’s tall. Holmes is also most interested in her qua jockey. Watson has noted that Moriarty never hires people who are tall for adults (he thinks this is because Moriarty likes lording over his lackeys), and is wondering whether the new hire fits this property. He asks Holmes (3) Is Moriarty’s new lackey tall? Holmes should say “No”. What matters is whether she is tall by the standards Watson cares about, not whether she is tall by the standards Holmes cares about (i.e. jockeys) or the standards Moriarty cares about (i.e. also jockeys), so (3) has SPEAKER. Lestrade wants information from Holmes about the lackey so his men can pick her up. They have this conversation. Lestrade: Is Moriarty’s new lackey tall? My men are looking for her. Holmes: Where are they looking? Lestrade: At her school. Holmes: Yes, she looks like she could be fourteen or fifteen. (2005), that all and only the terms in their “basic set” (apart from tense markers), i.e. the obvious indexicals, generate questions with NEGATIVE AGREEMENT. I think the fact that many terms generate questions with the SPEAKER, CLARIFICATION, and DIFFERENT ANSWERS properties tells more strongly against their view.

138

Brian Weatherson

Holmes quite properly asks for a clarification of the question so as to work out what standards for tallness are in play, so (3) has CLARIFICATION. And he properly gives different answers to Watson and Lestrade, so it has DIFFERENT ANSWERS. But note that “tall” doesn’t have NEGATIVE AGREEMENT. Holmes can’t answer (3) with “No, she’s tall”. I won’t repeat this observation for the other terms discussed in this section, but the point generalises. Next we’ll consider “empty”. Holmes and Watson are stalking Moriarty at a party. Watson is mixing poison, and Holmes is trying to slip the poison into Moriarty’s glass. Unfortunately Moriarty has just about finished his drink, and might be about to abandon it. Watson is absent-mindedly trying to concoct the next poison, but he seems to have run out of mixing dishes. Watson: Is Moriarty’s glass empty? Holmes: What do you want it for? Watson: I need something dry to mix this poison in. Holmes: No, it’s got a small bit of ice left in it. (Lestrade arrives, and sees Holmes holding a vial.) Lestrade: Why haven’t you moved in? Is Moriarty’s glass empty? Holmes: Yes. He should get another soon. Holmes behaves entirely appropriately here, and his three responses show that the question Is Moriarty’s glass empty? has the CLARIFICATION, SPEAKER, and DIFFERENT ANSWERS properties respectively. Note, in particular, that the glass being empty by the standards that matter to Moriarty and Holmes (i.e. it’s got not much more than ice left in it) doesn’t matter to how Holmes should answer until someone with the same interests, Lestrade, asks about the glass. Third, we’ll look at “ready”. Moriarty is planning four things: to rob the Bank of England, to invade Baker St and kill Holmes, to invade Scotland Yard to free his friends, and to leave for a meeting of criminals where they will plan for the three missions. Moriarty cares most about the first plan, Holmes about the second, and Lestrade about the third, but right now Watson cares most about the fourth because it’s his job to track Moriarty to the meeting. Holmes is tracking Moriarty through an installed spycam. Watson: Is Moriarty ready? Holmes: Yes. You should go now. (Watson departs and Lestrade arrives) Lestrade: Is Moriarty ready? Holmes: Who’s that? Oh, hello Inspector. No, he still has to plan the attack out.

Questioning Contextualism

139

Again, Holmes’ answers show that the question Is Moriarty ready? has the CLARIFICATION, SPEAKER, and DIFFERENT ANSWERS properties. Note that in this case the issue of whether Moriarty is ready for the thing he cares most about, and the issue of whether he is ready for the thing Holmes cares most about, are not relevant to Holmes’ answers. It is the interests of the different speakers that matter, which suggests that if one of the three types of pragmatism is true about “ready”, it is Type B pragmatism. Finally we’ll look at “everyone”. Lewis (1996) suggests that “know” is directly analogous to “every”, so the two words should behave the same way in questions. Moriarty’s gang just robbed a department store while the royal family, along with many police, was there. Holmes is most interested in how this affected the royals, Watson in how the public reacted, Lestrade in how his police reacted, and Moriarty merely in his men and his gold. The public, and the royals, were terrified by the raid on the store, but the police reacted bravely. Watson: Did Moriarty’s men terrify everyone? Holmes: Her majesty and her party were quite shocked. Oh, you mean everyone. Yes, the masses there were completely stunned. (Lestrade enters.) Lestrade: I just heard about the raid. How did they get through security? Did Moriarty’s men terrify everyone? Holmes: No, your men did their job, but they were outnumbered. Again, Holmes’ answers show that the question Did Moriarty’s men terrify everyone? has the CLARIFICATION, SPEAKER, and DIFFERENT ANSWERS properties. And again, what the quantifier domain would be if Holmes were to use the word “everyone”, namely all the royal family is irrelevant to how he should answer a question involving “everyone”. That’s the distinctive feature of expressions for which Type B pragmatism is true, and it suggests that in this respect at least “everyone” behaves as if Type B pragmatism is true of it.

3. Questions about Knowledge We have two reasons for thinking that if Type B pragmatism is true about “knows”, then questions about knowledge should have all the SPEAKER, CLARIFICATION, and DIFFERENT ANSWERS properties. First, the assumption that correct answers are true answers plus trivial facts about the environment (namely that environments are not always fully known and differ between speakers) implies that the questions have these properties. Second, many words that are either uncontroversial

140

Brian Weatherson

or controversial instances where Type B pragmatism is true, generate questions with these properties. So if “knows” is meant to be analogous to these controversial examples, questions about knowledge should have these properties. I’ll argue in this section that knowledge questions do not have these properties. Again we’ll work through a long example to show this. Last week Watson discovered where Moriarty was storing a large amount of gold, and retrieved it. Moriarty is now coming to Baker St Station to try to get the gold back, and Holmes is planning a trap for him. Moriarty has made educated guesses that it was Watson (rather than Holmes) who retrieved the gold, and that Holmes is planning a trap at Baker St Station. But he doesn’t have a lot of evidence for either proposition. Holmes has been spying on Moriarty, so he knows that Moriarty is in just this position. Neither Moriarty nor Holmes care much about who retrieved the gold from Moriarty’s vault, but this is very important to Watson, who plans to write a book about it. On the other hand, that Holmes is planning a trap at Baker St Station is very important to both Holmes and Moriarty, but surprisingly unimportant to Watson. He would prefer that he was the hero of the week for recovering the gold, not Holmes for capturing Moriarty. They have this conversation. Watson: Does Moriarty know that you’ve got a trap set up at Baker St Station? Holmes: No, he’s just guessing. If I set up a diversion I’m sure I can get him to change his mind. Watson: Does he know it was me who recovered the gold? Holmes: Yes, dear Watson, he figured that out. These answers sound to me like the answers Holmes should give. Because the trap is practically important to both him and Moriarty, it seems he should say “No” to the first question unless Moriarty has very strong evidence.6 But because it is unimportant to Holmes and Moriarty just what Watson did, the fact that Moriarty has a true belief that’s based on the right evidence that Watson recovered the gold is sufficient for Holmes to answer the second question with “Yes”. This shows that questions involving “knows” do not have the SPEAKER property.

6

The Type A pragmatist says that it is the importance of the trap to Moriarty that makes this a high-stakes question, and the Type C pragmatist (at least as I understand that view) says that it is the importance of the trap to Holmes that matters. Since Moriarty and Holmes agree about what is important here, the Types A and C pragmatists can agree with each other and disagree with the Type B pragmatist.

Questioning Contextualism

141

Some might dispute the intuitions involved here, but note that on a simple Type B pragmatist theory, one that sets the context just by the speaker’s interests (as opposed to the speaker and her interlocutors), Holmes should assent to (4) and (5). (4) Moriarty does not know that I’ve got a trap set up for him at Baker St Station; he’s just guessing. (5) Moriarty does know that Watson recovered the gold. And it is very intuitive that if he should assent to (4) and (5), then he should answer “No” to Watson’s first question, and “Yes” to the second. Some adherents of a more sophisticated Type B pragmatist theory will not say that Holmes should assent to these two, because they think he should take Watson’s interests on board in his use of “knows”. This is the “single scoreboard view” of Keith DeRose (2004).7 But we can, with a small addition of complexity, rearrange the case so as to avoid this complication. Imagine that Watson is not talking to Holmes, but to Lestrade, and that Lestrade (surprisingly) shares Watson’s interests. Unbeknownst to the two of them, Holmes is listening in to their conversation via a bug. When listening in to conversations, Holmes has the habit of answering any questions that are asked, even if they obviously aren’t addressed to him. (I do the same thing when watching sports broadcasts, though the questions are often mindnumbingly bland. Listening to the intuitive answers one gives is a good guide to the content of the question.) So the conversation now goes as follows. Watson: Does Moriarty know that Holmes has got a trap set up at Baker St Station? Holmes (eavesdropping): No, he’s just guessing. If I set up a diversion I’m sure I can get him to change his mind. Lestrade: I’m not sure. My Moriarty spies aren’t doing that well. Watson: Does he know it was me who recovered the gold? Holmes (eavesdropping): Yes, dear Watson, he figured that out. It is important here that Holmes is talking to himself, even though he is using Watson’s questions to guide what he says. So it will be a very extended sense of context if somehow Watson’s interests guide Holmes’ context. Some speakers may be moved by empathy to align their interests with the people they are speaking about, but Holmes is not such a speaker. So Type B pragmatic theories should say that Holmes should endorse (4) and (5) in this context, and intuitively if this 7

The failure of NEGATIVE AGREEMENT for knowledge questions suggests that if any contextualist theory is true, it had better be a single scoreboard view.

142

Brian Weatherson

is true he should speak in the above interaction just as I have recorded. But this is just to say that questions about knowledge do not have SPEAKER.8 One might worry that we’re changing the rules, since we did not use eavesdropping situations above in arguing that if Type B pragmatism is true of “knows”, then questions about knowledge should have SPEAKER. But it is a simple exercise to check that even if Holmes is eavesdropping, his answers to questions including “tall”, “empty”, “ready”, and “everyone” should still have the three properties. So this change of scenery does not tilt the deck against the Type B pragmatist. Cases like this one also suggest that questions involving “knows” also lack the CLARIFICATION and DIFFERENT ANSWERS properties. It would be odd of Holmes to reply to one of Watson’s questions with “How much does it matter to you?” as it would be in any case where a questioner who knows that p asks whether S also knows that p. Intuitions about cases where the questioner does not know that p are a little trickier, because then the respondent can only answer “Yes” if she has sufficient evidence to assure the speaker that p, and it is quite plausible that the amount of evidence needed to assure someone that p varies with the interests of the person being assured.9 But if Type B pragmatism were true, we should expect to find cases of CLARIFICATION where it is common ground that the speaker knows p, and yet a request for standards is in order, and no such cases seem to exist. Similarly, it’s hard to imagine circumstances where Holmes would offer a different answer if Lestrade rather than Watson asked the questions. And more generally, if two questioners who each know that p asks whether S knows that p, it is hard to see how it could be apt to answer them differently.10 But if Type B pragmatism about “knows” were true, knowledge questions would have these three properties, so it follows that Type B pragmatism about “knows” is false.

4. Objections and Replies Objection: The argument that Type B pragmatism implies that questions involving “knows” should have the three properties, assumes that correct answers are true answers. But there are good Gricean reasons to think that there are other standards for correctness. 8

As noted above, Types A and C pragmatists agree with this intuition, though for very different reasons. With the right supplementary assumptions, I believe this claim can be argued to be a consequence of any of our three types of pragmatism. 10 The above remarks about cases where the speakers don’t know that p also apply. I think that without a very careful story about how the norms governing assertion relate to the interests of the speaker and the audience, such cases will not tell us much about the semantics of “knows”. Best then to stick with cases where it is common ground that everyone, except the subject, knows that p. 9

Questioning Contextualism

143

Reply: It is true that one of the arguments for thinking that Type B pragmatism has this implication uses this assumption. But the other argument, the argument from analogy with indexicals and other terms for which Type B pragmatism is true, does not. Whatever one thinks of the theory, there are data here showing that “knows” does not behave in questions like other context-sensitive terms for which Type B pragmatism is the most plausible view. Moreover, Type B pragmatists have to be very careful in wielding this objection. If there is a substantial gap between correct answers to knowledge questions and true answers, then it is likely that there is a substantial gap between correct knowledge ascriptions and true knowledge ascriptions. As a general (though not universal) rule, truth and correctness are more tightly connected for questions than for simple statements. For example, some utterances do not generate all the scalar implicatures as answers to questions that they typically generate when asserted unprompted.11 But the primary “ordinary language” argument for Type B pragmatism about knowledge ascriptions assumes that correct knowledge ascriptions are, by and large, true. We can put this in more theoretical terms. If we are to use these considerations to generate a rebutting defeater for Type B pragmatism, we would need an argument that Holmes’ answers are correct iff they are true. And while that step of the argument is plausible, it is not beyond contention. But that isn’t the only use of the examples. We can also use them to undercut the argument from ordinary language to Type B pragmatism. We have a wide range of cases where ordinary usage is as if Type B pragmatism is false. And these cases are not peripheral or obscure features of ordinary usage. Answering questions is one of the most common 11

Here are a couple of illustrations of this point. Yao Ming is seven feet six inches tall. It would be odd (at best) to use (6) to describe him, but it is clearly improper to answer (7) with “No”. (6) Yao Ming is over six feet tall. (7) Is Yao Ming over six feet tall? For a second case, consider an example of Hart’s that Grice (1989) uses to motivate his distinction between semantics and pragmatics. A motorist drives slowly down the street, pausing at every driveway to see if anyone or anything is rushing out. It seems extremely odd to use (8) to describe him. (8) He drove carefully down the street. Is this oddity due to (8) being false, or it being otherwise infelicitous? Part of the argument that (8) is true, but infelicitous, is that intuitively it is correct to say “Yes” in response to (9), and incorrect to say “No.” (9) Did he drive carefully down the street? The implicatures associated with (8) are largely absent from affirmative answers to (9), though such an answer presumably has the same truth-conditional content as (8).

144

Brian Weatherson

things we do with language. So ordinary usage doesn’t provide an all-thingsconsidered reason to believe that Type B pragmatism is true. If ordinary usage is (or at least was) the best reason to believe Type B pragmatism, it follows that there is no good reason to believe Type B pragmatism. Objection: Sometimes when we ask Does S know that p? all we want to know is whether S has the information that p. In this mood, questions of justification are not relevant. But Type B pragmatism is a theory about the interaction between the subject’s justification and knowledge ascriptions. So these questions are irrelevant to evaluating indexicalism.12 Reply: It is plausible that there is this use of knowledge questions. It seems to me that this is a usage that needs to be explained, and isn’t easily explained on current theories of knowledge.13 But I’ll leave discussion of that use of knowledge questions for another day. For now I’ll just note that even if this can be an explanation of why we sometimes assent to knowledge questions, it can’t be an explanation of why Holmes denies that Moriarty knows about the planned trap at Baker St Station. Holmes agrees that Moriarty believes there is a trap planned, but insists that because Moriarty is “just guessing” this belief does not amount to knowledge. What really needs explaining is the difference between Holmes’ two answers, and this other use of knowledge questions doesn’t seem sufficient to generate that explanation. Objection: There is no explanation offered here for the data, and we should not give up an explanatory theory without an explanation of why it fails. Reply: The best way to explain the data is to look at some differences in our attitudes towards questions containing “knows” and questions containing other terms 12

A similar view is defended by Alvin Goldman (2002). One explanation, due to Stephen Hetherington (2001), is that all that is ever required for a true ascription of knowledge that p to S is that p is true and S believes it. This leaves open the question, as discussed below, of why we sometimes deny that true believers are knowers, but we can bring out familiar Gricean explanations about why we might usefully deny something that is strictly speaking true, or we could follow Hetherington and offer an explanation in terms of the gradability of knowledge claims. Another possible explanation is that “know” univocally means something like what most epistemologists say that it means, but there is a good pragmatic story about why speakers sometimes attribute knowledge to those with merely true belief. (I don’t know how such an explanation would go, especially if Type B pragmatism is not true.) Finally, we might follow Goldman and say the English word “know” is ambiguous between a weak and strong reading, and the strong reading has been what has concerned epistemologists, and the weak reading just requires true belief. Such an ambiguity would be a small concession to Type B pragmatism (since it is the interests of the speaker that resolve ambiguities) but we could still sensibly ask whether Type B pragmatism is true of the strong reading. 13

Questioning Contextualism

145

for which Type B pragmatism is plausibly true. I’ll just go over the difference between “knows” and “ready”; the point I’m making easily generalises to the other cases. Different people may be concerned with different bits of preparation, so they may (speaker) mean different things by X is ready. But neither will regard the other as making a mistake when they focus on a particular bit of preparing by saying X is ready. And this is true even if they think that the person should be thinking about something else that X is preparing. Knowledge cases are not like that. Different people may have different standards for knowledge, so perhaps they may (speaker) mean different things by A knows that p, because they will communicate that A has met their preferred standards for knowledge. But in these cases, each will regard the other as making a mistake. Standards for knowledge aren’t the kind of thing we say people can differ on without making a mistake, in the way we do (within reason) say that different people can have different immediate goals (e.g. about what to have for dinner) without making a mistake. That explains why we don’t just adopt our questioners’ standards for knowledge when answering their knowledge questions. Objection: In Section 2 it was argued that questions involving “knows” should have the three properties (SPEAKER, CLARIFICATION, and DIFFERENT ANSWERS) because questions involving other terms for which Type B pragmatism is true have the properties. But this argument by analogy may be flawed. All those terms are “indexical” in John MacFarlane’s sense (manuscript, 2005). That is, the content of any utterance of them varies with the context. But not all Type B pragmatist theories are indexical, and this argument does not tell against a nonindexical Type B pragmatism. Reply: The view under consideration says that all utterances of “S knows that p” express the same proposition, namely that S knows that p. But the view is Type B, because it says that proposition can be true relative to some contexts and false relative to other contexts, just as temporalists about propositions say that a proposition can be true at some times and false at other times, and the utterance is true iff the proposition is true in the context of the utterance. This is a Type B view, and I agree that the argument by analogy in Section 2 is powerless against it. But that argument was not the only argument that Type B views are committed to questions involving “knows” having the three properties. There was also an argument, at the end of Section 1, from purely theoretical considerations. It turns out that this argument is also complicated to apply here, but it does tell against this type of Type B pragmatist as well. Here are two hypotheses about how one should answer a question Does NP VP? where NP is a noun phrase and VP a verb phrase. Consider the proposition

146

Brian Weatherson

expressed by the sentence NP VP in the speaker’s context. First hypothesis: the correct answer is “Yes” iff that proposition is true in the speaker’s context. Second hypothesis: the correct answer is “Yes” iff that proposition is true in the respondent’s context. If propositions do not change their truth-value between contexts, these two hypotheses are equivalent, but on this version of Type B pragmatism that is not so, so we have to decide between them. The best way to do this is by thinking about cases where it is common ground that propositions change their truthvalue between contexts, namely cases where the contexts are possible worlds. Let’s consider the scenario I briefly discussed above of the television watcher answering whatever questions gets asked. Above I put the questioners in the same possible world as me, but we can imagine they are fictional characters in a different possible world. Imagine a show where it has just been revealed to the audience, but not all the characters, that the prime minister is a space alien. The following happens: Character (on screen): Is the Prime Minister a space alien? Me (in real world): Yes! Since the proposition that the prime minister is a space alien is true in their world, but not in our world, the propriety of this response tells in favour of the first hypothesis above. Now this is not a conclusive argument, because it might be that there is some distinctive feature of fiction that is causing this answer, even though the second hypothesis is in general correct. But it seems we have reasons to think that if this kind of Type B pragmatism were true, respondents would use the speaker’s context to work out what kind of answer to give. And as we saw above, this is not what respondents actually do; they either use their own context or (more likely) the subject’s. Objection: At most this shows that Type B pragmatism about “knows” is not actually true of English. But there might still be good epistemological reasons to adopt it as a philosophically motivated revision, even if the data from questions show it isn’t true. So even if hermeneutic Type B pragmatism is false, revolutionary Type B pragmatism might be well motivated. Reply: It’s true that the philosophically most interesting concepts may not map exactly on to the meanings of words in natural language. (Though I think we should be careful before abandoning the concepts that have proven useful enough to get simple representation in the language.) And it’s true that there are reasons for having epistemological concepts that are sensitive to pragmatic factors. But what is hard to see is what interest we could have in having epistemological terms whose application is sensitive to the interests of the person using them. Hawthorne (2004)

Questioning Contextualism

147

provides many reasons for thinking that terms whose applications are sensitive to the interests of the person to whom they are being applied are philosophically and epistemologically valuable. Such terms provide ways of expressing unified judgements about the person’s intellectual and practical reasoning. On the other hand, there is little to be gained by adopting Type B pragmatism. If there needs to be a revision around here, and my guess is that there does not, it should be towards Type A, not Type B, pragmatism.14

References Cappelen, H., & Lepore, E. (2005). Insensitive semantics: A defense of semantic minimalism and speech act pluralism. Malden, MA: Blackwell. Cohen, S. (1988). How to be a fallibilist. Philosophical Perspectives, 2, 91–123. DeRose, K. (1995). Solving the skeptical problem. The Philosophical Review, 104, 1–52. DeRose, K. (2004). Single scoreboard semantics. Philosophical Studies, 119, 1–21. Egan, A., Hawthorne, J., & Weatherson, B. (2005). Epistemic modals in context. In: G. Preyer & G. Peter (Eds), Contextualism in philosophy: Knowledge, meaning, and truth (pp. 131–169). Oxford: Clarendon Press. Goldman, A. (2002). What is social epistemology? A smorgasbord of projects. Pathways to knowledge: Private and public (pp. 182–204). New York: Oxford University Press. Grice, P. (1989). Studies in the way of words. Cambridge, MA: Harvard University Press. Hawthorne, J. (2004). Knowledge and lotteries. Oxford: Clarendon Press. Hetherington, S. (2001). Good knowledge, bad knowledge: On two dogmas of epistemology. Oxford: Clarendon Press. Lewis, D. (1996). Elusive knowledge. Australasian Journal of Philosophy, 74, 549–567. MacFarlane, J. (2005). The assessment sensitivity of knowledge attributions. Oxford Studies in Epistemology, 1, 197–233. MacFarlane, J. (manuscript). Non-Indexical Contextualism. Presented at the Rutgers Semantics Workshop, September 2005. Manuscript. Stanley, J. (2000). Context and logical form. Linguistics and Philosophy, 23, 391–434. Stanley, J. (2002). Nominal restriction. In: G. Peter & G. Preyer (Eds), Logical form and language (pp. 365–388.). Oxford: Clarendon Press. Stanley, J. (2005). Knowledge and practical interests. Oxford: Clarendon Press. Weatherson, B. (2002). Misleading indexicals. Analysis, 62, 308–310. Weatherson, B. (2005). Can we do without pragmatic encroachment? Philosophical perspectives, 19, 417–443.

14

Thanks to David Chalmers, Keith DeRose, Tamar Szabó Gendler, Stephen Hetherington, John MacFarlane, Ishani Maitra and Matthew Weiner for helpful discussions about earlier drafts.

This page intentionally left blank

148

PART B UNDERSTANDING KNOWLEDGE?

149

This page intentionally left blank

150

Chapter 10

Truthmaking and the Gettier Problem Adrian Heathcote

When we know something what we know is some fact or other. But we easily shift from the fact to the proposition that expresses that fact. And thus we come to say that what we know is a proposition, that the content of our belief is a proposition. But taking this step lets in certain anomalous situations, first identified by Russell, but given full expression by Edmund Gettier — the so-called Gettier counterexamples. These were taken by Gettier to show that the two conditions of a belief being both true and justified could not jointly be sufficient for knowledge, that something more was required. Some have argued (e.g. Lewis, 1996) that the constraints that Gettier places on the situations that he describes are simultaneously unsatisfiable, and thus the Gettier problem is unsolvable. This is meant to explain why the problem has long seemed so resistant to solution. I believe this is a mistake. The Gettier problems are not unsolvable, and to that effect I offer something that I take to be a complete solution. Moreover, the reason why the Gettier problems have seemed so resistant to solution is the phobia that philosophers have been possessed of, since the 1950s, for invoking facts and states of affairs to solve philosophical problems. This was the great mistake — for, as I will argue, the problem only arose in the first place from shifting attention from the fact that is the object of our knowledge to the proposition that is the object of our knowledge. Let me begin by going over Gettier’s two classic cases, so that we can make some fresh points about their logic and their structure.

