Oxford Studies in Epistemology: Volume 4 9780199672707, 9780199672714, 0199672709

Table of contents :
Cover
Contents
Contributors
PAPERS
1 The Epistemology of Conditionals
2 A Defense of Dogmatism
3 Rational Agnosticism and Degrees of Belief
4 Probability and Prodigality
5 Essence and Natural Kinds: When Science Meets Preschooler Intuition
6 Easy Knowledge, Transmission Failure, and Empiricism
7 Why Philosophy Can Overturn Common Sense
8 Could Evolution Explain Our Reliability about Logic?
9 Can Selection Effects on Experience Influence its Rational Role?
SYMPOSIUM
10 Knowledge as a Mental State
11 What Does Knowledge Explain? Commentary on Jennifer Nagel, ‘Knowledge as a Mental State’
12 Knowledge, Causal Explanation, and Teleology
13 Is Knowledge a Non-Composite Mental State?
Index
A
B
C
D
E
F
G
H
I
J
K
L
M
N
O
P
Q
R
S
T
U
V
W
Z

Citation preview

OXFORD STUDIES IN EPISTEMOLOGY

OXFORD STUDIES IN EPISTEMOLOGY Editorial Advisory Board: Stewart Cohen, University of Arizona Keith DeRose, Yale University Richard Fumerton, University of Iowa Alvin Goldman, Rutgers University Alan Hájek, Australian National University Gil Harman, Princeton University Frank Jackson, Australian National University Jim Joyce, University of Michigan Scott Sturgeon, University of Birmingham Jonathan Vogel, Amherst College Tim Williamson, University of Oxford Managing Editors: Julianne Chung, Yale University Alex Worsnip, Yale University

OXFORD STUDIES IN EPISTEMOLOGY Volume 4

Edited by Tamar Szabó Gendler and John Hawthorne

1

3

Great Clarendon Street, Oxford, OX2 6DP, United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © the several contributors 2013 The moral rights of the authors have been asserted First Edition published in 2013 Impression: 1 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer British Library Cataloguing in Publication Data Data available ISBN 978–0–19–967270–7 (hbk.) ISBN 978–0–19–967271–4 (pbk.) Printed by the MPG Printgroup, UK

E D I T O R S ’ P R E FA C E

It gives us great delight to share with our readers the fourth volume in the Oxford Studies in Epistemology series. Like its predecessors, this issue provides a showcase for some of the most exciting new work in the field of epistemology from throughout the English-speaking world. Published bi-annually under the guidance of a distinguished editorial board, each issue of Oxford Studies in Epistemology seeks to publish not only traditional work in epistemology, but also work that brings new perspectives to traditional epistemological questions, and that opens new avenues of investigation. We are particularly excited that this issue features the work of both young and established epistemologists, exploring a wide range of topics from a wide variety of perspectives. Three of our authors—Jane Friedman, Daniel Greco and Susanna Rinard—are newly-minted PhDs; the rest represent a range of other career stages. The questions addressed in the volume’s papers include the relation between knowledge and belief; appropriate responses to various sorts of skeptical challenges; the epistemology of conditionals; our epistemic relation to logical truths; the epistemic status of perception; and the reliability of philosophical intuition. Throughout the volume, there are papers that make use of the tools and insights of formal epistemology; decision theory; traditional “armchair” philosophical analysis and argumentation; evolutionary theory; experimental philosophy; and cognitive, comparative, developmental and social psychology. Two of the papers consider issues that arise when we try to model epistemic attitudes using formal tools. In ‘Rational Agnosticism and Degrees of Belief’, Jane Friedman argues against the view that traditional epistemology’s doxastic attitudes (believing p, disbelieving p and suspending judgment about p) can be reduced to precise degrees of belief for p. In ‘Probability and Prodigality’ Daniel Greco argues against the view that the right way to model knowledge is to assign it epistemic probability 1. Three of the other papers offer arguments on largely a priori grounds. In ‘A Defense of Dogmatism,’ Jeremy Fantl defends the view that it is sometimes epistemically legitimate to dismiss relevant counterarguments to a view that you hold, even if you cannot identify precisely where those arguments go wrong. In ‘Why Philosophy Can Overturn Common Sense’, Susanna Rinard argues that there may be cases where philosophical argumentation is sufficient to overturn even our most deeply-held anti-skeptical common-sense beliefs. And in ‘Easy Knowledge, Transmission Failure, and Empiricism’ Ram Neta proposes and defends a solution to one version of the so-called “problem of easy knowledge.”

vi

Editors’ Preface

The remaining papers make use of a range of resources from various empirical domains. In ‘Could Evolution Explain Our Reliability about Logic?’ Joshua Schechter looks to evolutionary reasoning for a possible explanation for the fact that our considered logical beliefs are by and large true. In ‘The Epistemology of Conditionals’, Igor Douven draws on experimental work to challenge a number of widely held views about the epistemology of indicative conditionals. In ‘Essence and Natural Kinds: When Science Meets Preschooler Intuition’, Sarah-Jane Leslie uses Kripke/Putnam-style natural-kind essentialism as a case study to suggest that certain philosophical intuitions may gain their subjectively compelling status, not because they track important truths about the world, but because they arise from deep-seated, early-developing conceptual biases. And in ‘Can Selection Effects on Experience Influence its Rational Role?’ Susanna Siegel examines empirical work in the psychology of perception to offer a partially affirmative, partially negative answer to her question. Finally, in ‘Knowledge as a Mental State’, Jennifer Nagel draws extensively on work in developmental, comparative and social psychology to explore the question of her title, namely whether knowledge is a mental state. Nagel’s account is discussed in a symposium, whose contributors include Stephen A. Butterfill (‘What Does Knowledge Explain?’), Johannes Roessler (‘Knowledge, Causal Explanation, and Teleology’) and Patrick Rysiew (‘Is Knowledge a Non-Composite Mental State?’). As in the past, some of the papers that appear here were brought to our attention by members of the editorial board, others were solicited directly from authors; all were refereed by the members of our Editorial Advisory Board and a small group of distinguished referees. Thanks are due our referees: David Christensen, Susan Gelman, Sally Haslanger, Nico Silins, Declan Smithies, Eric Swanson, Brian Weatherson, and three anonymous reviewers; and to our editorial board members: Stewart Cohen (University of Arizona), Keith DeRose (Yale University), Richard Fumerton (University of Iowa), Alvin Goldman (Rutgers University), Alan Hájek (Australian National University), Gil Harman (Princeton University), Frank Jackson (Australian National University and Princeton University), Jim Joyce (University of Michigan), Scott Sturgeon (University of Birmingham), Jonathan Vogel (Amherst College), and Tim Williamson (Oxford University). We are also indebted to our outstanding managing editors, Julianne Chung and Alex Worsnip, for their superb editorial assistance, and to Peter Momtchiloff, for his continuing support of this project. Tamar Szabó Gendler, Yale University John Hawthorne, Oxford University

CONTENTS

Contributors

ix PAPERS

1 The Epistemology of Conditionals

3

Igor Douven

2 A Defense of Dogmatism

34

Jeremy Fantl

3 Rational Agnosticism and Degrees of Belief

57

Jane Friedman

4 Probability and Prodigality

82

Daniel Greco

5 Essence and Natural Kinds: When Science Meets Preschooler Intuition

108

Sarah-Jane Leslie

6 Easy Knowledge, Transmission Failure, and Empiricism

166

Ram Neta

7 Why Philosophy Can Overturn Common Sense

185

Susanna Rinard

8 Could Evolution Explain Our Reliability about Logic?

214

Joshua Schechter

9 Can Selection Effects on Experience Influence its Rational Role?

240

Susanna Siegel

SYMPOSIUM 10 Knowledge as a Mental State

273

Jennifer Nagel

11 What Does Knowledge Explain? Commentary on Jennifer Nagel, ‘Knowledge as a Mental State’

309

Stephen A. Butterfill

12 Knowledge, Causal Explanation, and Teleology

321

Johannes Roessler

13 Is Knowledge a Non-Composite Mental State?

333

Patrick Rysiew

Index

345

This page intentionally left blank

CONTRIBUTORS

Stephen A. Butterfill University of Warwick Igor Douven University of Groningen Jeremy Fantl University of Calgary Jane Friedman University of Oxford Daniel Greco Yale University Sarah-Jane Leslie Princeton University Jennifer Nagel University of Toronto Ram Neta University of North Carolina Susanna Rinard University of Missouri (Kansas City) Johannes Roessler University of Warwick Patrick Rysiew University of Victoria Joshua Schechter Brown University Susanna Siegel Harvard University

This page intentionally left blank

PAPERS

This page intentionally left blank

1. The Epistemology of Conditionals Igor Douven Compared to most other areas of philosophy, epistemology is in good shape. Whatever fighting is going on in epistemology takes place amidst considerable agreement on a large number of fundamental issues. Not that epistemology is nearing completion: the questions that are still up for debate are too substantive for that. But it might be said that at least we see the contours of a final epistemology emerging, a claim that presumably no one would venture to make about, for instance, metaphysics or moral philosophy. This paper argues that even if epistemologists have reason for some general optimism regarding their discipline, they have barely scratched the surface insofar as the epistemology of conditionals is concerned. Not only are many specifically epistemological issues concerning conditionals still wide open, but the two epistemological claims concerning conditionals that virtually all philosophers deem uncontentious—claims about the probability as well as the assertability and acceptability of conditionals—are severely undermined by recent findings in experimental psychology. As a preliminary disclaimer, I note that the present paper deals exclusively with indicative conditionals. This is not to suggest that there are no epistemological problems related to subjunctive conditionals. Most probably there are, but they are beyond the scope of this paper. Unless specified otherwise, “conditional” refers to indicative conditionals throughout.

1.

WHY WE NEED AN EPISTEMOLOGY OF CONDITIONALS

Prima facie, conditionals do not appear to call for any special attention from epistemologists. It would seem that they can be believed, asserted, known, disbelieved, doubted, learned, in the same way that, for instance, conjunctions and disjunctions can be believed, asserted, and so on. The impression that conditionals are epistemologically unremarkable is reinforced by the fact that none of the accounts of rational acceptance, justification, probability, knowledge, assertion, or rational updating that are found in the literature make exceptions for conditionals. Nonetheless, the impression is false. It is not just that the lack of a generally agreed upon semantics of conditionals makes it difficult to say anything remotely definitive about the epistemology of conditionals; it could then still have been the case that there were semantics that yield perfectly satisfactory answers to all the important epistemological questions one might ask about conditionals. The bigger

4

Igor Douven

problem is that each of the better-known semantics of conditionals faces difficulties of a specifically epistemological nature. To illustrate this problem, let us consider what are arguably the three main candidate semantics: the material conditional account, according to which “If ϕ, ψ” has the same truth conditions as ϕ ∨ ψ; Stalnaker’s [1968] possible worlds semantics, which declares a conditional to be true (false) precisely if its consequent is true (false) in the closest possible world in which the conditional’s antecedent is true, where “closest” means “most similar to the actual world,” and provided there is a world in which the antecedent is true; and the non-propositional view, according to which conditionals never have a truth value, or, in a slightly less radical version, have the truth value of their consequent if their antecedent is true, and otherwise are neither true nor false.1 Consider the non-propositional view first. Given that knowledge, belief, doubt, and so on, are propositional attitudes, how can a conditional be known, or believed, or doubted (and so on) unless it expresses a proposition? Concomitantly, what is the norm of assertion for conditionals? There is widespread agreement that whether an assertion is proper depends on the asserter’s epistemic position vis-à-vis what he or she has asserted. The main divide is over how strong that epistemic position must be, whether the asserter must know what he or she has asserted, or whether something weaker, like rational credibility, suffices. If conditionals cannot be believed in the first place, then, it would seem, no such epistemic norm applies to them.2 The material conditional account is beset by epistemological problems of its own. For example, it is commonly held that high probability is at least close to being sufficient for rational credibility. To be more precise, most hold that high probability cannot be quite sufficient in view of Kyburg’s [1961] lottery paradox, but that, barring the kind of propositions that figure in Kyburg’s paradox, high probability is sufficient for rational credibility. Suppose, then, a conditional has the truth conditions of the corresponding material conditional. Given that a disjunction is at least as probable as its most probable disjunct, it follows that a conditional is highly probable if its antecedent is highly improbable. But consider this sentence:

1 Of these three semantics, the first was once the orthodoxy in philosophy. While that is no longer the case, it still seems to be the “default” position, in that it is the view tacitly, or even not so tacitly, assumed by most philosophers not working on conditionals. By contrast, the nonpropositional view seems to be endorsed by most philosophers who do work on conditionals. In psychology, the situation is slightly different in that even psychologists doing research on conditionals are divided. Those working in the mental models tradition of Johnson-Laird typically advocate the material conditional account, while those working within the Bayesian approach to reasoning seem more inclined toward the non-propositional view. 2 Adams [1998a:202 f], [1998b:190], one of the main champions of the non-propositional view, acknowledges that the view raises fundamental epistemological issues, though he does not go on to say whether he regards this as a cause for concern.

The Epistemology of Conditionals

5

(1) If Manchester United ends last in this year’s Premier League, they will give their manager a raise. Although it is exceedingly unlikely that Manchester United will end last in this year’s Premier League, we do not find (1) rationally credible; to the contrary, it strikes us as rather incredible! Because (1) is unrelated to the kind of propositions that give rise to the lottery paradox, it appears that—assuming the material conditional account—high probability is not even nearly enough for the rational credibility of conditionals. Arguably, the deeper problem here is that the assignment of a high probability to (1) is wrong to begin with. This is not an isolated fact about (1). It is a manifestation of a more general deficiency of the material conditional account, to wit, that—as will be seen further on—it makes systematically wrong predictions about people’s assignments of probabilities to conditionals. In fact, the same will be seen to hold for Stalnaker’s semantics. Moreover, in Douven [2012] it is shown that Stalnaker’s semantics, as well as the other semantics mentioned above, face further epistemological problems, related to the issue of belief change, specifically updating on conditionals. (See also Douven and Romeijn [2011].) Some of the information we receive is of an essentially conditional nature, as when we learn that (2) If Millie is not going to work harder, she will not pass her exam. or that (3) If it starts raining this afternoon, the match will be cancelled. Not only do we learn things like these, and thereupon come to believe them, such things also tend to impact what else we believe. Learning (2) may make me believe that Millie is not as zealous a student as I had thought she was. Equally, learning (3) may raise my confidence that if it starts snowing this afternoon, the match will be cancelled as well. What is more, we sometimes have clear intuitions about how one ought to respond to the receipt of conditional information. For example, barring special circumstances, it would seem wrong to lower one’s confidence that the match will be canceled if it starts snowing, upon learning (3). However, while we seem to routinely process incoming conditional information, the questions of how we process such information and to which normative constraints such processing is subject, have hardly been addressed by epistemologists.3,4 In making a beginning at 3 The latter question has been addressed, to some extent, by logicians and computer scientists; see, e.g. Boutillier and Goldszmidt [1995] and Kern-Isberner [1999]. But first, they have considered the question almost exclusively from the perspective of AGM theory, which tells at best only a very partial story about belief change. And second, they have mostly sidestepped the question of whether the resulting account of belief change is materially adequate, as far as it pertains to conditionals, that is, whether it does justice to the intuitions we have about how people should respond to the receipt of particular pieces of conditional information. 4 Cf. Adams [1998a:146], who calls updating on conditional information “an important open research problem.”

6

Igor Douven

answering these questions, Douven [2012] argues that on none of the standard semantics of conditionals do we get a satisfactory account of learning conditional information if learning conditionals is to be modeled in the way that we model learning non-conditional information (i.e. by standard updating procedures, like Bayesian conditionalization). Hence, either some special account of learning conditional information or a new semantics of conditionals is called for. I will not try to catalog here all open epistemological questions regarding conditionals. Instead, this paper focusses on what are unarguably the only two epistemological issues vis-à-vis conditionals that are generally regarded settled. One issue concerns the probabilities of conditionals, the other their assertability and acceptability. What is thought to be settled about the former is a negative claim, a claim regarding what the probabilities of conditionals cannot be. What is thought to be settled about the latter is a positive claim, a claim regarding what determines the degree of assertability/acceptability of a conditional. Uncontentious though these claims may appear, it will be argued that they need serious rethinking in light of recent experimental findings. Thus, not only do we face a number of open issues regarding the epistemology of conditionals, but the list of open issues may include issues previously thought to be closed.

2.

THE PROBABILITIES OF CONDITIONALS

Philosophers have been pondering the probabilities of conditionals for over forty years now.5 And while psychologists have also been working on conditionals for quite some time, they have only very recently—since around the beginning of the new millennium—started looking into the question of the probabilities of conditionals. As we shall see, there is a manifest tension between the newer empirical results that psychologists have obtained and the older conceptual work done by philosophers. In fact, the empirical findings seem to call into doubt what is universally regarded as the most central philosophical result concerning the probabilities of conditionals. At the same time, they motivate reconsideration of some technical work that has always remained somewhat in the margins of the debate. I start by giving my own account of the philosophical work on the probabilities of conditionals, and then turn to the newer experimental work. 2.1. Philosophical Work on the Probabilities of Conditionals Given that I believe ϕ to a degree of x, to what degree should I believe ϕ? Any epistemology that respects probability theory—perhaps in view of some 5 Or for over eighty years, if one counts a passing but influential remark in Ramsey [1929] as the starting point.