Aspects of Knowing: Epistemological Essays Edited by S. Hetherington Crown Copyright © 2006 Published by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

151

152

Adrian Heathcote

The Classic Gettier Cases According to the venerable, Platonic account of knowledge, to know A is to have a justified, true belief that A. These conditions are taken to be necessary, for if A is not true then it would not count as knowledge (this is the facticity condition); and if it is not a belief that we are justified in holding then it is merely a fortunate event that we happen to believe it. So both conditions are necessary for someone to be said to know A. But are these conditions sufficient? Edmund Gettier (1963) published a now famous paper, which purported to show that they could not be. He gave two counterexamples to the thesis that justified true belief is sufficient for knowledge. Here is the first counterexample. (I have kept to Gettier’s original numbering of these displayed sentences.) Smith and Jones have applied for the same job. Smith has (non-conclusive) evidence for the conjunctive proposition: (d) Jones will get the job, and Jones has 10 coins in his pocket; Smith’s evidence for (d) is that he has been told by someone who should know that Jones is going to get the job and (by some method best not enquired into) he has counted the coins in Jones’s pocket only a few minutes ago. This is not conclusive evidence for (d) but it is still evidence, sufficient to justify belief. Proposition (d) logically entails (e): (e) The person who will get the job has ten coins in his pocket; Smith sees that (d) entails (e) and accepts (e) on the basis of (d). The evidence for (d) is thus transmitted to (e); Smith therefore has good reason to believe that (e) is true. It will turn out, however, that he, Smith, will actually get the job, and also that he himself — all unknown to himself — has ten coins in his pocket. (e) is then true though the proposition (d) from which it was inferred is false. So (e) is true, Smith has good reason to believe (e), and does in fact believe it. But Smith couldn’t really be said to know (e) because what makes (e) true are the two facts that (1) he will get the job and (2) the number of coins in his own pocket is ten, and he is ignorant of both these things. Smith’s belief in (e) is based on the false, but well-supported, belief that Jones will get the job. So, Gettier concludes, since someone could have a justified true belief but not have knowledge, having knowledge must require the satisfaction of some extra condition over and above having a justified true belief. The structure of his argument here seems perfectly cogent, and it does not leave much room for complaint. Should we agree with the conclusion though? It is arguable that we shouldn’t. The proposition that Smith believes — that the person who will get the job has ten coins

Truthmaking and the Gettier Problem

153

in his pocket — is ambiguous, and it is on this ambiguity that the example depends. If we ask ourselves whom Smith means when he is believing this proposition, the answer is obvious: he means Jones. For sentences, philosophers of language would call this the speaker meaning. If, however, we ask ourselves who the referring expression (“the person who will get the job …”) objectively picks out — who it is in fact true of — then the answer is that it picks out Smith, not Jones. We could call this the objective referent meaning. Now usually the speaker meaning and the objective referent meaning coincide — but here they have come apart. If we consider the speaker meaning interpretation of (e) we see that it turns out to be a justified but false belief. On the other hand, if we consider the objective referent meaning then we have a true belief but one that is unjustified. So when we disambiguate (e) we either get a justified false belief or we get an unjustified true belief — but in neither case do we get a justified true belief. The example does not show therefore that justified true belief is insufficient for knowledge; it’s just a case where we do not have knowledge because we don’t have a justified true belief in the first place. Still, as I said, Gettier gave two counterexamples, and we have only considered the first of them. Perhaps the next one fares better. Here it is. Smith — for once again it is he — has strong evidence for proposition (f): (f) Jones owns a Ford; Smith’s evidence — which we will not rehearse — is strong though non-conclusive. Quite separate from this, Smith has a friend, named Brown, of whose whereabouts Smith is completely ignorant. Smith randomly chooses three place names: Boston, Barcelona, and Brest Litovsk. He then formulates the following three propositions: (g) Either Jones owns a Ford, or Brown is in Boston; (h) Either Jones owns a Ford, or Brown is in Barcelona; (i) Either Jones owns a Ford, or Brown is in Brest Litovsk; Each of these propositions is logically entailed by (f) — a fact that Smith is well aware of. Since Smith was justified in accepting (f) he is also justified in accepting (g), (h), and (i) — despite the fact that in each case he has no reason to believe that the second disjunct is true. As it turns out, however, (f) is false — Jones does not own a Ford — but Brown is in Barcelona, unbeknownst to Smith. Therefore, (h) is a justified true belief, but, says Gettier, Smith does not know that (h) is true. So again, says Gettier, justified true belief is not sufficient for knowledge. This example is different from the first in that it does not seem to do to make a distinction between speaker meaning and objective referent meaning. What Smith

154

Adrian Heathcote

believes is unambiguous and clear — and what he believes is both justified and true, even though it is not true in quite the way that he thinks. So what we said about the first example is not sufficient to solve the Gettier problems. Note that in both of these counterexamples Gettier gives no reason for his judgment that Smith does not have knowledge — he simply takes it to be obvious. One response that we could make to these cases would be simply to say that Gettier is mistaken: Smith does know. (I am setting aside my response to the first of them, just for the sake of this point.) The cases are strange and our intuitions are unprepared for them, but Smith does know after all. Or at least, Gettier has to say more than that it is obvious that he doesn’t. I do not think that this point can be pushed too hard. We possess fairly robust intuitions about knowledge and their verdict is clear: Smith does not have knowledge in either of the two examples. The only room left for negotiation with our intuitions would be in denying that Smith has a justified true belief — and whereas in the first case we could insist that he doesn’t, there doesn’t seem much room for denial in the second, unless we deny some tried-and-trusted principles of logical inference. If one is tempted to press the other way and say that in these cases Smith has knowledge then a caution must be added: it is not at all difficult to tweak the second case so that if Smith does know then his knowledge must be non-local. Thus suppose we change the true proposition to the following: (j) Either Jones owns a Ford, or Brown is right now doing a jig on Betelgeuse; and suppose we mean by right now “at a space-like separation”. Then we may further suppose that Brown is “right now” doing a jig on Betelgeuse. In this case, Smith has non-local knowledge — something that may be regarded to be in conflict with the spirit of special relativity. So we have a problem, and it is one that must be met. Before we begin setting out our answer we might add one further example that is current in the literature. We see what we think is a sheep in a field and we form the belief (k) We are seeing a sheep in the field; From this we infer (by existential generalization): (l) There is a sheep in the field; But what we saw was not a sheep but a cardboard prop, set up by the farmer to fool visiting city-slickers. However behind a rock, where we can’t see it, there is in fact a sheep. So our belief (l) is true, and made true by the sheep that we can’t see, though it is the prop sheep that provides the justification for the belief (l), through

Truthmaking and the Gettier Problem

155

(k). In this case it is plain that the prop sheep, which is our evidence, is not evidence of the circumstance that makes (l) true. In fact when we reflect, this example is really just a simplified version of the first, and again it seems plausible to say that speaker meaning diverges from the objective referent meaning. But, again, we know that that can’t be the whole story. It is these cases, the classic Gettier cases, that we wish to solve. There are other cases proposed by Carl Ginet and Alvin Goldman on perceptual knowledge that are of a different type to the true Gettier counterexamples. They require a different solution; I’ll come back to this at the end. To give our solution we need two concepts: the concept of a state of affairs making a proposition true, and the concept of a state of affairs being the state of affairs of which we have evidence. Both make sense within the general framework of truthmaker theory.

States of Affairs and Truthmaking The primary idea of truthmaker theory is that there are facts or states of affairs (or situations, or circumstances, or ways things are) that make propositions true — i.e. are such that if the proposition had not been true the fact could not have been obtained. This is intended to capture the fundamental insight of the correspondence view that propositions are not true or false in a way that floats free of what the world is like: it is the world that does the determining — it is the world that does the truth-making. Let us designate these facts, or, as we will now call them, truthmakers, by s, t, u, v (with primes if necessary). And let us designate the truthmaking relation by “➤”. Then the primary idea of truthmaker theory is captured by the following two claims: TRUTHMAKER DEFN: s is a truthmaker for proposition A, designated s➤ A, iff it is a necessary truth that if s obtains then A is true; TRUTHMAKER AXIOM: For any true proposition A, there is an s such that s➤ A. There are no truthmakers for false propositions. We will come back to the question of the existence of facts later; for now we just assume that there are some. Then in addition to these two claims, we have a number of subsidiary propositions that put flesh on the account.1 Here they are. 1

These axioms come proximally from Stephen Read (2000) but many go back to the original article by Mulligan et al (1984). I discussed them in Heathcote (2003).

156

Adrian Heathcote

1. For all s, there is some A, such that s➤ A. 2. If s➤ A and t➤ B then s ⫹ t➤ A & B, where s ⫹ t is the mereological fusion of the truthmakers s and t. Both of these propositions has a considerable claim to being part of the basic truthmaking intuition, though neither is completely unassailable.2 In contrast the following is very much less plausible: * For all truthmakers s and propositions A and B, if s➤ A and A → B then s➤ B. (Where “→” expresses the classical entailment relation.) This leads directly to every s being a truthmaker of every necessary truth, since all propositions, and therefore all true propositions, classically entail every necessary truth. We will call this Necessity Monism. NECESSITY MONISM: If s➤ A and A classically entails B then s➤ B. But then every s is a truthmaker for every necessary truth. Certainly, this is not an unarguably disastrous consequence — indeed it might be said that it is merely a result of the folly of trying to define a truthmaking relation for all truths, including necessary truths. For, one might think: it is the mark of the contingent truths that they are made (what, after all, does “contingent” mean?) but the necessary truths are no made at all, they are true regardless of the facts that obtain in this world. This idea that the necessary truths are true regardless of what facts obtain is well-captured, it might be said, by the consequence that since all facts make them true, no fact really makes them true at all. Were the facts different the necessary truths would sail on regardless. (This is clearly Wittgenstein’s position in the Tractatus.) But we need not submit to this counsel of despair. Perhaps necessary truths do have particular truthmakers after all. But if they do then * is clearly a mistake — because it will have the immediate effect of making the truthmakers for the tautologies all the same, passing the truthmaker for one to all the others like a contagion. Indeed * seems quite unmotivated: its only effect is to spread truthmakers about in an all too liberal fashion. It reminds us, forcibly, that logical equivalence is not equivalence tout court. There is a more fine-grained equivalence to be had, namely having the same truthmaker(s).

2

If we were to want sentences or statements in place of propositions then there would have to be considerable doubt about 1), as we will see in the next section.

Truthmaking and the Gettier Problem

157

I argued for this view in Heathcote (2003). The idea was as follows: in place of * approximate truthmaking equivalence more carefully, by first defining it for conjunction, disjunction, and negation. 1. For all truthmakers s and propositions A and B, s➤ (A & B) iff s➤ A and s➤ B; 2. If either s➤ A or s➤ B then s➤ (A ⵪ B); (Note that the converse does not hold.) 3. A is true iff there exist no truthmakers for ⬃A; 4. For all propositions A, either there exists an s such that s➤ A or (exclusive or) there exists a t such that t➤ A. Condition 4 is plausible for the truthmaking of propositions (because it is plausibly analytic that propositions are either true or false) — but it does not imply that every sentence is either true or false. Some sentences may not express propositions. We can then extend the truthmaking relation to other truth-functional connectives in the obvious way. Consider material implication: if there exists an s such that s➤B, regardless of whether there is a truthmaker for A, or there exists no truthmaker for either A or B, then there is a truthmaker for A
B. (The extension to the biconditional is obvious.) Incidentally, this extension of the truthmaker relation to (material) conditionals brings out the genuine oddities of the latter and also reveals the dangers of too simple-minded a reading of the idea that truthmaker theory vindicates the notion that “truth supervenes on being”. For the conditional A
B is guaranteed to have a truthmaker when there is no truthmaker for either A or B. This is contrary to the intuition that many people seem to have, that if being were to shrink to nothing the true propositions would shrink to nothing also. Not so! And it also serves to remind us that material conditionals are creatures of art: when we think that their characteristics are paradoxical or counterintuitive we are just forgetting that we are transferring our intuitions about “natural” conditionals onto these artifacts of logical theory. But to return to our main point. What we would like to do is to find the natural equivalence relation for truthmaking: the condition that must hold for two propositions to have exactly the same set of truthmakers. The traditional answer to this, by which we mean Wittgenstein’s answer, is that logically equivalent propositions have the same truthmakers. But this seems, on reflection, quite wrong: is it really plausible that 2 ⫹ 2 ⫽ 4 has the same truthmaker as either snow is white or snow is not white? We might also note that all false propositions have the same set of truthmakers, namely the empty set — but that means that the contingently false and the necessarily false have the same set of truthmakers, but they are by no means logically equivalent.

158

Adrian Heathcote

Truthmaking equivalence must then be a more fine-grained relation than logical equivalence. Let us denote it by “↕”.3 And yet, if two propositions have exactly the same meaning then it seems correct to say that they should have the same truthmakers. So we must try to ensure that the equivalence relation reflects that fact. We cannot give a sufficient condition for two propositions to have the same meaning (other than to say that they have the same meaning) but we can give a necessary condition that uses the correct core of the logical equivalence condition. Thus, let us take a set of replacement rules, as constraining our equivalence relation ↕. The set that I’ve chosen is from Copi (1979) — because it is likely to be familiar to many. This is a set of 11 rules that allow replacement of one proposition for another in a proof — but they also define equivalence of truthmakers. Thus consider De Morgan’s rules ⬃(A & B) ⬅ ⬃A ⵪ ⬃B ⬃(A ⵪ B) ⬅ ⬃A & ⬃B Suppose s is a truthmaker for ⬃(A & B), then there is no truthmaker for (A & B), and so either no truthmaker for A or no truthmaker for B, which gives us the RHS. Similarly if there is no truthmaker for ⬃(A & B) then there is one for (A & B), and so there must be no truthmaker for either ⬃A or ⬃B. Analogous considerations give us the second formula. Now if we run through the other rules we see that Commutation, Associativity, Tautology, and Double Negation, are perfectly obvious, and Distribution nearly so. Transposition, Material Implication, Material Equivalence, and Exportation we may take as defining truth functional implication. This defines a restricted logical equivalence relation. Additionally, we will take the conditional forms of Copi’s first nine rules (his elementary valid argument forms — modus ponens, modus tollens, etc.) as preserving the transmission of truthmakers from antecedent to consequent. Now, this set of rules has an important feature: it doesn’t allow us to derive an arbitrary tautology from any proposition — and so one cannot transmit a truthmaker for A to an arbitrary tautology B ⵪ ⬃B. Nor is there any way to substitute one arbitrary tautology for another, say B ⵪ ⬃B for A ⵪ ⬃A. But this has a surprising, and attractive, consequence. For suppose s➤A. Then s➤A ⵪ ⬃A. But imagine if A had not been true; then it would have had no truthmaker; but then ⬃A would have had a truthmaker, say t. And so in this case t➤A ⵪ ⬃A. 3 Note that I do not call it “truthmaker equivalence”, which is the condition for one fact to be the same as another: that is a different matter.

Truthmaking and the Gettier Problem

159

But this means that A ⵪ ⬃A would have had a truthmaker come what may, but perhaps a different one — and also it has a different truthmaker to B ⵪ ⬃B, whose truthmaker is either the truthmaker for B or ⬃B, whichever is actually true. Thus, we have truthmakers for these necessary truths, but they are of an entirely this-worldly character. Moreover we have different truthmakers for different necessary truths. Thus, our truthmaking equivalence relation is indeed more finegrained than the full logical equivalence relation, as promised.4 In summary, we now have the truthmaking relation defined in such a way that sameness of truthmakers can be approximated via the restricted logical equivalence relation, as defined above. Of course, the sameness-of-truthmakers relation should be taken as a primitive and its structure explored from within. Here I’ve just wanted to show how, starting from some natural considerations, we can get the intuitively correct result for the truthmaking of necessary truths — so that they are not made true by nothing, nor by everything, nor by something supernatural. And that is surely an advance in the right direction.5

True Sentences The fundamental idea of truthmaker theory is that the truth-value of a true proposition is conditionally dependent on what states of affairs obtain in the world. Truth does not float free of what the world is like. But the relationship between true sentences and propositions adds a level of complexity that has not been sufficiently appreciated. We can begin by defining the relation between true sentences and propositions. EXPRESSIBILITY: A sentence may be said to be true iff it expresses a true proposition. The first point to make is that not all states of affairs, or collections of states of affairs, are expressible in true sentences. Consider an infinite, randomly chosen, 4

I showed in Heathcote (2003) that, starting just from the definition of truthmakers, if s is the truthmaker for A ⵪ B and, ⬃A then it will also be the truthmaker for B. Thus, since disjunctive syllogism is valid in our account, the “logic of truthmaking” is not that of Anderson and Belnap’s relevance logic E. But it is possible that the logic of truthmaking might throw light on the elusive intuition that motivated that view. 5 One might suspect the use of a fragment of classical logic to be ad hoc in these circumstances. But I think that would be a mistake. Firstly, the fragment breaks along a natural faultline — it is logic without conditionalization. Secondly, we were looking to solve one problem, Necessity Monism, and the result of solving it is that we get a very attractive solution to another problem. If this is ad hoc then so is any solution to a problem in philosophy.

160

Adrian Heathcote

collection of states of affairs F. This large state of affairs is a mereological sum of smaller states of affairs, each one of which is the truthmaker for some proposition. We may call such states of affairs as F monster facts. But by the mereological fusion principle there is a proposition that is made true by the sum F. Since propositions are abstract entities this presents no difficulty. But sentences are constrained by finiteness requirements, and no sentence can express this true proposition. And of course there will be an infinite number of such propositions that are so inexpressible.6 The second point is the more appreciated point that not all sentences express propositions. A sentence is true in virtue of expressing a true proposition — but this means that sentences inherit the truth-values of propositions, not the other way round. Thus the liar sentence “This sentence is false” does not express a proposition; nor does “This sentence has no truthmaker” — both sentences do not get mapped to true propositions, or indeed false propositions.7 I am certainly not suggesting that this observation is sufficient by itself to pull the teeth of the Liar paradox, but it is important nonetheless. Both of these points have negative implications for our understanding of the so-called Tarskian biconditional schema (TBS) TBS: “p” is true iff p. (I say the “so-called TBS” because in my view it would be better to reserve that name for the Tarskian-style construction of the “truth” predicate in the 1935 paper, “The Concept of Truth in Formalised Languages”. As many philosophers, from John Etchemendy (1988) to Hilary Putnam (1994) and Tarski (1983 [1935]), have emphasized the Tarskian truth predicate is at best only extensionally equivalent to our ordinary truth predicate over a highly restricted fragment of natural language. Instances of the schema are formally provable from axioms of syntax and logic (plus set theory). For the ordinary notion of truth, and the corresponding ordinary schema it would be better if another name could be found, I suggest that a better name for the above “platitude” might be “Moorean”. But affixing “Tarski” to the schema is now entrenched, and there is no groundswell for reform. We might also

6

Non-random infinite collections of states of affairs of course may be expressible — and that is precisely what quantifiers are for. But we should not be misled into thinking that they make any but the smallest dent in the infinite. We are very fortunate that laws of nature involve (or imply) universal generalizations, for if they involved monster facts such as F we would find them inexpressible. 7 The second sentence has been given by Peter Milne (2005) as a counterexample to the Truthmaker Axiom — but this depends on taking the truthmaker relation as between states of affairs and sentences, rather than propositions.

Truthmaking and the Gettier Problem

161

note that instances of the platitudinous biconditional schema are contingent, whereas instances of the TBS are necessary — a difference if ever there was one.) The overwhelming consensus at present is that the TBS is true, and indeed platitudinously true: it expresses the fundamental fact about truth. Indeed for many there is nothing more to say about truth than this. But in the light of our preceding points we can see that the TBS is radically deficient as an account of truth — and thus certainly not complete even if in some sense platitudinous. For if we think of the RHS of the biconditional as giving “the facts” then we must note that it misses all of the states of affairs, such as F, that are not expressible in sentences. We easily miss this point because we are substituting the same sentence, say “Grass is green”, into the variable position in the left- and right-hand sides. Thus reading (1) “Grass is green” is true if and only if grass is green correctly, it is simply an asymmetrical explanation of how the sentence “Grass is green” comes to be true. It is true because grass is green — and because of the semantic content of the sentence. (And it is asymmetrical because we would not say that grass being green is explained by the fact that the sentence “Grass is green” is true.8) But not every state of affairs is invoked in this process, because we are only concerned to explain what it means for sentences to be true. Facts like F get invoked because there are no sentences that express the corresponding propositions. This means that there is a vast difference in the substitution class of TBS and (2) The proposition p is true if and only if p; For here, F is a possible substituand of the variable. Indeed substitution instances of (2) are plausibly necessary rather than contingent. This is because (2) is best read (again asymmetrically) as an explanation of what it means for the given proposition to be true. Thus in (3) The proposition grass is green is true if and only if grass is green; the explanation for the proposition grass is green being true is that grass is green. And in this case there will be a proposition that corresponds to the monster fact F. Considered in this light (2) is an alternative way of expressing the conditional 8

I argue for the asymmetrical character of the “if and only if” locution in Heathcote (2005), where I also give truth conditions for this, considered as an extension of a natural indicative conditional. But I do not rely on that analysis here.

162

Adrian Heathcote

dependence contained in the truthmaker axiom, and this conditional dependence is necessary (by the truthmaker definition). When we come to existential statements we can discern another kind of deficiency in TBS, though in this case it only shows up against the Truthmaker axiom. Suppose we take (4) “There is a sheep in the field” is true if and only if there is a sheep in the field. The individuation of facts on the RHS is rather coarse. The states of affairs that are the truthmakers for the proposition expressed by “There is a sheep in the field” are any of the sheep in the field. If there are a hundred sheep in the field then there are a hundred states of affairs that each stand as the truthmaker for the claim. Sentence (4) thus gives a rather misleading picture of the “correspondence” between sentences and facts. It was undoubtedly this that led Quine (1960, p. 247) to claim that “only indirection results from positing facts, in the image of sentences, as intermediaries [between sentences and the way we use words]”. But of course there is no reason why our individuation of facts should be “in the image of sentences” — and thinking so is the result of hanging too much importance on the TBS. But let us now ask a different question: what proposition does the sentence “The man who will get the job has ten coins in his pocket” express? Of course, in its full generality this is nothing less than the large question addressed by the philosophy of language. But if we restrict our focus we can see that there are two broad answers. If we ask “Who does Smith mean?” then the answer is that he means Jones, and his intended proposition refers to him, and is false. But if we ask “Who does the definite description in Smith’s sentence objectively pick out?” then the answer is that it picks out himself, and the resulting (true) proposition refers to himself. So in this case there are two answers to the question: what proposition does the sentence “The man who will get the job has ten coins in his pocket” express? One true proposition, and one false proposition. It is these two propositions — and the similar pairs in the other Gettier counterexamples — that we wish to look at in the next section.

Compass and Ruler in Truthmaking Under normal circumstances a person’s belief gets formed in such a way that the state of affairs that makes their belief true — i.e. its truthmaker — is the same as the state of affairs that the chain of justifications is ultimately grounded in. Thus if I see a sheep in the field and come to believe on the basis of my resulting

Truthmaking and the Gettier Problem

163

visual experience that the sheep in the field is black, then the truthmaker for the proposition the sheep in the field is black is the black sheep in the field and this is also the state of affairs that I see and that has caused me to come to believe the proposition. So the justification that I have for believing the proposition is traceable to a state of affairs that is also, in fact, the truthmaker for the proposition. Let us call this a “ruler situation” — and mnemonically it is also the “rule” in our knowledge of the world. It is illustrated in Figure 1 (the bottom diagram). In the classic Gettier cases the ruler situation doesn’t obtain. Rather there is a split, with two states of affairs, so that one is the truthmaker for the proposition, and the other is the state of affairs that the chain of justifications is grounded in. This split gives rise to what I will call a “compass situation” (see Figure 1, the top diagram). Let us examine the three Gettier cases in turn and see how this works. Firstly, Smith, Jones, and the coins. Smith believes the proposition the man who will get the job has ten coins in his pocket. This proposition is (e) of our first section. The truthmaker for this proposition is Smith himself, with the ten coins in his own pocket. But the state of affairs that the justification is ultimately grounded in is Jones with the ten coins in his pocket, combined with his (Smith’s) being told that Jones will get the job. Smith forms his belief on the basis of what he takes to be reliable testimony, and his direct acquaintance with the contents of Jones’s pocket. The belief that he forms is false — it is (d), Jones will get the job, and Jones has ten coins in his pocket. But despite (d) being false it logically implies the weaker true proposition (e). So if we assume that logical implication preserves justification of belief then Smith is justified in believing (e). But the truthmaker for Smith’s proposition is (mereologically) disjoint from the state of affairs that the justification is grounded in. Herein lies the source of its peculiarity. And indeed it is the same peculiarity that exists for all the classical Gettier counterexamples. The second counterexample was that of Smith, Brown, and Barcelona. The truthmaker for the proposition (h), either Jones owns a Ford, or Brown is in Barcelona, is that Brown is in Barcelona. But that is not the state of affairs that the justification is grounded in. That state of affairs that the justification is grounded in is that which provides the evidence for Jones owning a Ford. So the proposition that Smith believes has a justification that is grounded in a state of affairs that is disjoint from the state of affairs that is the truthmaker. And this disjointness can be increased, as we have seen, to space-like separation. Thirdly, the sheep in the field, the existential statement (l), there is a sheep in the field, is made true by the sheep behind the rock. But our evidence and justification for the statement is grounded in the prop sheep that we see and that we mistakenly believe to be a real sheep. Again these are mereologically disjoint states of affairs. We can thus see how the ambiguity of reference in the Gettier situations arises. In the first example, Smith takes himself to be referring to Jones, because that is

164

Adrian Heathcote

Figure 1: “Compass” and “ruler” states of affairs.

Truthmaking and the Gettier Problem

165

of whom he has evidence getting the job and having ten coins in his pocket. He has no such evidence about himself. But the description that he uses objectively picks out himself, not Jones. The ambiguity goes along with the split between the two states of affairs that are responsible for the problem. The second example has no such ambiguity of reference — Smith has a good reason to believe one disjunct, but it is the second disjunct that is true. So ambiguity of reference is not essential to the Gettier cases. Nevertheless it reappears in the third case: the proposition “There is a sheep in the field” can be thought to have a referential meaning (particularly if it is accompanied by pointing) and in this case it refers to the prop sheep that is seen rather than the sheep behind the rock that isn’t. In each case, then, we have disjoint states of affairs, one of which is the truthmaker for the statement, the other of which is the state of affairs which is the ground for the justification — this gives us the compass shape for these situations. Thus we can remedy the deficiency in the JTB definition of knowledge simply by insisting on the identity of the two states of affairs that play these different roles. This gives us the following: The K-conditions. X knows A iff X believes A; X is justified in believing A; A is true; and the evidence that X has which constitutes the justification is evidence of the very state of affairs that makes A true. This amendment solves the problem for knowledge that the classic Gettier counterexamples have opened up. Under normal circumstances — and this includes all three of the classic Gettier situations — justificatory evidence will stand in a causal relation to the state of affairs which constitutes its ground. However, I don’t wish to insist on the necessity of such causal relations in all situations. This is because I want to allow that mathematical knowledge can often be a matter of inference to the best explanation. (Thus when I know that there is a complex number field, I am not required to stand in a causal relationship to the complex number field — though I may of course stand in a causal relationship to a proof that certain objects form a field. My (fallible) knowledge that the complex numbers exist may come from an inference to the best explanation from the explanatory success of quantum theory.9) 9 In a sense inference to the best explanation is analogous to non-constructive proof methods like reductio ad absurdum. One finds out that something is without finding out important details about it. One could say that inference to the best explanation stands to non-demonstrative inference, as nonconstructive proof methods stand to demonstrative inference. This analogy would bear further investigation, I think. (A “transcendental argument” for the necessity of inference to the best explanation was given in Heathcote 1995.)

166

Adrian Heathcote

Missed Opportunities The idea of truthmaking comes ultimately from Russell, in 1918: “The first truism to which I wish to draw your attention is that the world contains facts… . When I speak of a fact … I mean the kind of thing that makes a proposition true or false” (1956, p. 182). However it is quite remarkable how, with very little argument and a great deal of derision, the idea has suffered in the intervening century. This has represented a substantial missed opportunity. For as we see, truthmaking is the key to unlocking the Gettier counterexamples. To have recognized this earlier would have forestalled our entry into the latest of the blind alleys that the Gettier cases have led philosophy: I refer to the idea that the problem is unsolvable because it is ill-formed. In fact, the logic of Gettier’s paper is strikingly clear, and he is perfectly correct in his demand for sufficient conditions for the concept of knowledge. But it is obvious that, as long as truth is a component of our concept of knowledge, philosophical problems with truth will carry over into philosophical problems with knowledge. We cannot expect it to be otherwise. One final point: I have argued that the classic Gettier counterexamples are solved by insisting on the identity of the relevant states of affairs. But that still leaves the Ginet–Goldman-style counterexamples. These are often grouped in with the classic counterexamples to make one large set. In my view this is a mistake; the two kinds of counterexamples are entirely different. Here is a typical Ginet–Goldman-style counterexample: You are passing a field in a car and you see a sheep and form the belief that you are seeing a sheep. But you would have formed that belief if you were not seeing a sheep but instead a cardboard prop. You certainly have a justified true belief — but do you know that you’ve seen a sheep? Alvin Goldman (1976) thinks not. Moreover, here we do not have two states of affairs playing two different roles. What to say? Here I agree with Stephen Hetherington (1998) that insisting on the fallibility of knowledge will solve these kinds of puzzles. You do know that you have seen a sheep, and correspondingly your belief entirely satisfies the K-conditions. Of course you could not be said to be certain that you’ve seen a sheep, but certainty is not a requirement of knowledge. If there had been a prop sheep in the field your belief that there is a sheep in the field would have been false after all, and so would not have counted as knowledge. But there was not a prop sheep, there was a real sheep, and so you have knowledge.10

10

It should be noted that the classic Gettier counterexamples depend upon taking knowledge to be fallible, so grouping these together with the Ginet–Goldman-type puzzles will only lead to confusion.