The Epistemology of Conditionals

7

Dutch book argument (e.g. de Finetti [1937]) or some accuracy domination argument (e.g. Joyce [1998])—will answer that you ought to believe ϕ to a degree of 1 − x. Here is an equally legitimate and seemingly also equally simple question: Given that I believe ϕ to a degree of x, and ψ to a degree of y, to what degree should I believe “If ϕ, ψ”? Or, if my degrees of belief for just ϕ and ψ are not enough to answer that question, given as many of my other degrees of belief (conditional or unconditional) as you like, what should my degree of belief for “If ϕ, ψ” be? Of the standard semantics, only the material conditional account makes the probability of a conditional come out as a function of other probabilities: on this account, Pr(If ϕ, ψ) = Pr(ϕ ∨ ψ) = 1 − Pr(ϕ ∧ ψ) = 1 − Pr(ϕ) Pr(ψ | ϕ). But, as is easy to see, and as we already noted, this account renders any conditional with an improbable antecedent highly probable. This is not only hard to reconcile with our idea that high probability is nearly enough for rational credibility, but, as we also noted, it is implausible in itself. If the point needs further underscoring, contrast (1) with (4) If Westham United wins next year’s Premier League, the players will receive a bonus on their salaries. The antecedents of (1) and (4) seem extremely unlikely. Thus, supposing the material conditional account, both conditionals must be highly likely. But that is not at all how it appears. Whereas (4) will strike most as very probably true indeed, (1) will strike most as probably false. Proponents of the material conditional account are aware that they have to explain away our intuitions about the probabilities of conditionals like (1). What they tend to do is claim that these intuitions are grounded in our mistaking probability for some closely related property that conditionals may have, like their degree of assertability or their degree of acceptability. As such, the move would seem to be on a par with explaining the well-known conjunction fallacy (Tversky and Kahneman [1983]) by arguing that people are inclined to misinterpret questions about probability as questions about, for instance, coherence (e.g. Bovens and Hartmann [2003: 85–88]). On further and more careful consideration, however, the parallel appears to be not quite so close. For empirical research has shown that people are much less likely to commit the conjunction fallacy once it has been brought to their attention that a conjunction can never be more likely than its conjuncts (Moutier and Houdé [2003]), or even when the questions about the probabilities of the conjunction and the conjuncts are phrased in terms of relative frequencies (Hertwig and Gigerenzer [1999]). By contrast, even if I attend carefully to the distinction between probability and assertability (or acceptability), I still deem (4) to be much more probable than (1). To be sure, I also deem (4) to be much more assertable, and also much more acceptable, than (1), but for all I can introspect, that is at least in the present case mainly or even solely due to the fact that the former is more probable than the latter. It thus seems that explaining our intuitions about

8

Igor Douven

the probabilities of conditionals will be even harder than explaining why people sometimes assign higher probabilities to a conjunction than to its conjuncts.6 At first glance, Stalnaker’s proposal might seem to do justice to our intuitions about the probabilities of both (1) and (4). After all, is it not much more probable that we inhabit a world whose closest Westham-ends-first world is a Westham-players-get-bonus-for-winning world than that we inhabit a world whose closest Manchester-ends-last world is a Manchester-managergets-raise world? Well, yes, but note that this is simply to rephrase, using the terminology of Stalnaker’s semantics, our intuition that (4) is much more likely than (1). In particular, it gives no indication how to compute the probabilities of these conditionals on the basis of the probabilities we assign to other propositions expressible in our language, and so does not suggest anything like the rules for calculating probabilities of negations, conjunctions, and disjunctions that probability theory offers us. In a sense, Stalnaker himself presented a rule for calculating probabilities of conditionals of precisely that kind when in his [1970] he proposed the following as an adequacy criterion for any semantics of conditionals: Stalnaker’s Hypothesis (SH) Pr(If ϕ, ψ) = Pr(ψ | ϕ), provided Pr(ϕ) > 0.7 While this hypothesis has undeniable intuitive attraction, it is easy to see that at least Stalnaker’s semantics does not validate it. In this semantics, the proposition expressed by a conditional may be thought of as corresponding to the set of worlds whose closest antecedent worlds (i.e. worlds in which the antecedent holds) are consequent worlds. So, where the arrows in Figure 1. 1 indicate closest ϕ-worlds, the gray areas indicate the propositions expressed by “If ϕ, ψ.” If we set Pr({wi }) = pi > 0 for all i ∈ {1, . . . , 4} in models I–III, then the probabilities of “If ϕ, ψ” in these models are, respectively, p1 , p1 + p3 , and p1 + p3 + p4 . On the other hand, the conditional probability of ψ given ϕ is p1 /(p1 + p2 ) in all three models. That is, in these models, Pr(If ϕ, ψ)  = Pr(ψ | ϕ), in violation of (SH). Figure 1.1 in effect illustrates a theorem proved in Lewis’ [1976: 147f], to wit, that the probability of a Stalnaker conditional “If ϕ, ψ” is the probability of ψ after imaging on ϕ (again provided there is a world in which ϕ holds), where imaging on ϕ consists of transferring the probability of any ϕ-world to its closest ϕ-world.8 The picture makes it easy to see why this must hold. After imaging ψ on ϕ, the ψ-worlds come to carry all the probability mass that, before the operation, was carried by the worlds that have a ϕ ∧ ψ-world 6 In this connection, it is not encouraging that almost thirty years after the discovery of the conjunction fallacy, there is still no consensus on what underlies it. 7 This thesis also goes by the names “the Equation“ (e.g. Edgington [1995: 271]) and “the conditional probability hypothesis“ (e.g. Gilio and Over [2012: 119]). 8 See Gärdenfors [1982] for a strengthening of Lewis’ result.

The Epistemology of Conditionals w1

I

y y

w2

w1

y y

w2

w1

j

j

j

j

j

j

w3

w4

II

w3

w4

III

y y

w3

9

w2

w4

Figure 1.1 The Stalnaker conditional “If ϕ , ψ ” in different possible worlds models

as their closest ϕ-world, which are the worlds that have an arrow going to a ϕ ∧ ψ-world, and which are precisely the worlds in the gray areas. As a result, the sum of the probabilities assigned to the ψ-worlds after imaging on ϕ— which is the probability of ψ after imaging on ϕ—cannot but equal the sum of the probabilities assigned to the worlds in the gray area before the imaging operation, which is the probability of the Stalnaker conditional “If ϕ, ψ.” Lewis’ aforementioned paper is actually famous for two other technical results—“triviality arguments,” as they came to be called—which have seemed to refute (SH) once and for good. (The reason for the hedged formulation will become apparent in Section 1.2.3.) It is no coincidence that the Stalnaker conditional fails to validate (SH), for, Lewis claims, there can be no semantics of conditionals which validate (SH). For later purposes, I state Lewis’ first triviality argument here in detail. Let Pr be a probability function for which (SH) holds, and let ϕ and ψ be such that both Pr(ϕ ∧ ψ) > 0 and Pr(ϕ ∧ ψ) > 0. Then, from the law of total probability, it follows that Pr(If ϕ, ψ) = Pr(If ϕ, ψ | ψ) Pr(ψ) + Pr(If ϕ, ψ | ψ) Pr(ψ), which, assuming (SH), can be rewritten as Pr(If ϕ, ψ) = Pr(ψ | ϕ ∧ ψ) Pr(ψ) + Pr(ψ | ϕ ∧ ψ) Pr(ψ), which simplifies to Pr(If ϕ, ψ) = 1 × Pr(ψ) + 0 × Pr(ψ). By applying (SH) once more to the left-hand side of the last equation, we get the result that, given very weak conditions, conditional probabilities equal unconditional probabilities. Surely that is absurd. Lewis’ second triviality result, and many similar results by Lewis and others that have later appeared, derive equally absurd conclusions from (SH). In view of these results, it is not surprising that advocates of (SH) are no longer to be found in the philosophical community. It is also worth briefly mentioning here Nolan’s [2003] variant of Stalnaker’s semantics for conditionals, according to which a conditional is true (false) precisely if its consequent is true (false) in all closest antecedent worlds (plural, where Stalnaker has singular), given that, toward the end of his paper, Nolan

10

Igor Douven

explicitly considers the issue of the probability of a conditional. Most of his discussion centers around the following concrete example: (5) If the coin was flipped, it came up heads. which is supposed to be asserted of a coin we know to be fair. Where we deem it only 20 per cent likely that the coin was flipped, the probability of (5), according to Nolan, is 10 per cent, namely, the probability that the coin was flipped times the conditional probability that the coin came up heads given that it was flipped. Nolan stops short of suggesting this as a fully general rule for computing the probability of a conditional. Nor is there any indication that he would be opposed to this rule, even though, as a general rule, it would seem pretty hopeless. For, suppose I have been informed that there is a current on the fence. That makes it highly likely to me that (6) If I touch the fence, I will get an electric shock. At the same time, the information makes it highly unlikely to me that I shall touch the fence. Reasonable though these two probability judgments seem, if Nolan’s proposal were meant to hold generally, they cannot both be correct, for then a conditional cannot be more probable than its antecedent. However, even if the proposal is only meant to apply to (5) and sentences that are somehow more similar to it than (6) is, it is at odds with how people in reality evaluate the probabilities of conditionals, as will be seen shortly. What about the non-propositional view of conditionals? It should be clear that our intuitive verdicts about (1) and (4) above are not only at odds with the material conditional account but also militate against the non-propositional view, and against both the most radical version of this view—how could a conditional be probably true at all (or probably false, for that matter) if it cannot be true (or false) to begin with?—and against the version according to which a conditional is true iff its antecedent and consequent are both true: by my lights, (4) is probably true, even though, by those same lights, the conjunction of its antecedent and consequent is probably false. Advocates of the non-propositional view are aware of this line of critique. Their reply typically involves the claim that the probability operator has a non-standard interpretation when it occurs in front of a conditional. This claim has its own problems, even apart from the fact that it implies that we are massively mistaken about our attitudes toward conditionals—which, it does seem, can be probable to varying degrees. We turn to these problems in Section 3. So, all things considered, conditionals better have probabilities of truth. On the other hand, if they do have probabilities of truth, we seem to lack any systematic account of them, in particular, of how they relate to our other probabilities. However, the three candidate rules for computing the probabilities of conditionals that we encountered—Pr(If ϕ, ψ) = Pr(ϕ ∨ ψ), Pr(If ϕ, ψ) = Pr(ψ | ϕ), and Pr(If ϕ, ψ) = Pr(ϕ) Pr(ψ | ϕ)—were dismissed on more or less a priori grounds, without attending to what probabilities people actually assign to conditionals and seeing whether there is any pattern to be

The Epistemology of Conditionals

11

discovered in those assignments. Psychologists have recently started studying the question of the probabilities of conditionals, and this has led to some surprising results. 2.2. Probabilities of Conditionals: The Empirical Turn Conditionals have been the subject of psychological studies for many decades. For instance, numerous studies have been conducted in an effort to find out why many people are prone to infer ϕ from “If ϕ, ψ” and ψ, or to infer ψ from the same conditional with the additional premise ϕ. However, in all these studies, it was simply assumed that the semantics of the conditional is given by the material conditional account. The point of the studies was then to find out whether people reason in accordance with this semantics, and if not, whether there is any system to be discerned in the mistakes they make. As mentioned, it has only been some ten years since psychologists have also turned to the question of the probabilities of conditionals. Pioneering work in this regard has been done by Evans and Over, whose earliest results have been gathered in their [2004].9 In view of the triviality arguments, the unarguably most surprising of these results is that people do generally judge the probability of a conditional to be equal to the corresponding conditional probability. What is more, this result was later replicated several times over, in experiments conducted both by Evans and Over and by other experimentalists.10 When taken at face value, the data obtained in these experiments show all semantics that have surfaced in our discussion so far to be empirically inadequate. We saw that, in Stalnaker’s semantics, (SH) does not hold.11 And given that Pr(ψ | ϕ) is equal to Pr(ϕ ∨ ψ) only under trivial conditions, the material conditional account would thereby be ruled out as well. Finally, the data

9 For some relevant earlier papers, see Hadjichristidis et al. [2001], Evans, Handley, and Over [2003], and Over and Evans [2003]. 10 See, e.g. Oaksford and Chater [2003], Oberauer and Wilhelm [2003], Weidenfeld, Oberauer, and Hornig [2005], Evans et al. [2007], Oberauer, Weidenfeld, and Fischer [2007], Over et al. [2007], Gauffroy and Barrioullet [2009], Douven and Verbrugge [2010], [2013], Pfeifer and Kleiter [2010], Politzer, Over, and Baratgin [2010], and Fugard et al. [2011]. A number of these studies found the conditional probability only as the modal evaluation of the probability of a conditional. In these studies, a minority of participants evaluated the probability of a conditional as the probability of the conjunction of antecedent and consequent. Evans et al. [2007] found that the conjunction response was more frequent among those of lower cognitive ability. Barrouillet and Lecas [1999] and Lecas and Barrouillet [1999] found that this response is even the predominant one in young children. However, there was no evidence for the conjunction response in studies that used “real world conditionals”—conditionals pertaining to events that have actually occurred or might still occur and about which participants may be expected to hold prior beliefs—or conditionals about fictional but ordinary situations; see Over et al. [2007] and Douven and Verbrugge [2010], [2013]. 11 Nor could it hold on Nolan’s semantics. Note that the models depicted in Figure 1.1 are also models for Nolan’s conditional, just models in which, as it so happens, all worlds have a unique closest ϕ-world. (It is no requirement of Nolan’s semantics that worlds have more than one closest antecedent world for any conditional.)

12

Igor Douven

would also give the lie to the view that conditionals do not express propositions, for apparently people do assign probabilities of truth to conditionals. Again, this is so when the said data are taken at face value. But perhaps the experiments are methodologically flawed, or their results are to be discounted for other reasons. Surely the triviality arguments pose a formidable obstacle to interpreting the experimental results as showing what they appear to show, notably, that people’s probabilities of conditionals equal their corresponding conditional probabilities, in accordance with (SH). As to methodology, three points are worth making. First, it is to be emphasized that in none of the experiments apparently supporting (SH) were participants simply asked whether they deem (SH) intuitively plausible. These studies did not ask participants about (SH) at all. Instead, they asked participants to rate the probabilities of conditionals as well as the corresponding conditional probabilities. To be more exact, in one type of experiments—what psychologists call “within subjects experiments”—the same participants who were asked to rate the probabilities of a number of conditionals were also asked to rate the corresponding conditional probabilities; in another type— called “between subjects experiments”—one group of participants was asked to rate the probabilities of conditionals while another group of participants was asked to rate the corresponding conditional probabilities. Despite the difference in design, in both types of experiments the same close match between probabilities of conditionals and conditional probabilities was found to exist. Second, there were also differences between the various studies on (SH) as regards the types of stimuli used. Some studies used as stimuli socalled causal conditionals (conditionals stating a causal connection between antecedent and consequent); others used various types of inferential conditionals (see below). Some studies used conditionals concerning actual matters of fact about which participants could be expected to have opinions; others offered fictional vignette stories and asked for the equally fictional probabilities of conditionals pertaining to those stories. In addition, different methods of eliciting conditional probabilities were deployed in the various studies: while some studies determined conditional probabilities via Kolmogorov’s well-known ratio definition, other studies relied on the so-called Ramsey test, that is, they asked participants to judge the probability of the consequent on the supposition that the antecedent holds true.12 Nevertheless, each time the results were in favor of (SH). Third, Johnson-Laird and co-authors have criticized the experimental work on (SH) because of how these studies elicited probabilities of conditionals. According to the critics, the probability operator takes narrow scope over conditionals, so that people naturally understand the phrase “the probability that [if ϕ, then ψ] equals x” as meaning “if ϕ, then [the probability of ψ equals x].” 12 According to some theorists, the Ramsey test provides the best definition of conditional probability. See, e.g. Edgington [1995: 266 f], [1997:108 f], Bennett [2003: 53], and Oaksford and Chater [2007: 109].

The Epistemology of Conditionals

13

As a result, when asked how probable “If ϕ, ψ” is, people hear this as a mere stylistic variant of a question asking for the probability of that conditional’s consequent given its antecedent. Thus—the critics contend—we should not be surprised to find a close match between what are supposed to be the probabilities of conditionals and the corresponding conditional probability: we have been asking for the same value twice over, just in slightly different words.13 Support for their claim concerning the scope of the probability operator is to come by analogy, so to speak, from considering how certain other operators behave in the face of conditionals. For instance, as Johnson-Laird and his coworkers point out, and as has been observed by other authors as well, the negation operator normally takes narrow scope when attached to a conditional. However, Politzer, Over, and Baratgin [2010] are surely right to protest that Johnson-Laird and his co-authors are much too quick to take this as evidence for their claim that the probability operator has narrow scope in the face of a conditional. As they note, it would be absurd to suggest that the necessity operator takes narrow scope in the face of a conditional: “Necessarily, if ϕ then ϕ” appears nothing more than a truism, whatever one’s exact view of the semantics of conditionals; but “If ϕ, then necessarily ϕ” seems plain false, apparently claiming that whatever obtains, obtains of necessity. While this response is well taken, it does not settle the question of whether the probability operator is to be grouped with the operators that take narrow scope over conditionals or with the operators that take wide scope over conditionals. To make progress on this question, Over, Douven, and Verbrugge [2013] conducted a number of experiments investigating scope ambiguities of the probability operator and closely related epistemic operators. The results of these experiment are consistent in indicating that the probability operator takes wide scope over conditionals. For example, one experiment investigated whether asking for the acceptability/credibility/plausibility of a conditional yielded the same value as asking for the acceptability (etc.) of the conditional’s consequent on the supposition of its antecedent. Asking these two questions for probability does yield the same values, as work on (SH) has amply shown, and if Johnson-Laird’s grounds for dismissing that work were to hold, one would expect the same to happen for the other operators. However, it was found that neither for acceptability nor for credibility, nor for plausibility, did the values match, not even nearly. And it would seem utterly ad hoc to maintain that the probability operator, as the only one from a family of similar and conceptually related operators, takes narrow scope over conditionals. Naturally, falsifiers of (SH) might still be lurking in types of stimuli that have not been considered so far. Most notably perhaps, all experiments that have been hitherto conducted were limited to simple conditionals, that is,

13 See Johnson-Laird and Byrne [2002], Girotto and Johnson-Laird [2004], [2010], and Byrne and Johnson-Laird [2010].

14

Igor Douven

conditionals whose antecedents and consequents are not themselves conditional in form nor are compounds which have one or more conditionals among their components. But even if (SH) were to hold only in the form limited to simple conditionals, that would be no better news for the semantics that, some paragraphs back, were said to be empirically inadequate in light of the empirical findings, nor would it help to diminish the tension between these findings and the triviality results. After all, the latter seem to show that (SH) cannot even hold true for simple conditionals. To at least resolve the tension with the triviality arguments (even if not to shield the various semantics from disconfirmation), one might note that the data obtained in the experiments are bound to be noisy to some extent, so that all they can be taken to show is that people’s probabilities of conditionals approximately match the corresponding conditional probabilities. Thus, perhaps we should replace (SH) with an approximate version of it, with ≈ replacing =. Unfortunately, this will not fly. It can easily be shown that from an approximate version of (SH), a variant of (for instance) Lewis’ first triviality argument can be obtained that has as a conclusion that, for all ϕ, ψ such that ϕ ∧ ψ and ϕ ∧ ψ have both positive probability, the conditional probability of ψ given ϕ is close in value to the unconditional probability of ψ.14 This is no harder to refute than the contention that conditional probabilities are always strictly equal to unconditional probabilities. A more radical way to neutralize the empirical data is to interpret them as manifesting people’s limited capabilities when it comes to probabilistic reasoning; perhaps the experiments concerning the probabilities of conditionals should be regarded as having unearthed something akin to the previously mentioned conjunction fallacy. However, before attributing massive error to people, we should consider whether there is not some more charitable interpretation of the data. The most charitable proposal is, of course, to take the data as showing that people’s bona fide judgments of the probabilities of conditionals are equal to their bona fide judgments of the corresponding conditional probabilities indeed, and then to explain that this is consistent with the triviality arguments. That may seem well-nigh impossible to accomplish, but, I want to argue, the triviality arguments do not discredit (SH) as unimpeachably as they are commonly held to do. 2.3. Triviality Undone A first crucial observation to be made is that the triviality arguments rely on more than just (SH) and standard probability theory. Look again at Lewis’ first triviality argument as presented in the previous section, and notice that in substituting Pr(ψ | ϕ ∧ ψ) and Pr(ψ | ϕ ∧ ψ) for, respectively, Pr(If ϕ, ψ | ψ) and Pr(If ϕ, ψ | ψ), we are really relying on the following generalization of (SH): 14

See Douven and Dietz [2011: 35]; also Hájek and Hall [1994: 102–105].