Truthmaking and the Gettier Problem

167

Much has been made of the fact that you yourself can’t distinguish the two states: the state that you are in as a result of seeing the real sheep and the state that you would have been in had there been only a prop sheep. So much the worse for purely subjective accounts of knowledge. To have knowledge one’s belief must be true, and this means that there must be a relation between the belief and an external state of affairs. When one sees a real sheep in the field, the sheep in the field is the truthmaker for the belief and the ground for one’s justification. In epistemology, it has long been clear that fallibilism is the only game in town (Feldman 1981). One only wishes that the same unanimity had prevailed on the question of truthmaking.

References Copi, I. M. (1979). Symbolic logic (5th ed.). New York: Macmillan. Etchemendy, J. (1988). Tarski on truth and logical consequence. Journal of Symbolic Logic, 53, 51–79. Feldman, R. (1981). Fallibilism and knowing that one knows. The Philosophical Review, 90, 266–282. Gettier, E. (1963). Is justified true belief knowledge? Analysis, 23, 121–123. Goldman, A. (1976). Discrimination and perceptual knowledge. The Journal of Philosophy, 73, 771–791. Heathcote, A. (1995). Abductive inference and invalidity. Theoria, 61, 231–260. Heathcote, A. (2003). Truthmaking and the alleged need for relevance. Logique et Analyse, 46, 345–364. Heathcote, A. (2005). Conditionals as functions (submitted manuscript). Hetherington, S. (1998). Actually knowing. The Philosophical Quarterly, 48, 453–469. Lewis, D. (1996). Elusive knowledge. Australasian Journal of Philosophy, 74, 549–567. Milne, P. (2005). Not every truth has a truthmaker. Analysis, 65, 221–224. Mulligan, K., Simons, P., & Smith, B. (1984). Truth-makers. Philosophy and Phenomenological Research, 44, 287–321. Putnam, H. (1994). A comparison of something with something else. In: J. Conant (Ed.), Words and life (pp. 330–350). Cambridge, MA: Harvard University Press. Quine, W. V .O. (1960). Word and object. Cambridge, MA: MIT Press. Read, S. (2000). Truthmakers and the disjunction thesis. Mind, 109, 67–79. Russell, B. (1956 [1918]). The philosophy of logical atomism. In: R. C. Marsh (Ed.), Logic and knowledge: Essays, 1901–1950 (pp. 175–281). London: George Allen & Unwin. Tarski, A. (1983 [1935]). Der Wahrheitsbegriff in den formalisierten Sprachen (translated as ‘The concept of truth in formalized languages’). In: J. Corcoran (Ed.), Logic, semantics, metamathematics: Papers from 1923 to 1938 (J. H. Woodger, Trans.) (2nd ed., pp. 152–278). Indianapolis: Hackett.

This page intentionally left blank

168

Chapter 11

Is Knowing Having the Right to be Sure? André Gallois

Surely not. Surely we have learned that much from Gettier. After all, Gettier established, as conclusively as anything could be established in philosophy, that knowledge is not to be analyzed as true justified belief, and A. J. Ayer’s analysis of knowing as having the right to be sure is one of the three paradigms of a truejustified-belief analysis mentioned at the beginning of Gettier’s (1963) paper. I have no wish to defend an analysis of knowledge as true justified belief. Despite that, I will defend Ayer’s analysis of knowing as having the right to be sure which, for convenience, I will refer to as the normative analysis. Hence, I do not take Ayer’s normative analysis of knowledge to be an example of a true-justified-belief analysis. Why defend the normative analysis? Suppose the normative analysis turns out not to be vulnerable to any counterexample so far proposed to any other analysis. Some would take that as a good enough reason to resuscitate the normative analysis. At this stage in the debate over the analysis of knowledge, I would not make the same assumption. So many analyses of knowledge that were able to cope with the then extant Gettier style counterexamples have fallen to subsequent ones. Some would conclude that we have, at least, strong inductive grounds that no Gettier-proof analysis of knowledge can be given. Whether or not that is so, there is reason to refrain from inflicting yet another analysis of knowledge on the philosophical public, unless the normative analysis has something special going for it. I will argue that the normative analysis does have something special going for it. While it is not conclusive, a strong argument can be given to show that the normative analysis is not vulnerable to Gettier counterexamples. If that argument succeeds, we have reason to repose confidence in the normative analysis despite the pessimistic induction generated by the failure of previous ones. Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

169

170

André Gallois

We might think that any analysis of knowledge worth taking seriously will come with an argument for its invulnerability to Gettierisation. The argument goes like this. If an analysis of knowledge is correct, it will not be open to Gettier counterexamples. Any analysis worth its salt will be able to, at least, cope with existing counterexamples. Its ability to do so constitutes an argument in favor of that analysis being correct. So, the ability of an analysis to cope with existing counterexamples constitutes an argument for an analysis being invulnerable to future counterexamples. Perhaps there is such an argument accompanying any analysis of knowledge that is not a non-starter. The point is that there is an additional, and stronger, argument for the imperviousness of the normative analysis to Gettierisation. Even if that argument fails, it will be beneficial to articulate and examine it. The analysis of knowledge has become something of a test case for analyses in general. The long history of failed analyses of knowledge has led some to believe that the project of giving philosophical analyses is misguided. It is premature to agree with them without having a better understanding of what a philosophical analysis should be expected to accomplish. Examining the reason for thinking that the normative analysis has a distinctive protection from Gettier counterexamples will, hopefully, promote that understanding. I will proceed as follows. First, after restating it, I will say something about the implications of the normative analysis. I will then say why there is reason to think that the normative analysis is invulnerable to Gettier style counterexamples. After responding to objections, I will comment on what the normative analysis may tell us about the enterprise of analysis.

1. I propose to defend the following analyses of knowledge: A subject S knows that P if and only if: (1) S has the right to be sure that P. The normative analysis is a descendant, but not identical with, the analysis of knowledge defended by A. J. Ayer in The Problem of Knowledge (1956, Chapter 1), which goes: A subject S knows that P if and only if: (1) S has the right to be sure that P; (2) S is sure that P; and (3) it is true that P. Ayer takes (1) to be necessary for knowing that P. Why take (1) to be both necessary and sufficient for knowing that P? Let us begin with the reason for omitting (3). There is near universal agreement that (3) is a necessary condition for

Is Knowing Having the Right to be Sure?

171

knowing. The reason for omitting (3) foreshadows the argument in the next section that the normative analysis is invulnerable to Gettier counterexamples. It is this. There are different ways I can be deprived of the right to be sure that P. One is because it is not true that P. If so, then (3) is implied by (1). So, it would be redundant to add (3) as a necessary condition for knowing alongside (1). (2) is being omitted for the opposite reason. (2) is not being omitted because it is redundantly implied by (1). On the contrary, (2) is being omitted because it is not implied by (1). One can, of course, have the right to something without having it. So, one can have the right to be sure that P without being sure that P. Moreover, it is, I suggest, a merit of the normative analysis that it does not take knowledge to require certainty, if only for the following reason. There are a number of plausible arguments, familiar from the literature, to the effect that knowledge does not imply certainty. For example, the unconfident examinee may fail to be certain, but yet know, that the Armada sailed in 1588. Some have taken this type of case to show that knowledge does not imply belief. Some have even argued that knowledge is inconsistent with belief. Does the normative analysis pronounce on either of these claims about the relation between belief and knowledge? So far as I can see, it does not. Consider the claim that knowing P implies believing P. If one can have the right to be sure that P without being sure that P, what is to prevent having the right to be sure that P without even believing that P? Perhaps something does, but, if so, that needs to be shown.

2. Later I will consider whether (1), S has the right to be sure that P, provides a necessary condition for knowledge. Let us first consider whether (1) provides a sufficient condition. One can see why in his famous paper Edmund Gettier viewed Ayer’s analysis of knowledge as a version of a true-justified-belief analysis in which being sure that P replaces believing that P. It is tempting to equate having the right to be sure with having a sufficient justification for believing. Despite that, some, including Ayer himself, have resisted that temptation. In his book on Ayer, John Foster (1985, part II) makes a convincing case for Ayer having endorsed the following. A necessary condition for S having the right to be sure that P is that S forms her belief in a truth-reliable way. That is, S has the right to be sure that P only if S’s belief that P is formed as a result of a process that reliably results in the formation of true beliefs.1 Should we agree with Ayer that 1

Foster is inclined to think that Ayer takes the reliability condition to be sufficient for knowledge. If he is right, Ayer would not have seen his analysis of knowledge as Gettier-proof.

172

André Gallois

having the right to be sure that P implies forming the belief that P in a truth reliable way? Of course, the answer to this question depends on the answer to the following one. What is it to have an epistemic, as opposed to a moral, legal, etc. right to be sure? In order to answer this last question, let us address a different one. What is it to have a right of any kind? Consider having a legal right: say the legal, as opposed to moral, right to give away all one’s money to support some absurd cause. What are the implications of having such a right? Here is one. If you have the legal right to give away all your money there is no consideration that would make it legally permissible for someone, without your consent, to prevent you from doing so. Of course, there may be some other consideration that would make it morally, aesthetically, or practically permissible for someone to prevent you from giving away your money. It is just that if the right you have to give away your money is a legal right, no such consideration could make it legally permissible to stop you. The concept of justification is, arguably, a normative concept. Suppose it is. In that case, if, as the true-justified-belief analysis invites us to, we wish to partially analyze knowledge in terms of justification, the type of justification at issue will have to be epistemic rather than legal, practical, or moral. Likewise, if we wish to analyze knowledge in terms of having a right, it must be an epistemic as opposed to a legal, practical, moral, or some other kind of right. What results when we combine this observation about the kind of right involved in analyzing knowledge as having the right to be sure with the above necessary condition for having a right? What results is this. Someone has an epistemic right to be sure that P only if there is no consideration that would make it epistemically permissible to prevent her from being sure that P. Let us now reconsider what is, according to Foster, Ayer’s claim that having the right to be sure that P implies forming one’s belief that P in a truth-reliable way: that is, in a way causally conducive to forming true beliefs. What would justify this claim? First, recall that the right, which is the subject of Ayer’s claim is an epistemic right. So, we can say the following. Someone has an epistemic right to be sure that P only if there is no consideration that would make it epistemically permissible to prevent her from being sure that P. Suppose Jayne has formed the belief that P in a way that is not truth-reliable. Suppose she has come to believe that P as a result of consulting her clairvoyant. In addition, suppose that Jayne reposes so much confidence in her clairvoyant that, unless Sally prevents her, she will be sure that P. Is there any consideration that would make it epistemically permissible for Sally to prevent Jayne from being sure that P? There is if Jayne has formed her belief that P in a way that fails to be truth-reliable. The fact that Jayne has formed her belief in such a way makes it epistemically permissible for Sally to prevent Jayne from being sure that P. We may say that from an epistemic, rather than legal, aesthetic, practical, or

Is Knowing Having the Right to be Sure?

173

moral, point of view it is permissible for Sally to prevent Jayne from being sure that P. Here is an argument for this last claim. It is not epistemically permissible for Sally to prevent Jayne from being sure that P only if it would be wrong, again from an epistemic point of view, for Sally to do so. Suppose we learn that Sally has prevented Jayne from being sure that P, and accuse her of epistemic wrongdoing. Surely, Sally could adequately meet our charge by pointing out that Jayne has come to believe P in a way that is not truth-conducive. Failure to form one’s belief in a truth-conducive way is one factor undermining knowledge. Consider other Gettierising factors that prevent one from knowing. Suppose Jayne has formed her belief that P in a causally impeccable way, but her justification for believing P is subject to a knowledge-undermining defeater. Suppose that Jayne has come to correctly believe there is a barn in front of her by seeing one in an area saturated by fake barns any one of which she could just as easily have seen. Call such an area a fake barn area. This time Sally prevents Jayne from being sure that there is a barn in front of her, and, again, is charged with epistemic wrongdoing. As in the previous example, Sally has the means to rebut the charge. She needs to only point out that, though Sally sees a barn in broad daylight, etc. she does so in a fake barn area. Hence, the fact that Jayne has formed her belief in a fake barn area is a consideration that renders it epistemically permissible for Sally to prevent Jayne from being sure that her belief is true. Call features of possible situations that enable those situations to serve as Gettier-style counter-examples to different analyses of knowledge Gettier features. Gettier features would include: being accidentally right, basing one’s belief on a false lemma, forming a belief that fails to track the truth, having a justification open to the right kind of defeater. Here is a claim I would make about extant Gettier features. All Gettier features so far mentioned in the literature have this in common. That a situation displays such a feature makes it permissible to prevent someone in that situation being sure that a suitably related belief is true. If this last claim is correct, what is the best explanation for it being so? I suggest the following. A situation having any P-related Gettier feature, including any that have yet to be discovered, supplies a consideration that makes it permissible to prevent someone in that situation from being sure that P. If so, it follows that if one is in a situation in which one has the right to be sure that P, then that situation manifests no P-related Gettier feature.

3. I do not expect this argument for the normative analysis to immediately convince. A number of objections can be raised to it of which the most serious, I believe, is

174

André Gallois

this. The normative analysis’s alleged strength is, in fact, its weakness. Suppose we concede that the normative analysis has the resources to automatically deflect any counterexample to it. All that means is that the normative analysis incorporates a condition that allows it to, in an illegitimately ad hoc fashion, rebut any counterexample. Moreover, the incorporation of such a condition renders the normative analysis vacuous. Call this the vacuity objection. The vacuity objection is closely related to a further one. It goes like this. I said that when a person’s justification for believing P depends on her holding a false belief, she does not have the right to be sure that P. Is that so? Someone might respond: we cannot say whether it is so until more is said about what it is to have the right to be sure. In the absence of that, confronted with a putative counterexample to the normative analysis, we simply cannot tell whether it succeeds. Call this the lack of determination argument. The vacuity and lack of determination arguments are closely related. Despite that, they have conflicting conclusions. According to the vacuity argument, what is wrong with the normative analysis is that it has a built-in guarantee against counterexamples. According to the lack of determination argument we cannot tell whether a putative counter-example to the normative analysis is a genuine counterexample or not. First, consider the vacuity argument. Is the normative analysis vacuously invulnerable to Gettierisation? Is it simply a matter of stipulation that any Gettier feature rules out having the right to be sure? The proper response to this objection is the following. Recall the argument for the normative analysis’s imperviousness to Gettierisation. It went like this. What extant Gettier features have in common is that their instantiation undermines having the right to be sure. The best explanation for that being so is that Gettier features in general undermine the right to be sure. This argument rests on two crucial premises. First, that Gettier features specified in the literature so far do undermine having the right to be sure. Second, the best explanation for this is that a situation instancing a property that makes it a Gettier feature tells against having the right to be sure in that situation. Both these premises are obviously open to refutation. For example, the first would be refuted if someone specified a Gettier feature that does not conflict with having the right to be sure. Hence, it is far from vacuously true that the normative analysis is invulnerable to Gettierisation. What of the lack of determination objection? It goes: without knowing more about what it is to have the right to be sure, we simply cannot tell whether a situation instancing a given Gettier feature conflicts with the relevant individual in that situation having that right. The objection rules out too much, and demands too much. It tells against giving a normative analysis of any epistemic concept. Consider a deontological analysis

Is Knowing Having the Right to be Sure?

175

of the concept of justification. Suppose it goes: X is justified in believing that P if and only if X ought to believe that P. Many objections could be raised to this analysis. But, surely, the following is not a good one. In stating the analysis “ought to believe” must be construed epistemically. If it is, we simply cannot tell whether someone, in the relevant sense, ought to believe something. If this objection against the above deontological analysis of justification is misguided, the counterpart objection to the normative analysis is equally so. In addition, the lack of determination objection is too demanding. It raises the complaint that we cannot tell whether the analysans of the normative analysis applies in a given situation. But, that we cannot tell whether the analysans of an analysis applies is, by itself, no objection to that analysis. It may be that we cannot tell whether the analysandum applies. In any case, we can tell, in a wide variety of situations, whether someone who believes something has the right to be sure that it is true. Someone who believes that P only because they want P to be true usually does not have the right to be sure that P is true. Moreover, we can, in many cases, tell whether a situation having a Gettier feature conflicts with having the right to be sure. For example, in the original Gettier counterexamples to the true justified belief analysis we can tell that the individual described does not have the right to be sure that, say, what they believe about car ownership is true. Still, it might be said, this misses the force of the lack of determination objection. The problem is that we can tell whether someone knows something in a situation in which we cannot tell whether that person has the right to be sure. In certain cases we can tell that the instantiation of a Gettier feature rules out knowledge in a case where we cannot tell whether the instantiation of the same feature rules out having the right to be sure. I will respond to this variant of the lack of determination objection by invoking a comparison with a similar objection to a well-known analysis of modal concepts. Call a possible world as a modal realist such as Lewis conceives of it a Lewis world. Now, consider the following modal claim that Lewis would analyze by appealing to Lewis worlds: there might have been a talking donkey. The analysis, call it the Lewis analysis, goes: there might have been a talking donkey if and only if there exists a talking donkey in some Lewis world. Given that there are no actual talking donkeys, how can we tell, runs the objection, that there is a Lewis world containing one? I take the proper response to this objection to be essentially the one given by Lewis. Either we have a sufficient reason to believe in the Lewis analysis, or we do not. Suppose we have no sufficient reason to believe the Lewis analysis. In that case, it does not count against the truth of that analysis that we can tell whether there could have been a talking donkey, but cannot tell whether there is a Lewis world containing one.

176

André Gallois

Suppose that there is sufficient reason to believe the Lewis analysis. In that case, we can tell whether there is a Lewis world inhabited by a talking donkey. We can, let us allow, tell that it is possible for there to have been a talking donkey. We have, we are supposing, sufficient reason to believe that it is possible for there to have been a talking donkey only if there is one in some Lewis world. So, we can tell that there is a talking donkey in some Lewis world. What goes for the Lewis analysis of modality goes for the normative analysis of knowledge. Can we tell whether, in a Gettier situation, someone has the right to be sure that P? Suppose we can tell that the relevant individual in that situation does not know that P. Again, there are two cases to consider. First, we lack any sufficient reason to believe the normative analysis. As with the Lewis analysis, in that case it counts not at all against the normative analysis that we cannot tell whether the relevant individual has the right to be sure in the situation in question. Second, we do have sufficient reason to believe the normative analysis. In that case, since we can tell that there is absence of knowledge in the situation under consideration, we can tell that there is no right to be sure in the same situation. One objection to the normative analysis goes like this. The analysans of the normative analysis involves the concepts of having a right and being sure. Both these concepts stand in as much need of philosophical analysis as the concept of knowledge. So, we should not give an analysis of the last concept in terms of the former two. As with the previous one, this objection rules out too much. Consider Goldman’s (1967) analysis of factual knowledge, which utilizes the concept of causation. There are a number of objections that can be raised against Goldman’s causal analysis. However, it is no objection to a causal analysis that the concept of cause stands in as much need of analysis as the concept of knowing something. If it were, almost any analysis of knowledge would be ruled out including the original true-justified-belief analysis. A third objection goes like this. Let us allow that it is no objection to the proposed analysis that nothing more has been said about what it is to have the right to be sure. That, by itself, is no objection to the normative analysis. At most it indicates that further work needs to be done. The objection to the normative analysis is that nothing more can informatively be said about what it is to have the right to be sure. The last claim is just false. It is commonly said that rights divide into liberty and claim rights. In turn, claim rights are customarily divided up into positive and negative claim rights. One is said to have a positive claim right to something if others have a duty to assist one in getting it. One has a negative claim right to something if others have a duty not to prevent one from getting, or having, it. Here is an issue that certainly deserves further investigation. Is an epistemic right to be sure a liberty or a claim right? If it is a claim right, is it a positive or

Is Knowing Having the Right to be Sure?

177

negative claim right? Suppose, as I would be inclined, we decide that the right to be sure is a negative claim right. In that case, we can certainly say more about what that right consists in. One has an epistemic right to be sure that P if and only if others have an epistemic duty not to prevent one from being sure that P. This, it might be said, underestimates the force of the above objection. Agreed, we can hope to say something using normatively loaded language about what it is to have the right to be sure. What we cannot do is to say something more about what it is to have the right to be sure by confining ourselves to the use of nonnormative vocabulary. It all depends. Can we give an analysis of having a right in non-normative terms? If we can, there is no reason to think that we are precluded from spelling out the analysans of the normative analysis in non-normative terms. If we cannot give a non-normative analysis of having the right to be sure, it does not follow that we cannot non-normatively specify the facts on which having the right to be sure supervenes. Having the right to be sure supervenes on non-normative facts about the holding of relevant counterfactuals, non-accidentally being correct, the holding of a causal connection between a belief and what is believed. Here is a different objection to the normative analysis. Compare the normative analysis with a true-justified-belief analysis. We can break up the analysans of the true-justified-belief analysis into truth, belief, and justified belief. It is far from clear that we can do likewise with the normative analysis. Agreed, it is far from clear that we can decompose the analysans of the normative analysis as we can the analysans of a number of other candidate analyses of knowledge. However, it is not obvious why this counts against the normative analysis. Consider Unger’s (1968) analysis of knowledge as non-accidentally acquired true belief. The analysans of Unger’s analysis appears no more open to decomposition than its normative counterpart. Here is what I take to be one of the more serious objections to the normative analysis. Why should we analyze knowing as having the right to be sure rather than as having, say, the right to believe?2 Here, I hope, is, at least the beginning of an answer to this objection. What confers a right to something? One answer to this question can be given by appealing to the consequences of individuals in general exercising such a right. If the consequences of most individuals exercising it are on balance, good then, and only then, is the right possessed by any given individual. Call this Rights Consequentialism. Is Rights Consequentialism a good way to account for the possession of moral rights? I am not sure. However that may be, Rights Consequentialism is more

2

I am indebted to Chris Daley for raising this objection.

178

André Gallois

clearly applicable to epistemic rights. If we want to know whether an individual has a given epistemic right, we need to know whether its general exercise would, from an epistemic point of view, be beneficial. The question before us is this. Why is knowing having the right to be sure rather than the right to believe? Here is the main reason for identifying it with having the right to be sure. For an individual to believe that P in a situation free from Gettier features, she must have the right to be sure that P. Moreover, in order to have the right to be sure, she must be in a situation free of Gettier features. So, does an individual have the right to be sure only in situations free of Gettier features? In order to answer that question, we need to answer the following one. Is it, on the whole, beneficial for individuals to exercise the relevant right only in situations free of Gettier features? Whatever the answer to this last question may be, let us ask the counterpart one about having the right to believe. Call situations free of Gettier features Gettierfree situations. Would individuals exercising the right to believe only in Gettierfree situations have, for the most part, beneficial consequences? Surely, the answer is no. An individual will be completely justified in forming a belief in many situations that are not Gettier-free. So, if individuals exercise their right to believe only in Gettier-free situations, they will, very often, fail to form completely justified beliefs. Not a good consequence from an epistemic point of view. Hence, if we accept Rights Consequentialism, a compelling case can be made for denying that someone has the right to believe only in a situation that is Gettier-free. But, knowing requires being a Gettier-free situation. So, having the right to believe does not imply knowing. Here is a further objection to the normative analysis. Let us allow that knowledge is a state: perhaps, as Tim Williamson argues, a mental state. A right to be sure is not a state. So, how can we identify knowledge with the right to be sure? There is a quick and obvious reply to this objection that may not do it sufficient justice. If any identification of states is in question, it is not an identification of knowing with the right to be sure. Instead, it is an identification of the state of having knowledge with the state of having the right to be sure. This reply may miss the point of the objection. Reformulated, it runs like this. It is not just having knowledge, which is a state, but knowing itself. Even if, in some broad sense, having a right is a state, it is not the same kind of state as having the right to be sure. For example, though contestable, it is not implausible to take knowing to be a state of mind. It is completely implausible to take having the right to be sure, as opposed to being sure, as a state of mind.3

3

Tim Williamson (2000, Chapter 1) defends the view in his chapter that knowledge is a state of mind.

Is Knowing Having the Right to be Sure?

179

We should recall that the normative analysis is an analysis. If we analyze what it is for there to be something of kind k in terms of there being things of kind k*, it does not follow that things of kind k are even of the same kind as things of kind k*. For better or worse, a phenomenalist may attempt to analyze statements about physical objects as statements about sensory experiences. It does not follow that the phenomenalist is committed to identifying physical objects with sensory experiences. Likewise, if an advocate of the normative analysis analyzes statements about knowledge via statements about epistemic rights, it does not follow that she is committed to identifying a state of knowing with a state of having a right. Does the normative analysis collapse into some more familiar one such as a virtue theoretic or reliabilist analysis of knowledge? The main conclusion defended in this paper is that the normative analysis enjoys a prima facie immunity from Gettierisation. That prima facie immunity alone distinguishes the normative analysis from any other we might be tempted to equate it with. Finally, let me say something about the implications of the normative analysis for skepticism about knowledge. It does not decisively count against an analysis of knowledge if it implies skepticism about knowledge. Still, it is a strike against such an analysis if it does imply skepticism about knowledge. Does the normative analysis have that implication? To see that it does not, consider just one way of attempting to defuse skepticism about knowledge. Ignoring refinements, an anti-skeptical contextualist about knowledge will say this.4 The standard that has to be met for an attribution of knowledge to be correct will vary with the context of attribution. In a context where skepticism is at issue an attribution of knowledge may be out of place, which is not out of place in some other context. So, contra the skeptic, attributions of knowledge may be correct in many, perhaps most, contexts. Notice that an advocate of the normative analysis can simply take over this contextualist reply to skepticism. Does Sadie know that P? If we are in a context where skepticism is at issue, the answer is no. Does Sadie have the right to be sure that P? Again, if we are in a context where skepticism is at issue, the answer will be no. Suppose we are not in a context where skepticism is at issue. In that case, we may very well be right to attribute to Sadie knowledge that P. But, in the same nonskeptical context, we may also be right to attribute to Sadie knowledge that P. So far as I can see there is no standard response to skepticism about knowledge that an advocate of the normative analysis cannot take over. What this means is that the normative analysis is neutral on the issue of skepticism. Such neutrality

4

For an exposition and defense of the contextualist response to skepticism see Keith DeRose (1995).

180

André Gallois

is not a black mark against the normative analysis. Indeed, some would say it speaks in favor of that analysis.

4. Does the normative analysis supply a necessary, as well as sufficient, condition for knowledge? Suppose it does not. In that case, for some P, it is possible for someone to know that P, but lack the right to be sure that P. How could that be so? We can immediately rule out one way of its being so. It could not be that an individual who knows that P lacks the right to be sure that P because she is in a situation with a P-related Gettier feature. If she is in such a situation, she would fail to know that P. Could there be some factor other than the presence of a Gettier feature that undermines having the right to be sure that P without undermining knowing that P? For the following reason, it is unlikely we shall find one. The property of having the right to be sure is a normative property. Compare that normative property with the psychological one of being sure that P. Someone may lack the psychological property of being sure that P for all sorts of reasons that do not, or, at least, do not obviously, tell against knowing that P. Someone may fail to be sure that P because they are generally unconfident about their opinions in a certain area. If that person’s lack of confidence is unwarranted, it does not follow that she lacks knowledge that P. But, her lack of confidence does not count against her having the right to be sure that P, unless it is warranted. Moreover, if it is warranted then she does fail to know that P. Any feature that counts against having the right to be sure that P will have to be a normatively relevant feature. In order to show that having the right to be sure that P is not a necessary condition for knowing that P, we need to find, where the norms are epistemic, a normatively relevant feature that counts against having the right to be sure that P without counting against knowing that P. Perhaps, one can be found. But, again, the onus is on the opponent of the normative analysis to find one. This last observation highlights an earlier observation about the normative analysis. It highlights the observation that, despite being Gettier-invulnerable, the normative analysis is not made trivially true by stipulation. If it were, there would be no point in looking for counterexamples to it as a necessary condition for knowing. Despite all of the preceding arguments for the normative analysis, it may seem too insubstantial to qualify as an informative analysis. On this view the normative analysis’s prima facie imperviousness to Gettierisation is bought at too high a price. It is bought at the price of the normative analysis, even if correct, telling us what an analysis of knowledge should tell us.