The Epistemology of Conditionals

15

Generalized Stalnaker’s Hypothesis (GSH) Pr(If ϕ, ψ | χ) = Pr(ψ | ϕ ∧ χ), provided Pr(ϕ ∧ χ) > 0. From this, we obtain (SH) as a special instance by letting χ be an arbitrary tautology. (GSH) is not a basic assumption of Lewis’ argument. Rather, Lewis derives (GSH) from (SH) in conjunction with what is a basic assumption of the argument, namely, that the class of probability functions for which (SH) holds is closed under conditionalization, meaning that for any probability function Pr in the class, the probability function Prϕ that results from conditionalizing Pr on ϕ, for any proposition ϕ such that Pr(ϕ)  = 0, is in the class as well. With this extra premise, Lewis derives (GSH) as follows: Let Pr be an arbitrary probability function in an arbitrary class P of probability functions such that (i) P is closed under conditionalization and (ii) (SH) holds for all probability functions in P . Then, by (i), where Pr(χ)  = 0, Prχ is in P . Thus, by (ii), for any ϕ and ψ such that Prχ (ϕ)  = 0, Prχ (If ϕ, ψ) = Prχ (ψ | ϕ). And so, given that, by definition, Prχ (If ϕ, ψ) = Pr(If ϕ, ψ | χ) and Prχ (ψ | ϕ) = Pr(ψ | ϕ ∧ χ), it follows that (GSH) holds for Pr. Because Pr was an arbitrary element of P , (GSH) holds for all probability functions in P . Finally, because P was chosen arbitrarily, (GSH) holds for all probability functions in any class of such functions for which (i) and (ii) hold. I should note at once that the assumption that the class of probability functions for which (SH) holds is closed under conditionalization is fairly innocuous, even if not entirely uncontentious. For we are interested in (SH) only insofar as it might pertain to probability functions that could represent rational degrees of belief functions; if irrational people violate (SH), then surely that should not be held against the thesis. So then the assumption is that conditionalization preserves rationality of degrees of belief. It is to be emphasized that the assumption is not that only conditionalization will take one from one rational degree of belief function to another: it is that, possibly along with several other update rules, conditionalization will do so. Indeed, the closure assumption is perfectly consistent with the supposition that rational people sometimes just give up their probability function (throw away their priors, as it is also called) and arbitrarily pick another one. And even those philosophers who do not buy into the dynamic Dutch book arguments (e.g. Teller [1973]) or accuracy domination arguments (e.g. Rosenkrantz [1992])—which implore us to believe that conditionalization is the only rational update rule—typically have no difficulty accepting conditionalization as a rational update rule.15 If I think, nonetheless, that the triviality arguments are not as airtight as they have appeared to be, where is the rub? There is yet another assumption 15 Appiah [1985: 94 ff] may be the only dissenter here. According to him, rationality mandates that one assign probability 1 only to tautologies. Because conditionalization always results in assigning probability 1 to a contingent proposition, it is, in Appiah’s view, not a rational update rule. Clearly, though, if rationality were really to require that we reserve probability 1 for tautologies, then very few of us, if any, would qualify as rational.

16

Igor Douven

underlying Lewis’ derivation of (GSH), one that is easily overlooked, namely, the assumption that the con-di-tion-al-forming operator has an interpretation that is independent of belief states. For suppose the interpretation of “if” were allowed to vary with a person’s belief state so that, in effect, for each Pr we might have a different, tacitly indexed, conditional-forming operator. Thus, let “if” denote the conditional corresponding to the probability function Pr and let “ifχ ” denote the conditional corresponding to the probability function Prχ that comes from Pr by conditionalizing on χ. Then from the assumption that (SH) holds for each pair consisting of a probability function and a corresponding conditional, we get that Prχ (Ifχ ϕ, ψ) = Prχ (ψ | ϕ). By the definition of conditionalization, we have both Prχ (ψ | ϕ) = Pr(ψ | ϕ ∧ χ) and Prχ (Ifχ ϕ, ψ) = Pr(Ifχ ϕ, ψ | χ). But this only gives us (GSH) if we can assume that Pr(Ifχ ϕ, ψ | χ) = Pr(If ϕ, ψ | χ), which we can not assume if the interpretation of conditionals may vary across belief states. We have so far considered only one triviality argument. After the publication of Lewis’ triviality arguments in his 1976 paper, many more triviality arguments have appeared, not all of which rely on (GSH).16 But whatever the details of these further arguments, due to a result by van Fraassen, it is known that all of them—as well as any possible future such arguments— are committed to the assumption that the conditional-forming operator has a fixed interpretation across belief states (unless one is willing to impose certain restrictions on the probability functions representing those belief states; more on this below). More specifically, van Fraassen [1976] proves not only that (SH) is tenable provided that the interpretation of “if” may depend on a person’s belief state, but also that this remains the case if, in each interpretation, “if” is required to satisfy certain logical principles generally agreed to be characteristic for conditionals.17 16 However, a number of these arguments do rely on (GSH), or even on the so-called Import– Export principle (e.g. the triviality arguments offered in Blackburn [1986], Jeffrey [2004: 15 f], and Douven [2011: 394]), according to which “If ϕ, then if ψ, then χ” is logically equivalent to “If ϕ and ψ, then χ,” which in the face of (SH) is an even stronger assumption than (GSH). After all, from Import–Export and the fact that probability theory respects logic, it follows that

Pr(If ϕ, then if ψ, χ) = Pr(If ϕ and ψ, then χ). Applying (SH) to both sides of the above equation yields Pr(If ψ, χ | ϕ) = Pr(χ | ϕ ∧ ψ), which is (GSH). The triviality arguments that rely neither on (GSH) nor on Import–Export include the ones to be found in Hájek [1989], [1994], Hall [1994], and Etlin [2009]. (Hájek’s 1989 argument will be discussed at length further on; for some comments on his 1994 argument, see note 25.) 17 To be still more exact, van Fraassen’s paper shows that the result holds good if “if” is to satisfy the principles of the conditional logic CE. It further shows that if (SH) is restricted to simple conditionals, simple left-nested conditionals (i.e. conditionals whose antecedent is a simple conditional), and simple right-nested conditionals (conditionals whose consequent is a simple conditional), then “if” will even satisfy the stronger conditional logic C2. The latter result is particularly noteworthy, given that, as mentioned, the empirical data so far only support the hypothesis that people judge the probabilities of simple conditionals in accordance with (SH).

The Epistemology of Conditionals

17

In short, all triviality arguments have as a premise, not only (SH), but also the fixedness of the interpretation of the conditional (with the proviso still to be discussed), and these premises, when conjoined with the assumption that the class of probability functions we are considering is closed under conditionalization, entail (GSH). The hypothesis that rational people’s probability functions validate (GSH) should be testable in much the same way in which (SH) has been tested. Yet, whereas much experimental work has been devoted to testing the empirical adequacy of (SH), the question of the empirical adequacy of (GSH) has been largely neglected. In fact, (GSH) has only very recently been subjected to empirical testing in experiments reported in Douven and Verbrugge [2013]. The main experiment in Douven and Verbrugge’s paper is a between subjects experiment in which one group of participants was asked to rate the conditional probabilities of a number of conditionals and another group was asked to rate the corresponding conditional probabilities. The experimental set-up is made clearer by an example of a question that was presented to the first group: Suppose that the global economic situation stabilizes in the next year. Then how probable is the following sentence? If the British government is unsuccessful in creating more jobs, there will be more riots in the UK like the ones we saw a few months ago. Highly improbable 1 2 3 4 5 6 7 Highly probable18 The corresponding question posed to the participants in the second group was this: Suppose that both of the following sentences are true: 1. The global economic situation stabilizes in the next year. 2. The British government is unsuccessful in creating more jobs. Then how probable is the following sentence? There will be more riots in the UK like the ones we saw a few months ago. Highly improbable 1 2 3 4 5 6 7 Highly probable In this study, Douven and Verbrugge found a significant difference between people’s judgments of the probability of a conditional, conditional on a given

18 The numbers 2–6 were also labeled, in the obvious way (2 = Improbable, 3 = Somewhat improbable, etc.).

18

Igor Douven

proposition, and their judgments of the probability of the conditional’s consequent, conditional on the same given proposition in conjunction with the conditional’s antecedent. Thus, their study presents clear evidence against (GSH). Given that (GSH) follows from three premises—(SH), the closure-underconditionalization assumption, and the fixedness of the conditional—we seem to be confronted by a veritable case of Quinean underdetermination. How are we to identify the culprit? Note that the statuses of these premises were not equal to begin with. We saw that during the past decade or so, empirical support has piled up for the first premise. And, as intimated, the second premise should be largely uncontroversial, given that we are throughout concerned with the class of rational probability functions only. As to the third premise, Lewis [1976: 138] thought that denying it would be tantamount to denying that there could be any genuine disagreements about conditionals, given that, being in different belief states, people ostensibly disagreeing about a conditional would in effect be talking past each other. But, as critics have pointed out, that was rash: meaningful disagreement does not require that parties have exactly the same proposition in mind, and, as far as van Fraassen’s result goes, differences between the propositions expressed by a conditional relative to different belief states may be exceedingly small.19 What is more, the idea that the interpretation of conditionals is sensitive to belief states has some independent plausibility. Those who are attracted by something along the lines of Stalnaker’s semantics and hold that similarity relations between possible worlds are involved in the truth conditions of conditionals seem committed to a version of relative semantics for conditionals, given that similarity judgments are known to be sensitive to background beliefs.20 In fact, theorists have of late been proposing semantics that make not only the interpretation of conditionals relative to belief states.21 Thus, the best explanation of why (GSH) 19 See Hájek and Hall [1994: 96] and Douven and Dietz [2011: 32]. Hájek and Hall [1994:99] think that van Fraassen still owes us an explanation of what it is for two propositions to be close to each other in the relevant sense. Here, van Fraassen could usefully appeal to work done in the area of truth approximation, where some authors have developed general accounts of one theory being close to another; see Oddie [2007]. Another problem that van Fraassen’s result faces, according to Hájek and Hall (ibid.), is that it fails to specify a rule that governs the change from one interpretation of a conditional to another as we move from one belief state to another. It is true that neither van Fraassen, nor anyone else for that matter, has been able to specify such a rule. But, to refer back to the issue of updating on conditionals mentioned in Section 1.1, we at most know some rules for updating on conditionals under special circumstances. For a broad class of circumstances, we do not know how we update on conditionals. Yet, we do update on conditionals even in such circumstances, and we also have clear intuitions about when such updates go right and when they go wrong. If the absence of explicit rules for updating on conditionals is no impediment to updating on conditionals, nor to discriminating between such updates that are intuitively correct and intuitively incorrect, then why should the absence of explicit rules for going from one interpretation of the conditional to another be an impediment to carrying out such transitions, and for doing so in an intuitively correct way? 20 Though, as we saw, Stalnaker’s semantics itself does not validate (SH). 21 See MacFarlane [2012] and references given therein.

The Epistemology of Conditionals

19

is empirically inadequate is that the third premise involved in its derivation— the assumption that conditionals have a fixed interpretation across belief states—is false. It is presumably fair to say that most theorists working on conditionals find van Fraassen’s tenability result technically impressive but otherwise of academic interest at most. At least that would explain why, in the literature on conditionals, the result has received but scant attention. Already the empirical data relating to (SH) that have been gathered in the past ten years should have been reason to take van Fraassen’s result more seriously, given that it follows from this result that the said data need not be explained away and can be taken to show what they seem to show. The experimental data reported in Douven and Verbrugge [2013] lend still further support to van Fraassen’s idea that the interpretation of the conditional may be belief-sensitive, in view of the fact that these data show to be empirically inadequate a thesis—(GSH)— that follows from the conjunction of (i) the negation of van Fraassen’s idea, (ii) (SH), a thesis for which considerable support has accrued over the years, and (iii) a fairly uncontentious assumption about the class of rational probability functions (closure under conditionalization). To be sure, these data do not thereby constitute conclusive evidence for the correctness of van Fraassen’s idea. But they support it strongly enough, I submit, to warrant thorough rethinking of what are generally regarded as some of the most solid results of twentieth-century analytic philosophy. It is time to return to the proviso I made when I summarized van Fraassen’s tenability result, that, in view of that result, all triviality arguments, present and future, must assume that the interpretation of the conditional-forming operator is independent of belief states, unless they impose certain restrictions on the probability functions representing those belief states. That van Fraassen’s result can be circumvented in this way follows from a number of arguments which are all conveniently surveyed in Hájek and Hall [1994]. Here, I consider only the argument originally put forward in Hájek [1989], but the general point to be made in response to this applies to the other arguments as well. Hájek’s 1989 paper presents a triviality argument that does not assume the fixedness of the interpretation of the conditional though it does assume that at least some people’s belief states can be represented—at least some of the time—by means of a probability function that is definable on a finite set of possible worlds. Instead of going through the various steps of this argument, I give a simple illustration of it.22 Let a person’s degrees of belief function be representable by means of three equiprobable possible worlds, w1 , w2 , and w3 . Then all unconditional probabilities must be multiples of 1/3. By contrast, some conditional probabilities have the value 0.5. For instance, where ϕ := {w1 , w2 } and ψ := {w1 }, Pr(ψ | ϕ) = .5. As a result, there can be no proposition expressed by “If ϕ, ψ” such that its unconditional probability matches the 22

I owe the following illustration to Alan Hájek.

20

Igor Douven

corresponding conditional probability. Hájek’s argument is a generalization of this simple idea, leading to the conclusion that if a person’s degrees of belief function can be represented by means of a probability assignment to n ≥ 3 possible worlds, then there will be conditional probabilities that cannot be matched with any unconditional probabilities. How problematic is it that, if we want to keep to (SH) in light of the empirical support it has received, we must assume rational people’s belief states to be representable only by infinite algebras? It is certainly not a priori that people’s belief states are ever representable by means of probability functions defined on only finitely many possible worlds. Naturally, for certain practical purposes, it may be convenient to consider only part of a person’s belief state, and it may be possible to model that part by reference to only finitely many possible worlds. But, for all anyone has shown, a full representation of a person’s belief state requires an assignment of probabilities to all models of the person’s language, which will realistically mean to uncountably many models. That this is true for all anyone has shown is no guarantee that it is true. However, if we accept the methodological tenets of traditional Bayesianism, then we do have independent grounds for holding that our degrees of belief functions are to be modeled by infinite means. For according to traditional Bayesian thinking, we will, when prompted, be able to come up with a fair betting rate for (say) the proposition that tomorrow the sun will set between six o’ clock and one minute past six in Amsterdam even if we have never actively considered this proposition. Still according to traditional Bayesian thinking, this fact evidences a pre-existing disposition to bet on, and thus degree of belief in, the designated proposition. Given that the same will hold for each proposition that differs from the aforementioned one only in referring to a different time interval, it follows that we have degrees of belief in uncountably many propositions concerning at what time the sun will set in Amsterdam tomorrow, whether or not these propositions have ever consciously crossed our minds. Those who disagree with the broadly behaviorist assumptions of the foregoing argument may still be persuaded to accept its conclusion by the observation that, if for no other reason, the fact itself that—as the data show— people’s degrees of belief accord with (SH) but not with (GSH) seems, in view of Hájek’s argument, to constitute a legitimate reason for holding that those degrees of belief are to be represented by infinite means. Indeed, it is worth putting this point in more general terms, so as to make it apply equally to the other arguments that have sought to maintain triviality in the face of van Fraassen’s result. In these more general terms, the point is that if the simplest account of a wide range of data—some of which pertain to (SH), some to (GSH)—represents people’s degrees of belief functions as belonging to a certain class of functions, where we have neither empirical nor theoretical reasons for believing that they do not belong to that class, then that fact itself provides indirect empirical support for

The Epistemology of Conditionals

21

holding that people’s degrees of belief functions do belong to the given class. Some readers may have as little sympathy for the abductive reasoning underlying this second argument as they have for the behaviorist bent of the first. Perhaps the following “transcendental” argument, which appeals to a version of the principle of charity, can sway even these readers. Van Fraassen’s result and the responses it elicited show that if people’s (nontrivial) degrees of belief conform to (SH), then these degrees of belief are probabilities only if they require (uncountably) infinite means for their representation. Rationality requires people’s degrees of belief to be probabilities. So, given that people’s degrees of belief do conform to (SH), people are either irrational or their degrees of belief are representable by infinite means only. And hence, finally, given that charity dictates that we assume people to be rational if at all possible, we must assume that people’s degrees of belief functions are representable by infinite means only.23 There are a number of more detailed points worth making concerning the responses to van Fraassen’s result. Here, I confine myself to one comment on the argument from Hájek [1989]. The comment is to the effect that even if it had to be granted that at least some belief states can be modeled by finite means, Hájek’s argument would still not give reason to think that the empirical results concerning the probabilities of conditionals exhibit flawed probabilistic reasoning, and so should be grouped with the experiments, mentioned earlier, which showed that people sometimes give more credibility to a conjunction than to its conjuncts. To see why not, first note that, given n equiprobable worlds, the divergence between the probability of a conditional and the corresponding conditional probability need never exceed 1/2n. So, for instance, if n = 99, then even though there is no proposition that has a probability of exactly 0.5, there are propositions that have a probability that is within 0.005 of 0.5. That divergence is certainly smaller than any experiments of the kind that have been hitherto conducted, and perhaps simply any experiments, could detect.24

23 Note that there is no tension between this argument and Hájek’ s [2001: 380] remark that it is hard to see why rationality should require people to have degrees of belief functions that can only be modeled by infinite means. What is involved in the argument is not the claim that rationality requires people’s degrees of belief functions to be of the said sort but rather the weaker claim that if people’s (nontrivial) degrees of belief functions conform to (SH), then rationality requires these functions to be of that sort. The latter claim follows directly from van Fraassen’s results and Hájek’s and others’ responses to it in conjunction with the, at least in the context of the current debate, uncontested claim that rational degrees of belief are probabilities; like Hájek, I am not aware of any argument supporting the former claim. 24 In this connection, it should be noted that many of the experiments concerning the probabilities of conditionals used Likert scales (labeled, typically, from “highly probable” to “highly improbable,” as in the example from Douven and Verbrugge [2013] on page 17) to measure participants’ probabilities. Those experiments that did not use Likert scales asked participants to indicate their probabilities in terms of percentages, which will have invited these participants to indicate their probabilities up to a hundredth only.

22

Igor Douven

Of course, it is no assumption of Hájek’s argument that all worlds are equiprobable. But if they are not, then that does not necessarily mean that the divergence between probabilities of conditionals and conditional probabilities must be, or at least must sometimes be, greater than 1/2n. In fact, quite the opposite may hold! Thus, unless the assumption is not so much that some belief states are representable by finitely many possible worlds but rather that some belief states are representable by just a handful of possible worlds, we should not expect that experiments conducted to compare the probabilities of conditionals and the corresponding conditional probabilities show a significant difference between these variables. So, even given the assumption that at least some belief states can be modeled by finite means—an assumption which, for all Hájek has told us, we are under no obligation to make—Hájek’s argument leaves unscathed at least the approximate version of (SH). And that, as said earlier, is all that the empirical results support anyway.25

3.