Is Knowing Having the Right to be Sure?

181

Even if this suspicion is borne out, the normative analysis may still perform the following function. Considering it gives urgency to the following question. What differentiates the normative analysis from one that is substantive? Be that as it may, it is ironic that, in Gettier’s justly celebrated paper, Ayer was taken to be proposing an analysis of knowledge vulnerable to the original Gettier counterexamples whereas, if I am right, he could equally well have been taken to be advancing one that is Gettier-proof.

References Ayer, A. J. (1956). The problem of knowledge. London: Macmillan. DeRose, K. (1995). Solving the skeptical problem. The Philosophical Review, 104, 1–52. Foster, J. (1985). Ayer. London: Routledge & Kegan Paul. Gettier, E. L. (1963). Is justified true belief knowledge? Analysis, 23, 121–123. Goldman, A. I. (1967). A causal theory of knowing. The Journal of Philosophy, 64, 357–372. Unger, P. (1968). An analysis of factual knowledge. The Journal of Philosophy, 65, 157–170. Williamson, T. (2000). Knowledge and its limits. Oxford: Clarendon Press.

This page intentionally left blank

182

Chapter 12

Knowledge by Intention? On the Possibility of Agent’s Knowledge Anne Newstead

1. Introduction Action theorists are typically interested in those actions done with, or in some cases preceded by, a certain kind of knowledge. Let us call such knowledge “agent’s knowledge” to signify that it is an agent’s knowledge of her own action.1 Famously, in Intention Anscombe claimed that (AK) Agents have non-observational knowledge of their own intentional actions.2 How is (AK) to be understood? Consider the simple example of a toddler drawing a figure on paper. The toddler may answer the question “What are you doing?”

1

Anscombe’s term for such knowledge in Intention is “practical knowledge”. However, I have avoided that term as I wish to defend the idea that agents have a distinctive non-observational knowledge of their actions without being committed to her claims about the existence of a kind of knowledge with a reverse direction of fit from ordinary knowledge. 2 The idea originates in English speaking philosophy with Wittgenstein’s qualified claim that “in a large class of cases it is the impossibility of taking an observant attitude towards a certain action that characterises it as a voluntary one” (Wittgenstein, 1958, p. 153). Anscombe makes a stronger claim with regard to intentional actions, that “the class of things known without observation is of great interest because of the class of intentional actions is a sub-class of it” (Anscombe, 2000 [1958], p. 14).

Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

183

184

Anne Newstead

by explaining, “I am drawing a house”. Presumably, the toddler knows what she is doing not by looking at the figure on the paper, which may indeed barely resemble a house. Rather, she knows what she is doing in virtue of knowing her intention and trusting that it is being implemented. Though there may be other possible descriptions of her action, she counts as knowing what she is doing so long as she is aware of her action under some description that fits it. Is (AK) true? How far can we go in defending (AK)? Many theorists have thought that (AK) is obviously false or else in need of some important restrictions. Searle (1983, p. 90) agrees that “at any given point in a man’s conscious life he knows without observation the answer to the question ‘what are you now doing?’ ” However, he allows that knowing the answer to the question “what are you now doing?” may consist simply in knowing without observation what one is trying to do. To see what one is actually doing, however, requires observation in Searle’s view. Unless we identify tryings with actions, this falls short of the thesis that agents know their intentional actions non-observationally. More radical opposition to (AK) stems from opposition to the idea of privileged self-knowledge generally. Gopnik (1993) follows Ryle in arguing that the idea that subjects have privileged knowledge of their intentional states is an illusion. Theorists influenced by this line of thought will claim that agents can only know about their intentional actions in the same way as everyone else: by observation. In order to account for the qualitative difference between an agent’s knowledge of her action and observer’s, we can say that the agent has more data (as Gopnik does) or perhaps different data — such as sensory feedback from one’s own body (as Pickard (2004) does) — than observers do. Despite the objections, there is a stubborn intuition that counts in favour of (AK). The idea is that so long as one is acting intentionally, one will (potentially) have access to one’s intention and interpret one’s action in the light of that intention. To view an action as a mere observer would be to view it blankly, without awareness of the intention it fulfils. It would then become questionable to what extent the movement constituted one’s own intentional action at all. Movements that one undertakes but cannot interpret as fulfilling some intention of one’s own are not one’s intentional actions, but mere happenings. So it does seem that agents must be able to become aware of their own intentional actions in a way that differs from being mere observers.3 Obviously, a defence of (AK) requires more than an appeal to intuition. Imagine someone who accepts that agents have some non-observational awareness of their intentional actions, but disputes that this awareness amounts to 3

This should not be misinterpreted as the stronger, false claim that agents cannot in principle observe what they are doing. The point is just that an agent’s awareness of her action is qualitatively different from (and certainly more first personal in character) than that of a mere observer.

Knowledge by Intention? On the Possibility of Agent’s Knowledge

185

knowledge. This is the kind of sceptic whose position I address in this essay. Whereas the debate in action theory has by and large focused on the issue of whether an agent’s awareness of her action is based on observation or not, I will focus on the issue of whether such non-observational awareness could be knowledge. To see just how far the defence of (AK) can go, I will construct two models of agent’s knowledge: a causal-reliabilist model (presented in Section 2) and a teleological, reasonsbased model (presented in Section 3). Both are undeniable cases where an agent is apparently aware of her action without observation. My contention will be that these cases are also — in the right circumstances — cases of knowledge. It turns out that progress in epistemology — in the form of an appreciation of the fallibilist response to scepticism — allows us to defend (AK) from some significant objections.4

2. A Causal Approach Can the causal theory of knowledge be extended to knowledge of action? In the case of action, the object of knowledge is that one is ␸-ing (in the present case) or that one is going to ␸ (in the future case), where “␸” represents some suitably basic description of one’s action.5 In the present case, the causal approach yields: (CTAK-present) The causal theory of agent’s knowledge for the present case. S knows that she is ␸-ing just in case: (i) S is ␸-ing; (ii) S believes she is ␸-ing; and (iii) S’s belief that she is ␸-ing is caused in the right way by the fact that she is ␸-ing. What counts as the right way for S to acquire her belief that she is ␸-ing? S’s belief that she is ␸-ing must be “directly” (non-deviantly) caused by her action of ␸-ing. Presumably, if as on (CTAK-present), the action of ␸-ing is the direct cause of the subject’s belief that she is ␸-ing, then the agent’s belief and knowledge comes after her action.6 But the kinds of actions we are interested in are ones in which such knowledge comes before completion of the action. In Anscombe’s favourite example, 4 I am indebted to Stephen Hetherington for helping me to see how an epistemologist’s toolbox (and especially fallibilism) might be fruitfully brought to bear on an issue in action theory. 5 The basic description of the action should present the action as occurring early in the causal genesis of the action, but it is also important that it not be a recondite description of the action in terms of inner mechanisms. Rather, a “suitably basic” description of the action will describe the action in terms recognisable by the agent as something that she knows how to do directly without having to do anything else first (e.g. “raising my arm”). This may mean that a suitably basic description is “teleologically basic” but not “causally basic”. On the distinction, see Hornsby (1980, Chapters 5 and 6). 6 That is, we generally assume that effects are temporally posterior to their causes. This assumption has been challenged by believers in immanent causation. For example, a cloth may bleed its dye into

186

Anne Newstead

someone knows what she is doing (e.g. writing a certain word) before she has completed the action. Furthermore, the delay between performance of an action and the receipt of sensory feedback from the action of some 50–250 ms makes it unlikely that such knowledge of the action is derived from feedback from the actual performance of the action.7 This suggests we amend our theory as follows: (CTAK*-present). S knows that she is ␸-ing just in case: (i) S is ␸ing; (ii) S believes she is ␸-ing; (iii*) S’s belief that she is ␸-ing is caused by her intention of ␸-ing. This approach will restore the anticipatory character of agent’s knowledge provided we can assume that the agent’s action is caused by a prior intention to perform that action. However, in some cases, the intention of ␸-ing can merely be embodied in the action of ␸-ing without being preceded by a conscious prior intention to ␸.8 Nonetheless, we do think that an agent knows what she is doing in such a case. Where there is apparently no (conscious) prior intention to ␸, the agent’s knowledge that she is ␸-ing will be roughly contemporaneous with her performance of the action. Our formulation of (CTAK*-present) should leave it open whether the agent’s knowledge reflects just an intention-in-action or also a prior intention. Figure 1 present a possible configuration of the temporal and causal relations between intentions, actions, and knowledge consistent with (CTAK*-present). The optional part of the figure is the agent’s prior intention to ␸ and the agent’s anticipatory knowledge. In those cases where the prior intention is present, the agent will also have some anticipatory warning that she is going to ␸ before she actually does. Nonetheless, one might say her knowledge that she is ␸-ing is still derived from her intention-in-action, because only once the intention is embodied in action can she know she is performing the intentional action.9 some boiling water, with the bleeding of the cloth being the immanent cause of the discoloration of the water. For a full discussion, see Newton-Smith (1980). Even if causes and effects can be simultaneous, however, the anticipatory character of agent’s knowledge would not be restored. 7 The estimation of 50–250 ms is from Trevarthen (1984, p. 224). Could such anticipatory knowledge be based on a prediction about what will happen, using a feed-forward model? Yes, it could and this account would be compatible with the one offered here. 8 On the distinction between prior intentions and intentions-in-action, see Searle (1983). The existence of actions without prior intentions is one reason Anscombe advances for rejecting the causal analysis of intentional action in Anscombe (1983). The problem of causally deviant chains is another problem for the causal theory of action, but see Goldman’s (1976) solution. 9 The assumption that the link between the intention-in-action and the action is causal will bother some theorists, since efficient causes precede their effects. It may well be that the nature of the link is not that of an efficient causal one. This difficulty might persuade some to abandon a purely causal approach in favour of a reliable one, on which intentions-in-action are reliably correlated with (but need not cause) corresponding actions.

Knowledge by Intention? On the Possibility of Agent’s Knowledge Anticipatory knowledge

Agent’s knowledge that she is ϕ-ing

Agent’s prior intention to ϕ

Agent’s intention of ϕ-ing

187

Agent’s action of ϕ-ing

Figure 1: The causal theory of agent’s knowledge, present case. (The arrows indicate direct causal links.) In the future case, matters are different. Knowing what one is going to do in the future requires having a prior intention to ␸ before one actually does ␸, where this prior intention is an efficient cause of the action of ␸-ing. So in the future case: (CTAK*-future). S knows (at time t) that she is going to ␸ just in case: (i) S is in fact going to ␸ at some later time t* (later than t); (ii) S believes (at t) she is going to ␸ and (iii*) S’s belief that she is going to ␸ is caused by her intention to ␸ at some point in the future. The future case raises deep issues concerning the philosophy of time, specifically whether there can be a fact of the matter as to what someone is going to do if they have not yet done it. Proponents of an “open future” view will reject the assumption. But this leads to the counterintuitive consequence that one can never know what one is going to do until one actually starts doing it. So suppose there can be facts about what one is going to do. Even so S’s knowledge of what she is going to do looks incorrigible, since if she does not end up ␸-ing in the future, she can nevertheless say that she was going to ␸ until she changed her mind, or something intervened. This contrasts with knowledge of what one is currently doing, which can be undermined by pointing out that, although the agent believes she is ␸-ing and intends to be ␸-ing, she is not actually ␸-ing. In the future case, the intention to act precedes both the knowledge and the action it describes. This relationship is presented in Figure 2. The crucial element common to both (CTAK*-present) and (CTAK*-future) is that they place an intention as the common cause of both the agent’s knowledge and her action. In the present case, the intention-in-action causes both the knowledge and the action. In the future case, the prior intention causes both the foreknowledge and the action. This way of viewing matters gives a new sense to Anscombe’s claim (2000 [1958], p. 87) that an agent’s knowledge of what she is doing, now understood as knowledge of intention, is “the cause of what it understands”. Does the amendment also restore the anticipatory character claimed for agent’s knowledge? As we saw, it

188

Anne Newstead Agent’s knowledge: I am going to ϕ

Agent’s intention to ϕ

Agent’s action of ϕ-ing

Figure 2: The causal theory of agent’s knowledge (future case). preserves the sense in which agent’s knowledge is anticipatory only in those cases where the intentional action is preceded by a conscious prior intention. 2.1. Objections to the Causal Approach The causal theory makes several important assumptions that will strike some epistemologists as objectionable. First, the causal theory as stated is unabashedly externalist. Second, the theory requires a belief in the reality of mental causation. Third, the causal theory is in no way sceptic-proof. Let me consider these objections and reply to them in turn. The causal theory is externalist. It allows that an agent can have a true belief that she is ␸-ing, but it says nothing about the agent’s reasons for this belief. Instead the causal theory concentrates on the aetiology of the belief: it is somehow caused by the agent’s having the intention to ␸.10 But it does not require that the agent is conscious of an intention to ␸ as a reason for believing that she is ␸-ing (or is going to ␸ shortly). Therefore the causal theory allows that the agent may lack an understanding of why her belief is true. For epistemologists of an internalist bent, this lack of understanding on the subject’s part may be enough to disqualify the subject from having knowledge. Though the causal view presented here is externalist, there is no reason why an internalist cannot add requirements to the causal theory. For the present case, an internalist could simply add the requirement that S must be aware, or capable of becoming aware, that she has an intention-in-action of ␸-ing. Awareness of the intention does provide her with some evidence that she is ␸-ing, since ceteris paribus she would not have that impression if she were not actually ␸-ing. However, in order to really appreciate the evidence, she would have to reflect that intentions-in-action of ␸-ing are usually accompanied by actions of ␸-ing. This level of reflection might only be appropriate if the agent wishes to achieve a theoretical understanding of

10

Here, for simplicity, I have assumed that there is a prior intention causing the action. I think generally there will be a prior intention causing the action, whether conscious or not.

Knowledge by Intention? On the Possibility of Agent’s Knowledge

189

her knowledge. For the purpose of just having a rational belief that one is ␸-ing, awareness of the intention of ␸-ing may be enough. For the future case, an internalist could require that the subject be aware, or be capable of becoming aware, that whenever she has had intentions of the ␸-type in type of context C, those intentions generally result in successful actions of ␸-ing in C. Her past success in implementing ␸-intentions provides a kind of general evidence that she will ␸ in a similar situation in the future. What she lacks, but does not need, is specific evidence that she is now ␸-ing, which could indeed be provided by observation or perceptual feedback from her action.11 So the account can be fleshed out so as to be made acceptable to those who advocate an internalist, evidentialist view. It is worth pointing out, too, that the causal theory even in its externalist form still imposes a considerable warrant requirement on agent’s knowledge.12 It thus differs from the radical view that one can have weak knowledge that one is going to ␸ (or that one is ␸-ing) merely in virtue of having a true belief.13 What seems wrong with such a view is that there is now nothing relevant to the agent’s knowledge — neither evidence available to the agent nor some external factor in the agent’s situation — to raise the likelihood that the agent’s belief is true. In such a case, we have a hard time differentiating the agent’s true belief from one that is merely luckily true. And most epistemologists — with some notable exceptions — think that one cannot have knowledge simply because one is lucky enough to acquire a true belief.14 Our second objection notes that it is clearly an assumption of the causal theory that mental states such as intentions are causally efficacious. The theory therefore requires a belief in the reality of mental causation. Philosophers who suspect that mental states are epiphenomenal will conclude that the necessary conditions for possessing agent’s knowledge according to the causal theory are never in fact met. Such scepticism about mental causation may take the form of scepticism about the role of conscious intention in causing action. The work of Libet (1985) is sometimes cited in this connection. In Libet’s experiment, subjects were told to move whenever they felt the urge to do so after the experiment started. They were told to note the time when they felt an “urge”, intention or will (W) to move by noting the position of a moving spot on a clock dial. Their brains’ physical preparation for movement was measured using a readiness potential (RP). The time at which they 11

On the distinction between general and specific evidence, see Velleman (1989). I follow standard usage in using “warrant” as a neutral term for whatever turns a true belief into knowledge. 13 If the warrant requirement is lifted, then it is indeed the case that “agent’s knowledge” would be a kind of true belief not at all based on evidence. Though some, such as Goldman (1999) and Hetherington (2001), would countenance such true belief as a weak kind of knowledge, most epistemologists would want to maintain some justification requirement on knowledge. 14 For an argument against this common view, see Hetherington (2001, Chapter 4).

12

190

Anne Newstead

actually moved was recorded using an electromyogram (EMG) designed to measure the activation of a particular muscle. The time of actual movement provided a zero reference point. Correlating the measurements, Libet claimed that whereas W preceded the time of actual movement by about 200 ms, the RPs preceded the time of actual movement by much more, by some 300–500 ms. One of the conclusions that a sceptic about causation by intention might draw from Libet’s experiments is that they demonstrate that conscious, introspectively accessible, intentions to act cannot be what causally determine agents to perform voluntary actions. The unconscious cerebral processes detected by RP must be the true causes of the action, because they occur before conscious awareness of the intention. Such a conclusion can be challenged on several grounds. First, it is not clear that Libet’s measurement procedure should be trusted. It is possible that making subjects attend to the position of a dot — giving them an additional mental task beyond simply attending to their intention — has the effect of increasing cognitive load and delaying when subjects can note their conscious intention. Second, it is not obvious that the sorts of movements studied by Libet — movements that one performs whenever one feels like it — are comparable to the kinds of actions for which agent’s knowledge should be claimed. Perhaps proponents of agent’s knowledge should claim that such knowledge only applies to deliberate actions preceded by conscious prior intentions. This would be a significant concession to Libet. A less concessive response would simply deny that Libet has measured intentional actions at all. Perhaps the instructions to “move whenever one feels like it” do not produce intentional actions, but mere sporadic, random movements.15 The third objection is that the causal theory is in no way sceptic-proof. The causal theory requires that an agent actually be ␸-ing at some point in order to know that she is ␸-ing. But the agent will not be in a position herself to know for certain that she is ␸-ing. It is always possible for someone to have the intention of ␸-ing, form the belief that she is ␸-ing, and yet fail to actually be performing the action in question. This does not seem to be a decisive objection to the theory, any more than it is a decisive objection to a causal theory of perceptual knowledge. Any theory of knowledge that incorporates a robust, objective conception of truth and that allows that subjects do not have total control over their beliefs and experiences is going to have to acknowledge the possibility that knowledge may elude subjects. That is, if the truthmakers for beliefs are beyond the control of the subject’s mind, then in aiming to have true beliefs, the subject can simply miss the target. In the case of

15

I am indebted to a conversation with David Coutts here.

Knowledge by Intention? On the Possibility of Agent’s Knowledge

191

agent’s knowledge, the truthmaker is the fact concerning the action itself. Although the agent exerts some control over whether or not she manages to perform the action, and her mental state (of intending to act) is usually causally efficacious, agents do not always have total control over their actions. Indeed, these considerations suggest that, on the causal approach, the objects of agent’s knowledge should plausibly be restricted to those actions over which agents have control — so-called “basic actions”. I will return to the issue of sceptical objections to agent’s knowledge in Section 4, after considering one more model of agent’s knowledge.

3. The Reasons-Based Model I turn now to a different model of agent’s knowledge, one based on knowledge of practical reasons for acting. The model is much more closely suggested by Anscombe’s remarks in Intention than the efficient causal model.16 It is based on a teleological model of action explanation. The causal model portrays intentional actions as the effects of certain efficient psychological causes. The teleological model sees actions as performed for the sake of achieving a certain purpose, end or goal. The purpose of the action, if you like, exerts an attraction on the agent who performs it. On Anscombe’s view, intentional actions are those that are done for some reason. They are actions for which the question “Why?” has application (Anscombe, 2000 [1958], p. 9). Such actions are rationally intelligible to the agent as moving her towards her goals. Anscombe’s special term for an agent’s knowledge of her action is “practical knowledge”. And, Anscombe says, we cannot understand the nature of practical knowledge without knowing about practical reasoning (ibid., p. 57). Practical reasoning consists in means-end calculation about what to do in order to achieve a goal. Thus, for example, someone may have the overall goal of building a sand castle. In order to build the castle, some intermediate steps are necessary: one has to pile up sand, shape it into towers, connect the towers with walls, and so on. Let the general action aimed at be “A” and each of the intermediate steps needed to reach A be “X”, “Y”, and “Z”. Then if someone asks the agent “Why are you Xing?”, she can answer “In order to A”. The sense in which she knows her reason for acting is just that she knows her goal is to A and understands how all of her sub-actions X, Y, and Z are directed towards the goal of accomplishing A. The

16 It is also indebted to some recent Anscombe exegesis by Moran (2004) and Hursthouse (2000). However, it requires, I think, a correspondingly greater commitment to Anscombe’s peculiar views than the other model.

192

Anne Newstead

explanation is easily adapted to the future case. In the future case, someone may ask “Why are you going to X?”, and the answer is as before “In order to A”. According to the reasons-based model, an agent’s knowledge of what she is doing is bound up with her “practical knowledge”, that is, her knowledge of her practical reasoning. The idea is that having practical knowledge of a certain sort is constitutive of performing that very intentional action. Without the practical knowledge, one would not be acting in the light of certain reasons. One’s action would not actually have the same intentional, purposive character without being backed up by such practical reasoning. So practical knowledge determines what it is that one is doing. This suggests that practical knowledge exerts a backwards pull. Whereas on the causal model, an intention to ␸ gave rise to knowledge of ␸-ing, on the teleological model, the knowledge that one is aiming at ␸-ing makes it the case that one is intentionally ␸-ing. It is in light of this reasons-based model that we should understand one of Anscombe’s (2000 [1958], p. 87) most important passages on agent’s knowledge: Surprising as it may seem the failure to execute intentions is necessarily the rare exception … What is necessarily the rare exception is for a man’s performance in its more immediate descriptions not to be what he supposes. Further it is the agent’s knowledge of what he is doing that gives the descriptions under which what is going on is the execution of an intention. This passage would be badly misunderstood if it were read simply as supporting the idea that there is an efficient causal, reliable link between an intention to ␸ and ␸-ing. Rather, what makes it the case that an action is intentional under the description ␸-ing (rather than under some other description) is the agent’s knowledge that she is ␸-ing. So an agent’s knowledge of what she is doing is constitutive of the very identity of her intentional action.17 How can this consideration about intentional actions yield knowledge? Anscombe writes (ibid.): If we put these considerations together, we can say that where (a) the description of an event is of a type to be formally the description of an executed intention (b) the event is actually the execution of an intention … then the account given by Aquinas of the nature of practical knowledge holds: Practical knowledge is “the cause of what

17

This point is emphasised in Moran (2004).

Knowledge by Intention? On the Possibility of Agent’s Knowledge

193

it understands”, unlike speculative knowledge, which is “derived from the objects known”. There is much compressed in this passage. In the first part, Anscombe lays out the requirements for an agent to know what she is doing. For example, in order for an agent to know that she is ␸-ing, it must be the case that (a) the agent takes herself to be ␸-ing (where her intention is to be ␸-ing); and (b) the agent’s action does fall under the description of ␸-ing. In the second part, Anscombe claims that when these conditions hold, we should think of the agent’s knowledge of what she is doing as the cause (in a sense to be explained) of the agent’s intentional action itself. The kind of cause that would be appropriate in this context is not an efficient cause, but something like a formal and final cause.18 The knowledge that one is ␸-ing gives shape and form to the intentional action of ␸-ing. In the final part, Anscombe states that an agent’s knowledge of her action, qua practical knowledge, differs from observational, speculative knowledge in not being derived from the objects known. That is, agent’s knowledge is not based on detecting the action performed and then inferring one is performing the action. Rather, such practical knowledge is spontaneously generated by the agent herself and makes the intentional action what it is. It’s clear from Anscombe’s remarks above and elsewhere (2000 [1958], section 32), that she intends the difference between practical knowledge and speculative knowledge to be very radical. She notoriously claims that practical knowledge differs from speculative knowledge in having a reverse direction of fit: whereas in speculative knowledge, the belief is true if it matches the world, in practical knowledge, the belief is “practically true” just in case the world matches the belief.19 Many theorists find this part of Anscombe’s theory obscure or downright false. I think such theorists would be right to do so, on the grounds that all knowledge involves the concept of truth, where propositions are true just in case they fit with how the world is. The fact that a particular proposition concerns an agent’s action does not exempt it from this basic characteristic of true propositions.

18

That the notion of cause is not efficient is a point that should be obvious given Anscombe’s reference to Aquinas, but Anscombe (2000 [1958]) should have credit for pointing it out. There is evidence of opposition to the causal analysis of intentional action in Intention (Hursthouse, 2000, Sections 11–19), and the rejection of the causal analysis was a lifelong theme, appearing again in Anscombe (1983). 19 “Direction of fit” is Searle’s term in Searle (1983), but the idea (as he acknowledges) derives from Anscombe (2000 [1958]).

194

Anne Newstead

Even if we reject Anscombe’s idiosyncratic conception of practical knowledge, though, we can appreciate the idea that an agent’s knowledge could differ from observational knowledge in being based on knowing one’s practical reasoning and plans for action. So understood the theory of agent’s knowledge resembles the medieval doctrine of “maker’s knowledge”.20 According to this doctrine, makers occupy a superior epistemic position with respect to what they produce or make themselves, when compared with mere users of a product. The difference is that only makers, but not users, have direct introspective access to their intentions or plans for making a certain product. In the same way, the theory of agent’s knowledge implies that agents have a superior knowledge of their actions compared with mere observers, because agents know their plans for acting in a relatively a priori fashion. 3.1. Mind the Gap On the broadly Anscombean model developed above, an agent has knowledge of what she is doing “from the inside”, by knowing her intentions (reasons and plans) whereas a mere observer does not. However, this can make it look as though Anscombe slides from a plausible view — that one can have non-observational knowledge of one’s intention for acting — to the implausible view that this same knowledge is knowledge of one’s intentional action. It is true that the overarching intention — which is specified in the practical reasoning — is indispensable for the intentional action. But it’s not clear why practical knowledge should yield knowledge of the intentional action, not just the intention. There seems to be a gap — epistemic (if not also metaphysical) — between the intention and the action. In fact the objection applies not just to the teleological model, but also to the modified causal view. (The modified causal view is just the causal theory made more acceptable to internalists.) On the modified causal theory: S knows that she is ␸-ing just in case: (i) S is ␸-ing; (ii) S believes she is ␸-ing; and (iii) S’s belief that she is ␸-ing is caused and justified by her intention-in-action of ␸-ing. Here the immediate basis for S’s belief that she is ␸-ing is not direct awareness of the action itself, but awareness of her intention-in-action of ␸-ing. So as with the teleological model, there is an epistemic gap between the subject’s intention and her intentional action. Again it is not clear why knowledge of the former suffices — non-inferentially, no less — for knowledge of the later.

20

The observation is due to Hintikka (1974).

Knowledge by Intention? On the Possibility of Agent’s Knowledge

195

Both models face the problem that there appears to be an epistemic gap between having an intention and performing a certain intentional action. In each case, we have no difficulty with the idea that agents know their intentions without observation (by introspection), but the conclusion falls short of (AK), the claim that agents know their intentional actions without observation. I shall first consider how the objection is to be met using the resources and sticking to the confines of Anscombe’s approach to intentional action. I shall then discuss the objection in the context of a causal approach; and how the objection can be met using fallibilism is shown in Section 4. The fallibilist approach provides welcome relief from the objection regardless of which model of agent’s knowledge (causal or teleological) we adopt. Anscombe’s approach to the problem of the gap is simply to deny that the gap exists. She denies that the gap exists in every sense: both metaphysically and epistemically. More specifically, because there is no real metaphysical gap, there is no epistemic gap either. There are two ways in which a metaphysical gap could interpose between an intention and an action. First, the gap could be temporal and causal, as when a prior intention fails to eventuate in the corresponding action. Second, the gap might consist in the fact that the intention itself is conceived of as something that falls short of the whole intentional action. Here the intention would stand to an action as a sense-datum does to an object of perception. Anscombe denies both sorts of metaphysical gaps. First, as we saw, Anscombe says that the failure to execute intentions to perform basic intentional actions is “necessarily rare”. Consider once again this (Anscombe, 2000 [1958], p. 87) passage: What is necessarily the rare exception is for a man’s performance in its more immediate descriptions not to be what he supposes. Further, it is the agent’s knowledge of what he is doing that gives the descriptions under which what is going on is the execution of an intention. One important point made in this passage is a point that we are now inclined to recognise as a Davidsonian point about the unintelligibility of massive errors (in this case massive errors about what we are doing). Our identification of our actions as intentional under some description depends on there being a reliable link between intending and doing. In the absence of such a reliable link between the intention to ␾ and the action of ␾-ing, we would cease to be able to identify actions as intentional under that description (␾) at all. This means that the link between intentions and actions is not merely reliable, but constitutive. The consideration appears powerful for simple cases. Consider what would be the case if a neuroscientist (with a perverse sense of humour) interfered with

196

Anne Newstead

someone’s brain so that whenever she intended to build a sand castle, she wound up eating sand instead. Repeated attempts to build the castle resulted only in more sandeating. The subject’s behaviour would be very puzzling to her. She would claim: “That’s not what I meant to do at all”. Eventually, perhaps she would adjust her expectations, so that whenever she felt like she used to feel when intending to build a sand castle, she would expect that she would eat sand. Anscombe’s point would be that we begin to lose a grasp on the content of the original intention or plan apart from its manifestation in action. The intelligibility of the action (to the agent) depends on seeing it as the expression of intention; likewise, the identification of the intention reciprocally depends on viewing a certain action as its fulfilment. Still, one might worry that this sort of claim has to allow for occasional exceptions. And an occasional exception is all the sceptic needs to reinstitute the gap. Second, Anscombe denies that a gap can open up between an intention-in-action and the intentional action itself. Anscombe (2000 [1958], Section 29) considers the temptation to split an intentional action into a mental component (such an intention, volition) and a physical component (the movement). It is then natural to infer that whereas the mental component is known without observation (by introspection), the physical component can only be known by observation. Her arguments against such a dual component view of action, however, are very compressed and not entirely convincing. She dismisses the view that only the mental component of the action is known non-observationally as “a mad account”, on the grounds that it is hard to see what this mental component amounts to apart from its physical effect. However, a behaviourist fear of unmanifested mental states is no argument against them. More obscurely, but more plausibly, she dismisses the view that all we ever do intentionally is what we think we are doing as “nonsense” (ibid., p. 52). Presumably, this is because this view leads to scepticism about whether anyone ever acts intentionally. Though such scepticism runs counter to common sense, it is not incoherent and Anscombe owes an argument against it. Anscombe’s remarks against a dual component view of action are suggestive, but not sufficient.

4. Fallibilism and the Gap The objection raised against (AK) based on the gap between intention and action instantiates a pattern of argument wholly familiar from certain sceptical arguments against our knowledge of the external world. The external world sceptic (EWsceptic) might argue: 1. To know there really are material, physical objects, I need to know that my perceptions as of objects are veridical;

Knowledge by Intention? On the Possibility of Agent’s Knowledge

197

2. I do not know that my perceptions are veridical, since I cannot certainly distinguish them from non-veridical perceptions; 3. So, I do not know that there really are material objects — at least not by perception. Two standard moves made in response to EW-scepticism are: (A) to allow that one can in fact distinguish veridical and illusory perceptions, albeit only fallibly, and (B) to espouse a form of direct realism about perception on which perceptions reach all the way out to the objects themselves.21 I am going to advocate (A) in preference to (B). But I will briefly and programmatically indicate why I do not think (B) is the most promising response to the sceptical argument. Option (B) is often advocated as an answer to scepticism, particularly that motivated by the argument from illusion. But such an approach is actually powerless to answer the sceptical argument mounted above, which we might call “the argument from the subjective indistinguishability of veridical and non-veridical perceptions”. It is wholly compatible with direct realism that a subject not be able to introspectively distinguish the content of her veridical and non-veridical perceptions. Given externalism about mental content, the content of these perceptions will in fact be different. But this difference need not be transparent to the subject.22 So this kind of direct realist approach can at best persuade us not to take up the kind of Cartesian standpoint from which an argument like the one above is raised. So the direct realist approach does not offer a straight response to that sceptical argument. Other things being equal, a straight reply to an argument is preferable to an evasive one. A straight response is possible by attacking the sceptic’s underlying epistemological assumptions as in (A). The quest for certainty must be given up if knowledge is possible. Though I cannot know with certainty that my perceptions are veridical, I can know with less than certainty — fallibly — that my perceptions are veridical when they are. So I prefer a fallibilist motivation for rejecting (2). To be sure, the concept of fallible knowledge is not without its detractors and a full development of a theory of fallible knowledge remains in its infancy.23 For some theorists, the intractability of the Gettier problem is taken to indicate that fallible warrant is incompatible with knowledge. There is promising work that opposes this claim, so I shall not attempt a rebuttal here.24 For other theorists, 21

For developments, see McDowell (1986). McDowell (1986) is of course well aware of this point. 23 For a comprehensive introduction to fallibilism, see Hetherington (2005). For the application of fallibilism to sceptical problems, see Feldman (2003, pp. 122–129). For more specific considerations relating the KK principle and fallibilism, see Feldman (1981). 24 See Zagzebski (1994). For arguments against the idea, see Howard-Snyder, Howard-Snyder, and Feit (2003). 22

198

Anne Newstead

the idea of fallible knowledge just seems intuitively wrong. For example, Lewis (1996, p. 549) says that “To speak of fallible knowledge, of knowledge despite uneliminated possibilities of error, just sounds contradictory”. Lewis may be right about how odd, to Cartesian trained ears, attributions of fallible knowledge sound. But the perceived oddness might be the result of a confusion. Once we recognise the confusion for what it is, we have reason to be optimistic about the coherence of the concept of fallible knowledge. It is an axiom of Cartesian epistemology that: (*) If S knows that p, then S cannot be mistaken that p. In one sense, this is trivially true: S knows that p entails that p is true. So if S knows that p then S cannot have gone wrong in picking up a false belief. Fallibilists of course deny (*). But they are not committed to denying the truism that knowledge requires truth. Fallibilists point out that S’s knowing p does not logically require there is no way in which S could have been mistaken about p. All that is required in this respect for S to know p is that S is not actually mistaken about p. There are plenty of ways in which S could have gone wrong, but thankfully in the case where S knows that p, S did not go wrong. The perceived oddness of the claim that S’s knowing that p is compatible with the possibility of S’s having been wrong about p must stem, I think, from a tendency of those who hear the claim to reduce possibilities to actualities. On the reductionist view, every possibility must be actual at some point in time (the Aristotelian view) or realised in some actual world (the Lewisian view). That is why a possibility of error is such a threat to knowledge for reductionists. For if there is a possibility of error, then at some point, in some world, there is an error and knowledge is lacking. But of course the fact that knowledge is lacking in that other world does not mean that it is lacking in our actual world. Reductionism, by not allowing mere (unrealised) possibilities, encourages a focus just on the actual. That reductionist tendency could lead those to read claim (*) wrongly as just the claim: (**) If S knows p, then S is not in fact actually wrong about p. It would be contradictory to deny this later claim.25 However, as we have seen, fallibilists do not deny (**). So fallible knowledge is free from the charge of 25

We need not convict Lewis of such confusion: he is just reporting the perceived oddness of claiming that S knows p despite uneliminated possibilities. However, it is then puzzling why, if he is not confused, Lewis is not more sympathetic to fallibilism. It is an interesting question for further research to what extent Lewis’s strong modal realism — his belief that all worlds are actual — influences his rejection of fallibilism.

Knowledge by Intention? On the Possibility of Agent’s Knowledge

199

incoherence. I suspect that many objections to fallible knowledge will be seen ultimately to rest on such confusion. So I conclude the way is wide open to affirm fallible knowledge. Allowing fallible knowledge, far from encouraging scepticism, enables an elegant and mature response to scepticism. This type of response is available, moreover, not just in the case of sceptical threats to perceptual knowledge of the external world. Fallibilism can be used to defuse sceptical objections to the claim that agent’s know what they are doing by intention. One sceptical argument against agent’s knowledge has the following structure: (A1) To know what she is doing, S must know that her intention-in-action is actually being executed; (A2) S does not ever know that her intention-in-action is actually being executed, because S cannot distinguish between (i) the case where the intention-in-action corresponds to the actual action, and (ii) the case where S has an illusion of acting with that intention — a mere illusion of having an intention-in-action; (A3) So, S does not ever know that she is performing a given intentional action — at least not by awareness of her intention. As with the argument for EW-scepticism, one could deny the second premise in two different ways: (A*) by adopting a fallibilist approach to knowledge of actions, or (B*) by adopting a direct approach to knowledge of actions. This argument focuses on intentions-in-action. But since intentions-in-action are always embodied in actions, there is some optimism for thinking that one cannot have the relevant intention-in-action without performing the corresponding intentional action. This means that in this case approach (B*) is viable: one’s intentionin-action does reach all the way out to include one’s action. But again, it is not clear what help this closure of the metaphysical gap between intention and action is in meeting the sceptical argument. For it is possible that the intentions with which subjects act are not transparent to themselves. This is a possibility that Anscombe did not see.26 As before, though, there is an alternative fallibilist motivation for rejecting (A2). S does fallibly know that her intention is being executed. Her awareness of her intention (and her lack of awareness that anything is wrong) provides a good fallible reason for thinking that she is acting as described by her intention. In 26

The approach will not, of course, work either if the sceptical argument is recast in terms of a subject wondering whether her prior intention to ␸ will give rise to an action of ␸-ing. Here there is an undeniable metaphysical gap between intention and action, so that knowledge of intention will not yield knowledge of the action immediately.

200

Anne Newstead

those cases where S is suffering an illusion or delusion about the efficacy of her intentions, S might falsely claim to have knowledge. But the fact that S might go wrong in that case does not detract from S’s right to claim knowledge in the normal case, where her intention is actually being executed.

5. Conclusion I have presented two distinct models of agent’s knowledge: a causal-reliabilist model and a teleological, reasons-based account. Both approaches face a parallel difficulty about how the gap between an intention and an action gets crossed. In each case it might seem that closing the epistemic gap requires closing the metaphysical gap between intention and action. On a causal theory the metaphysical gap gets closed only if (i) prior intentions perfectly reliably cause their corresponding intentional actions, or (ii) intentions-in-action are always embodied in the corresponding intentional actions. In each case, knowledge of an intention would suffice (on an externalist, reliabilist theory) for knowledge of the corresponding intentional action. However, the claim that prior intentions perfectly reliably give rise to actions needs further defence. So it looks as though the causal theory cannot deliver the result that agent’s knowledge is anticipatory, being derived from knowledge of prior intentions. On the teleological approach, what is required is a defence of why knowledge of a plan yields knowledge of what one is actually doing. As we saw, on Anscombe’s theory of intentional action, the metaphysical gap is closed by making knowledge of one’s plan for acting constitutive of one’s intentional action. But again the closure of the gap needs more motivation. I have suggested that we could learn to live with a metaphysical gap between intentions and intentional actions by adopting a more fallibilist conception of agent’s knowledge. On the causal model, awareness of one’s intention-in-action provides fallible evidence for thinking that one’s intention is being executed. On the teleological model, awareness of one’s plan for acting and awareness that one has started to act on that plan provide fallible evidence for thinking that one is acting as planned. If this is right, then progress in epistemology — in the form of an improved understanding of fallible knowledge — will lead to progress in the theory of action.

References Anscombe, G. E. M. (1983). The causation of action. In: C. Ginet & S. Shoemaker (Eds), Knowledge and mind: Philosophical essays (pp. 174–190). New York: Oxford University Press.

Knowledge by Intention? On the Possibility of Agent’s Knowledge

201

Anscombe, G. E. M. (2000 [1958]). Intention. Cambridge, MA: Harvard University Press. Feldman, R. (1981). Fallibilism and knowing that one knows. The Philosophical Review, 90, 77–93. Feldman, R. (2003). Epistemology. Upper Saddle River, NJ: Prentice-Hall. Goldman, A. (1976). A theory of human action. Princeton: Princeton University Press. Goldman, A. (1999). Knowledge in a Social World Oxford: Oxford University Press. Gopnik, A. (1993). How we know our own minds: The illusion of first-person knowledge of intentionality. Behavioral and Brain Sciences, 16, 1–14. Hetherington, S. (2001). Good knowledge, bad knowledge: On two dogmas of epistemology. Oxford: Clarendon Press. Hetherington, S. (2005). Fallibilism. The internet encyclopedia of philosophy, http:// www.iep.utm.edu/f/fallibilism.htm. Hintikka, J. (1974). Practical vs. theoretical reason — An ambiguous legacy. In: Knowledge and the known: Historical perspectives in epistemology (pp. 80–97). Dordrecht: Reidel. Hornsby, J. (1980). Actions. London: Routledge & Kegan Paul. Howard-Snyder, D., Howard-Snyder, F., & Feit, N. (2003). Infallibilism and Gettier’s legacy. Philosophical and Phenomenological Research, 66, 304–327. Hursthouse, R. (2000). Intention. In: R. Teichman (Ed.), Logic, cause and action: Essays in honour of Elizabeth Anscombe (pp. 83–105). Royal Institute of Philosophy Supplement 46. Cambridge: Cambridge University Press. Lewis, D. (1996). Elusive knowledge. Australasian Journal of Philosophy, 74, 549–567. Libet, B. (1985). Unconscious cerebral initiative and the role of conscious will in voluntary action. Behavioral and Brain Sciences, 8, 529–566. McDowell, J. (1986). Singular thought and the extent of inner space. In: P. Pettit & J. McDowell (Eds), Subject, thought, and content (pp. 137–168). Oxford: Clarendon Press. Moran, R. (2004). Anscombe on ‘practical knowledge’. In: J. Hyman & H. Steward (Eds), Agency and action (pp. 43–69). London: Royal Institute of Philosophy. Newton-Smith, W. (1980). The structure of time. London: Routledge & Kegan Paul. Pickard, H. (2004). Knowledge of action without observation. Proceedings of the Aristotelian Society, 104, 205–230. Searle, J. (1983). Intentionality. Cambridge: Cambridge University Press. Trevarthen, C. (1984). How control of movement develops. In: H. A. Whiting (Ed.), Human motor actions — Bernstein reassessed (pp. 223–261). North Holland: Elsevier Science Publications. Velleman, J. D. (1989). Epistemic freedom. Pacific Philosophical Quarterly, 70, 73–97. Wittgenstein, L. (1958). The blue and brown books. Oxford: Blackwell. Zagzebski, L. (1994). The inescapability of Gettier problems. The Philosophical Quarterly, 44, 65–73.

This page intentionally left blank

202

Chapter 13

Gettier’s Theorem John Bigelow

The Manifest Image Once upon a time there was a theory of knowledge, the theory that knowledge is true justified belief. There was also a metatheory, a theory about this theory. The metatheory said that native speakers of English all know the meaning of the verb “to know”; and in virtue of this, they have tacit knowledge of the definition of the word “knowledge”; and in virtue of this, they have tacit knowledge that knowledge is true justified belief. It is true “by definition” that knowledge is true justified belief, and this means that it is something we can know a priori. To illustrate this theory of knowledge, suppose you think something to be the case, which also happens to be an accepted part of science or common sense, as for instance that the Earth is round. Suppose you think this belief has, to a very high degree, whatever kinds of justification it is possible to obtain either in the sciences or in everyday life. Then — according to the theory that knowledge is true justified belief — you think that you know that the Earth is round. Anyone who hears you claim that you know that the Earth is round, if they understand English, should interpret you as claiming that the Earth is round, and as implying that you believe the Earth to be round, and as claiming that you have some appropriate kind of justification for making this claim. Bertrand Russell once inadvertently refuted this theory of knowledge, and seemed not to have realized what he had done. He was intending only to illustrate the thesis that knowledge is something more than true belief, and he cited an example of a concrete case in which a person has a true belief but manifestly does Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

203

204

John Bigelow

not have knowledge. Yet, the case was one in which a person also has a justification for this true belief — a justification of precisely the same sort that we ordinarily have in everyday or scientific instances of knowledge. In experimental work in the sciences, we rely on instruments like clocks and rulers. We should check our instruments before we rely on them. So, for instance, we should measure our meter rule to check that it really is a meter long. This can be tricky, of course. Which meter rule do you use to measure the length of the meter rule that you are checking? Often we just trust the manufacturers. Similarly we should check a clock to see if it is working accurately, presumably using another clock that we have already tested. Suppose we have checked some particular clock as thoroughly as it is ever possible to check any clock when engaged in experimental work in the sciences, and it works fine. Now imagine that, for reasons you could not possibly have foreseen, three minutes ago it malfunctioned and its digital display jumped forward three minutes in time and then stopped working altogether, frozen at the time 12:21. In the midst of an experiment you look at this clock and write down your observation in your experimental records: “Scintillation occurs at 12:21 plus or minus 3 seconds”. The belief you record is true, and it is justified. Yet it is not knowledge. Russell’s clock could have been cited as a counterexample to the theory that knowledge is true justified belief, at least if the kind of justification required for knowledge is of the kind we can obtain in the sciences and everyday life. Russell himself set the standards of justification higher than that, and concluded that we simply do not have knowledge — properly so-called — in the sciences and everyday life, but only “probable opinion”. However, many were unwilling to follow Russell to this skeptical conclusion. Many acknowledged Russell’s observation, that science and common sense rest on justifications that are, as we may say, fallible: it is possible to have a very good justification of this kind and yet to have bad luck, and to be mistaken. Nevertheless they wanted to say that this kind of justification is, after all, sufficient for something to qualify as knowledge, properly so-called, except, of course, in the cases where we have bad luck and the belief is not true. Knowledge just is probable opinion that happens also to be true. However, the significance of Russell’s counterexample was little noticed at the time. His clock example seemed to suggest that probable opinion that happens also to be true may not qualify as knowledge. Then one day, a young man called Gettier published in Analysis a short paper that presented a paradigm pair of counterexamples to the theory that knowledge is true justified belief. The counterexamples consisted in descriptions of possible states of affairs within which any native speaker of English would be confident in a judgment that an imagined person (whom Gettier calls “Smith”) does not have knowledge. Yet, the theory that knowledge is true justified belief would entail that

Gettier’s Theorem

205

this person “Smith” does have knowledge. These examples purported to show that the definition of “knowledge” that is tacitly known by native speakers of English is not that knowledge is true justified belief. When it was wondered how native speakers of English could recognize that the imagined case was not an example of knowledge, it was said that they relied on “intuition”. This meant that they were confident they were right even though they could articulate no reasons to support their judgment. Their confidence on this matter was in this respect like the confidence we have in our judgment that the sentence I am written is ungrammatical. Linguists around the time of Gettier’s paper were speaking frequently of the role of “intuitions” of native speakers of a language concerning the grammar of that language. Philosophers of language used the same term in a very similar sense. It was thought that native speakers have tacit knowledge of the definitions of the words in their language, and that this tacit knowledge revealed itself — not in an ability to articulate those definitions — but in “intuitions” about what descriptions would fit a variety of actual or merely imagined concrete cases. In the decades that followed Gettier’s watershed paper, no consensus emerged about the upshot of Gettier’s arguments. Philosophers in the territory began to speak of the “one patch per puncture” methodology in analytic philosophy (David Lewis for instance), and of “the dull thud of conflicting intuitions” (Tim Oakley for instance). The failure to reach consensus was giving analytic philosophy a bad name. Some (Quineans) were arguing that native speakers of English do not have tacit knowledge of any definition of knowledge. Yet, even these anti-definitional philosophers were left with unanswered questions in the wake of Gettier’s persuasive arguments that knowledge is not justified true belief. Even if you think we have no tacit knowledge of definitions, you can still be compelled to acknowledge in various concrete cases, if you are honest, that you do confidently believe that a given particular person does not have knowledge, even though the theory under scrutiny clearly entails that this person does have knowledge. Then, these concrete cases do falsify the theory by a simple valid argument: the theory entails something false (about this concrete case); hence, the theory is false. So, even putting aside any dubious metatheories about philosophical method, and theories about tacit knowledge, there nevertheless remains a strong case for the thesis that knowledge is not true justified belief. Not all justified true beliefs are knowledge. And it is not necessarily true that every justified true belief is knowledge. And if knowledge is not true justified belief, then what is knowledge? Attempts to answer this question were regularly plagued by Gettier-style counterexamples; those rival theories that seemed to be immune to Gettier-style counterexamples proved to be vulnerable to various objections of various other kinds; and over the years no consensus seemed to be emerging. When seen from some angles, under the wrong lighting, the endlessness of these debates could be depressing.

206

John Bigelow

Taking Heart I take heart from an analogy between this crisis in epistemology and a crisis in the foundations of mathematics that emerged more than half a century before Gettier. Just before the beginning of the twentieth century, Gottlob Frege and others were making monumental leaps forward in working out new ways in which mathematicians could avoid contradictions when they were reasoning about collections of things, like numbers and lengths, when those collections were infinitely large or when the things in those collections were infinitely small. Along the way, they created what is called set theory. Frege laid down five axioms that seemed to be self-evident. Each of these axioms had deep roots in the practice of mathematics. If you denied one of these axioms, that would be like pulling a weed out of the ground and leaving all the roots behind: the plant would just grow back almost immediately, only tougher. Then Bertrand Russell presented a short argument, “Russell’s Paradox”, that demonstrated that Frege’s five axioms, together with very simple axioms of logic, entailed a contradiction. Hence, we must deny either one of these axioms or else one of a very small number of very basic principles of logic. In the century following Russell’s paradox, no firm consensus has emerged on the solution to the problems that emerged in the wake of his demolition of Frege’s foundations for mathematics. A large number of theorists came to the conclusion that we have to deny Frege’s fifth axiom, the one that is often called Comprehension, and to replace it by some other axiom that will do all the useful mathematical work that Comprehension was supposed to do, but without landing us in a contradiction. The mathematician Zermelo, and others, settled on an axiom called Separation, as a replacement for Comprehension, and many mathematicians have followed that path. Yet although theories built on Separation are accepted across a broad range of mathematicians, the consensus is still not universal. When seen from some angles, under the wrong lighting, the endlessness of these debates could be depressing. Yet that is just how things are in a priori disciplines like logic and pure mathematics — and philosophy. We used to think we could find unshakeable foundations for the a priori disciplines; now we find that even the pure mathematicians and logicians, just like everyone else, will have to rely in the end on a leap of faith and a pinch of epistemic luck, and they will have to stop hankering after a timeless universal consensus. The best we can hope for is rational belief in what seems to be the “best” overall explanation of all the things that seem to stand in need of explanation. Notice that Russell’s paradox did not demonstrate that Frege’s theory had misarticulated something that is “tacitly known” by all native speakers of a natural language like German or English. Russell showed that a contradiction followed from

Gettier’s Theorem

207

propositions that seemed to be “self-evident”, whatever that means. It was tempting to say that those propositions rested on “intuitions”. Yet those “intuitions” — the ones supporting Frege’s axioms — were not naturally construed as consisting in tacit knowledge of definitions, as possessed by all native speakers of a natural language. Frege’s “intuitions” were like those underlying the putatively self-evident axioms of Euclidean geometry, rather than like the “intuitions” trafficked in by linguists and analytic philosophers. It is perhaps better to think of Frege’s axioms as propositions that have a central explanatory role in a range of theories. If you deny any one of these key propositions, that has enormous ramifications in numberless nooks and crannies in numerous theories. Yet, Russell’s paradox demonstrates that we do have to deny at least one of these key propositions — and when we do, this will entail a lot of consequent work, either digging out all the old roots and re-rooting a replacement, or else perhaps grafting a new stem onto the old roots. I urge that Gettier’s paper on knowledge should be seen as analogous, in many key respects, to Russell’s paradox. Gettier did for epistemology what Russell’s paradox did for the foundations of mathematics.

Three Axioms In his watershed paper, Gettier in effect derived a contradiction from three propositions. The first axiom that Gettier relies upon is not explicitly stated, but it emerges from something that was commonly granted on all hands: that (loosely put) knowledge is something more than just true belief. To say someone knows something is to express more than just your own belief that it is so, along with your belief that they believe it too. It is granted on almost all hands, furthermore, that what differentiates knowledge from true belief has to be something that is of relatively deep and pervasive explanatory significance of some kind or other. The kind of explanatory significance that is in question may be called epistemic explanatory significance: more should be said, but it can wait. I submit that this widely accepted assumption, that knowledge is “more than true belief ”, supports the following proposition: A.1. If some person both believes and knows something, p, and some person believes but fails to know something, q, then this belief that p and this belief that q must differ to some significant degree in some epistemically significant respect other than truth.