A S S E R TA B I L I T Y A N D A C C E P TA B I L I T Y: A D A M S ’ T H E S I S A N D BEYOND

Before Stalnaker proposed (SH) as an adequacy condition for semantics of conditionals, Adams [1965] had already proposed a thesis that looked almost identical, and which now goes by his name: Adams’ Thesis (AT) Where “If ϕ, ψ” is a simple conditional, Pr(If ϕ, ψ) = Pr(ψ | ϕ), provided Pr(ϕ)  = 0; if Pr(ϕ) = 0, then Pr(If ϕ, ψ) = 1. One difference between (AT) and (SH) is immediately obvious: the former is restricted to simple conditionals, while the latter is not. This restriction was motivated by Adams’ skepticism about the idea that conditionals express propositions. Were (AT) to hold unrestrictedly, then, where ϕ or ψ is a conditional, and given the ratio definition of conditional probability, the probability of “If ϕ, ψ” would equal Pr(ϕ ∧ ψ)/ Pr(ϕ), according to (AT). That is, the probability of the conditional would equal something in the numerator of which occurs a conjunction at least one of whose conjuncts is a conditional. But to admit that conditionals can figure as conjuncts in conjunctions commits one 25 It is also worth mentioning Hájek’s [1994] so-called perturbation argument, another attempt to challenge van Fraassen’s result. This perturbation argument does require uniformity of the meaning of “if” across belief states, but Hájek takes this argument to show that there is no way to ensure that (SH) holds if the meaning of “if” is allowed to vary other than simply to impose (SH) as a requirement on probability functions. For our present concerns, however, this is unproblematic. It is not as though we have to delineate on a priori grounds the class of degrees of belief functions for which (SH) holds. Rather, the task we face is to make sense of the fact that, for whatever precise reason, people’s degrees of belief functions do seem to satisfy (SH), or at least an approximate version thereof, and to do so in a way that does not imply that—supposing these degrees of belief functions to be probability functions—they must have features that they are known not to have as a matter of empirical fact (for example, that unconditional probabilities equal conditional probabilities).

The Epistemology of Conditionals

23

to the view that conditionals express propositions: the conjunction operator is, after all, a propositional operator. Adams’ non-propositional view of conditionals has a further consequence, however, one that is not apparent from (AT) as usually stated. As noted earlier, it is hard to make sense of the idea that conditionals could have a probability of truth if they cannot be true to begin with. Therefore, as Adams realized, given non-propositionalism, the probability operator cannot apply literally to conditionals. He thus proposed that we think of Pr(If ϕ, ψ) as indicating the degree of assertability of “If ϕ, ψ”; this, he thought, was measured by the corresponding conditional probability Pr(ψ | ϕ).26 Later, Adams came to think of Pr(If ϕ, ψ) as indicating rather the degree of acceptability of the conditional. Thus, it might be clearer to state Adams’ Thesis as As(If ϕ, ψ) = Pr(ψ | ϕ), as Jackson [1987: 11] does, or as Ac(If ϕ, ψ) = Pr(ψ | ϕ), as in Douven and Verbrugge [2010: 305], where, of course, the provisos of (AT) are supposed to hold. The non-propositional view of conditionals protects (AT) against all extant triviality arguments. Nearly all of these arguments are blocked by the fact that conditionals cannot occur in Boolean combinations if they do not express propositions. For instance, in that case the law of total probability—–or an analog of that for Adams’ assertability/acceptability operator—cannot be applied to conditionals, which immediately blocks Lewis’ first triviality result rehearsed in Section 2.1. And although Hájek’s [1989] argument does not require any embedding of conditionals in Boolean combinations, it too is blocked because if conditionals do not express propositions, they cannot be equated with sets of possible worlds whose degree of assertability might somehow equal the sum of the degrees of assertability assigned to their members. A fortiori, Hájek cannot argue that, assuming that a person’s belief state can be represented by means of a finite algebra of propositions, the number of distinct values of the person’s conditional probabilities must exceed the number of degrees of assertability that this person attaches to the various propositions. (AT) not only escapes the triviality arguments; according to all commentators, it also enjoys strong intuitive and empirical support. The following quotes are representative: There is a great deal of evidence for [(AT)]. There is head-counting evidence. Very many philosophers of otherwise differing opinions have found [(AT)] highly intuitive. There is case-by-case evidence. Take a conditional which is highly assertible . . .; it will invariably be one whose consequent is highly probable given its antecedent. (Jackson [1987: 12]) [(AT)] describes what English speakers assert and accept with unfailing accuracy. (McGee [1989: 485])

26 We are supposed to abstract here from Gricean and other broadly social considerations. The thesis could in effect be said to concern what Jackson [1987: 10] calls “assertibility.”

24

Igor Douven

The numerical values [Adams assigns to conditionals] accurately measure the assertability and acceptability of conditionals. (McGee [1989: 485)

Indeed, philosophers seem to agree that (AT) is true as much as they agree that (SH) is false.27 Despite what the above citations might suggest, they are not based on any systematic investigation of (AT)’s empirical adequacy. Until very recently, experimentalists have paid no attention to (AT). Perhaps they thought that it would make no difference to ask people to judge the degrees of assertability or acceptability of conditionals instead of the probabilities. Adams’ own work may even have suggested as much, given that he sticks to the use of the probability operator in front of conditionals, which is then to be interpreted as an acceptability or assertability operator. Still, the assertion that asking for a conditional’s probability and asking for its degree of acceptability or assertability will yield the same result is itself an empirical claim, which it would seem well worth checking. Douven and Verbrugge [2010] report the first experimental findings that explicitly pertain to (AT). For the stimuli of their experiments, they make use exclusively of what linguists call “inferential conditionals,” that is, conditionals that express a reasoning process, having the conditional’s antecedent as a premise and its consequent as a conclusion.28 Here are some examples of such conditionals: (7) a. If a + 1 = 7, then a = 6. b. If John lives in Chelsea, he is rich. c. If Tim and Hank are jogging together, they must have patched up their recent quarrel. Douven and Verbrugge propose a typology of inferential conditionals on the basis of the type of inference a conditional reflects, where the relevant types are deductive, inductive, and abductive inference, and where the type may depend on the background assumptions that are being made in the context in which a conditional is asserted or evaluated. For instance, (7a) is (what Douven and Verbrugge call) a deductive inferential (DI) conditional. On the other hand, (7b) is plausibly thought of as reflecting an inductive inference, leading to the conclusion that John is rich on the basis of the premise that he lives in Chelsea, given the background assumption that most people living in Chelsea

27 However, many also agree with McGee [1989: 485] that (AT) is disappointingly weak, given that it says nothing about compounds of conditionals. McGee’s paper is actually an attempt to extend (AT) to right-nested conditionals, though, according to Lance [1991 and Dietz and Douven [2010], not a successful one. 28 See, e.g. Dancygier [1998], Declerck and Reed [2001], and Haegeman [2003]. Inferential conditionals have also been discussed in the psychology of reasoning literature; see, e.g. Politzer and Bonnefon [2006] and Verbrugge et al. [2007].

The Epistemology of Conditionals

25

are rich; such conditionals are called “inductive inferential conditionals” (“II conditionals,” for short). And in a context in which it is believed that old friends Tim and Hank recently had a serious quarrel which, it seemed, ended their friendship for good, (7c) would qualify as an abductive inferential (AI) conditional, given that, in the said kind of context, their having patched up their quarrel best explains the (presumed) fact that they are jogging together. Douven and Verbrugge’s main experiment is again a between subjects experiment: The participants in  one group were given thirty items, consisting of context–conditional pairs Ci , "If ϕi , ψi " (1 ≤ i ≤ 30), with ten DI, ten II, and ten AI conditionals; participants were asked to rate the acceptability of each conditional “If ϕi , ψi ” light of context Ci . The thirty context–conditional pairs were used to construct  thirty context–sentence pairs of the schematic form Ci + "Suppose ϕi ", ψi . These were presented to a second group of participants, who were asked to rate the probability of each ψi in light of context Ci + "Suppose ϕi , " thereby yielding conditional probabilities ratings of ψi given ϕi in Ci , for all i. The results reported in Douven and Verbrugge’s paper are mixed, but mostly negative. When taken over all inferential conditionals, the results manifestly refute (AT), even in the “approximate” version of that thesis. In fact, the data refute a weaker thesis still, to wit, that whenever the acceptability of a conditional is high, so is the corresponding conditional probability, and similarly when the acceptability of a conditional is middling and when it is low. The only general thesis supported by the data was that there is a high correlation between acceptability judgments and judgments of corresponding conditional probabilities. This, as said, is when taken over all inferential conditionals. When the types of inferential conditionals are considered separately, the results are partly better, partly worse. For DI conditionals, (AT) does hold (or at any rate the approximate version of that thesis). For AI conditionals, the most that can be said is, again, that acceptability of conditionals highly correlates with conditional probability. However, for II conditionals, not even that much is true. Given that inferential conditionals do not exhaust the class of conditionals, the picture may come to look still more fragmented when further types of conditionals are taken into account.29 Ironically, while (SH) had never been tested specifically in relation to inferential conditionals, Douven and Verbrugge, in one of their control experiments, used the same stimuli for testing (SH) that they had used for testing (AT), with the result of finding more support for (SH). That is to say, when 29 Linguists contrast inferential conditionals with (what they call) content conditionals, which are defined as describing relations between states of affairs or events as they happen in reality (see note 28 for references). Whether the distinction between inferential and content conditionals is well founded, and in particular whether at least some of the conditionals classified as content conditionals in the literature are not better thought of as reflecting an inference from antecedent to consequent, will not concern us here. In any case, it seems safe to assume that many, but not all, conditionals encountered in colloquial speech fall into the class of inferential conditionals.

26

Igor Douven

they asked a third group of participants to rate the probability (instead of the acceptability) of each “If ϕi , ψi ” in light of Ci , they found a close to perfect match between those ratings and the corresponding conditional probability ratings of their second group of participants. The results reported in Douven and Verbrugge [2010] in effect create an even more fundamental problem for Adams and those who, like him, wish to maintain that we can make sense of a probability operator as pertaining to a conditional even if conditionals do not express propositions—namely, by interpreting it as indicating the degree of acceptability of the conditional. For Douven and Verbrugge also found a significant difference between the probability and the acceptability ratings of conditionals. So, apparently, people do not interpret the probability operator when it occurs in front of a conditional as indicating degree of acceptability.30 (AT) is a quantitative thesis, relating the degree of acceptability of a conditional to the corresponding conditional probability. But although we do talk and think in terms of degrees of acceptability, we also use “acceptability” and “acceptable” (and kindred terms) in a categorical sense: statements can be said to be acceptable tout court as well as unacceptable tout court. Accordingly, we can ask under what conditions a conditional is acceptable, categorically speaking. For those who were (or still are) attracted to (AT), the most natural answer may be provided by the following qualitative version of that thesis: Qualitative Adams’ Thesis (QAT) A simple conditional “If ϕ, ψ” is assertable/acceptable iff Pr(ψ | ϕ) is high. Among the notable proponents of (QAT) are Lewis [1976] and Jackson [1987: 31]. Natural and plausible though (QAT) may appear, there is an a priori reason to doubt its material adequacy. Perhaps the most fundamental intuition about the acceptability of conditionals is that, for a conditional to be acceptable, there must be some sort of connection between antecedent and consequent. And while (QAT)’s requirement of high conditional probability may seem to secure the presence of just such a connection, it does not really do that. Any proposition that is highly probable unconditionally is also highly probable conditional on any other proposition that is probabilistically independent of it, and often the absence of probabilistic relevance signals the absence of relevance in a broader sense. We may suppose that (7) If Obama is reelected, there will be some rainfall in Boston in 2013. satisfies the condition of (QAT). Yet many will find it intuitively unacceptable. Plausibly, this is because the truth of the antecedent is entirely irrelevant to

30 Here in particular it is worth mentioning that Douven and Verbrugge also conducted a control experiment to verify that their participants had understood the notion of acceptability in the intended (epistemic) sense, and not as, for example, social or political acceptability.

The Epistemology of Conditionals

27

the truth of the consequent, which is reflected in the fact that the antecedent and consequent are probabilistically independent of each other. Motivated by this kind of consideration, Douven [2008] hypothesizes that (QAT) goes wrong in lumping together all cases in which the probability of the consequent, given the antecedent, is high regardless of whether that probability is higher than the unconditional probability of the consequent, and that a conditional is assertable/acceptable only if the antecedent is evidence (in the Bayesian sense of the word) for the consequent, that is, only if it makes the consequent more probable. Specifically, the proposal is that Evidential Support Thesis (EST) A simple conditional “If ϕ, ψ” is assertable/acceptable iff Pr(ψ | ϕ) is not only high but also higher than Pr(ψ).31,32 To convince oneself that ordinary usage provides evidence for (EST), one may consider any number of examples of conditionals and register one’s responses to them. But a general lesson to be learned from the empirical work on (AT) discussed above is that philosophers’ intuitions regarding the acceptability or unacceptability of conditionals are anything but failsafe. The way to gain more definite insight into the status of theses such as (QAT) and (EST) is to subject them to the same kind of testing to which (SH) and (AT) have been subjected. This has recently been done and the results of these tests have been reported in Douven and Verbrugge [2012]. In that paper, Douven and Verbrugge not only consider the descriptive accuracy of (QAT) and (EST) but also the possibility that the notion of evidential support is key to distinguishing between the acceptability of an indicative conditional and the acceptability of the corresponding concessive conditional, that is, the conditional with “if” replaced by “even if” (or with “still” inserted in the consequent). In particular, they propose the following thesis as a complement to (EST): Concessive Absence of Support Thesis (CAST) A simple concessive conditional “If ϕ, still ψ,” or “Even if ϕ, ψ,” is assertable/acceptable iff Pr(ψ | ϕ) is high but no higher than Pr(ψ).

31 The proposal was actually a bit more complicated in that a condition was added to prevent conditionals like “If there is only one winner, then your ticket is a loser” from qualifying as acceptable in a situation in which it is known that either the relevant fair and large lottery has a unique winner or half of the tickets will end up being winners, and that the chances of either of these possibilities obtaining are equal. For present purposes, this complication can be sidestepped. 32 A referee pointed out that (EST) may be in trouble if we accept the principle that “when a conditional is unacceptable, the possibility of a counterexample is acceptable: if ‘If ϕ, ψ’ is unacceptable then ‘Maybe ϕ and ψ’ is acceptable.” However, it seems to me that this “not-ifto-maybe” principle is to be rejected. We may find “If Arsenal ends first in this year’s Premier League, then 2 + 2 = 4” unacceptable for essentially the same reason we find (8) unacceptable. Yet no one will find “Maybe Arsenal ends first in this year’s Premier League and 2 + 2  = 4” acceptable.

28

Igor Douven

They even consider the possibility of distinguishing between cases in which the conditional probability is strictly lower than the unconditional probability of the consequent and those in which the two are equal. They conjecture that at least some of the latter cases may fall into the class of non-interference conditionals, conditionals that are meant to emphasize the inevitability or obviousness of the consequent, regardless of whether the antecedent obtains (see Bennett [2003: 122 ff]). Douven [2008: 32] argues that these conditionals may be characterized by the fact that their assertability/acceptability is either not at all or positively affected if “even if” is substituted for “if” in them, or if the word “still” is inserted in their consequent clause. However, Douven and Verbrugge also note that other cases in which the consequent is high given the antecedent and just as high as it is unconditionally will be cases in which the indicative form is simply unacceptable. Douven and Verbrugge conducted a within subjects experiment designed to test all three of (QAT), (EST), and (CAST). In this experiment, the participants were presented with a number of items, each centering around a different conditional. The items consisted of three parts. The first part asked for the unconditional probability of the consequent of the conditional corresponding to the item. The second part asked for the conditional probability of the consequent given the antecedent of the same conditional. And the third part showed the indicative conditional together with the corresponding concessive conditional and asked whether the participant accepted only the former, only the latter, both, or neither; for instance, the following is the third part of one of the items used in the study: Consider the following statements: 1. If some second-rate actor plays the role of James Bond in the next Bond movie, that movie will become a box office hit. 2. Even if some second-rate actor plays the role of James Bond in the next Bond movie, that movie will become a box office hit. Of these statements:    

I accept only 1. I accept only 2. I accept both. I accept neither.

The results Douven and Verbrugge obtained are both evidence against (QAT) and for (EST) and (CAST). While participants did find some indicatives with a high corresponding conditional probability acceptable, they found many others with an equally high or even higher corresponding conditional probability not acceptable, thereby casting doubt on (QAT). The data were also consistent with the hypothesis that the vast majority of participants regarded

The Epistemology of Conditionals

29

high conditional probability as a necessary or at least close-to-necessary condition for the acceptability of the corresponding indicative or concessive, as predicted by (EST) and, respectively, (CAST). Finally, the data confirmed the prediction, following from the conjunction of (EST) and (CAST), that if people judge Pr(ψ | ϕ) to be high, their preference for either “If ϕ, ψ” or “Even if ϕ, ψ” depends on whether or not they judge ϕ to be evidence for ψ.

4.

CONCLUDING REMARKS

That (SH) is false and that (AT) is true seemed to be the two pillars on which any future epistemology of conditionals was to be built. Neither claim was backed by empirical evidence, and neither may have seemed to need such backing: the falsity of (SH) seemed to follow from indubitable formal results, and the truth of (AT) seemed a matter of course. It is more than ironic that recent empirical results suggest that philosophers have gotten things wrong with respect to both claims: (SH) has been experimentally confirmed several times over, and in one experiment (AT) was refuted (which is refutation enough). As explained, the data apparently in favor of (SH) might really betray a systematic error in how people judge the probabilities of conditionals, or in how they determine conditional probabilities, or in both. But further empirical research was seen to support a more charitable interpretation of those data, one that seeks fault rather with an assumption underlying the formal results that had been thought to undermine (SH). While (AT) turned out to be false, there still appeared to be a high correlation between the degrees to which people judge conditionals to be acceptable and their conditional probabilities. On the other hand, when the empirical results were split out for different types of conditionals, the picture was messier, and it may be messier still if more types of conditionals are taken into account. If all this sounds somewhat disconcerting, then note that if (SH) is tenable after all, perhaps it can serve the purpose which it was originally meant to serve, namely, guiding our quest for a semantics of conditionals. And even though the relation between the acceptability of conditionals and the corresponding conditional probabilities is not nearly as straightforward as (AT) suggests, at least we now have an account—even if perhaps only a partial one—that is buttressed by empirical research. In addition, we have proposals about the categorical acceptability conditions of indicative and concessive conditionals that also enjoy empirical support. And lastly, to make a positive point on a more general level, the foregoing underscores the importance of empirical research on conditionals for the epistemology of conditionals, and suggests that progress on this front may call for a closer and more sustained collaboration between philosophers and experimental psychologists than has hitherto been the case.

30

Igor Douven

ACKNOWLEDGMENTS

I am greatly indebted to two anonymous referees for valuable comments on a previous version. I am also grateful to David Etlin, David Over, and Sara Verbrugge for many helpful discussions on conditionals over the past years and for introducing me to empirical approaches to the study of conditionals. Versions of this paper were presented at the universities of Bochum, Lund, Milan, and Utrecht. I thank the audiences present on those occasions for useful questions and discussions.