208

John Bigelow

Call this the Axiom of Warrant. This principle A.1 is stronger than just the claim that knowledge is “more than” just true belief. It is like a parallel claim in ethical theory. If one person’s action is right and another person’s action is wrong then there must be some further significant difference between these two actions in addition to the one’s being right and the other’s being wrong. This other difference needs to be something with the right kind of moral significance. For instance, imagine that I say that my action is right and yours is wrong, and I am challenged to say what the difference is between the cases. Imagine that I point to the obvious difference: that the first action was performed by me and the second action was performed by you. Yet that difference, by itself, is not of the right kind to mark a moral difference between the cases: not unless there is also some other difference between me and you that can carry enough moral weight to justify my action and not yours. As in moral theory, so too in epistemology: an epistemic difference between two cases must be accompanied by some other difference, and this other difference must be one that has the right kind of explanatory significance in epistemology. Differences in knowledge supervene on differences with the right kind of explanatory significance. The name I have given this axiom, “Warrant”, draws on useful terminology that is vividly explained by Merricks, following Plantinga (1993). Warrant is that, whatever precisely it is, which makes the difference between knowledge and mere true belief. (Merricks, 1995, p. 841) If there are true beliefs that are not knowledge, then there is something that all beliefs that are knowledge share, and all merely true beliefs lack. This is warrant. Merricks (ibid.) says this is a “purely formal characterization of warrant”: we are just giving a technical label to whatever it is that must be added to a true belief in order to turn it into knowledge. This definition of “warrant” proceeds by what is sometimes called “logical subtraction”: take knowledge, subtract the component that is constituted by the possession of a true belief, and then what remains (whatever it may be) is what we will call “warrant”. Humberstone has shown, however, that logical subtraction is not as straightforward as you might have expected. From a conjunction “p and q” it may appear to be perfectly straightforward to “subtract” p, leaving the “remainder” q. Yet other cases will not even appear to be so easy. Consider an example. Necessarily when something is red, then it is coloured; but not everything coloured is red. Hence, being red involves something more than

Gettier’s Theorem

209

being coloured. In other words, being red involves being coloured plus something more. So, suppose we try to logically-subtract being coloured from being red, to see what remains. It is not easy to see what remains. For discussion and references see Humberstone (2000, Section 3), where two citations are given for the “red”/ “coloured” example to work by Ronald Jaeger and Frank Jackson. The same problem might arise in the Merricks definition of warrant by the process of “subtracting” true belief from knowledge and giving a name to “whatever remains”. Hence, it may have been misleading for Merricks to have said that he has given a “purely formal characterization of warrant”. Nevertheless, despite an awareness of these warning flags it can be helpful to use the term “warrant” in the manner Plantinga and Merricks have suggested. We should recognize, however, that this term does not rest on a “purely formal” definition. It rests on a substantive claim, namely A.1. The claim is that there is some epistemically significant property other than truth, which distinguishes cases of knowledge from cases of merely true belief. Considerations equivalent to the principle of Warrant, A.1, are explicitly articulated by philosophers who are cited by Gettier in the opening of his 1963 paper (namely, Plato, Chisholm and Ayer). However, A.1 is not explicitly articulated by Gettier himself. There are several points in Gettier’s argumentation at which he needs to be relying on something with the same import as A.1. Yet he does not cite anything like A.1 explicitly as a theoretical assumption. It is at these points in his argument that it is natural to say that the argument is relying on “intuitions”. Yet I say that it is misleading to use the term “intuitions” here. In relying on principle A.1, Gettier’s argument is relying on something a little like one of Frege’s axioms of set theory. It is relying on a principle that has an air of “self-evidence” and may indeed be said to be supported by “intuition” in some sense — but this principle is also supported in that more nebulous way (whatever it is) in which all a priori principles in pure mathematics come to be progressively more and more strongly supported or undermined, in the cumulative and collective process of the evolution of theories. Now turn to two further propositions, upon which Gettier’s argument crucially turns. These further propositions are explicitly articulated by Gettier (1963, p. 121): I shall begin by noting two points. First, in that sense of “justified” in which S’s being justified in believing P is a necessary condition of S’s knowing that P, it is possible for a person to be justified in believing a proposition that is in fact false. Secondly, for any proposition P, if S is justified in believing P, and P entails Q, and S deduces Q from P and accepts Q as a result of this deduction, then S is justified in believing Q.

210

John Bigelow

The first of these “points” can be re-articulated in the following proposition: A.2. It is possible for q to be false, and yet for a person’s belief that q to possess epistemically significant properties such that, necessarily, whenever a belief possesses those properties and is true then that belief qualifies as knowledge. Call this the Fallibility Axiom. It says in effect that a “warrant” is “fallible”. This principle trades on an obvious feature of many senses of “justification”, and it adds the supposition that truth plus this sort of justification is sufficient for knowledge. Note that it seemed obvious to Russell and others that some “justifications” leave open the possibility that the justified belief might not be true. Yet Russell assumed that any justification that leaves open the possibility of error cannot be a good enough to support a claim of knowledge — properly so-called. The Fallibility principle A.2 is the bold claim not only that some “justifications” fall short of ruling out all possibilities of error, but also that these “justifications” plus truth do constitute a sufficient condition for knowledge — properly so-called. In epistemology, one of the primary motives for this Fallibility claim is — I submit — to leave room for the possibility that science and common sense attain not only probable opinion but also knowledge — properly so-called. Russell was skeptical that science and common sense can attain much knowledge strictly speaking. Yet many have wanted to evade that skeptical conclusion. The Fallibility claim furnishes one tempting way of vindicating scientific and common sense claims not only to probable opinion but also to knowledge. Yet note that this Fallibility claim is virtually equivalent to the negation of Warrant, A.1. Fallibility, A.2, is the claim that a belief might have all the epistemically significant properties apart from truth, and yet it might nevertheless be false. This entails that there are two possible cases: one in which a belief has all the epistemically significant properties including truth; and the other in which the belief has all the very same epistemically significant properties apart from truth. These two cases differ in no epistemically significant respect apart from truth. Hence, Fallibility, A.2, flatly contradicts Warrant, A.1. So if Warrant, A.1, is “intuitively” compelling, it should seem relatively obvious that Fallibility, A.2, is false. I say that in the right frame of mind this is indeed intuitively obvious. Before Gettier, Russell and others seem often to have assumed it to be obvious. Yet in contexts in which we are defending science and common sense against the skeptic, we can be drawn towards Fallibility and lose sight of Warrant — and lose sight of the fact that Fallibility contradicts Warrant. I trust I am not alone in finding

Gettier’s Theorem

211

these matters confusing. I am indebted to David Chalmers for pointing out in discussion that A.2 contradicts A.1 even without the mediation of a third principle, A.3, “Closure”, that Gettier in fact drew upon in his original paper. What Gettier did, in effect, was to reactivate the “intuitions” supporting Warrant, A.1, and to do so using vivid examples. Gettier in fact “teased out” the intuitively elusive incompatibility between A.1 and A.2 with the help of a third principle, which he calls his second preliminary “point”. The second of the “points” that Gettier makes explicit is called the closure of knowledge under obvious and known entailments. We can call this third principle the Closure Axiom. It says that if you are justified in believing P and you draw a deductively valid inference from P to Q, then you are justified in believing Q. If you are justified in believing P, and you know that it is not possible for P to be true and Q false, then you are justified in believing Q. I think this Closure principle, in full generality and without qualifications, is demonstrably false. When we have a long chain of deductions, as for instance in Euclid’s Elements, and then even if we are completely confident about some of the early theorems, we may be very uncertain indeed about the later theorems in, say, Book X. Even when our reasoning was in fact perfectly valid at every single step along the way, we may not be justified in being one hundred percent confident that we have made no mistakes along the way. Small chances of errors in reasoning will accumulate across a long chain of deductions, until we will no longer be justified in believing certain remote consequences — even though they were, in fact, validly deduced from things that we were completely justified in accepting, and were deduced by deductive inferences that we were in fact completely justified in drawing at each step of the way. Hence, the degree of justification for a conclusion can sometimes be lower than that for the premise. If there is some threshold of strength of justification required for knowledge, then it is possible for a premise to lie above that threshold and the conclusion to lie below it. Then that will falsify Gettier’s claim that if you are (in the relevant sense) justified in the premise of a valid deduction then you must always be (in the relevant sense) justified in the conclusion. Yet, although you are not always as justified in the conclusion of a deduction as you are in the premises, sometimes you are. Consider for instance fairly simple deductions from a relatively strong premise to a weaker conclusion. For instance: Consider a deduction from a conjunction (p and q) to one of the conjuncts: p. Or consider a deduction from a proposition p to a disjunction: (p or q).

212

John Bigelow Or consider a deduction from a singular proposition F(a) to an existentially quantified claim: (Some x)(F(x)).

These deductions are so simple and obvious that you lose very little justification from the merely Cartesianly skeptical possibility that you are making an error in reasoning. What you lose in confidence, as a result of this merely Cartesian possibility of error in reasoning, is compensated by the fact that the conclusion is in any case a weaker claim than the premise. Furthermore, in some instances the degree of justification for the premise is well above the threshold of strength of justification that is required for knowledge; and the loss of strength of justification from a short and simple deduction is nowhere near significant enough to take the degree of justification down below that threshold. All Gettier really required, for the purposes of the argument he was mounting, was a principle very much weaker than the strong principle of deductive closure as he articulates it. All he requires is that sometimes the epistemically important properties of the premise of a very obviously valid deduction will be shared by the conclusion of that deduction. This, I submit, can be summed up in the following proposition: A.3. For some cases of beliefs that p and that q, where p obviously entails q, these two beliefs will not differ to any significant degree in any epistemically significant properties apart from truth. Call this principle Weak Closure. With the three general principles A.1 (warrant), A.2 ( fallibility) and A.3 (closure) in hand, we can proceed to what I will call: Gettier’s theorem: The three propositions A.1 and A.2 and A.3 jointly entail a contradiction. From Fallibility, A.2, it follows that there can be a belief that p, where p is false, and so the belief does not count as knowledge: and yet nevertheless for that belief to have epistemically significant properties that would make any belief count as knowledge if only that belief were true. From Weak Closure, A.3, it follows that it is possible for a belief in some obvious deductive consequence of p, namely q, not to differ to any significant degree in any epistemically significant properties from the belief that p. And yet it is possible that the consequence q is true, even though p is false. Hence it is possible that the belief that q is true, and also possesses epistemically significant properties that will ensure that any belief, which has them will qualify as knowledge provided it is true. Hence, the belief that q does qualify as knowledge.

Gettier’s Theorem

213

Yet this entails that the beliefs that p and that q differ, in that one qualifies as knowledge and the other does not, even though they do not differ to any significant degree in any epistemically significant properties other than truth — and that contradicts the principle of Warrant, A.1. Hence, the three propositions A.1–A.3 entail a contradiction, which is what was to be proved. Gettier’s paper illustrates this theorem with two thought experiments: Cases 1 and Case 2. Case 1: We have a false premise of roughly the form F(a), where a is Jones — “Jones will get the job and has ten coins in his pocket”. This premise has, to a very high degree, epistemically significant properties that ensure that it would count as knowledge if only it were true. This premise deductively entails, as a conclusion, a true existential claim of roughly the form (Some x)(F(x)) — “Someone with ten coins in his pocket will get the job”. So Smith, who believes that Jones will get the job, draws this obvious conclusion, and so believes that someone with 10 coins in his pocket will get the job. The conclusion possesses both truth and epistemically significant properties that entail that a belief with those properties will qualify as knowledge provided that it is true; so by A.2 the conclusion qualifies as knowledge. Yet “intuition” says the case does not count as knowledge. Why? Because by A.3 the premise and conclusion do not differ, to any significant degree, in any epistemically significant properties other than truth. A.1 says that it cannot be that one qualifies as knowledge and the other does not unless they differ to some significant degree in some epistemically significant property other than truth. Yet, as we have seen, they do not. Thus, the contradiction between the “theory”, namely A.2 and A.3, and “intuition” corresponds closely to a contradiction between the two propositions A.2 and A.3 and a third assumed proposition A.1. I submit, therefore, that the role of “intuitions” in Gettier’s argumentation really reflects an implicit recognition of the theoretical weight of a principle like A.1, rather than tacit knowledge of any “definition” of knowledge. Case 2: A false premise p — “Jones owns a Ford” — deductively entails, as a conclusion, a true disjunction (p or q) — “Either Jones owns a Ford or Brown is in Barcelona”. So Smith, who believes that Jones owns a Ford, draws this obvious conclusion, and so believes that either Jones owns a Ford or Brown is in Barcelona. The premise is false so it does not qualify as knowledge; and yet nevertheless it possesses, to a very high degree, properties that would ensure that it would qualify as knowledge if only it were true. By A.3 the conclusion does not differ, to any significant degree, in the epistemically significant properties of the premise, which are properties that would ensure

214

John Bigelow

that any belief would qualify as knowledge if only it were true. Yet the conclusion is true, and so it follows that this conclusion does qualify as knowledge. Yet “intuition” says that belief in the conclusion does not qualify as knowledge. This is, I submit, because the premise and conclusion do not differ, to any significant degree, in any epistemically relevant property other than truth. A.1 says that it cannot be that one qualifies as knowledge and the other does not unless they differ to some significant degree in some epistemically significant property other than truth. And yet, as we have seen, they do not. Thus, again we find that the contradiction between the “theory”, namely A.2 and A.3, and “intuition” corresponds closely to a contradiction between the two propositions A.2 and A.3 and a third, assumed proposition A.1. Again, I submit that the role of “intuitions” in Gettier’s argumentation really reflects an implicit recognition of the theoretical weight of a principle like A.1, rather than tacit knowledge of any “definition” of knowledge.

Zagzebski, Merricks and Lewis The argument I have rehearsed above is structurally similar to arguments mounted by Zagzebski (1994) and Merricks (1995). Suppose that, in keeping with the Gettier Fallibility principle A.2, you say that knowledge is true belief plus some further property that leaves room, in principle, for the kind of “bad luck” that might make the belief false. Following Plantinga and Merricks we may call this further property (whatever it is) “warrant”. Suppose, then, that warrant does not entail truth, but that a true belief with warrant necessarily qualifies as an instance of knowledge. In this case it will be possible for a warranted belief to be subject to “bad luck” of some kind, this meaning that there may be some freakish events that bring it about that, despite being fully warranted, the belief turns out in the end to be false. Call this piece of “bad luck” an instance of first-luck. It is “luck” because it is the sort of thing that has no significant explanatory role in epistemic theory: it is not the sort of difference that has enough theoretical weight to mark a difference between cases that qualify as knowledge and cases that do not qualify knowledge. Whenever it is possible for a warranted belief to be subject to first-luck of this kind, it will also be possible for this first-luck to be subject to a freakish reversal of fortunes, which we might call second-luck, and which will cancel the effect of the first-luck and bring it about that the belief is true after all. Think of an animated cartoon like ones featuring a short-sighted “Mr. Magoo”, who continually has bad luck (like stepping off a high ledge at a building site) counteracted by good luck (an I-beam suspended from a crane swings around and happens to align itself right under his foot, just as he steps forward).

Gettier’s Theorem

215

Think of the example of Russell’s stopped clock that happens to be telling the right time. The first-luck is that the clock has stopped. The second-luck is that the time when you consult the clock happens to exactly match the time at which the clock’s register is frozen. The second-luck is “luck” because it involves something that has no more theoretical weight than the first-luck did — it contributes nothing of explanatory significance apart from truth. In the case where you have only first-luck, although your belief is warranted alas it is not true, and so it does not qualify as an instance of knowledge. Yet your belief does have a warrant, that is, it has properties that would have entailed than any belief with those properties would qualify as knowledge if only it were true. Warrant includes all the epistemically relevant properties apart from truth. We are supposing that warrant does not entail truth. That is, warrant is logically independent of truth. Hence, there must be factors that can sometimes determine that a warranted belief is true, and can sometimes determine that a warranted belief is false, and these factors will have no epistemic significance — apart from determining whether the belief is true or false. If there are such factors in an instance of first-luck, and they determine that the belief is false, then there can be further such factors, in an instance of secondluck, that determine that the belief is true. The difference between the two cases, however, will consist entirely of the factors that have no epistemic significance — apart from determining whether the belief is true or false. Thus, in the case where you have only first-luck, you do not have knowledge. In the case where you have double-luck you have the same warrant, and there is no change in any of the epistemically relevant factors apart from truth. In the second case, your belief has a warrant, and it is true and hence, according to the theory under scrutiny, it must qualify as knowledge. Yet it is absurd to say in the double-luck case that you have knowledge whereas in the first-luck case you do not — since the two cases do not differ, to any significant degree, in any epistemically significant properties apart from truth. Thus, I submit, the reasoning of Zagzebski and Merricks is recapitulated in my derivation of a contradiction from the three Gettier principles of Warrant, Fallibility and Weak Closure.

Where Does That Leave Us? Merricks argues that the only way to block the Gettier contradiction is by ensuring that every difference between a first-luck and a double-luck case is guaranteed to be an epistemically significant one. Yet this forces us to the conclusion that, as Merricks succinctly puts it in the title of his paper (1995): warrant entails truth.

216

John Bigelow

That is, Merricks concludes that the way to avoid the Gettier Paradox is by denying the Fallibility principle A.2. In effect, Merricks concludes that knowledge is true belief that could not be mistaken. The “warrant” of a true belief is something that necessarily confers on the belief the property of being such that “it could not be mistaken”. The belief could have been mistaken, because it could have failed to have this property; but it could not be mistaken at the same time that it does have this property. For any belief, having this property does entail truth. David Lewis was influenced by very similar arguments advanced a little earlier by Zagzebski (1994) — Lewis wrote to me in response to a draft of an unpublished paper I had sent him, saying “you’ve been scooped” and referring me to Zagzebski. Like Merricks, Lewis concluded in effect that warrant entails truth. Lewis argued (1999, p. 425) that S knows that P iff S’s evidence eliminates every possibility in which not-P — Psst! except for those possibilities that we are properly ignoring. (Psst: Lewis adds that when a possible source of error becomes actual, when it actually does bring about an erroneous judgment, then it can never be “properly” ignored.) Warrant in that sense manifestly does entail truth. Hence Lewis, like Merricks, evades the Gettier Paradox by denying Fallibility, A.2. There is a rack of other theorists who, whether wittingly or unwittingly, evade the Gettier Paradox by, whether implicitly or explicitly, denying principle A.2. Some of them advance theories with many features in common with Lewis’s theory, and these theories may all be classed as versions of contextualism. (Lewis trades on the idea that in some practical or conversational contexts it is perfectly “proper” to ignore some far-fetched possibilities of error, and it is this feature that makes his theory count as a version of “contextualism”.) There are also alternative ways of denying principle A.2, in ways that are so different from Lewis’s that they should not be counted as versions of “contextualism” but should be given some other name. According to Lewis, for a person to have a warrant is for that person to have excluded all possibilities of error. Other theorists say that for a person to have a warrant is for the world to have excluded all possibilities of error. According to these theories, knowledge can be conferred when something “external” to the knower has eliminated relevant possibilities of error: so these theories are classed as versions of externalism. In responding to Russell’s Paradox, most theorists look for substitutes for the Comprehension Axiom; likewise in responding to Gettier’s Paradox, most theorists look for substitutes for the Fallibility Axiom A.2. Nevertheless, there are other theories that are worth exploring. Some settheorists have tried to sustain the Naïve Comprehension Axiom, and to avoid

Gettier’s Theorem

217

Russell’s contradiction by tinkering with some of the axioms of logic. Likewise some epistemologists may reasonably explore ways of preserving Fallibility, A.2, and avoiding the Gettier’s contradiction by tinkering with A.1 or A.3. Some might explore ways of replacing Weak Closure, A.3, with some principle that will block all Gettier’s problems. Yet it is unlikely that this strategy will solve the problem: arguably, there is an incompatibility between A.1 and A.2 that will remain, even if we abandon Gettier’s use of A.3 to make this incompatibility more vivid. Some might have the courage to contradict “intuitions” — which entails denying Warrant, A.1: good recent examples are found in Weatherson (2003), and in various works by Hetherington (such as 1998, 1999, 2001, Chapter 3). This strategy might be more promising. Yet the most promising strategy, I submit, is to abandon Fallibility, A.2, and to admit that knowledge is belief which in some epistemically significant sense “could not be mistaken”.

Moral The principal moral is to see the role of Gettier’s paper in epistemology as analogous to that of Russell’s paradox in the foundations of mathematics. We should not be impatient to “solve the problem” by articulating a “definition” of knowledge, which we are to discover by a procedure that seeks to dredge up “intuitions” that reflect “tacit knowledge” of native speakers of the language that contains the verb “to know”. As it happens, I believe that native speakers do have tacit knowledge of a definition, and that this definition can be found. I am quite sympathetic to David Lewis’s proposed definition: “You know when you have eliminated all relevant possibilities of error”. I am also sympathetic to externalist alternatives more like the reliabilisms of Armstrong and others: “You know when there are no relevant possibilities of error”. Nevertheless, even if we were to settle on some definition of this kind, or some other, we would still need to ask whether, under that definition, general principles like A.1, A.2 and A.3 turn out to be true or false. Gettier has shown that at least one must turn out to be false. I have urged that it is Fallibility, A.2, that we should conclude to be false. Knowledge is true belief that has some further property that entails truth — some property like that of its being in some significant sense impossible for this belief to have been mistaken. Yet if we just tear Fallibility, A.2, out of the ground like a weed, leaving all its roots behind, then it will just grow back. When we notice the arguments for

218

John Bigelow

scepticism, we are likely to fall back into presupposing Fallibility, abandoning our previous conviction that warrant entails truth. We need either to kill Fallibility root and branch and then plant something completely new in its place — or else to graft onto those roots something new and demonstrably consistent with viable versions of the principles of Warrant, A.1, and Weak Closure, A.3. This is no easy task.

References Gettier, E. (1963). Is justified true belief knowledge? Analysis, 23, 121–123. Hetherington, S. (1998). Actually knowing. The Philosophical Quarterly, 48, 453–469. Hetherington, S. (1999). Knowing failably. The Journal of Philosophy, 96, 565–587. Hetherington, S. (2001). Good knowledge, bad knowledge: On two dogmas of epistemology. Oxford: Clarendon Press. Humberstone, L. (2000). Parts and partitions. Theoria, 66, 41–82. Lewis, D. K. (Ed.) (1999). Elusive knowledge. Papers in metaphysics and epistemology (418–445). Cambridge: Cambridge University Press. Merricks, T. (1995). Warrant entails truth. Philosophy and Phenomenological Research, 55, 841–855. Plantinga, A. (1993). Warrant: The current debate. New York: Oxford University Press. Weatherson, B. (2003). What good are counterexamples? Philosophical Studies, 115, 1–31. Zagzebski, L. (1994). The inescapability of Gettier problems. The Philosophical Quarterly, 44, 65–73.

Chapter 14

Knowledge that Works: A Tale of Two Conceptual Models Stephen Hetherington

1. A Methodological Question In this paper I describe two models of knowledge — two templates that might be used in fashioning a concept of knowledge, in assessing the presence or absence of knowledge. One of these models should be familiar, because it already guides much philosophical thinking about knowledge. But should it have that influence? It is epistemological orthodoxy; need it be? It amounts to a methodological choice, made at the beginning of many epistemological inquiries, as to how to approach thinking about knowledge in the first place. An implicit reliance upon this standard model is why epistemologists find so natural many of their claims as to when knowledge is, or is not, present. Yet, crucially, this usual methodological choice is not conceptually mandatory; an alternative model is available, as we will see. There is a real possibility of our having been thinking about knowledge in a way that is more conceptually restricted than we have realised. Moreover, it will transpire, that model springs from a historically explicable methodological choice, one we need never have made, and which we need not continue making.

2. Sceptical Possibilities and the Not-Yet Model For reasons that will become apparent, I call the usual model the Not-Yet model. It reflects a way in which epistemologists habitually manifest a wariness when Aspects of Knowing: epistemological essays Edited by S. Hetherington Copyright © 2006 by Elsevier Ltd. All rights of reproduction in any form reserved. ISBN: 0-08-044979-4

219

220

Stephen Hetherington

attributing knowledge. A wariness of what? A wariness of attributing knowledge when there is even a chance of being mistaken in one’s attribution. We ask to be convinced that knowledge is present, before attributing it. Nowhere is this clearer than in some discussions of sceptical possibilities. Consider this Cartesian sceptical doubt: Although you seem, to yourself, to be sitting on a chair, it is possible for this inner experience to be part of your dreaming sitting on a chair. In Stroud’s (1984, Chapter 1) reconstruction of Descartes’ dreaming argument, you are required to know that no such possibility obtains. If you lack that knowledge, then (infer this sceptic), you fail to know that you are sitting on a chair. That sceptical thinking directs us to this partial picture of how to conceive of knowledge: 1 Regardless of whatever else is involved in your knowing that p, this knowledge is not to be accepted as being present unless and until various sceptical possibilities are somehow removed. It is a picture routinely accepted by non-sceptics, too. This is clear when we ask what “removed” means in 1. The Cartesian possibility might be removed, for example, by your knowing that you are not dreaming: you would be satisfying this sceptic’s demand. Or perhaps someone provides a philosophical dismissal of the sceptic’s demand, showing its not needing to be satisfied. Sceptics challenge us either to satisfy or dismiss their demands; non-sceptics strive to meet that challenge. Both sceptics and these non-sceptics thus concur in not attributing knowledge that p unless and until sceptical possibilities are removed in some appropriate way — maybe by the epistemic subject, possibly by some “onlooker”. Now we may precisify the rest of 1 (where XYZ is a conjunction of conditions, which we may leave unspecified): 2 For any XYZ: your XYZing is to be accepted as your knowing that p, only once (i) various sceptical possibilities are somehow removed, and (ii) their-being-removed is included in XYZ.1 1 Suppose that XYZ is the traditional justified-true-belief analysis. Cartesian external world sceptics, for instance, allow that your knowing that p (for an external world p) is your having a well-justified true belief that p, only once you eliminate the independently specifiable possibility of your dreaming that p, and only once your eliminating this possibility is included in your justification for your true belief that p.

Knowledge that Works

221

And 2 instantiates what I call the Not-Yet model of knowledge.2 It is standard for an epistemologist to muse as follows: “Is this XYZ knowledge? Not yet may I say so. Not unless and until these sceptical possibilities are removed may I say so. (Or so I must assume, if I am to avoid begging the question against sceptics.)” Thus, a sceptic’s further condition — requiring the removal of a pertinent possibility — becomes accepted as an enabling or constituting condition of knowledge that p. Only then are other conditions accepted as being able, along with it, to constitute knowledge that p. In the meantime, the sceptic’s condition’s not being satisfied is claimed by sceptics (and often conceded, even if hypothetically, by non-sceptics) to prevent other conditions from constituting knowledge that p.3

3. Gettier Circumstances and the Not-Yet Model The Not-Yet model has also guided epistemological reactions to Gettier cases. The concept of such a case arises with Edmund Gettier (1963).4 Almost universally,5 epistemologists say that each Gettier case is an actual or possible situation containing a belief which is true and well (although fallibly) justified — yet without being knowledge, due to some odd circumstance. That interpretation unhesitatingly assumes 3: 3 Knowledge that p is accepted to be absent from any pertinent Gettier situation unless and until some pertinent aspect of the situation is somehow removed.6 2

David Chalmers suggested this modification of 2: 2* For any XYZ: Your believing that p is to be accepted as your knowing that p, in virtue of its being XYZ, only once (i) various sceptical possibilities are somehow removed, and (ii) their-being-removed is included in XYZ.

I prefer 2’s greater generality, to cover the possibility (mentioned more fully in footnote 15) that not all knowing is believing. 3 We might claim that only once any given condition within XYZ is satisfied are the remaining ones enabled, along with it, to constitute knowledge that p: if all are needed, each has this enabling power. But epistemologists generally proceed as though there is something methodologically special about the sceptic’s condition. Sceptical arguments treat supposedly troublesome further possibilities as independently specifiable tests to be imposed upon any XYZ purporting to describe all there is to knowing that p. 4 For more on the concept and its history, see Shope (1983) and Hetherington (2005b). 5 Exceptions are Hetherington (1998, 1999, 2001a, Chapter 3) and perhaps Weatherson (2003). 6 In what ways might this occur? One simple mode I have endorsed elsewhere (Hetherington, 2001b) involves becoming aware of the troublesome Gettier circumstance.