REFERENCES

Adams, E. W. [1965] “The Logic of Conditionals,” Inquiry 8:166–197. [1998a] A Primer of Probability Logic, Stanford: CSLI Publications. [1998b] “The Utility of Truth and Probability,” in P. Weingartner, G. Schurz, and G. Dorn (eds.) The Role of Pragmatics in Contemporary Philosophy, Vienna: HölderPichler-Tempsky, pp. 176–194. Appiah, A. [1985] Assertion and Conditionals, Cambridge: Cambridge University Press. Barrouillet, P. and Lecas, J. F. [1999] “Mental Models in Conditional Reasoning and Working Memory,” Thinking and Reasoning 5:289–302. Bennett, J. [2003] A Philosophical Guide to Conditionals, Oxford: Oxford University Press. Blackburn, S. [1986] “How Can We Tell Whether a Commitment Has a Truth Condition?” in C. Travis (ed.) Meaning and Interpretation, Oxford: Blackwell, pp. 201–232. Boutilier, C. and Goldszmidt, M. [1995] “On the Revision of Conditional Belief Sets,” in G. Crocco, L. Fariñas del Cerro, and A. Herzig (eds.) Conditionals. From Philosophy to Computer Science, Oxford: Oxford University Press, pp. 267–300. Bovens, L. and Hartmann, S. [2003] Bayesian Epistemology, Oxford: Oxford University Press. Byrne, R. M. J. and Johnson-Laird, P. N. [2010] “Conditionals and Possibilities,” in M. Oaksford and N. Chater (eds.) Cognition and Conditionals, Oxford: Oxford University Press, pp. 55–68. Dancygier, B. [1998] Conditionals and Predictions: Time, Knowledge and Causation in Conditional Constructions, Cambridge: Cambridge University Press. de Finetti, B. [1937] “La Prévision: Ses Lois Logiques, ses Sources Subjectives,” Annales de l’Institut Henri Poincaré 7:1–68. Declerck, R. and Reed, S. [2001] Conditionals: A Comprehensive Empirical Analysis, Berlin/New York: Mouton de Gruyter. Dietz, R. and Douven, I. [2010] “Adams’ Thesis, Ramsey’s Test, and Left-Nested Conditionals,” Review of Symbolic Logic 3:467–484. Douven, I. [2008] “The Evidential Support Theory of Conditionals,” Synthese 164:19–44. [2011] “Indicative Conditionals,” in L. Horsten and R. Pettigrew (eds.) A Companion to Philosophical Logic, London: Continuum Press, pp. 383–405. [2012] “Learning Conditional Information,” Mind and Language 27:239–263.

The Epistemology of Conditionals

31

and Dietz, R. [2011] “A Puzzle about Stalnaker’s Hypothesis,” Topoi 30:31–37. Douven, I. and Romeijn, J. W. [2011] “A New Resolution of the Judy Benjamin Problem,” Mind 120:637–670. Douven, I. and Verbrugge, S. [2010] “The Adams Family,” Cognition 117:302–318. [2012] “Indicatives, Concessives, and Evidential Support,” manuscript Thinking and Reasoning 18:480–499. [2013] “The Probabilities of Conditionals Revisited, “ Cognitive Science, forthcoming. Edgington, D. [1995] “On Conditionals,” Mind 104:235–329. [1997] “Commentary,” in M. Woods, Conditionals (edited by D. Wiggins), Oxford: Clarendon Press, pp. 95–137. Etlin, D. [2009] “The Problem of Noncounterfactual Conditionals,” Philosophy of Science 76:676–688. Evans, J. St. B. T., Handley, S. J., Neilens, H., and Over, D. E. [2007] “Thinking about Conditionals: A Study of Individual Differences,” Memory and Cognition 35:1759–1771. Evans, J. St. B. T., Handley, S. J., and Over, D. E. [2003] “Conditionals and Conditional Probability,” Journal of Experimental Psychology: Learning, Memory and Cognition 29:321–355. Evans, J. St. B. T. and Over, D. E. [2004] If, Oxford: Oxford University Press. Fugard, A., Pfeifer, N., Mayerhofer, B., and Kleiter, G. [2011] “How People Interpret Conditionals,” Journal of Experimental Psychology: Learning, Memory, and Cognition 37:635–648. Gärdenfors, P. [1982] “Imaging and Conditionalization,” Journal of Philosophy 79:747–760. Gauffroy, C. and Barrouillet, P. [2009] “Heuristic and Analytic Processes in Mental Models for Conditionals: An Integrative Developmental Theory,” Developmental Review 29:249–282. Gilio, A. and Over, D. E. [2012] “The Psychology of Inferring Conditionals from Disjunctions: A Probabilistic Study,” Journal of Mathematical Psychology 56:118–131. Girotto, V. and Johnson-Laird, P. N. [2004] “The Probability of Conditionals,” Psychologica 47:207–225. [2010] “Conditionals and Probability,” in M. Oaksford and N. Chater (eds.) Cognition and Conditionals, Oxford: Oxford University Press, pp. 103–115. Hadjichristidis, C., Stevenson, R. J., Over, D. E., Sloman, S.A., Evans, J. St. B.T., and Feeney, A. [2001] “On the Evaluation of ‘if p then q’ Conditionals,” in J. D. Moore and K. Stenning (eds.) Proceedings of the 23rd Annual Meeting of the Cognitive Science Society, Edinburgh, pp. 381–386. Haegeman, L. [2003] “Conditional Clauses: External and Internal Syntax,” Mind and Language 18:317–339. Hájek, A. [1989] “Probabilities of Conditionals—Revisited,” Journal of Philosophical Logic 18:423–428. [1994] “Triviality on the Cheap?” in E. Eells and B. Skyrms (eds.) Probability and Conditionals, Cambridge: Cambridge University Press, pp. 113–140. [2001] “Probability, Logic, and Probability Logic,” in L. Goble (ed.) The Blackwell Guide to Philosophical Logic, Oxford: Blackwell, pp. 362–384. and Hall, N. [1994] “The Hypothesis of the Conditional Construal of Conditional Probability,” in E. Eells and B. Skyrms (eds.) Probability and Conditionals, Cambridge: Cambridge University Press, pp. 31–46.

32

Igor Douven

Hall, N. [1994] “Back in the CCCP,” in E. Eells and B. Skyrms (eds.) Probability and Conditionals, Cambridge: Cambridge University Press, pp. 141–160. Hertwig, R. and Gigerenzer, G. [1999] “The ‘Conjunction Fallacy’ Revisited: How Intelligent Inferences Look Like Reasoning Errors,” Journal of Behavioral Decision Making 12:275–305. Jackson, F. [1987] Conditionals, Oxford: Blackwell. Jeffrey, R. [2004] Subjective Probability: The Real Thing, Cambridge: Cambridge University Press. Johnson-Laird, P. N. and Byrne, R. M. J. [2002] “Conditionals: A Theory of Meaning, Pragmatics, and Inference,” Psychological Review 19:646–678. Joyce, J. [1998] “A Nonpragmatic Vindication of Probabilism,” Philosophy of Science 65:575–603. Kern-Isberner, G. [1999] “Postulates for Conditional Belief Revision,” in T. Dean (ed.) Proceedings of the Sixteenth International Joint Conference on Artificial Intelligence, San Francisco: Morgan Kaufmann, pp. 186–191. Kyburg, H. [1961] Probability and the Logic of Rational Belief, Middletown CT: Wesleyan University Press. Lance, M. [1991] “Probabilistic Dependence Among Conditionals,” Philosophical Review 100:269–276. Lecas, J. F. and Barrouillet, P. [1999] “Understanding Conditional Rules in Childhood and Adolescence: A Mental Models Approach,” Current Psychology of Cognition 18:363–396. Lewis, D.K. [1976] “Probabilities of Conditionals and Conditional Probabilities,” Philosophical Review 85:297–315. (Reprinted, with postscript, in Jackson (ed.) [1991], pp. 76–101; the page references are to the reprint.) MacFarlane, J. [2012] “Relativism,” in D. Graff Fara and G. Russell (eds.) The Routledge Companion to the Philosophy of Language, New York: Routledge, forthcoming. McGee, V. [1989] “Conditional Probabilities and Compounds of Conditionals,” Philosophical Review 98:485–541. Moutier, S. and Houdé, O. [2003] “Judgement under Uncertainty and Conjunction Fallacy Inhibition Training,” Thinking and Reasoning 9:185–201. Nolan, D. [2003] “Defending a Possible-Worlds Account of Indicative Conditionals,” Philosophical Studies 116:215–269. Oaksford, M. and Chater, N. [2003] “Conditional Probability and the Cognitive Science of Conditional Reasoning,” Mind and Language 18:359–379. [2007] Bayesian Rationality, Oxford: Oxford University Press. Oberauer, K., Weidenfeld, A., and Fischer, K. [2007] “What Makes Us Believe a Conditional? The Roles of Covariation and Causality,” Thinking and Reasoning 13:340–369. Oberauer, K. and Wilhelm, O. [2003] “The Meaning(s) of Conditionals: Conditional Probabilities, Mental Models and Personal Utilities,” Journal of Experimental Psychology: Learning, Memory and Cognition 29:688–693. Oddie, G. [2007] “Truthlikeness,” The Stanford Encyclopedia of Philosophy (Fall 2008 Edition), available at . Accessed 7 November 2012. Over, D. E., Douven, I., and Verbrugge, S. [2013] “Scope Ambiguities and Conditionals,” Thinking and Reasoning.

The Epistemology of Conditionals

33

Over, D. E. and Evans, J. St. B. T. [2003] “The Probability of Conditionals: The Psychological Evidence,” Mind and Language 18:340–358. Over, D. E., Hadjichristidis, C., Evans, J. St. B.T., Handley, S. J., and Sloman, S.A. [2007] “The Probability of Causal Conditionals,” Cognitive Psychology 54:62–97. Pfeifer, N. and Kleiter, G. D. [2010] “The Conditional in Mental Probability Logic,” in M. Oaksford and N. Chater (eds.) Cognition and Conditionals, Oxford: Oxford University Press, pp. 153–173. Politzer, G. and Bonnefon, J.-F. [2006] “Two Varieties of Conditionals and Two Kinds of Defeaters Help Reveal Two Fundamental Types of Reasoning,” Mind and Language 21:484–503. Politzer, G., Over, D. E., and Baratgin, J. [2010] “Betting on Conditionals,” Thinking and Reasoning 16:172–197. Ramsey, F. P. [1929] “General Propositions and Causality,” in his Philosophical Papers, edited by D. H. Mellor, Cambridge: Cambridge University Press, 1990, pp. 145–163. Rosenkrantz, R.D. [1992] “The Justification of Induction,” Philosophy of Science 59:527–539. Stalnaker, R. [1968] “A Theory of Conditionals,” in N. Rescher (ed.) Studies in Logical Theory, Oxford: Blackwell, pp. 98–112. [1970] “Probability and Conditionals,” Philosophy of Science 37:64–80. Teller, P. [1973] “Conditionalization and Observation,” Synthese 26:218–258. Tversky, A. and Kahneman, D. [1983] “Extension versus Intuitive Reasoning: The Conjunction Fallacy in Probability Judgment,” Psychological Review 90:293–315. van Fraassen, B. C. [1976] “Probabilities of Conditionals,” in W. L. Harper and C. A. Hooker (eds.) Foundations of Probability Theory, Statistical Inference, and Statistical Theories of Science (Vol. I), Dordrecht: Reidel, pp. 261–301. Verbrugge, S., Dieussaert, K., Schaeken, W., Smessaert, H., and Van Belle, W. [2007] “Pronounced Inferences: A Study on Inferential Conditionals,” Thinking and Reasoning 13:105–133. Weidenfeld, A., Oberauer, K. and Horning R. [2005] “Causal and Non-Causal Conditionals: An Integrated Model of Interpretation and Reasoning’, Quarterly Journal of Experimental Psychology 58:1479–1513.

2. A Defense of Dogmatism Jeremy Fantl

To defend one’s mind against these follies a man must have an adamantine faith, so that, even if he is not able to detect the precise trick by which the illusion is produced, he at any rate retains his conviction that the whole thing is a lie and an impossibility.

Lucian1

1.

F L AT D I S M I S S A L

Sometimes you know that evidence is misleading—that the conclusion of an argument2 is false—without knowing what’s wrong with the evidence or argument. You know that the illusionist did not burn to death even if you don’t know how the trick was done. You know that things move even if you don’t know how Zeno’s arguments fail. In some such cases, you could figure out how the evidence or argument is misleading. But you need not. It is “legitimate” to “flatly dismiss” the evidence or argument in the following sense: you know that the evidence or argument is misleading without knowing how it is. I expect that this latter point is uncontroversial and so I assume it without argument.3 Can these uncontroversial cases teach us anything about more controversial ones? Richard Dawkins (2006) thinks so: when it comes to magic tricks, [t]here is a perfectly good explanation. It is just that I am too naïve, or too unobservant, or too unimaginative, to think of it. That is the proper response to a conjuring trick. It is also the proper response to a biological phenomenon that appears to be irreducibly complex. Those people who leap from personal bafflement at a natural phenomenon straight to a hasty invocation of the supernatural are no better than the fools who see a conjuror bending a spoon and leap to the conclusion that it is “paranormal.” (129)

In the history of philosophy, the prize for most controversial flat dismissal goes to David Hume. In Section X of the Enquiry, Hume argues that he does 1 This passage is taken from Lucian’s Alexander; or, the False Prophet, paragraph 17, and the translation is Joseph Jastrow’s (1902: 49), though he doesn’t cite the source. Though relatively pithy, Jastrow’s translation is not particularly faithful to the text. For a more respectful translation, see A. M. Harmon’s. 2 I’ll be construing “argument” broadly enough to include evidence. So, flat dismissal of evidence for p will, in my sense of “argument,” count as flat dismissal of an argument for p. 3 For an argument for a kind of dogmatism regarding propositions like the ones subject to Zeno’s critique, see Kelly 2011.

A Defense of Dogmatism

35

not need to consider the details of particular instances of testimony that a miracle occurred to know that the testimony is misleading. It is far less likely that the miracle occurred than that the testifier is lying or deluded in some unknown way. As Hume (1748) says, favorably commenting on the Cardinal de Retz’s skepticism regarding a purported miracle in Saragossa, “He considered justly, that it was not requisite, in order to reject a fact of this nature, to be able accurately to disprove the testimony, and to trace its falsehood, through all the circumstances of knavery and credulity which produced it.” (108) A close second to Hume’s flat dismissal of testimony for miracles is G. E. Moore’s (1959) flat dismissal of skeptical arguments. After outlining the four assumptions Russell needs to mount his skeptical argument, Moore says, It seems to me more certain that I do know that this is a pencil and that you are conscious, than that any single one of these four assumptions is true, let alone all four . . . Nay more: I do not think it is rational to be as certain of any one of these four propositions, as of the proposition that I do know that this is a pencil. (226)

He concludes that he can dismiss the skeptical argument without a diagnosis in the same way that Hume thinks he can dismiss testimony for miracles without a diagnosis. Hume and Moore, then, can be taken as relying on the same general principle: when evaluating whether to accept a piece of evidence or argument for a proposition, the plausibility or probability of the claim that the evidence or argument is in some way misleading should be weighed against the plausibility or probability of the proposition itself. The targets of this general principle need not be limited to testimony for miracles or to skeptical arguments. They can include arguments regarding evolution, psychic phenomena,4 convoluted conspiracy theories, repugnant moral positions, the existence of God, and whether the Holocaust occurred. In all these domains, there is actual and broad disagreement about their central propositions. Intuitively, we cannot know whether these controversial propositions are true unless we figure out how the many arguments on the other side go wrong. If knowledge requires justified belief, this view is entailed by Scott Aikin’s (2011: 17) “dialectical requirement for justification”: “If S is justified in holding that p, then S can successfully address the standing cases for not-p and against p,” where “successfully address” does not just mean that S’s evidence that p defeats the evidence against p; it means that S can say where the case for not-p goes wrong. Likewise, with respect to controversial moral matters, Sarah McGrath (2008) argues that in a battle between two non-fallacious arguments for opposite sides, neither side can know that the conclusion of their argument is true: of course, the premises of my argument seem more compelling to me than the premises of Alice’s argument; but by the same token, the premises of Alice’s argument seem more compelling to her than the premises of my argument . . . Can

4

See, e.g. Price 1955.

36

Jeremy Fantl

I break the symmetry, then, by assuring myself that the reasons that I have are more compelling than hers? This seems no better than simply privileging my judgment about a given shade of green over Alice’s contrary judgment. (103)

Flat dismissal seems least legitimate when it comes from the opposition. It is illegitimate for my opponent (say, you) to ignore my argument or, possibly worse, hear me out and then dismiss my argument without finding a flaw. Suppose that the best you can do with my argument is tell me that, while all the steps are compelling, something just seems a bit fishy. I press you further: “Can you help me out? What exactly seems fishy?” You make some suggestions—this premise, that use of a concept—and, suppose, I answer them, to your satisfaction. But, you continue: “I guess I can’t put my finger on what’s wrong. I just don’t buy it.” This response is wholly unsatisfying, especially when it comes to the controversial issues about which I feel strongest. And if you can’t do it to me, why would I get to do it to you? That’s in theory. In practice I feel much more comfortable flatly dismissing your arguments than having you flatly dismiss my arguments. While I usually feel that, with a little time, I’d be able to figure out how your arguments go wrong or, at least, have some idea where the weak spots are, I wouldn’t be too concerned if I didn’t. There is a certain amount of guilt involved—in this as in all things—when I am reflective about my practices. But that doesn’t change the fact that the practice is there. To the extent that others are like me, we all of us act as if flat dismissal is often legitimate. The good news is that at least some of us don’t have to feel so guilty about flatly dismissing the opposition’s arguments. Even when we lack special training, some of us know controversial propositions. For those of us who do, flat dismissal of arguments against such propositions is often legitimate. Briefly, the reason is this: for any controversial proposition you know, there could easily be an apparently sound argument that the proposition is false. If that is the case, it will often not make a difference to whether you know if you are exposed to an apparently sound counterargument. Therefore, you will often retain knowledge even if you are exposed to an apparently sound counterargument.

2.

F O R WA R D - L O O K I N G D O G M AT I S M

When is flat dismissal legitimate and when is it not? One simple answer is this: flat dismissal of an argument is legitimate only when hardly anyone (reasonable5 ) would be convinced by the argument. Hardly anyone—and no one reasonable—is convinced by magic shows or Zeno’s arguments. So flat dismissal can be legitimate in those cases. The same can’t be said for many arguments about controversial philosophical, scientific, religious, and moral propositions. 5

I’ll usually keep this requirement implicit.

A Defense of Dogmatism

37

This simple answer is incorrect. Suppose that you are a murderer and that you find out that someone else’s fingerprints are on the murder weapon. This surprises you. You have no idea how those fingerprints came to be there. In this case, you retain knowledge that you committed the murder without having any reasonable idea what’s wrong with the argument to the contrary. It is legitimate for you to flatly dismiss the argument even though lots of people would be convinced by the argument: they would think you’re not guilty. The difficulty with this case is that those who would be convinced of your innocence do not have all the evidence you do that you are guilty. Were they aware of your evidence—your memory of committing the murder—they would not be convinced by the argument that you’re innocent. This can only go so far, of course. Let enough evidence come in that you’re innocent and even you could be convinced to disregard your apparent memories. Nonetheless, the mere fact that lots of people would be convinced by an argument isn’t sufficient for your flat dismissal of the argument to be illegitimate. You might have access to a decisive argument they don’t. A more promising line is that flat dismissal of an argument is illegitimate if lots of people would be convinced by the argument, and would remain convinced if they had your evidence to the contrary.6 Call such an argument a “relevant counterargument.” How many people must a counterargument be convincing to in order to be a relevant counterargument; how many people is a “lot”? More than one and allowably fewer than all. Fortunately, being more precise isn’t necessary. That’s because the domains of interest here are paradigmatically controversial domains—domains in which the number of people on both sides isn’t in the gray areas of “lots.” In these paradigmatically controversial domains, on each side there are lots of people who would be convinced by similar arguments so I’ll grant that, for every side on these issues, there are relevant counterarguments. Nonetheless, flat dismissal of these arguments is often legitimate. Commitment to the legitimacy of flat dismissal is a kind of dogmatism, though not the kind defended, for example, by James Pryor (2000), according to which you can know that a proposition is true even if your belief is not justified by any non-question-begging reasons. Pryor’s dogmatism is “backward-looking” in the sense that it comments on the sources of belief. Backward-looking dogmatism is suggested in Moore’s proof of an external world, because Moore takes himself to know the premises in his proof (e.g. that here is a hand) though he cannot prove those premises. But Pryor notes Moore’s additional “idea”—“that the proposition that there is a hand, though

6 Not included in the evidence you would share with your opponents are any feelings of plausibility you might have when considering your arguments. It is plausible enough that if lots of people are convinced by an argument and would remain convinced even if they were aware of all your arguments then it is illegitimate to flatly dismiss their argument even if they don’t share the feeling of plausibility you have regarding your arguments.