222

Stephen Hetherington

More precisely (with 4 standing to Gettier circumstances as 2 does to sceptical possibilities): 4 For any XYZ: your XYZing is to be accepted as your knowing that p, only once (i) any Gettier circumstance is removed (however this is to occur), and (ii) its-being-removed is included in XYZ. The standard interpretation of Gettier cases thus perpetuates the Not-Yet model of knowledge. Epistemologists offer hypotheses as to what constitutes knowledge: they propose initial values of XYZ. Then they encounter Gettier cases, before debating how best, exactly, to eliminate the cases’ odd circumstances.7 Meanwhile, knowledge is deemed absent from the cases. Knowledge is to be attributed only after Gettier circumstances are removed. And by seeing how to achieve this removal, epistemologists discover knowledge to be a variation on the previously proposed XYZ.8 So, the analytical investigation assumes that the epistemic significance of Gettier circumstances is constitutive: knowledge is absent if they are present; removing them would enable knowledge to be present.9

4. Some Failings of the Not-Yet Model So far, the Not-Yet model is the conjunction of the easily recognisable 2 and 4. Nonetheless, familiar though it is, we should be dissatisfied with it. This section describes two respects in which that model is needlessly narrow in what it allows, from the outset, to be modelled about knowledge. (1) The Not-Yet model does not do justice to the concept of some knowledge’s withstanding a threat (such as from a sceptical possibility). We often talk of knowledge in that way. Yet how well can we understand it via the Not-Yet model? On that model, we find ourselves inquiring as if there is no knowledge that p — there is something epistemically lesser; we are asking whether it is knowledge — until the possible “approach” of p’s being false, for example, has been independently evaded. But if no knowledge that p is allowed already to be “in place”, no 7

The “best, exactly” elimination of Gettier circumstances would apply to all Gettier cases, while satisfying other theoretical desiderata (such as being simple, explaining further aspects of knowing, and so forth), yet hopefully without setting so high a standard for knowledge as to imply a scepticism about knowledge’s ever being present. 8 For example, if XYZ is again the justified-true-belief analysis, and if a solution of the Gettier problem talks of there being no defeaters of one’s justification, this no-defeaters condition would be added to any prior specification of XYZ. 9 Having noticed this, however, I will mostly discuss sceptical possibilities, not Gettier circumstances, in the rest of the paper.

Knowledge that Works

223

such knowledge can be interpreted as having the opportunity, strictly speaking, to withstand that possible falsity. (What does not exist has no metaphysical chance to withstand threats to its existing.) Consequently, we need a way of inquiring that allows some knowledge to be present even when being assessed in relation to such threats. The Not-Yet model has conceptual room only for the idea of knowledge being recovered — returned to us — in the face of the threats. It does not allow us to understand the knowledge as coexisting while the threats are being assessed. Prior to the knowledge’s being recovered, the threats are to be removed. So, they are to be removed independently, without the knowledge’s contributing. It is usual for epistemologists to try disposing of sceptical doubts independently, trusting that this will then allow the knowledge to be recovered.10 (I think of this as a Cartesian recovery program for our knowledge. Conceptually, we relinquish the knowledge while independently disposing of threats; only then do we feel confident in reclaiming the knowledge — from where?) (2) Use of the Not-Yet model also fails to do justice to what it is to know fallibly that p.11 On that model, you are allowed to know that p only once sceptical possibilities, say, are somehow overcome or removed. In the meantime, you are deemed not to know that p. Hence, no instance of some XYZ can be regarded as knowledge which is yet to have overcome those threatening possibilities. In other words, any knowledge which the Not-Yet model allows to exist will have overcome all such distinctively philosophical knowledge threats. But knowledge like that has a fair claim upon being classified as infallible. So, the Not-Yet model does not permit fallible knowledge to be understood as present.

5. Towards an Alternative Model Section 4 reveals two desiderata for any alternative to the Not-Yet model. What unites these two is the idea of being able to assess a piece of knowledge even as it is in place as knowledge. Desideratum (1): only knowledge already in place could withstand sceptical threats, say. Desideratum (2): sometimes, knowledge needs to be able to be understood as fallible — in the sense of being in place, even if it has not 10

Non-sceptical epistemologists might not accept this description of the knowledge’s being recovered. “It is there all the time”, they will say, “although, in order to avoid begging the question, we must inquire as if it is absent. Only after besting the sceptic may we triumphantly display it again”. But when I say “recovered”, I mean something like “included in our best epistemological analysis, which we hope to use when attributing or denying knowledge”. 11 Or, more generally, failably that p. This concept comes from Hetherington (1999, 2001a, Chapter 2). In this paper I use the simpler, more standard notion.

224

Stephen Hetherington

yet overcome all sceptical possibilities, for example. We need a model that portrays — a way of conceiving of — knowledge as able to be present as knowledge, even while being tested, even in ways which might dispose of it. Maybe we must include its capacity for being tested within our conception of the knowledge’s normal functions. How might we do this? On the Not-Yet model, we consider a composite of conditions, XYZ, asking whether this is enough for knowledge, while assuming (so as to avoid begging various questions) that it is not. We ask whether specific further components are needed (such as knowledge of a sceptical possibility’s not obtaining, or some Gettier circumstance’s absence), if there is to be knowledge. (If these further components are needed, the previous specification of XYZ is to be revised accordingly.) In place of the Not-Yet model, we now see, a model of knowledge is needed which, at least for argument’s sake, allows the putative components described in the analysis XYZ to be knowledge. Then we will be asking, not whether specific further components are needed if there is to be the knowledge, but what further features such knowledge could have, without ceasing to be knowledge. Within this altered methodological setting, new challenges to knowledge—attributions may well arise. Will traditional sceptical doubts and Gettier challenges retain the impact many epistemologists currently accord them? That is yet to be seen. (And if they do not, there could be a correlative change in the list of core epistemological challenges.) I will call the alternative model the Working Knowledge model.

6. An Analogy We may understand the basic idea behind my alternative model through an analogy, between classifying something as knowledge and hiring a new employee. Ideally, the employer hires someone with all the skills, who knows all the details, to be required within the position. (Perhaps this person would have previously worked in a very similar position, needing either-no-or-minimal further training.) Alternatively, the employer could hire someone who shows enough signs of being able to learn the job’s details efficiently. Obviously, this is the usual method. It results in someone’s being hired who is yet to know everything which the employer will expect him to know. Maybe the employee is therefore hired on a tentative or probationary basis. Even so (other things being equal), he will be accorded full employment status, in spite of not possessing every skill or piece of knowledge ultimately wanted within the job. He is accorded the title and status of a full-time employee anyway. And significantly, he is thereby more likely, other things being equal, to become as good an employee as is desired. With the opportunity, the new responsibility,

Knowledge that Works

225

and the trust being shown in him, he can “grow into” the job — doing it more and more as, ideally, it should be done. Indeed, the onus is upon him to improve. This would occur as he learns the job while in the position. He would be using his potential, in building upon what was already enough to be doing the job. Promotions and salary increases, too, might ensue as this improvement occurs. All of this can happen within the scope of a single, continuing, job title and description. Does this amount to guaranteeing the person a job “for life”? No. Is it impossible for the person to lose the job? Of course not. Anyone could make bad mistakes in a job. And a person could do so before ever becoming the perfect employee (who could have been hired as someone needing no improvement or training). Nevertheless, it will be most evident whether the person is capable of doing the job, only once he is doing it. In general, the best test of his capacity to do the job is how he does do it. The best way to test him is to let him do the job. This does not entail that just anyone should be hired, letting him or her be tested in that best way possible — namely, “on the job”. Even without knowing for sure in advance when we are about to make a mistake in hiring, we apply discriminative criteria. Then we live with the results until we cannot do so any more: “He has to go. He isn’t up to the job”. And all of that is analogous to our epistemic cases, with epistemological inquirers functioning analogously to the potential employers. Satisfying a prima facie apt description XYZ is like having at least the objective potential to be an excellent or even perfect employee:12 one has done enough to be hired with that ideal in mind. But the Not-Yet model’s way of assessing whether an instance of XYZ is knowledge is like a search for an already perfect employee, who is not to be hired unless and until he is already a perfect employee. In contrast, the Working Knowledge model is like the initially less-demanding employer. That model is willing to treat being knowledge as like being an imperfect-but-trainable employee, who needs to develop within the job and who will be allowed to do so (even while risking failing to do so). On the Not-Yet model’s way of thinking, we would hire a person or accept an XYZ as being knowledge, only once no new development is needed in order for the person to do the job or for the XYZ to be knowledge. In other words, the knowledge is to be perfect, as that employee is to be perfect. On the Working Knowledge model’s way of thinking, though, we may hire someone, or we may accept an XYZ as being knowledge, while accepting that assorted challenges of training or development (such as are posed by the distinctively philosophical knowledge threats) still lie

12

“Perfect” here (and in what follows) is to be understood as “perfect, for the job in question”, not as “perfect, for any job”.

226

Stephen Hetherington

ahead, if the person is to do the job or if the XYZ is to be knowledge. The knowledge need not yet be what we ultimately want it to be as knowledge, just as that employee need not yet be what we ultimately want him to be as an employee. He will learn much of the job in the job. Even if he does not fully realise in advance that this will be so (“I’m qualified. I’ve got all the skills now!”), an experienced employer is likely to realise it. She should therefore not shy away from hiring the imperfectly experienced person. And we may approach assessments of knowledge analogously. By something’s being XYZ, it has the objective potential for being knowledge; so we should accept it as knowledge, at least for now.

7. Fallibly Working Knowledge The suggestion in Section 6 readily provides a way of conceiving of fallible knowledge. The simplest claim we could make towards that end is this: fallible knowledge is knowledge, with its fallibility just being part of it, one of its properties. Yet even that truism, as we saw in Section 4, is not obviously accommodated by the Not-Yet model. On this model (as Section 6 explained), knowledge is to be hired only as a perfect employee is to be hired: once knowledge is hired, therefore, it is not fallible; whatever potential there was for failure will already have been eliminated. But if knowledge is to be fallible, it must retain some chance of failure, even after it has been hired. In effect, its fallibility is now a capacity no longer to be knowledge. Its fallibility is thus its potential, even as something in place as knowledge, to cease being knowledge. Fallible knowledge, by being knowledge, is performing jobs that knowledge performs; however, it does so even while having the potential to be “dismissed” from its role as knowledge. It is knowledge which performs its job qua knowledge as well as its fallibility permits. There is a correlative risk of its ceasing to perform that job. Hence, it is knowledge by satisfying some description XYZ, even if there are as-yet-unremoved threats to its continuing to be knowledge in that way. Schematically, this is how we should conceive of fallible knowledge. Imagine being confronted with a putative instance of knowledge, satisfying some prima facie apt description XYZ.13 Suppose we accept that if it is knowledge, it is most likely fallible. The Working Knowledge model tells us that we will best discover whether this instance of XYZ is fallible knowledge, by first hiring it as knowledge (hiring it as a cognitive employee), and by then looking for whatever 13

I have not commented on how the supposedly apt description XYZ is initially to be derived. Perhaps it takes into account many non-philosophical kinds of knowledge–threat. It might be scientific in nature; it could be more “everyday”; maybe it has a conjectural aspect; and so on.

Knowledge that Works

227

fallibility it has. Only thus could we be finding its fallibility qua knowledge — that is, its being fallible knowledge. The Working Knowledge model is pragmatist. We accept and use the instance of XYZ as knowledge while accepting that, by being fallible, it could cease being knowledge. That moment could arrive suddenly. Like an unanticipated Humean alteration to the course of nature, what has been knowledge might no longer be knowledge. One day, its fallibility could be realised. Nonetheless, we need not — by acknowledging that the instance of XYZ could suddenly cease being regarded as knowledge — regard it already as not being knowledge. Sceptics in particular would have us do so;14 congruently, their Not-Yet model impels them towards that interpretation. By using that model, however, at the crucial interpretive moment they already treat the instance of XYZ as not being knowledge. So, it is not that sceptics hire — and then fire. They cannot even hire in the first place. Applications of the Not-Yet model begin by assuming that the knowledge in question within a given case is absent. The Working Knowledge model reverses that methodological procedure. We assume that the knowledge is present (once an XYZ is present), in order to be testing this assumption (such as against distinctively philosophical knowledge–threats like sceptical possibilities). At least sometimes, the assumption could be what Jacquette (2004) calls a working hypothesis. This is an attempt to formulate a hypothesis so that it can be an “official” hypothesis, to be tested as an hypothesis: “There is less commitment to such a pre-formulation of an assumption even as an hypothesis” (p. 185). If this is the spirit in which an XYZ is being deemed knowledge, it would be correspondingly unfair for sceptics to complain that the question is being begged against them. All that is not being conceded is the XYZ’s not being knowledge — which is apt, because it is what the sceptics want us to test. The fact of our doing so by treating the XYZ as knowledge (at least for now) is not unfair. We will be asking whether such knowledge withstands particular sceptical possibilities, for instance. We will ask whether, given its presumed fallibility, it does so. And we will not follow the Not-Yet model in assuming that only once a given sceptical doubt is somehow removed is the knowledge wholly constituted — and hence to be deemed present. The Working Knowledge model portrays us as (1) testing independently (even if fallibly) constituted knowledge, rather than as (2) showing (even proving) that something is knowledge in the first place. The latter approach 14

So too, I have urged (Hetherington, 1998), do epistemologists in general when responding to Gettier cases. Instead (I argued), we may interpret those cases as situations where someone knows but luckily so, because he could easily have failed to possess the knowledge (see also Hetherington, 1999, 2001a, Chapter 3; 2005b, Section 13).

228

Stephen Hetherington

gives us the Not-Yet model, on which no instance of XYZ is accepted to be knowledge unless and until sceptical possibilities are overcome — with the overcoming of them being deemed to be a constitutive part of the knowledge (being added to, or modifying, the description XYZ). Alternatively, the former approach gives us the Working Knowledge model, one statement of which is as follows: WK Any prima facie apt XYZ can be accepted as an instance of knowing that p, although various circumstances (including distinctively philosophical knowledge-threats, such as sceptical possibilities) could test XYZ’s capacity to remain as that knowledge that p. In effect, knowing would be treated as something other than a “final” cognitive state, relative to p.15 It would be accepted as something “alive” — developing, ongoing, and testable. It would be attributed as a work in progress. It need not be attributed with finality. All is testable, even something’s being knowledge.

8. Gradualism and Scepticism We might wonder whether the Working Knowledge model really does allow the testing of whether instances of XYZ are cases of knowledge. Is it a dogmatic model, in the sense of attributing knowledge too easily? Popperian destroyers. To assume that some instance of XYZ is knowledge is not to assume that it will survive all testing of whether it is knowledge.16 Many people withdraw claims and lose beliefs when encountering sceptical thoughts or when deciding that they have been (and could still be) in a Gettier situation. In this sense, whatever is knowledge need not remain knowledge. Employees are tested even while they are employees. Sometimes they fail the tests, thereupon ceasing to be employees. They fail as employees, thereupon ceasing to be employees — just as some knowledge could fail as knowledge, thereupon ceasing to be knowledge. So, in principle sceptical doubts can destroy knowledge. This is how we should think

15

Elsewhere (Hetherington, 2006), I argue for its not being a state at all. It is an ability, a kind of knowledge-how. This paper’s point blends smoothly with that one. 16 Presumably, some of these tests will be subtle; any prima facie apt description XYZ should already have passed comparatively obvious tests. For instance, because one component in any prima facie apt description XYZ is a truth requirement, I am not saying just that something we think is knowledge might turn out to be false, hence not knowledge.

Knowledge that Works

229

of their potential to cause epistemic damage.17 We are used to thinking of them as knowledge-preventing threats. But that is not mandatory. It reflects our being trained to think of knowledge in terms imposed by the Not-Yet model. Bypassing that model allows us to reconceptualise sceptical threats. (And if this seems to weaken their general epistemic impact, so be it.) We can think of them as akin to potential Popperian destroyers of knowledge which has been present so far (rather than as potential Cartesian preventers of there being knowledge in the first place). On the Working Knowledge model, attributions of knowledge can be discarded: what had previously been accepted as knowledge would no longer be regarded as knowledge (because now a sceptical possibility has not been overcome, say); and this could occur without entailing that knowledge never really had been present. The Working Knowledge model allows that there can be knowledge now, which fails to be knowledge later — by later ceasing to be knowledge, by later losing a battle with some threat. In effect, on the Not-Yet model, “x knows that p” means that, at least in theory, x can be proved to know that p; whereas on the Working Knowledge model, “x knows that p” means that x has not yet been proved not to know that p.18 Improved knowledge. Moreover, even if we decide that a sceptical threat is withstood by a piece of knowledge (some instance of XYZ), with the knowledge surviving that threat, we need not infer that the threat lacks all epistemic significance. For we might wonder whether the surviving of it has had a qualitative impact upon the piece of knowledge. The Working Knowledge model reveals how we may conceive of this as being so. The Not-Yet model allows us to regard sceptical possibilities only as having absolutist consequences: either we accept that they prevent some XYZ’s being knowledge; or (if they are overcome) it is allowed to be knowledge. But the Working Knowledge model permits us to accord such possibilities a further way, a non-absolutist way, of making an epistemic impact. We may think of such possibilities as akin to job-threatening challenges arising at work, threatening an employee with losing her position. Then we should call to mind this popular saying: whatever does not kill a person makes her stronger. By applying what is true in that saying to the present case, we derive this moral: whenever the employee survives one of these threats, she could well (all else being equal) be somewhat better as an employee for having engaged with, and bested, the threat. And the same is true, mutatis mutandis, whenever some knowledge survives a sceptical challenge. The Working Knowledge model permits us to interpret this

17

In this sense, they are worries especially for those who are thinking about them — a theme pursued by Lewis (1996) and other contextualists. It was also investigated in Hetherington (1992). 18 Andy Clark helped me with this formulation.

230

Stephen Hetherington

survival as strengthening that knowledge (other things being equal). On the Not-Yet model, the knowledge is deemed enabled to be present if a sceptical threat is overcome; nothing else is thought to be achieved. (Notice how often epistemological focus is on whether there can be knowledge — in some sense supposedly established by undermining sceptical arguments.) On the Working Knowledge model, the knowledge can again be seen as enabled — but it can also be deemed ennobled — by overcoming a sceptical threat. Remember that, on the Not-Yet model, the knowledge itself cannot be conceived of as overcoming the threat, strictly speaking. Independently, the threat is to be overcome, thereby enabling the knowledge to exist. But on the Working Knowledge model we may say that the knowledge itself is what does battle with the threat. So, if it wins, it can be strengthened, ennobled, as the knowledge it continues to be. It can have “grown” as that knowledge, a possibility not envisaged on the Not-Yet model. That conceptual possibility of ennobling, rather than just enabling, a piece of knowledge flows naturally into a theory I have elsewhere (Hetherington, 2001a) labelled a gradualism about propositional knowledge. On that theory’s central idea, any instance of knowledge that p (once constituted by satisfying some apt XYZ) is more or less qualitatively good as knowledge that p. This qualitative dimension is epistemic. Any case of knowledge that p can, equally well, be described as being of some epistemically lesser or greater degree or quality of knowledge that p. And this status need not be static or unchangeable. For instance, if you improve your evidence for p, then (other things being equal) you thereby improve your knowledge that p, insofar as the evidence’s presence is part of the knowledge’s presence. One instance of knowledge that p can be epistemically better than another, insofar as the former’s epistemic core (such as some form of justification) is better than the latter’s. (And it is epistemologically standard to accept that justification can be gradational.) One person’s knowledge that p can be better than another’s. A single person’s knowledge that p can be better at one time than another. There might be no limit to how much a case of knowledge that p could be improved in such ways. As sceptical possibilities mount, so do dangers to our knowledge — to knowledge we have (allows the Working Knowledge model). If that knowledge withstands those threats, it is correlatively stronger than it might have been. Even so, it never loses its fallibility. Accordingly, it is not only more or less fallible as knowledge; it is endlessly so. The Not-Yet model, we found, has a contrary implication: when it is applied to the evaluation of a sceptical possibility’s impact, knowledge is treated as disappearing — at best, as going into hiding until danger passes. (So, as we also saw, on the Not-Yet model the knowledge itself cannot be interpreted as withstanding the sceptical possibility. Nor, therefore, can it rightly be thought to be fallible when doing so. It cannot be regarded as strengthened

Knowledge that Works

231

by that experience, either.) But the Working Knowledge model allows us to accept that any piece of knowledge, while ever it is being used, is open to being destroyed, even by philosophically distinctive knowledge–threats. Nonetheless, we may say, while ever the knowledge has not been destroyed, it continues surviving; and it can thereby be strengthened, improved, by the number and seriousness of the tests it has passed. In theory, too, there is no upper limit to how many tests it could undergo. In principle, therefore, it is endlessly fallible and endlessly improvable. More epistemic bounty can be gained from doing battle with scepticism than we might have anticipated gaining.19 Worsened knowledge. Still, there is no guarantee of such improvement. The Working Knowledge model allows a sceptical challenge also to be conceived of as lowering some knowledge’s quality. At present, through the Not-Yet model, epistemologists generally treat sceptical challenges as being coherent and applicable in advance of — and hence as playing a part merely in conceptually constituting — whether a particular XYZ is knowledge. But this is like questioning whether a person should be hired, given only that there are pertinent tests which he cannot yet pass; of course he cannot pass all of them before he is hired. Thus, the Working Knowledge model becomes significant. On it, coherent sceptical challenges to knowing arise only once there is a working presumption of knowing. They are like tests arising at work, subjecting an actual employee to actual pressure. We cease worrying so much about whether a potential employee would survive that kind of pressure.20 And we should bear in mind that an employee who fails a test, even if she does not lose her job, might thereby be revealed to be less good at the job than had been assumed (by herself or others). The same is true, mutatis mutandis, of knowledge. Another gradualist moral appears among our conceptual options: Just as surviving a sceptical challenge might strengthen a piece of knowledge, failing the challenge might lower that knowledge’s quality. For example, suppose that you cannot eliminate the possibility of your dreaming. Do you thereby lack the knowledge that you are sitting in a chair? Sceptics say so,

19 Was Moore’s (1959) famous “hand waving” reply to the sceptic applying the Working Knowledge model? Certainly, it has been thought of as dogmatic. Certainly, too, it did not apply the Not-Yet model. Elsewhere (Hetherington, 2001a, pp. 169–178), I provide a gradualist use of Moore’s reasoning, one which could be regarded as applying the Working Knowledge model. 20 It is not that we should never conduct this sort of pre-emptive imaginative test. Often, we must take a chance on hiring someone, recognising that although we can think of ever more demanding tests, these need not become initial tests, used when considering whether to hire the person in the first place. Sceptical challenges are indeed rarefied, appropriate only for a later stage in the cognitive employment process.

232

Stephen Hetherington

and the Not-Yet model abets that interpretation. But on the Working Knowledge model, we are free to infer that you might pay only the price of knowing less well (all else being equal) that you are sitting in a chair, perhaps less well than you assumed you did.21 Thus, my proposal need not deny sceptics all possible conceptual victories. But it does conceive anew what forms their victories could take, hence what significance there could be in sceptical thinking. In general, the Working Knowledge model promotes a conception of knowledge that is ultimately less accepting of sceptical worries than is existing epistemological thinking, based upon the Not-Yet model. In effect, armed with the Not-Yet model, epistemologists treat knowledge that p as ultimate or unimprovable knowledge that p: only once the “needed” improvements are completed is there accepted to be knowledge. (These improvements are made in one’s overall epistemic situation, regarding p — such as if one gains knowledge of not dreaming that p. Only then is this summed up as … well, as one’s knowing that p.) In contrast, with the Working Knowledge model, epistemologists treat any instance of knowledge that p as potentially working towards being ultimate or unimprovable knowledge that p: in the meantime, however, almost all cases of knowledge that p are accepted as being improvable — nonultimate. (Those improvements deemed by the Not-Yet model as being made only in one’s overall epistemic situation regarding p are now able to be seen to be made in the knowledge that p itself, along the way.) By the same token, though, the Working Knowledge model allows epistemologists the conceptual option of seeing any case of knowledge that p as doubly fallible — able, alternately, to be lost or to be lesser. There could well be more to knowing that p than the Not-Yet model allows us to notice.

9. Working Knowledge in Gettier Cases Does the Working Knowledge model offer us new conceptual freedom in interpreting Gettier cases? I believe so. The standard epistemological interpretation of these is that, within each, some justified true belief fails to be knowledge. And epistemologists say that this interpretation gives voice to their “intuitions” about the cases. But we noted in Section 3 how the Not-Yet model, a substantive and theoretical model of knowledge, underlies those so-called intuitions: they are not

21

Elsewhere (Hetherington, 2001a, Chapter 2; 2002), I explain more fully how my gradualism defuses sceptical challenges. (“Does this beg the question against the sceptic?” No, as I have also argued: Hetherington, 2001a, pp. 38–40; 2002, pp. 95–97; 2004.)

Knowledge that Works

233

so “intuitive”. One indication of this is how differently we might wish to interpret Gettier cases once we adopt the Working Knowledge model. I will illustrate this briefly, via a representative case. The sheep in the field (Chisholm, 1966, p. 23n). You gaze into a field, seeing what looks like a sheep. So, you believe there is a sheep in that field. You are right, because there is one there — hidden from your sight, behind a hill. (What you are seeing is a disguised dog.) Hence, your belief is true. It is also justified, through a standard use of your senses, followed by a normally good inference. Is your justified true belief-that-there-is-a-sheep-in-the-field knowledge? “Of course not”, implies the Not-Yet model. But if we put that model to one side, we may interpret the case afresh. Thus, for a start, your belief within the case is fairly well qualified to be hired as knowledge. By being well justified and true, it satisfies a prima facie apt description XYZ of a sufficient condition of knowledge — in practice, of working knowledge. It is fallibly justified, though: even with its justification in place, the belief neither had to be true nor has to continue being true. (As soon as the unseen sheep wanders out of the field, the belief is false.) This implies that the belief neither had to be knowledge nor has to continue being knowledge. Is it therefore not knowledge, right now? The Working Knowledge model does not commit us to thinking so. Rather, we may regard this as an instance of knowledge, which is close to not being knowledge. For it is close, and remains so, to ceasing to be a belief that is both justified and true. At present, it is like an employee who almost made a bad mistake but did not, and who remains close to making that mistake but is not making it, all the while somehow continuing to work professionally and reasonably. If the mistake had occurred, she would have been fired; it could easily have occurred, due to circumstances beyond her control; in the meantime, she is working well — even if more tenuously so than she realises. To insist that there is a lack of knowledge in this case is like refusing to hire anyone in the first place unless and until no luck will ever be needed in her performing the job without mishap. That insistence is like saying, unrealistically, that a person is not fit to be employed initially in the job if she would ever (unwittingly) need to be somewhat lucky for a desired outcome within the job to occur (even when performing the job professionally and responsibly). Just as we hire people while accepting that at times luck might well (even in unseen ways) attend their good efforts, we may attribute knowledge, even quite fallible knowledge, in that same spirit. Just as we do not fire a person as soon as such luck arises, we should not deny knowledge as soon as the corresponding luck arises. In each case, we were allowing for some fallibility from the beginning; and fallibility, in turn, allows for some such luck. Once such luck does arise, then (given that no mistake or harm thereby occurs) it would be unfair to fire the employee. Similarly, it would be

234

Stephen Hetherington

unfair to deny the presence of knowledge. (Remember that no mistake or harm occurs; the relevant belief within a Gettier case is true.)22

10. The Not-Yet Model’s Reach It might be objected that not all analytical epistemologists apply the Not-Yet model: “We also discuss people whom we describe as having knowledge”. However, even then, contemporary epistemologists implicitly accept that there is such knowledge only so long as threats such as sceptical possibilities, and/or such actual or possible Gettier circumstances, have been overcome or removed (in the sense described in Section 2). In this way, the Not-Yet model remains at the core of contemporary analytic epistemology. Accordingly, even when attributing knowledge, epistemologists do so within the shadows of a way of thinking that is overly concerned with possible ways of lacking knowledge. It is a way of thinking that is fundamentally reluctant to attribute knowledge when there are “nearby” ways of lacking it (ways that involve luck, accidentality, fallibility, and so on). In particular, modern epistemologists worry deeply about possible ways of being mistaken. So, even when attributing knowledge, these theorists are attributing something that is meant to be understood in terms of the Not-Yet model — this way of not attributing that “something”, of delaying an attribution until any epistemic dangers are found to be overcome. Descartes is probably the paradigm exemplar of this way of thinking about knowledge, at least as it has influenced modern epistemology. But because I have presented an alternative to it,23 I have shown how the Not-Yet model is conceptually optional.