38

Jeremy Fantl

only defeasibly justified, is more certain than any of the premises which might be used in a skeptical argument that he does not know that there is a hand” (518). The dogmatism suggested by this second idea is “forwardlooking” in the sense that it comments on what might require belief-revision. Of the two, forward-looking dogmatism seems closer to the ordinary use of the term. As such, I’ll refer to forward-looking dogmatism simply as “dogmatism”: (Dogmatism) It is often legitimate to flatly dismiss relevant counterarguments.

You legitimately flatly dismiss a counterargument to a belief when your knowledge7 survives flat dismissal of the counterargument.8 To dismiss a counterargument to a belief is to maintain that belief even when faced with an argument that the belief is false. To flatly dismiss a counterargument is to dismiss it in such a way that it is irrelevant to the dismissal whether any specific failures are noticed in the argument. According to dogmatism, relevant counterarguments are “often” legitimately flatly dismissed in this sense. How often is “often”? This hides two questions. First, how many beliefs are such that it is legitimate to flatly dismiss relevant counterarguments to them? Second, what kinds of relevant counterargument are such that it is legitimate to flatly dismiss them? As I argue in section 3, the answer to the first question is this: as many beliefs about controversial matters as are known, but not “just barely” known. The answer to the second question is more difficult. Dogmatism does not entail that it is ever legitimate to flatly dismiss all possible or relevant counterarguments. I tentatively explore the second question in the conclusion. Arguments can be flatly dismissed subject to varying levels of scrutiny. You can dismiss arguments without having heard them out, having heard them out but without understanding them, having understood them but without taking the time to evaluate them, and having taken the time to evaluate them but without exposing flaws in them. Perhaps it’s easier to legitimately flatly dismiss arguments you haven’t taken the time to understand or critique than it is to legitimately flatly dismiss arguments you have taken time with but still

7 Alternatively, “legitimacy” could be defined, not in terms of knowledge, but in terms of justified belief: you legitimately flatly dismiss a counterargument to a belief just in case your belief is justified even though you flatly dismiss the counterargument. Construed in this way, the dogmatism that results is weaker than that argued for here—though still true and supported by nearly identical arguments to those offered below. What’s crucial for establishing the stronger form is the claim (below) that if the mere fact of disagreement allows for justified belief, then it allows for knowledge as well. To the extent the argument for this claim is unsuccessful, this paper will succeed only in demonstrating the weaker form of dogmatism. However, even the weaker form of dogmatism might be sufficiently counterintuitive to make its establishment interesting on its own. 8 My thanks to an anonymous referee for suggesting this way of putting the definition.

A Defense of Dogmatism

39

can’t flaw. If you haven’t spent any time with an argument, there’s a much greater chance that engagement would reveal a flaw than if you have spent time with the argument and still can’t flaw it.9 I will argue that flat dismissal of relevant counterarguments is often legitimate even if you have investigated the counterarguments, understood them, devoted time to figuring out what’s wrong with them, and still find the premises and moves in the argument compelling enough that you can’t locate a flaw. Say that, when you are in this position with respect to a relevant counterargument, you are “stymied” with respect to that counterargument. The kind of dogmatism that licenses flat dismissal when you are stymied, call “stymied dogmatism”: (Stymied Dogmatism) It is often legitimate, after understanding the details of some relevant counterargument, to dismiss the counterargument, even if, after reasonably lengthy attempts, a flaw has failed to be exposed.

Often, when we’re evaluating a counterargument, we are pretty sure that we’ve spotted the weak part—the part we’d press on. Or, if we’re not pretty sure, we have hunches about which steps or premises are the problematic ones: for example, when presented with arguments that there is no motion, we might have a hunch that the problem lies with the argument’s use of the concept of infinity—always a troublesome concept. Even Moore, who didn’t much care where Russell’s argument for skepticism broke down, thought that he’d isolated at least one false premise: premise 4—that what is “based on an analogical or inductive argument is never certain knowledge” (225 ff).10 As Moore says, he is “inclined to think” that this premise is false (226). Does a hunch that some premise or step is problematic constitute “exposure” of the argument’s flaw? Is it sufficient to be “inclined to think” that some premise is false? Must it be pretty sure that a step is fallacious? Must we have knowledge of what the flaw is in order to legitimately dismiss the argument? We need not to take a stand on what precisely it means to “expose a flaw” in an argument. Instead, we can say that stymied dogmatism itself comes in stronger or weaker grades depending on how “exposed” is cashed out. The degree to which a flaw is exposed can vary along two dimensions. First, it can vary according to how many and what kinds of people the exposure is convincing to. After reasonably lengthy attempts, you may not have managed to 9 There are a number of reasons why you might take time to engage with an argument that you have already flatly dismissed. You might have purely intellectual interest in how the argument fails, you might have rhetorical interest in convincing some interlocutor, you might have made a bet on whether you can flaw the argument, etc. None of these reasons for investigating an argument amount to a concession that the argument has a significant potential for being sound. Lots of people devote a significant amount of time to Zeno’s arguments without the slightest worry that the conclusion of the arguments is true. 10 My thanks to an anonymous referee for emphasizing this point.

40

Jeremy Fantl

come up with reasons for the counterargument’s failure that would convince anyone offering the counterargument, even if those reasons do convince you and would convince those on the fence. Or, you may have failed to come up with reasons that would convince anyone on the fence, even if those reasons do satisfy you. Or, you may have failed to come up with reasons that convince even you that this is the crucial flaw in the counterargument. The second dimension that exposure of a flaw can vary over is specificity. After reasonably lengthy attempts, you may have failed to show even to your satisfaction exactly what the flaw is in the counterargument. But you may have a pretty good idea what, roughly, the flaw consists in. Or, lacking a pretty good idea, you may have an “inkling” what the flaw consists in. Or you may lack any clue at all what it consists in: every step seems really compelling.11 Even on this extreme end of the continuum, though, you may have found yourself dismissing counterarguments. The claim that it is often legitimate to do so is super-stymied dogmatism: (Super-Stymied Dogmatism) It is often legitimate, after understanding the details of some relevant counterargument, to dismiss the counterargument, even if, after reasonably lengthy attempts, there is no clue at all about where or what the flaw in the counterargument is.

With respect to some of your most strongly held beliefs you might not have had the opportunity to flatly dismiss such counterarguments. Perhaps you usually can find a hole. But this doesn’t mean that you’re not fully prepared to flatly dismiss such counterarguments if they ever come your way. For many of the issues you have strong positions on, you might not much care whether, when faced with a relevant counterargument, you end up stymied. You might happily maintain your belief even if you can’t find a flaw. This disposition is only legitimate if some significantly strong grade of stymied dogmatism is true. For, if stymied dogmatism is not true then, in having this disposition, you are disposed to believe in an illegitimate way. And, to misquote Clifford, it is not possible so to sever the disposition from the belief it suggests as to condemn the one without condemning the other. Finally, there is an even stronger grade of dogmatism according to which, not only is it sometimes legitimate to flatly dismiss many relevant counterarguments, but it is sometimes the case that there is no actual or possible counterargument that it would be illegitimate to flatly dismiss, even if known about and understood. According to this kind of dogmatism, some beliefs are immune to rational revision in the light of new evidence. We can label this kind of dogmatism anti-Quinean dogmatism:

11 What I’m calling “specificity” is really two further distinct dimensions: the nature of the flaw, and the certainty with which it is believed that the flaw is there. You can be certain, pretty sure, suspicious, or clueless that X is the exact flaw in the argument. Or you can be certain, pretty sure, suspicious, or clueless that something involving something like feature F is what the argument does wrong.

A Defense of Dogmatism

41

(Anti-Quinean Dogmatism) It is sometimes the case that it is legitimate to dismiss all actual and possible counterarguments, even were those arguments understood and even were a flaw not found.12

I discuss anti-Quinean dogmatism no further, except to make explicit that super-stymied dogmatism does not entail anti-Quinean dogmatism.

3.

T H E C A S E F O R S T Y M I E D D O G M AT I S M

I argue that at least a very strong grade of stymied dogmatism—if not superstymied dogmatism—is true. Suppose you often dismiss relevant counterarguments for which you can’t, even after reasonably lengthy attempts, produce more than some general and vaguely convincing idea of where the argument might break down. When it comes to those controversial beliefs for which you have substantial epistemic support—the controversial beliefs that count as knowledge—it is legitimate for you to dismiss these counterarguments nonetheless. After all, what’s the greater miracle—that your wellsupported belief is wrong or that some clever person has come up with an unsound argument in which you cannot find a hole? The latter happens all the time. Let’s make this Humean argument more precise. Suppose that you know some controversial proposition, p, and that p is a “lay” proposition—a proposition about whose subject matter you have no special training. Then there could easily be “apparently sound” arguments that not-p—arguments in which, even after understanding the details and even after reasonably lengthy attempts, you would not expose a flaw. After all, there are apparently sound arguments that nothing moves and that I am not bald. If there are apparently sound arguments for propositions as obviously false as the propositions that nothing moves and that I am not bald, then for any controversial proposition, p, there could easily be apparently sound arguments that not-p. Because p is controversial, there are smart people who disagree with you. These smart people would presumably be convinced by apparently sound arguments that not-p even were they aware of all of your arguments for p. That makes this apparently sound argument a relevant counterargument. Therefore, for any controversial lay proposition you know, there could easily be an apparently sound relevant counterargument. Call this the Principle of Modesty. As Tom Kelly (2005) puts a similar point In deciding how to respond to any argument which appears to be flawless, one is in effect in the position of performing an inference to the best explanation . . . If . . . the 12 This specification of anti-Quinean dogmatism should be tweaked to require that the positive epistemic support for the belief in question be held constant. It might be that it is legitimate to dismiss all actual and possible counterarguments to a belief, only given the kind and strength of evidence in favor of the belief. Perhaps this kind of dogmatism is true of some beliefs about simple and obvious conscious episodes (like, “I am in pain”). The untweaked version wouldn’t be.

42

Jeremy Fantl

better explanation of one’s failure is one’s own cognitive limitations, then one should remain unmoved in the face of the argument . . . Notice that, if this is dogmatism, there is a respect in which it is an unusually modest variety. For when one reasons in this way, one’s refusal to change one’s beliefs is due to the weight that one gives to one’s own cognitive limitations.

Whether there could easily be apparently sound arguments for not-p is an epistemic matter. If your epistemic position excludes the possibility that there are apparently sound arguments for not-p, then not only couldn’t there easily be such arguments, there couldn’t be such arguments, period. If your epistemic position is consistent with there being such arguments, but makes the existence of such arguments extremely unlikely, then though there could be such arguments, there couldn’t easily be such arguments. How epistemically likely must it be that there are such arguments for it to be the case that there could easily be such arguments? Certainly if your epistemic position leans even slightly toward the existence of such arguments—if your rational credence even somewhat favors the existence of such arguments—then there could easily be such arguments. But even if your epistemic position leans against the existence of such arguments, but allows some significant chance that there are such arguments, then there still could easily be such arguments. On this epistemic construal of “could easily,” whether there could easily be an apparently sound relevant counterargument might depend on how versed you are in the literature. If you have made the Holocaust or evolution your life’s work and have seen pretty much all the contrary arguments that are likely to be on offer and feel justly confident you have exposed all their flaws, then it might not be the case that there could easily be apparently sound relevant counterarguments. But this is not so for most people’s controversial beliefs, because most people are not extremely well versed in the literature about most of the controversial propositions they believe. p, though, is a controversial lay proposition. So, there could easily be an apparently sound relevant argument that not-p. Does the fact that there could easily be an apparently sound argument that not-p mean that you don’t know that p? It had better not. Because if it does, then many people fail to know that things move simply because they can’t figure out how Zeno’s arguments go wrong. For many people, there is an apparently sound argument that nothing moves. If knowledge can survive an apparently sound counterargument, then knowledge can survive it being the case that there could easily be an apparently sound counterargument. So, the mere fact that there could easily be an apparently sound argument that some controversial proposition is true does not by itself mean that you don’t know that the proposition is false: your knowing p is consistent with it being the case that there could easily be an apparently sound argument that not-p. Nothing in the nature of beliefs about controversial propositions entails that—controversiality aside—it is impossible to have a very strong epistemic position with respect to those propositions—a strength of epistemic

A Defense of Dogmatism

43

position that in many other cases would suffice for knowledge.13 Controversial propositions—again, controversiality aside—can enjoy very high levels of evidential support, be true, be unGettiered, etc. Some of the propositions in question are theoretical or abstract, but that alone doesn’t preclude having a very strong epistemic position with respect to those propositions. We all know a range of theoretical and abstract propositions. In any case, many controversial propositions are not theoretical. They are about, say, the age of the Earth, whether any human has ever manipulated a spoon with his or her mind, and the number of people killed in the Holocaust. Nor does anything in the nature of beliefs about controversial propositions entail that—controversiality aside—you are prevented from having an extremely strong epistemic position just because you are a layperson. Laypeople are extremely strongly epistemically positioned with respect to many propositions about which they lack the relevant expertise. You have a strong enough epistemic position, for instance, to know that the Earth revolves around the Sun, that the green color in plants is the result of chlorophyll, and that dead animals in rivers increase the likelihood of disease transmission downstream.14 Controversiality aside, then, you can be very strongly positioned epistemically with respect to the controversial lay proposition, p. But, of course, we can’t just put controversiality aside. The controversiality of p provides two kinds of evidence that might outweigh any strong positive epistemic support you have for p. First there is the mere fact that there is disagreement about whether p. Second, there is the fact that there could easily be an apparently sound argument that not-p. I consider the first kind of evidence below. With regard to the second, is the fact that there could easily be an apparently sound argument that not-p guaranteed to be sufficient to outweigh your otherwise very strong epistemic position with respect to p? When it comes to controversial propositions, that there is an apparently sound argument that not-p is not in general very strong evidence for not-p. There are often apparently sound arguments for both sides of controversial propositions. But arguments on at most one of the sides have true conclusions. So, a very high proportion of apparently sound arguments about controversial propositions have false conclusions. If that there is an apparently sound argument for not-p is not in general very strong evidence for not-p, then that there merely could easily be an apparently sound argument for not-p is even worse evidence for not-p. The mere fact that 13 Here and throughout the paper I do not assume that there is any fixed level of evidential support such that when a proposition enjoys that level of support for a believer, the proposition is true, and the believer’s belief is based in the right way, the proposition is known. There are, as always, other Gettierish features that can interfere. But I do assume that in any particular case there is some minimal (perhaps vague) degree of evidential support that is required for knowledge, so that falling below this level can be the reason why a proposition fails to be known. 14 Thanks to Elizabeth Brake for suggesting these examples.

44

Jeremy Fantl

you have some evidence for not-p does not preclude your knowing that p.15 What then should we say about your total epistemic position with respect to p when, other than the ease of there being an apparently sound argument that not-p, you are in an extremely strong epistemic position with respect to p? At the very least, this: it is not guaranteed that what is otherwise an extremely strong epistemic position with respect to p is inevitably reduced to a level insufficient for knowledge. On the contrary, often your strong epistemic position with respect to p will swamp the quite weak evidence for not-p. Wouldn’t the same argument that shows that the fact that there could easily be an apparently sound argument that not-p is not very strong evidence for not-p also show that the fact that you have seemingly strong evidence for p is not very strong evidence for p? It might in some cases. But, of course, your evidence for p is not just that you have seemingly strong evidence for p. You also have the first-order evidence itself. And, the point here is that this first order evidence, if knowledge-level strong, will often swamp the weak evidence for not p provided by the fact that there could easily be an apparently sound argument for not-p. Knowing p is consistent with it being the case that there could easily be an apparently sound argument that not-p. Now suppose you become familiar with an apparently sound counterargument—you “find that a counterargument is apparently sound” in the sense that you come to understand the details of the counterargument but, after reasonably lengthy attempts, you fail to expose a flaw. With respect to many such arguments, this shouldn’t be very exciting news to you: there’s nothing too surprising about finding out that something that could easily be, in fact is. It’s not that finding out that there is such a counterargument can never make a difference to epistemic status. As David Christensen (2007) says, “it makes a difference whether my friend actually does disagree with me, or whether I just know she could have disagreed with me” (208). In general, evidence that could easily exist can be epistemically more forceful when you find out it actually exists. But at least in a lot of cases, the fact that it could easily exist means that it won’t make enough of a difference, when it

15 It’s not clear what to say about cases in which what would otherwise count as evidence for not-p is swamped by evidence for p so that you retain knowledge that p. Do we call the evidence for not-p merely prima facie—but not ultima facie evidence for not-p? Or do we call the evidence for not-p ultima facie evidence for not-p, but ultima facie evidence that is outweighed by the ultima facie evidence for p? Example: your student tells you he didn’t plagiarize—that his paper was written completely independently of any web research. That’s evidence that the paper isn’t plagiarized. But you have found a word-for-word version of the paper online. In this case, you know that the paper is plagiarized. Do we want to say that you have even some evidence—the student’s denial—that the paper is not plagiarized? If so, then we’ll also want to say that the evidence is so strongly outweighed by the evidence of plagiarism that you know that the paper is plagiarized. If not, then we’ll want to say that the so-called “evidence” is only prima facie evidence that the paper isn’t plagiarized. This issue, though, should not trouble us. Whatever stand one wants to take on this issue is consistent with the soundness of the argument in the body of this section of the paper, as long as the evidence that there could easily be an apparently sound counterargument for not-p is taken to be prima facie evidence that not-p.

A Defense of Dogmatism

45

is actually discovered, to destroy knowledge if that knowledge existed in the first place. Of course, you did not expect this argument. If the argument presents an entirely new way of thinking about an issue that makes it just obvious that p is false or if it presents evidence for not-p that is of a more convincing kind than the evidence you had for p, then perhaps being made aware of the specific content of the argument can make a difference to whether you know that p. But this is often not what happens when you become familiar with novel, apparently sound counterarguments. It is not, for example, what happens when you become familiar with apparently sound arguments that there is no motion. It’s not what happens at magic shows. That the illusionist is not really burning to death and that there is motion, while true, are not controversial. But the controversiality of the matter shouldn’t make a difference. As long as you often can be epistemically quite well situated with respect to controversial truths—as long as you can often know them—the weighing should be favorable to maintaining belief that p in the face of apparently sound arguments that not-p, even if those arguments are relevant counterarguments. Put another way, an epistemic position strong enough to allow you knowledge despite the fact that there could easily be an apparently sound relevant counterargument will often be strong enough to allow you knowledge despite the fact that you have become familiar with an apparently sound relevant counterargument and found it apparently sound. In slogan form: unsurprising arguments are often impotent. This principle is not in conflict with the standard solution to the so-called “dogmatism paradox” suggested by, for example, Gilbert Harman (1973), Carl Ginet (1980), Roy Sorensen (1988), and Earl Conee (2001). According to the dogmatism paradox (usually, following Harman, attributed to Saul Kripke), if you know, say, that Tom stole the book, then you know that any evidence (argument) that Tom did not steal the book is misleading. If skepticism is off the table, you can know that Tom stole the book. Therefore, you can know ahead of time that any evidence that Tom did not steal the book is misleading. What happens if, knowing this, you subsequently get evidence that Tom did not steal the book? Are you thereby justified in ignoring that evidence—in flatly dismissing it? Seemingly not. You don’t get license to ignore all future evidence that Tom did not steal the book just by now having enough evidence to know that Tom did steal it. On the standard solution, you can indeed know, before acquiring evidence that Tom did not steal the book, that the evidence is misleading, even if you are not thereby justified in dismissing that evidence when it comes along. That’s because, when the evidence comes along, you lose knowledge that Tom stole the book, and so lose knowledge that the evidence is misleading. As Harman puts the point, Since I now know that Tom stole the book, I now know that any evidence that appears to indicate something else is misleading. That does not warrant me in simply

46

Jeremy Fantl

disregarding any further evidence, since getting that further evidence can change what I know. In particular, after I get such further evidence I may no longer know that it is misleading. For having the new evidence can make it true that I no longer know that Tom stole the book. (149)

When it comes to the dogmatism paradox, there is no presumption that there could easily be evidence that Tom did not steal the book. There is only a presumption that such evidence is possible. More importantly, there is no presumption that there could easily be a relevant counterargument to the claim that Tom stole the book: there is no presumption that there could easily be evidence that would convince lots of people that Tom did not steal the book and would convince those people even if they had your evidence that Tom did steal the book. Once that presumption is added in, it’s much harder to see how knowledge could standardly be lost, if it is there in the first place.16 Suppose that you are Tom, that you stole the book, and that you know you did. Even so, there is evidence that, were it to become available, would convince even you that you didn’t steal the book—bizarre evidence that suggests that you are prone to hallucinations, that you were videotaped thousands of miles away at the time of the theft, etc. According to the standard solution, you can know ahead of time that such evidence is misleading even though you won’t know when it comes in that it is misleading because you won’t know, when it comes in, that you stole the book. But, when you know that you stole the book, there couldn’t easily be such evidence. Knowledge may very well be destroyed by discovering surprising evidence. This is consistent with it being impossible (or just difficult) for knowledge to be destroyed by the discovery of unsurprising evidence. It’s this latter impossibility (or, at least, difficulty) that is required here: if you should already suspect, before exposure, that the evidence is there, then exposure to the evidence will often not change whether you know. One final proviso: we need to allow for the possibility both that there can be close calls when it comes to knowledge and that exposure to an apparently sound relevant counterargument, even if it leaves knowledge intact, always reduces your strength of epistemic position.17 If both of these things are the case, then if you “just barely” know a controversial proposition, exposure to any apparently sound relevant counterargument will reduce your strength of epistemic position below the level required for knowledge. Therefore, it’s not the case that whenever you know a controversial proposition, your

16 Is it possible for you to know that Tom stole the book once it is no longer surprising that there is evidence that Tom did not steal the book? Yes, if what’s unsurprising is that there is prima facie evidence that Tom did not steal the book. Knowledge is compatible with having prima facie evidence to the contrary, so knowledge is compatible with it being unsurprising that there is prima facie evidence to the contrary. See note 15. 17 Though I want to allow for its possibility, I do not endorse this latter claim. As I note in the conclusion, it is plausible that one’s first confrontation with Zeno’s arguments reduces the strength of epistemic position not one iota for the proposition that there is motion.

A Defense of Dogmatism

47

knowledge will often survive exposure to an apparently sound relevant counterargument. But if you can know lots of controversial propositions, much of this knowledge won’t be of the close-call variety. Therefore, it will often be the case that, if there could easily be an apparently sound relevant counterargument, finding the argument apparently sound won’t weaken your epistemic position below the level required for knowledge. Therefore, we have two premises that can figure in an argument for a strong grade of stymied dogmatism. The first follows from the Principle of Modesty: 1. If you know some controversial lay proposition, then there could easily be an apparently sound relevant counterargument. The second is a version of the principle that unsurprising arguments are often impotent: 2. Often, if there could easily be an apparently sound relevant counterargument, then whether you find that some relevant counterargument is apparently sound won’t make a difference to whether you know some controversial proposition. If many of the cases that make 2 true are also cases in which you have knowledge, it follows from 1 and 2 that, often, if you know some controversial lay proposition, it won’t make a difference to whether you know that proposition if you were to find a relevant counterargument apparently sound.18 And it follows from this that 3. Often, if you know some controversial lay proposition, you would know that proposition even if you were to find a relevant counterargument apparently sound. If the antecedent of this conclusion is necessarily false—if it is impossible to know controversial lay propositions—then 3 is trivially true. Is it possible to know controversial lay propositions? Is the mere fact of significant disagreement about p guaranteed to undercut whatever positive epistemic support you have for p? Some of those who work on the epistemology of disagreement might be interpreted as arguing that you can’t know anything when there is significant disagreement. As Richard Feldman (2007) argues, I see that another person, every bit as sensible and serious as I, has an opposing reaction. Perhaps this person has some bit of evidence that cannot be shared or perhaps he takes the evidence differently than I do. It’s difficult to know everything about his mental life and thus difficult to tell exactly why he believes as he does. One of us must be making some kind of mistake or failing to see some truth. But I have no basis for thinking that the one making the mistake is him rather than me. And the same is true of him. And in that case, the right thing for both of us to do is to suspend judgment. (229) 18

This assumes, of course, that belief is preserved after exposure to the argument.

48

Jeremy Fantl

Others who take similar lines include David Christensen (2007) and Adam Elga (2007). Following Lackey (2010), call those philosophers who think confrontation with significant disagreement standardly requires significant revision of belief “conformists.” Assuming that the revision should be significant enough, conformists will also think that confrontation with significant disagreement standardly entails loss of knowledge.19 If conformism is committed to the claim that you cannot reasonably believe—and hence cannot know—propositions about which there is significant disagreement, this seems like a difficulty for conformism. At least, so say the conformists themselves. Elga calls it “the problem of spinelessness”— “the problem that a [conformist] view on how to respond to disagreement will recommend suspension of judgment on virtually all controversial issues” (492). Christensen (2011), though not emphasizing controversiality, likewise considers it a significant worry for conformist views if they always require giving up extremely well-supported beliefs whenever faced with disagreement: The first sort of hard cases are ones where an agent begins with extremely high rational confidence in her belief. In various such cases, it seems wrong to hold that she should revise her belief much at all, even if the agent’s friend disagrees sharply, and even if, before discovering the disagreement, she would have considered the friend her epistemic peer on the sort of issue in question.

One way for conformists to respond to such cases is to bite the bullet and say that in all such cases, conformism requires significant belief-revision. But this is not what actual conformists do. Rather, what they do is find ways to accommodate within their view the result that you can reasonably maintain strong confidence even when faced with disagreement: contrary to initial impressions—the equal weight view does not require one to suspend judgment on everything controversial. (Elga 494) we can have reasonable beliefs (but perhaps not knowledge) about some political, scientific, philosophical, or religious matters, even if disagreement sometimes undermines our justification for those beliefs. (Feldman 2006: 222) The personal information I have about myself . . . provides a perfectly reasonable basis for continuing to think that our shares of the bill are much more likely to be $43 than $45, despite discovering the disagreement of my heretofore-equally-reliable friend. (Christensen, 2011)

19 We shouldn’t be too quick to saddle any of these authors—Feldman included—with the view that knowledge is impossible in the face of disagreement. In addition to the reasons offered in the body of the text, below, some of the authors (e.g. Christensen and Elga) require merely that you move your credence toward the credence of the disagreeing interlocutor. This movement need not result in suspension of belief if, for example as Christensen points out, “my friend has a degree of confidence barely sufficient (given the context) for rational belief that not-P, but that I have a degree of confidence much greater than that required for believing P” (214).

A Defense of Dogmatism

49

Allowing for reasonable belief in the face of disagreement is still short of allowing for knowledge in the face of disagreement (and Feldman’s parenthetical explicitly separates the issues). But if conformists can allow reasonable or justified belief in controversial propositions, there is no principled reason why they can’t also save knowledge. After all, if controversial propositions can be believed with the strength of justification sufficient for knowledge— knowledge-level justification—then it’s perfectly possible that those propositions satisfy all the other conditions on knowledge: they can be true and unGettiered. So the only way to insist that controversial propositions can be justifiably believed but not known would be to insist that controversial propositions can be justifiably believed, but not knowledge-level justifiably believed. Why think that? The degree to which a belief is justified is a function of the positive support the belief enjoys and the strength of the countervailing evidence. Again, there is no principled limit on the degree of positive support you might have for a controversial proposition. And if conformists can make plausible the position that reasonable belief can be maintained in the face of significant disagreement, that means they have found a way to minimize the strength of the countervailing evidence—the evidence, that is, that there are seemingly reasonable people who disagree. If they have found such a way, then depending on the situation, there is no principled bar to that countervailing evidence being weakened to the extent that the belief is knowledge-level justified and, hence, potentially known. Can conformists allow reasonable belief to be maintained in the face of significant disagreement? They think they can. One way to allow this is to allow you to fail to regard the disagreeing party as an epistemic peer. Elga, for example, maintains that in “messy, real-world cases,” one’s reasoning about the disputed issue is tangled up with one’s reasoning about many other matters . . . As a result, in real-world cases one tends not to count one’s dissenting associates—however smart and well-informed—as epistemic peers. (492)

A different strategy is suggested by the conformist Tomas Bogardus (2009). According to Bogardus, in the problematic cases, you can “just see” that the disagreeing party’s view is wrong and thereby acquire evidence that the disagreeing party lacks: In philosophy-speak, we might say Smith comes to have knowledge from direct acquaintance in the problematic cases. . . A relevant piece of evidence is intellectually obvious to Smith; she has unmediated cognitive access to the truth of a pertinent proposition. Her knowledge does not rely on any report, indication, or representation. (331)

Christensen, alternatively, emphasizes the possibility of knowing that your belief-forming methods were not unreliable in various ways that, for all you know, plague the disagreeing party. He licenses the following reasoning: The belief my friend announced was incompatible with the one at which I arrived. This is strong evidence that one of us did not arrive at his or her belief in a highly

50

Jeremy Fantl

reliable way, or that one of us is not sincerely announcing his or her belief. I can eliminate (via personal information) many of the ways that I could have failed to use a reliable method, as well as the possibility that my announcement was not sincere. But I cannot eliminate analogous possibilities for my friend. So it’s likely that she did not sincerely announce a belief that was formed by a highly reliable method.

What are the ways one might fail to have used a reliable method? Christensen mentions exhaustion, lack of care, drugs, insincerity, and psychotic attacks. These are all relatively temporary interferences with reliability that will not realistically affect large numbers of people who disagree with you. But we might add to the list racism, wish-fulfillment, Hume’s “passion of surprize and wonder,” iconoclastic motivations, sexual desire, and other psychological pressures to form beliefs that do not track the truth. To the extent that when your belief is not formed on the basis of methods influenced by such factors, it will be more accessible to you that it isn’t than that the ill-formed beliefs of others aren’t influenced by such factors, Christensen’s strategy will allow you to believe reasonably in the face of disagreement from one you would otherwise regard as an epistemic peer. Finally, both Christensen and Elga mention another way to legitimately disregard a disagreeing party. In order for significant belief-revision to be mandated, it’s not enough that your evidence be neutral on the subject of whether the disagreeing party is an epistemic peer. You must have strong positive evidence that the disagreeing party is an epistemic peer. Elga is more implicit on this issue. In the text immediately prior to the passage cited above, Elga says “in the clean cases one is in a position to count one’s associates as peers based on reasoning that is independent of the disputed issue. But in the messy real-world cases one is rarely in a position to do so” (492). Why, according to Elga, can you maintain reasonable belief in the face of messy real-world disagreement? It is because, in the messy real-world cases, you are so rarely in a position to conclude that a subject is an epistemic peer. Christensen is quite explicit about the requirement of positive evidence of epistemic peerage in his rejection of the following principle of belief revision: (A) Insofar as the dispute-independent evaluation fails to give me good reason for confidence that I’m better informed, or more likely to have reasoned from the evidence correctly, I must revise my belief in the direction of the other person’s. In place of (A), Christensen recommends (B): (B) Insofar as the dispute-independent evaluation gives me good reason to be confident that the other person is equally well-informed, and equally likely to have reasoned from the evidence correctly, I must revise my belief in the direction of the other person’s. When it comes to the controversial lay propositions with respect to which we are in the strongest epistemic position, it is not obvious that

A Defense of Dogmatism

51

dispute-independent evaluation would provide strong positive evidence that the disagreeing parties are our epistemic peers. At the very least, it doesn’t follow from the mere fact that a proposition is a controversial lay proposition that dispute-independent evaluation would provide such evidence. On the contrary, with respect to many of the relevant issues—Holocaust denial, for instance—it seems that there would be positive evidence that the disagreeing parties were not epistemic peers, especially if Christensen’s “personal information” were allowed to count as relevant evidence. Do these strategies save conformism from the objection that conformism doesn’t allow reasonable belief in controversial propositions? For the purpose of defending dogmatism no answer is required. The point is not that any of the conformists’ strategies for allowing reasonable strongly held belief in the face of apparent peer-disagreement will be successful. The point is that conformists feel the need to have the strategies in the first place: they do not shrug and insist that reasonable belief in controversial lay propositions is impossible. They think it’s important that conformism not have the consequence that reasonable belief in controversial propositions is impossible. It might be that reasonable belief in—and, thus, as I argue above, knowledge of—controversial propositions is possible but doesn’t happen very often. If it turns out that knowledge of controversial lay propositions is, while possible, rather uncommon, then the dogmatism defended in this paper would have to be weakened to simply the claim that it is possible to legitimately flatly dismiss relevant counterarguments. That should be a surprising result on its own. After all, if it’s even possible to legitimately flatly dismiss relevant counterarguments then there can be situations in which you are faced with an interlocutor convinced of some controversial proposition and offering you an argument which you cannot flaw. As in McGrath’s argument, above, in this case presumably the premises and steps in your argument seem more compelling to you than the premises and steps in your interlocutor’s argument. But, by the same token, the premises and steps in your interlocutor’s argument seem more compelling to her than the premises and steps in your argument. It should be surprising to learn, as argued for here, that this situation allows one of you to retain knowledge. Certainly you’d feel that your interlocutor would have offered an inadequate response to your argument if she said, “Well, I find every step in your argument compelling and I can’t see where it goes wrong. But I just don’t buy it.” You’d feel entitled to ask, “But what’s wrong with my argument?” If you’re entitled to ask this, it’s surprising to learn that your interlocutor isn’t. Still, though surprising, the claim that it’s possible for such situations to arise is not the strong dogmatism that is true if knowledge of controversial propositions is, not only possible, but common. A full defense of the stronger conclusion requires a case-by-case examination of controversial propositions—an examination in which it is shown that a significant proportion are potentially known. Such an examination is, to quote the seer, “beyond the scope of this paper.” But if knowledge of controversial lay propositions

52

Jeremy Fantl

is possible, there is nothing in principle standing in the way of it being the case that you often can know controversial lay propositions, because there is nothing in principle standing in the way of your having a very strong epistemic position with respect to controversial lay propositions. Therefore, let’s grant that you often can know controversial lay propositions. Assuming that there is substantial overlap in the cases that make 3 true and the cases in which you can know controversial lay propositions, then, we can derive: 4. Often, you would know some controversial lay proposition even if you were to find a relevant counterargument apparently sound. To find a relevant counterargument apparently sound is just to, after understanding the details and making reasonably lengthy attempts, fail to expose a flaw. A relevant counterargument is legitimately dismissed just in case knowledge is retained even when faced with the argument. Therefore, 4 entails that it is often legitimate to dismiss relevant counterarguments, even when you understand the details and, after reasonably lengthy attempts, fail to expose a flaw. That’s stymied dogmatism. None of this is to say that it won’t often be of interest to investigate relevant counterarguments—even the ones we can legitimately flatly dismiss. Nor is it to say that studying such arguments can’t teach us interesting things about the relevant objects of study nor even about how best to justify our own beliefs. Zeno’s arguments, after all, are of interest even though they pose no threat to the belief that there is motion. By figuring out where they go wrong, or even trying to, we might figure out legitimate and illegitimate uses of the concept of infinity, or we might refine our concept of motion, or we might simply find engagement with the arguments fun. We are not, of course, required either intellectually or morally to investigate Zeno’s arguments, either in order to retain belief in motion or for any other reason. Some people are interested in such things and others aren’t. The same might be said for legitimately flatly dismissed arguments in more controversial domains, though the situation is not so clear: there might be moral requirements to fully investigate arguments for false moral beliefs. But there is no requirement to fully investigate such arguments for the purpose of retaining knowledge and, hence, justified belief about the relevant propositions.

4.