11. A Platonic Precedent and a Cartesian Constriction That model is also historically optional. Many epistemologists, I suspect, feel indebted to Descartes for his clear-minded use of what I am labelling the Not-Yet model. To them, it feels like a natural way of conceiving of knowledge, whenever they seek to reflect analytically and detachedly upon the nature and possibility of 22 I hope it is clear that my case for the interpretive feasibility of thinking there is knowledge within Gettier cases is not stemming from an “intuition” to that effect. The argument is more theoretical. If Section 3 is right, though, so are the putative intuitions with which epistemologists standardly deny knowledge to anyone featured within a Gettier case. That is, use of the Not-Yet model is a theoretical move. 23 Would-be fallibilists will say that they already have one. But we have seen that they are wrong about that until they discard the Not-Yet model. Simply claiming to be a fallibilist is not sufficient for being one.

Knowledge that Works

235

knowledge. Nonetheless, their reaction might reflect a historically implanted and reinforced conceptual limitation. After all, epistemology’s history has not presented us with only that model. No less a philosophical predecessor than Plato used something like the Working Knowledge model. Why was it ever discarded? Plato’s endorsement of at least part of a Working Knowledge model of knowledge is attested to by Annas (1981, p. 193). In her interpretation of the Republic (Book 10, 601b–602b), she presents this as his view: Knowledge is not opposed to scepticism. The craftsman’s beliefs are true, and fine as far as they go; Plato never suggests that they might be false, or that we should try doubting them. The user’s state is better than the maker’s, not because he is more sure of anything, but because he has understanding of the subject-matter and its point in a way that the maker lacks. … [T]he person with knowledge is contrasted not with the sceptic but with the person who, for practical purposes, takes over true beliefs in an unreflective and second-hand way. To know is to understand (ibid., p. 212), which is what a user of knowledge does. Moreover, such knowledge can involve improvement. This is improvement on true belief, an improvement in “the knower’s relation to the objects of his or her true belief” (ibid.). And this “kind of improvement … is not one familiar from the post-Cartesian tradition” (ibid., p. 192), because “[i]t is not an increase in certainty, or a relief from doubt” (ibid., pp. 192–193). Thus (ibid., p. 212), for Plato, he has no reason to think that knowledge will be achieved by a negative, corrosive doubting of all our previous beliefs. Rather than undercutting the beliefs we have, we advance to understanding their significance and finding them intelligible. Of course, when we have knowledge, we may find that some of the beliefs we accepted are in fact false. But we do not begin by trying to doubt the truth of particular beliefs in a wholesale way, as Descartes recommends that we do. Descartes’s method of doubt applied a special case of the Not-Yet model. Hence, in conceptual terms it is far removed from this Platonic conception of knowledge. Annas tells us (ibid., p. 200) that in Book 5 of the Republic, Plato finds it natural to think of knowledge as coming in degrees which vary with the intelligibility of its object; and this is because he is

236

Stephen Hetherington not thinking of knowledge as the result of excluding sceptical doubt. … [F]or him, the advance to knowledge is a progress to increased understanding, and this comes about not by focussing ever more sceptically on one’s grounds for a particular belief, but by setting the belief in a wider context of one’s other beliefs and their mutually explanatory relationships. It is because knowledge comes with an increase of explanation and of understanding that Plato finds it natural to think of it as a matter of degree, rather than as an all-or-nothing matter of sheer certainty.

Was it epistemological progress for Descartes to replace such a model of knowledge with his own — the Cartesian one, which has so markedly shaped the epistemological values and methods to which contemporary philosophers are introduced when reflecting upon knowledge? Not if this paper is correct. It is hardly the case that Descartes showed that there was a philosophical failing or inadequacy in instances, such as Plato’s, of the Working Knowledge model of knowledge. What seems to have prompted Descartes’s version of the Not-Yet model was more an attitude than a demonstration — specifically, an attitude of methodological caution. From where did this attitude come? When and where Descartes was writing, knowledge was a theologically charged prize. Certainly, knowledge as an object of Descartes’s philosophical attention was like that, because he wished to prove that Christian knowledge could survive 17th-century scientific advances and Pyrrhonian sceptical worries.24 What would such Christian knowledge encompass? Part of it was to be knowledge of essences in the world — this being knowledge of how nature has to be in its details, given its necessarily existing and involved Creator. And this requirement upon the knowledge’s content had a significant metaphysical implication regarding the knowledge’s nature. It meant that the knowledge itself would have that same kind of metaphysical stability. In other words, instances of knowledge were to be essentially those instances of knowledge. That is, knowledge was conceived of (in advance, on doctrinally inspired grounds) as being such that any instance of it has to be the knowledge it is. Yet how could this be so, especially whenever there is knowledge of a contingent p? (If p had been false, the belief that p would not have been knowledge.) The answer is that nothing is to be knowledge if it has the potential within itself to cease being that knowledge. The Church would not have accepted that susceptibility, that kind of constitutive frailty, as ever being a feature of some knowledge. But one instance of that kind of susceptibility is

24

On this motivation of Descartes’s, see Popkin (1979, Chapter IX).

Knowledge that Works

237

what we now call epistemic fallibility, with fallible knowledge being so constituted that even with its other components being in place, its truth component need not have been. (Hence, even a belief’s continuing to be justified as it presently is will not entail its remaining true.)25 Fallibility would thus have been deemed to be incompatible with the kind of knowing which is needed if there is to be Churchapproved religious knowledge. Such knowledge is knowledge essentially.26 It is not surprising, then, that from the outset Descartes applied a model of knowledge which reflected that expectation of what knowledge would need to be like. It was a methodologically cautious model, such caution being religiously appropriate — meaning that knowledge would be attributed only with complete confidence, having satisfied that model. Nothing was to be accepted in a tentative way as being knowledge, therefore, because no such mode of acceptance is appropriate unless the something in question is essentially knowledge. To use this paper’s description: No case of XYZ could be “hired” as knowledge unless it was already seen to be fully or unimprovably formed and developed as knowledge. There could be no acceptance of the possibility of developing a piece of lesser knowledge that p, say, into what it was wanted to be — perfect or unimprovable knowledge that p. For there could never be an acceptance of the possibility of lesser or improvable knowledge that p in the first place. How could the existence of God, for example, be thought to be known of in that way? This would be unthinkable. Thus, an assumption of anti-gradualism — absolutism — about knowledge was spawned. In that way, too, the Not-Yet model was adopted. And it was adopted prior to ascertaining whether any particular belief was knowledge. A Not-Yet attitude, as it may be called, was already in place. Descartes was being constricted from the start of his inquiry in his thinking about any putative instance of knowledge. He was being constricted, not only by the Pyrrhonian sceptical doubts being heatedly discussed at the time, but by the Church’s demands upon people as knowers.

25

I said that such fallibility is one instance of the constitutive frailty that the Church would not have accepted within a piece of knowledge. If we generalise that notion of fallibility, we derive the concept of failability (referred to in footnote 11): a piece of knowledge is failable insofar as any one of its components (not only its truth component) need not have been present, even given the presence of the other components. Accordingly, the concept of failability more accurately captures the deeper constitutive frailty that Descartes would not have accepted within knowledge. For simplicity, though, in this paper I talk just of fallible knowledge, not of failable knowledge, as being anathema to Descartes’s Church-formed investigative environment. 26 This is not to deny the role of faith in religious thinking. But it was the Pyrrhonist sceptics who were taken to be opening the door to faith (Popkin, 1979, pp. 52–53). In contrast, Descartes sought religious knowledge.

238

Stephen Hetherington

12. Modern Epistemology and Unwitting Cartesianism That was then, though; this is now. And epistemologists no longer believe themselves to be beholden to the kind of doctrinally authoritarian power that, in effect, restricted how Descartes was allowed to conceive of knowledge. In general, philosophers do not seek to establish the presence or the absence of knowledge in accord with how some Church ordains knowledge to be. Moreover, although science’s influence remains intellectually strong, its 17th-century emphasis upon finding natural essences does not. And, in keeping with that shift of focus, the fact of our fallibility has become generally accepted, even scientifically so. Most epistemologists believe that even when we know that p, we do this fallibly — with our evidence, say, not having guaranteed the truth of our belief that p. In such ways, I take it, most philosophers believe that epistemology has progressed since Descartes’s time. Those antecedent doctrinal restrictions will be deemed to have been his, not ours, leaving us free to adopt a fallibilist conception of knowledge and to apply it when attributing or denying knowledge.27 Even so, we have seen why contemporary would-be fallibilists have not succeeded in moving as far from Cartesian epistemology as most probably believe has occurred. They have not fully cast off Descartes’s impliedly infallibilist methodological shackles. They will believe themselves to have done so, with the Cartesian quest for certainty no longer being seen as central to epistemological thinking. Nevertheless, the underlying methodological key to Descartes’s epistemology was not his search for certainty. As is widely accepted, a more fundamental element was his method of doubt. And we should now appreciate that his method of doubt depended in turn upon a prior receptivity to something more fundamental still — namely, the Not-Yet model of how to think about knowledge at all. Descartes was presuming that nothing could be determined only tentatively to be knowledge. Doubts would therefore need to be removed first of all. By thus applying the NotYet model, Descartes was methodologically bound to respect doubt and certainty. To expunge the strongest doubt would be to establish the strongest anti-doubt — certainty. Yet although contemporary epistemologists generally seek to be fallibilists in their conceptions of knowledge, and although they do not routinely and overtly embrace the method of doubt, we have seen how they do reach for the Not-Yet model of knowledge. This underlying aspect of Cartesian methodology has persisted. And Section 4 showed that anyone applying the Not-Yet model is committed (even if unwittingly) to an infallibilism about knowledge. Professed fallibilists still cleaving to the Not-Yet model are therefore confused, in a way of which the admitted infallibilist Descartes was innocent. 27

For more on the nature of fallibilism, see Hetherington (2005a).

Knowledge that Works

239

But there is no good reason for us to be relying upon the Not-Yet model. We have noted its conceptual limitations and conceptual optionality. Now we are noting its historical optionality. The surrounding social reasons that made it natural for Descartes to rely upon the Not-Yet model are no longer in place. For a start, contemporary epistemologists in general are not doctrinally answerable to any Church. Nor need they be methodologically constrained by any Church, having to regard knowledge with the exaggerated and pre-emptive caution that guided Descartes’s assessments of knowledge’s presence. So, in continuing to apply the Not-Yet model contemporary epistemologists seem mired in a philosophical habit of thought. It is time to break that habit. A first step would be our recognising it as a habit. A second step involves remembering that habits can run deep, becoming “intuitive” and reflexively attitudinal, without always feeling like habits. We have taken to evincing a Not-Yet attitude of caution towards attributing knowledge. And this can feel like something other than such a habit. (1) It can feel unlike an attitude of caution, as we reach confidently for “intuitions” regarding particular cases. (2) It can also feel as though it is not just a habit, because it can feel like a process of reflecting, not reflexing — autonomous and free thinking, applied carefully and even surprisingly anew on different occasions of inquiry. Nonetheless, it can still be a methodologically cautious habit of thought — an eminently replaceable one, too. For what underlies it is our having adopted the correlative Not-Yet model of knowing; and this is not an inescapable model. Over time, its implicit use has come to feel natural, even unavoidable, for epistemologists. But some theses and issues about knowledge arise only because the Not-Yet model has been guiding inquiry; and given how second-nature that model has become for epistemologists, some of those implied theses and issues have come to feel unavoidable, even intuitive. What are actually doctrinal implications of adopting the Not-Yet model have been widely discussed without epistemologists noticing the presence and sustaining role of the substantive presupposition that is the Not-Yet model. Yet if this paper is right, we are free to discard the Not-Yet model.28 And the heartening news is that once we do discard that model, we need not be left directionless, stranded nobody-knows-where on our epistemological journey. A replacement model, the Working Knowledge model, is ready to hand, even if schematically so. We may replace our professionally inculcated assumption of the Not-Yet model’s inescapability, therefore, with a recognition of the Working Knowledge model’s availability — all the while acknowledging the real possibility 28 There is some methodological similarity between this paper and Williams’s (1991) project of uncovering the substantive-and-therefore-not-purely-intuitive-and-unavoidable presuppositions of external world sceptical reasoning.

240

Stephen Hetherington

of some professionally trained ideas and expectations having been formed by our implicit acceptance of that replaceable Not-Yet model. Knowledge might well not look quite the same to us after we begin “hiring” it in accord with the Working Knowledge model rather than the Not-Yet model.

References Annas, J. (1981). An introduction to Plato’s Republic. Oxford: Clarendon Press. Chisholm, R. M. (1966). Theory of knowledge. Englewood Cliffs, NJ: Prentice-Hall. Gettier, E. L. (1963). Is justified true belief knowledge? Analysis, 23, 121–123. Hetherington, S. (1992). Epistemology’s paradox: Is a theory of knowledge possible? Savage, MD: Rowman & Littlefield. Hetherington, S. (1998). Actually knowing. The Philosophical Quarterly, 48, 453–469. Hetherington, S. (1999). Knowing failably. The Journal of Philosophy, 96, 565–587. Hetherington, S. (2001a). Good knowledge, bad knowledge: On two dogmas of epistemology. Oxford: Clarendon Press. Hetherington, S. (2001b). A fallibilist and wholly internalist solution to the Gettier problem. Journal of Philosophical Research, 26, 307–324. Hetherington, S. (2002). Fallibilism and knowing that one is not dreaming. Canadian Journal of Philosophy, 32, 83–102. Hetherington, S. (2004). Shattering a Cartesian sceptical dream. Principia, 8, 103–117. Hetherington, S. (2005a). Fallibilism. The Internet Encyclopedia of Philosophy, http://www.iep.utm.edu/f/fallibil.htm. Hetherington, S. (2005b). Gettier problems. The Internet Encyclopedia of Philosophy, http://www.iep.utm.edu/g/gettier.htm. Hetherington, S. (2006). How to know (that knowledge-that is knowledge-how). In: S. Hetherington (Ed.), Epistemology futures (pp. 71–94). Oxford: Clarendon Press. Jacquette, D. (2004). Editor’s page: Working hypotheses. American Philosophical Quarterly, 41, 185–186. Lewis, D. (1996). Elusive Knowledge. Australasian Journal of Philosophy, 74, 549–567. Moore, G. E. (Ed.) (1959). Proof of an external world. Philosophical papers (pp. 127–150). London: George Allen & Unwin. Popkin, R. H. (1979). The history of scepticism from Erasmus to Spinoza. Berkeley: University of California Press. Shope, R. K. (1983). The analysis of knowing: A decade of research. Princeton: Princeton University Press. Stroud, B. (1984). The significance of philosophical scepticism. Oxford: Clarendon Press. Weatherson, B. (2003). What good are counterexamples? Philosophical Studies, 115, 1–31. Williams, M. (1991). Unnatural doubts: Epistemological realism and the basis of scepticism. Oxford: Blackwell.

Author Index Alston, W. P., 23, 53, 64, 71, 109 Anderson, A. R., 36 Annas, J., 235 Anscombe, G. E. M., 183, 186, 191, 193, 195, 196 Aristotle, 3, 51 Armstrong, D. M., 104 Austin, J. L., 9, 115, 116, 128 Ayer, A. J., 171 Belnap, N. D., 36 Bender, J., 18 BonJour, L., 21, 32 Burge, T., 71 Burnyeat, M., 110 Cappelen, H., 136 Carroll, L., 41 Cavell, S., 98 Chisholm, R. M., 71, 233 Cohen, S., 72, 119, 134 Conant, J., 97 Copi, I. M., 158 Cottingham, J., 99 Cruz, J., 71 Davidson, D., 128 Davies, M., 71 DeRose, K., 134, 141, 179 Dretske, F., 86 Edgley, R., 105 Egan, A., 134

Engel, M., 23 Etchemendy, J., 160 Everitt, N., 21 Feit, N., 197 Feldman, F., 28 Feldman, R., 21, 167, 197 Fisher, A., 21 Fogelin, R. J., 119 Foster, J., 171 Fumerton, R., 21, 23, 24 Garfinkel, H., 126 Gettier, E. L., 152, 209,221 Ginet, C., 71 Goldman, A., 144, 166, 176, 186, 189 Gopnik, A., 184 Grice, P., 143 Harman, G., 49, 85 Hawthorne, J., 86, 93, 122, 134, 146 Heathcote, A., 155, 157, 159, 161, 165 Hetherington, S., 7, 11, 68, 69, 130, 132, 144, 166, 189, 197, 221, 223, 227–232, 238 Hintikka, J., 194 Hornsby, J., 185 Howard–Snyder, D., 197 Howard–Snyder, F., 197 Humberstone, L., 36, 209 Hume, D., 106, 126 Hursthouse, R., 191, 193 241

242

Author Index

Jackson, F., 37, 47, 49, 57, 71 Jacquette, D., 227 Joske, W., 63 Kaplan, D., 39 Kaplan, M., 116 Koppelberg, D., 27 Korcz, K. A., 26 Kripke, S.A., 39 Kyburg, H., 64 Lepore, E., 136 Levi, I., 64 Lewis, D., 50, 119, 126, 134, 139, 151, 198, 216, 229 Libet, B., 189 Lycan, W., 67 Macarthur, D., 101 MacFarlane, J., 134 McDowell, J., 197 Merricks, T., 208, 214 Milne, P., 160 Montague, R. L., 39 Moore, G. E., 9, 67, 77, 231 Moran, R., 98, 105, 191, 192 Mulligan, K., 155 Murdoch, D., 99

Pollock, J., 71 Popkin, R. H., 236, 237 Post, J. F., 21 Pryor, J., 71 Putnam, H., 49, 99, 121, 160 Quine, W. V., 1, 2, 21, 111, 162 Read, S., 155 Reiter, D., 23 Russell, B., 21, 166 Schiffer, S., 72 Searle, J., 184, 186, 193 Sellars, W., 49 Shope, R. K., 221 Simons, P., 155 Smith, B., 155 Sosa, E., 120 Stanley, J., 134, 136 Stoothoff, R., 99 Strawson, P., 63, 112 Stroud, B., 97, 130, 220 Tarski, A., 160 Travis, C., 116, 128 Trevarthen, C., 186 Unger, P., 97, 177

Nagel, T., 110, 111 Neurath, O., 8 Newton–Smith, W., 186 Oakley, I. T., 19, 23 Pastin, M., 28 Peacocke, C., 71 Pickard, H., 184 Plantinga, A., 54, 208 Pojman, L., 21, 31

Velleman, J. D., 189 Vogel, J., 84, 91 Weatherson, B., 134, 135, 217, 221 Williams, M., 99, 109, 239 Williamson, T., 178 Wittgenstein, L., 107, 112, 183 Wright, C., 69, 71 Zagzebski, L., 197, 214, 216

Subject Index absolutism, 237 action basic, 191 intentional, 184, 186, 188, 192–196, 199, 200 agency epistemic, 103, 109 rational, 38, 44, 46, 103, 106, 112 analytic epistemology, 3, 234 analyticity, 21 a priori consequence, 36, 40, 44, 45, 50 a priori truth, 45, 46, 50 assertability conditions, 48–50 basing, 8, 26, 27, 31, 173 belief basic, 17–21, 26, 28, 30, 31, 112 -system, 25, 26, 31, 32, 38, 39 see also basing; voluntarism brain in a vat, see envattment causation, 176, 190 mental, 188, 189 certainty, 7, 55, 62, 166, 171, 197, 235, 238 Christianity, 54, 55, 61, 62, 236–237 closure, 46, 70, 71, 84, 85, 87, 99, 130, 199, 200, 211, 212, 215, 217, 218 cognition, 6, 7, 10, 12 cognitive employee, 226 cognitive psychology, 2, 10 cognitive science, 1, 12

commitment, 8, 35, 44, 49, 105–107, 109, 112, 227 implicit, 36, 40, 43, 45, 51 ontological, 36, 37, 41, 43, 46, 47, 50 reason-sensitive, 103, 110 commonsense, 21, 63, 67, 98, 196, 203, 204, 210 context, shift of, 87 contextualism, 9, 10, 115–126, 130, 133, 134, 216 see also context, shift of correspondence, 155, 162 counterfactuals, 177 defeasibility, 119, 122 defeater, 72, 115, 143, 173 definition, 20, 31, 162, 165, 203, 205, 207–209, 213, 214, 217 deliberative stance, 98, 103–107, 112 demons, 55, 63, 83, 126 dependence, see epistemic dependence disquotation, 123 dogmatism, 71, 120 doubt, 5, 9, 17, 20, 26, 37, 41, 42, 47, 49, 55, 69, 70, 72–76, 78, 89, 94, 95, 97, 99–101, 103, 107, 110, 115, 119, 126, 127, 220, 223, 227, 228, 234, 235–238 see also demons; dreaming; envattment; illusion; scepticism dreaming, 4, 55, 70–74, 76, 220, 231, 232 243

244

Subject Index

entailment, 35–41, 43, 46, 50, 51, 71, 75, 78, 156, 211 envattment, 55, 60, 63 epistemic challenge, 117–129 see also contextualism epistemic dependence, see basing epistemic gap, 194, 195, 200 see also metaphysical gap epistemic luck, 206 epistemic reasonableness, 120, 121, 123, 124, 129 epistemic right, 108, 172, 176–179 epistemic standard, 155, 123–126 epistemological progress, 236 ethics, 3, 12, 124, 125 evidence, 2, 4, 6–10, 23–27, 29, 43, 50, 53, 56, 62, 63, 74, 76, 78–80, 83, 87, 90, 103, 105, 119, 126, 135, 140, 142, 152, 153, 155, 163, 165, 188, 189, 200, 209, 216, 230, 238 see also justification externalism, 98, 197, 216 see also warrant external world, 9, 67–80, 101, 106, 109, 112, 196, 199 facts, 10, 104, 105, 119, 124, 128, 129, 137, 139, 151, 152, 155, 156, 160–162, 166, 177, 187 failability, 223, 237 fallibilism, 7, 11, 167, 195, 196, 199 fallibility, 6, 9, 11, 166, 210, 212, 214–218, 226, 227, 233, 234, 237, 238 see also fallibilism; knowledge, fallible foundationalism, 17–21, 30, 31 foundations of mathematics, 206, 207, 217

Gettier cases, see also knowledge, 152, 155, 163, 165, 166, 221, 222, 232, 233 Gettier counterexamples, see Gettier cases Gettier problem, 4, 83, 151, 154, 197 God, 56, 62, 63, 99, 237 gradualism, 228, 230, 237 see also knowledge, gradability of; knowledge, improved hinge propositions, 107 illusion, 70, 73, 77–80, 184, 197, 199, 200 indexicals, 135, 136, 143 infallibilism, 6, 238 see also fallibility inference to the best explanation, 98, 165 infinite regress, 19, 21, 22, 107 intentions, 118, 119, 124, 186, 188–190, 192, 194, 195, 199, 200 see also knowledge, agent’s internalism, see externalism intuition, 37, 38, 43, 117, 120, 121, 137, 141, 142, 154, 156, 157, 184, 205, 207, 209, 211, 213, 214, 217, 232, 239 justification see also epistemic dependence; evidence justified true belief, 152–154, 166, 205, 232, 233 see also Gettier cases KK principle, 197 knowledge absolute, 2

Subject Index agent’s, 183–195, 199, 200 causal theory of, 185 concept of, 2, 3, 6, 7, 83, 124, 166, 176, 219 fallible, 11, 165, 197–200, 223, 226, 227, 233, 237 gradability of, 144 improved, 229–231 inductive, 90–95 methodological, 227 normative analysis of, 169, 176 of action, 185, 199 perceptual, 155, 190, 199 perfect, 225, 237 practical, 191–194 religious, 237 speculative, 193 tacit, 203, 205, 207, 213, 214, 217 -threats, 223, 225, 227, 228, 231 see also doubt; facts; failability; justified true belief; lottery; scepticism knowledge-how, 228 liar paradox, 160 logic, 1, 11, 44–46, 50, 51, 151, 160, 166, 206, 217 logical equivalence, 156, 158, 159 lottery, 9, 83–96 luck, 204, 206, 214, 215, 233, 234 materialism, 37, 41 meaning, 20, 43, 48, 53, 103, 127, 146, 158, 214, 237 speaker, 153, 155 see also assertibility conditions; commitment, implicit memory, 63, 64, 106, 107 mental state, 104, 178, 189, 191, 196

245

metaphysical gap, 195, 199, 200 modal realism, 37, 41 naturalism, 9, 54, 55, 101, 102, 111, 112 naturalized epistemology, 1–2, 8 natural language, 146, 160, 206, 207 naturalistic stance, 9, 98, 102, 104–112 necessary truth, 40, 41, 155, 156, 159 necessitation, 36–47, 50, 51 nonsense, 125, 127–129, 196 observation, 3, 19, 22, 47, 83, 84, 90, 92, 96, 102, 104, 138, 160, 172, 180, 184, 185, 189, 195, 196, 204 see also perception open question argument, 119 ordinary language, 9, 10, 20, 21, 126, 127, 143 other minds, 19, 20, 30, 101, 106, 109, 115, 123 perception, 63, 64, 101, 104–106, 109, 195–197 see also perceptual experience perceptual experience, 73, 74 philosophy of time, 187 possible worlds, 36, 37, 41, 42, 118, 146 see also modal realism practical reasoning, 124, 147, 191, 192, 194 pragmatics, 47, 48, 50, 123, 125, 133, 135, 141, 146 pragmatism, 134–137, 139, 140, 142–147 semantic, 133 probability, 8, 53, 55–60, 62, 63, 65, 79, 80, 83, 88, 90–93, 118

246

Subject Index

probable opinion, 204, 210 proof, 9, 40, 44, 67–69, 71–77, 80, 158, 165, 169, 181, 188, 190 quietism, 98 rational action, 130 rationality, 6, 9, 10, 36, 38, 41, 44–47, 73, 119 reductionism, 123, 198 relativism, 121–123 reliabilism, 30, 217 see also warrant reliability, see reliabilism responsibility, 97, 103, 106, 108–112, 124, 224 rights consequentialism, 177, 178 Russell’s paradox, 206, 207, 216, 217 salience, 119, 125–127 sceptical challenges [skeptical challenges], 4, 231 see also scepticism [skepticism] sceptical possibilities [skeptical possibilities], 8, 11, 87, 212, 219–224, 227–230, 234 see also sceptical challenges [skeptical challenges] scepticism [skepticism], 8, 9, 26, 48, 63, 116, 125–127, 129, 130, 185, 189, 196, 197, 199, 218, 228, 231, 235 ancient, 108, 110, 112 Cartesian, 97

Humean, 97 Kantian, 97 see also contextualism; doubt; external world; lottery; quietism science, 1–3, 6, 7, 9, 10, 12, 50, 101, 102, 116, 117, 203, 204, 210 see also cognitive science scientific realism, 62 self-awareness, 102, 103 self-deception, 100 self-knowledge, 97, 98, 105, 107, 110, 112, 184 sense-datum, 195 state of affairs, see facts teleology, 185, 191, 192, 194, 195, 200 testimony, 63, 64, 163 tranquility, 108 transcendental argument, 98 true justified belief, see justified true belief truth see also truthmaker truthmaker, 155–160, 162, 163, 165, 167, 191 see also logical equivalence voluntarism, 109 warrant, 8, 53, 54, 56, 64, 189, 197, 208–218 working hypothesis, 227