CONCLUSION

Your knowledge can often survive confrontation with arguments whose flaws are only guessed at. This result is uncontroversial (I would hope), and manifests itself in Zeno’s arguments and magic shows. What’s more interesting, and what I have argued for, is that controversial matters can often survive confrontation with arguments that those who disagree with you would be convinced by, even if you can’t muster up more than a vague inkling of where

A Defense of Dogmatism

53

the flaws in those arguments might be. It is this result that I take to be the controversial and important one. It is not, however, always legitimate to flatly dismiss relevant counterarguments. Sometimes exposure to relevant counterarguments destroys knowledge. When does knowledge survive and when doesn’t it? Plausibly, whether it is legitimate to flatly dismiss relevant counterarguments depends at least in part on your strength of epistemic position with respect to the denials of the conclusions of those arguments. For example, when you have enough evidence for a proposition you can flatly dismiss at least a great range of arguments that the proposition is false. Whether you get to flatly dismiss any given relevant counterargument also depends on the nature of the relevant counterargument. Stymied dogmatism does not entail anti-Quinean dogmatism, nor does it even entail that, when it is legitimate to flatly dismiss some relevant counterargument, it’s legitimate to flatly dismiss every relevant counterargument. You might be especially willing to flatly dismiss arguments about important moral issues that involve highly abstract principles or complicated mathematical formulae. And your immediate perception of moving objects seems to completely trump whatever theoretical considerations might be brought in to show that there is no motion. If we’re looking for a general rule regarding what kinds of arguments it is legitimate to flatly dismiss, we might start by placing perceptual arguments in a safer place than theoretical ones. This is too hasty. First, a blanket prejudice against theoretical arguments smacks of George W. Bush’s repeated charge in the first 2000 US presidential debate that Al Gore’s calculations about Bush’s tax cut plan were just “Washington fuzzy math.” Second, it’s not easy to separate theoretical from perceptual components of arguments. Magic shows, for example, seem to involve a tension between theoretical arguments about physical possibility and perceptual considerations about, say, illusionists on fire—a tension in which the theoretical arguments win. But the perceptual considerations import theoretical ones: they aren’t just reports of sense data. And the theoretical ones are based on a lifetime of perceptual training. So, it’s not easy to come up with a list of features that arguments have when, and only when, they are immune from legitimate flat dismissal: to say, for example, that it is never legitimate to flatly dismiss arguments with perceptual premises.20 Bayesian defenses of Humean principles 21 suggest that whether it is legitimate to flatly dismiss a relevant counterargument (A) whose conclusion is the denial of a controversial proposition (p) depends on the three probabilities needed to calculate the probability that p is true conditional on there being an apparently sound argument like A: the prior probability that p, the probability 20

Thanks to Matt McGrath for helpful discussion on this issue. See Owen (1987), Sobel (1987), Dawid and Gillies (1989), Gillies (1991), and Hájek (1995), though Bayesianism may not capture the sort of reasoning Hume had in mind (see, e.g. Wilson (1989) and Gower (1990)). 21

54

Jeremy Fantl

that there is an apparently sound argument like A, conditional on p being false, and the probability that there is an apparently sound argument like A, conditional on p being true. In these terms, my claim is that the prior probability of p, and the probability that there is an apparently sound argument like A, conditional on p being true, are often sufficiently high that the probability that p is true conditional on there being an apparently sound argument like A is high enough for knowledge that p despite the presence of A. Sometimes, though, it’s not, depending on the nature of A. In those cases, flat dismissal of A is illegitimate. This proposal is only tentative, and for two reasons. First, we should allow that some of the controversial propositions in question—in particular, truths about morality and, perhaps, God’s existence—are necessary truths or necessary falsehoods. It is not straightforward how to apply the Bayesian calculus to such propositions: what’s the probability of someone having an apparently sound argument like A, conditional on p being false, when p is a necessary truth? Second, though the Bayesian calculus might show us how to maintain knowledge that p in the face of an apparently sound relevant counterargument, it still requires, unless the prior probability of p is 1, that the posterior probability of p goes down after conditionalizing on the fact that there is an apparently sound argument like A. But in many of the cases in question, not only does the apparently sound relevant counterargument leave knowledge intact, it doesn’t seem to lower rational credence. Certainly that’s the way it is with Zeno’s arguments that nothing moves. Perhaps, in these cases, the prior probability of p is so high that the resultant decrease in rational credence isn’t noticeable, or perhaps the proposition that things move has probability 1 for us. Nonetheless, these issues mean that the Bayesian suggestion cannot be more than tentative. A somewhat different way to determine when dogmatic belief is legitimate is hinted at when we notice that contrary evidence can be evaluated at different levels of specificity. A scientist evaluating the evidence might have to look at the details in order to appropriately form beliefs on the basis of that evidence. But a layperson, for whom the details of the evidence are inaccessible or, if accessible, not easily evaluated, might be required simply to rely on the fact that there’s a body of scientific evidence.22 So, sometimes, when evaluating evidence, we’re required to look at the details of the evidence. Other times, when evaluating evidence, we’re required not to look at the details. Which it’s legitimate to do can sometimes depend, in part, on how versed we are in the relevant literature and how likely it is we’d be duped by sophistical

22 The relationship between this proposal and the total evidence requirement—the principle that the evidential support enjoyed by a proposition is a function of your total evidence—is fraught. It is not obvious to me that the proposal can’t be reconciled with the requirement, but the prima facie tension between the two means that this proposal, as well, can’t be more than tentative.

A Defense of Dogmatism

55

maneuvers. If it turns out that in some controversial domains, some people are required to evaluate the evidence provided by apparently sound arguments to the contrary, not by looking at the details, but by looking at the general fact that there’s an apparently sound argument offered by my opponent, then it looks like in those domains, those people may very well be legitimately dogmatic, as long as they have strong evidence to support their own side. No matter what, it seems that the legitimacy of flat dismissal of arguments that not-p depends in large part on how much support you have for p. But suppose that you and I are on opposite sides of an issue. We each think that our respective arguments are strong enough to license flat dismissal of the other’s argument and suppose that our judgments equally well cohere with our other beliefs and attitudes. Can we avoid saying that if one of us legitimately flatly dismisses the arguments of the other, then we both legitimately flatly dismiss the arguments of the other? The issues here aren’t distinctive to the legitimacy of flat dismissal. For even when it comes to mere justified belief, there can be disagreeing parties, both of whom coherently judge that they are in a position to know that their beliefs are true, but only one of whom we’d like to say is correct. Must their beliefs be equally well justified? Must both parties be in a position to know? If not, how? There are different things to say, and epistemologists have said them all. It might be hard to figure out whether your epistemic position is strong enough to license dogmatic belief. But it’s not in principle harder than figuring out whether your epistemic position is strong enough for knowledge. One of us has that strength. The other one doesn’t. Which one has it? I think it’s me.23

REFERENCES

Aikin, Scott F. 2011. Epistemology and the Regress Problem. New York: Routledge. Bogardus, Tomas. 2009. “A Vindication of the Equal-Weight View.” Episteme 6: 324–35. Christensen, David. 2007. “Epistemology of Disagreement: The Good News.” Philosophical Review 116 (2): 187–217. 2011. “Disagreement, Question-Begging and Epistemic Self-Criticism.” Philosophers’ Imprint 11 (6): 1–22. Conee, Earl. 2001. “Heeding Misleading Evidence.” Philosophical Studies 103: 99–120. Dawid, Philip and Gillies, Donald. 1989. “A Bayesian Analysis of Hume’s Argument Concerning Miracles.” The Philosophical Quarterly 39: 57–65. Dawkins, Richard. 2006. The God Delusion. Boston: Houghton Mifflin Harcourt. Earman, John. 1993. “Bayes, Hume, and Miracles.” Faith and Philosophy 10: 293–310. Elga, Adam. 2007. “Reflection and Disagreement.” Noûs 41 (3): 478–502.

23 My thanks to Stewart Cohen, Berislav Marusic, Matthew McGrath, Mark Migotti, and two anonymous referees for Oxford University Press.

56

Jeremy Fantl

Feldman, Richard. 2006. “Epistemological Puzzles About Disagreement.” In Stephen Cade Hetherington (ed.) Epistemology Futures. Oxford: Oxford University Press: pp. 216–36. 2007. “Reasonable Religious Disagreements.” In Louise M. Antony (ed.) Philosophers Without Gods: Meditations on Atheism and the Secular Life. Oxford: Oxford University Press: pp. 194–214. Gillies, Donald. 1991. “A Bayesian Proof of a Humean Principle.” The British Journal for the Philosophy of Science 42: 255–6. Ginet, Carl. 1980. “Knowing Less by Knowing More.” Midwest Studies in Philosophy 5: 151–61. Gower, Barry. 1990. “David Hume and the Probability of Miracles.” Hume Studies 16: 17–29. Hájek, Alan. 1995. “In Defense of Hume’s Balancing of Probabilities in the Miracles Argument.” Southwest Philosophy Review 11: 111–18. Harman, Gilbert. 1973. Thought. Princeton, NJ: Princeton University Press. Hume, David. 2007 (1748). An Enquiry Concerning Human Understanding. Oxford: Cambridge University Press. Jastrow, Joseph. 1902. “Belief and Credulity.” Educational Review 23: 22–49. Kelly, Thomas. 2011. “Following the Argument where it Leads.” Philosophical Studies 154: 105–24. 2005. “Moorean Facts and Belief Revision, or Can the Skeptic Win?” Philosophical Perspectives 19: Epistemology: 179–200. Lackey, Jennifer. 2010. “A Justificationist View of Disagreement’s Epistemic Significance.” In Adrian Haddock, Alan Millar, and Duncan Pritchard (eds.) Social Epistemology. Oxford: Oxford University Press: pp. 298–325. Lucian. 1925 (c. 150 ad). “Alexander, or the False Prophet.” In A. M. Harmon (Trans.) Lucian, volume 4. Cambridge, MA: Harvard University Press, Loeb Classic Library: 173–254. McGrath, Sarah. 2008. “Moral Disagreement and Moral Expertise.” Oxford Studies in Metaethics 3: 87–107. Moore, George Edward. 1983 (1959). “Four Forms of Scepticism.” Philosophical Papers. London: Allen & Unwin: 196–226. Owen, David. 1987. “Hume Versus Price on Miracles and Prior Probabilities: Testimony and the Bayesian Calculation.” The Philosophical Quarterly 37: 187–202. Price, George R. 1955. “Science and the Supernatural.” Science 122: 359–67. Pryor, James. 2000. “The Skeptic and the Dogmatist.” Noûs 34: 517–49. Sobel, Jordan Howard 1987. “On the Evidence of Testimony for Miracles: A Bayesian Interpretation of David Hume’s Analysis.” The Philosophical Quarterly 37: 166–86. Sorensen, Roy. 1988. “Dogmatism, Junk Knowledge, and Conditionals.” The Philosophical Quarterly 38: 433–54. Wilson, Fred. 1989. “The Logic of Probabilities in Hume’s Argument against Miracles.” Hume Studies 15: 255–75.

3. Rational Agnosticism and Degrees of Belief∗ Jane Friedman

1.

PRELIMINARIES

“Traditional” epistemology presents us with a quite spare doxastic taxonomy. This taxonomy is typically thought to include just belief, disbelief, and suspension of judgment. And given that in this context disbelieving p is thought identical to believing ¬p, traditionalists present us with just two attitudes: belief and suspension of judgment. “Formal” epistemology presents us with an expansive doxastic taxonomy: a subject faces a continuum-many different doxastic options with respect to p. These doxastic states are thought to be something like states of confidence in the truth of the relevant propositions. These states of confidence come in degrees and those degrees of belief (credences) are taken to bear some intimate relation to the probability calculus. A number of interesting questions arise about the relationship between these two doxastic taxonomies. An obviously pressing one is which (if any) are accurate: do both of these doxastic taxonomies accurately describe our doxastic lives, or does just one (or does neither)? If only one gives a correct description, which one, and what should we say about the other? Many have thought that only one of these taxonomies is accurate and that the attitudes described in the other should be identical or reducible to the attitudes described in the correct taxonomy. Those keen on such a reduction typically take the formalist’s taxonomy to correctly describe our doxastic lives and aim to reduce the traditionalist’s doxastic attitudes to degrees of belief. The formalist’s taxonomy can seem much richer than the traditionalist’s, and as such the former is often thought capable of simply subsuming the latter. Let’s define the Straightforward Reduction as follows. The Straightforward Reduction is a reduction of the traditionalist’s doxastic attitudes to the

∗ Thanks to audiences at a conference on the ontology of the doxastic attitudes in Odense, Denmark and at the Pacific APA in San Diego for discussion of some of the arguments here. Thanks to Alan Hájek and an anonymous reviewer for Oxford Studies in Epistemology for extremely helpful written comments. I am grateful to Roger Clarke, John Hawthorne, Phil Kremer, Nikolaj Jang Lee Linding Pedersen, Gurpreet Rattan, Scott Sturgeon, Ralph Wedgwood, Jonathan Weisberg, and Roger White for discussion and/or written comments on different parts and permutations of this paper. Special thanks to Jennifer Nagel and Tim Williamson for feedback on multiple versions of many of the ideas here. Finally, I would like to thank the Social Sciences and Humanities Research Council of Canada for their generous support.

58

Jane Friedman

formalist’s degrees of belief that says that believing p, disbelieving p, and suspending about p are just matters of having (some specified) standard credences for p.1 What is a standard credence? Credences will count as standard for our purposes if the following two assumptions hold: 1. A subject’s total doxastic standing is represented with a single credence function. 2. The standard (Kolmogorov) axioms of the probability calculus are normative for credence functions. Here is the basic picture of standard credences with which I will be working. Take all the propositions for which a subject S has some degree of confidence or other. S’s confidence in these propositions is modeled or represented with a single real-valued function that takes these propositions as arguments. This is S’s credence function (Cs (·)). The numbers in the range of this function are S’s degrees of belief or levels of confidence in the relevant propositions.2 This function represents all of S’s degrees of belief at a given time—what I am calling her total doxastic standing. The standard axioms of the probability calculus are normative for this credence function.3 A subject with a credence function that fails to be a probability function is (at least to some extent) irrational. Credence functions aren’t always probability functions, but they ought to be probability functions.4 In this paper I want to argue that the Straightforward Reduction simply cannot succeed. Most discussions of the traditional-to-formal reduction (and the relationship between the two taxonomies in general) have focused on traditional belief. Instead, I want to focus on the traditionalist’s other attitude: suspended judgment (agnosticism).5 I want to argue that attempting to reduce suspension about p to having a standard credence for p (in some range) 1 ‘Suspending judgment about p’ is ungrammatical if ‘p’ is to be replaced with a declarative complement. ‘Suspend judgment about’ can embed interrogative complements and noun phrases, but not declaratives. In these sorts of constructions (and some others throughout) ‘p’ should be replaced with sentences like, ‘the proposition p’. 2 This is a very informal account of the relevant set-up, but it should do just fine. One thing worth flagging is that there are constraints placed upon the set of propositions that is the domain of a subject’s credence function. Let’s call this set F. F should contain all of the truth-functional combinations of its members: if p and q are in F, then ¬p, ¬q, (p ∧ q), (p ∨ q), (p → q), and (p ↔ q), and so on are all also in F. I will just assume that F has these properties, but one can think of them as normative constraints instead if one prefers. 3 These standard norms are:

3 Non-Negativity: Cs (p) ought to be a non-negative real number. 4 Normalization: Cs (p) ought to be 1 if p is a tautology ( p). 5 Countable Additivity: for any countable sequence of mutually exclusive outcomes   p1 . . . pn . . ., Cs ( j pj ) ought to equal j Cs (pj ). The basic picture is then one according to which a single real-valued function that is normatively bound by (3)–(5) represents a subject’s total doxastic standing at a time. 4 Although, I will assume that suspension is closed under negation: that S suspends about p at t iff S suspends about ¬p at t, and (for ease of exposition) that Cs (p) = x iff Cs (¬p) = 1 − x (even if I sometimes do not say so). 5 For a small, but good discussion of suspension and degrees of belief, see van Fraassen (1998), Hájek (1998), and Monton (1998).

Rational Agnosticism and Degrees of Belief

59

guarantees the failure of the Straightforward Reduction, and shows why suspension about p is not (just) a matter of having a standard credence for p. I will also say a bit about how we might extend some of the arguments in this paper to make more trouble for common credence-theoretic accounts of belief and other credence–theoretic accounts of suspension. In general, thinking about the p-credences of a rational p-agnostic can significantly impact one’s thinking about the relationship between the two doxastic taxonomies.6 My focus then is on the traditionalist’s second attitude, suspension of judgment. But should we even think of suspension as an attitude? I think that we should.7 Suspended judgment is, or at least involves, a proper attitudinal doxastic commitment. Agnosticism about p is not merely failing to believe both of p and ¬p. We have the property of neither believing nor disbelieving all sorts of propositions about which we are not agnostic: propositions we cannot grasp, or those we can but have never contemplated or had in mind in any way. Suspension requires some sort of decision about or commitment with respect to the truth of p; it isn’t a state that we are in in virtue of being opinionless, rather it is a state of opinion. It is in this sense that suspension is, or at least involves, a proper doxastic commitment about the truth of p on the part of the subject. The most natural way of understanding this commitment is as an attitude. What sort of attitude? A subject who suspends is effectively neutral or undecided about whether p is true. Her attitude then is one that represents or expresses or simply is her neutrality or indecision with respect to the relevant content.8 At first glance it might seem as though we have a good chance of finding this “indecision-representing” attitude in the formalist’s expansive taxonomy. After all, according to that taxonomy there are many intermediate doxastic attitudes: states of confidence that fall between a subject’s being absolutely confident in the truth of p and her being absolutely confident in the truth of ¬p. In order to show that the Straightforward Reduction cannot succeed and that suspending about p is not (just) a matter of having a standard credence for p, I will rely on a very basic principle of rational suspension of judgment. I will argue that given this norm for suspension, if we try to say that suspending about p is just a matter of having a standard credence for p, we will have to say that it’s a matter of having any standard credence for p at all; that is, we will have to say that one suspends about p iff Cs (p) ∈ [0, 1].9 6 The Straightforward Reduction (or something quite like it) was previously besieged by the Lottery Paradox (see Foley (1992) for a good discussion). The worrying attack though relied on a closure principle for rational belief that some chose to give up or deny. If anything, abandoning that principle has become more, rather than less acceptable with time. Some of the arguments here then can be thought of as a new (and more dangerous) attack on the Straightforward Reduction that requires no such principle. 7 For some agreement see (e.g.) Bergmann (2005) and Sturgeon (2010). 8 For more on these and other arguments for thinking that suspension of judgment is a proper doxastic attitude, see Friedman (2013). 9 Unless I say otherwise, when I talk about S’s credence for p, I mean her unconditional credence for p.

60

Jane Friedman

The norm for rational suspension that I want to focus on can be called the absence of evidence norm. This norm says roughly that in the absence of evidence for or against an ordinary contingent proposition p, it is epistemically permissible to suspend judgment about p. Let me clarify and defend this norm. I want to rely on a norm for suspension that is relatively uncontroversial: that suspension about p is epistemically permissible if you have no evidence relevant to whether p is true or false. But in order for this norm to be largely uncontroversial we need first to limit the class of propositions it applies to. It would be fairly controversial to claim that suspension about (say) ‘2 + 2 = 4’ is epistemically permissible absent evidence relevant to whether it is true or false. There are other sorts of propositions for which the absence of evidence norm might be more controversial as well. Some immediate perceptual propositions, or introspective ones might not be good candidates for falling under the norm. And some other sorts of contingent propositions might be able to escape as well: Kripke-style superficially contingent a priori propositions, and perhaps even some deeply contingent ones.10 Nonetheless, it is easy to see that for a wide class of propositions the absence of evidence norm is fairly uncontroversial. The propositions to focus on are what I’ll call ordinary contingent propositions. These propositions are “deeply” contingent for the relevant subjects—subjects will have no semantic guarantee of their truth— and they simply describe mundane (possible) facts about the physical world. Here is one helpful way to think about the absence of relevant evidence. Cases in which S lacks evidence relevant to p are cases in which S has no evidence either for or against p. In these sorts of cases, while S may have information (e.g. semantic or conceptual) about p, none of that information favours one of p or ¬p over the other. What’s important here is not that S’s total evidence fail to favour one of p, ¬p over the other. This can happen when that total body of evidence has some bits that favour p over ¬p and other bits that favour ¬p over p, and these bits are evenly weighted. These sorts of evidential circumstances also look like circumstances in which S is permitted to suspend judgment, but they aren’t circumstances in which S has no relevant evidence. This is precisely because some bits of S’s total evidence do favor one of p, ¬p over the other. Subjects who lack evidence for or against p possess no information that supports one of p or ¬p over the other. This idea is sometimes fleshed out in probabilistic terms. Evidence that is relevant to p is sometimes thought to be the sort of evidence that induces a change in a rational subject’s degree of belief for p.11 One way to think of a subject who has no evidence for or against p is as a rational subject whose credence for p has not been moved at all by the evidence she has acquired thus far; p has simply not felt the impact of her evidence. While this way 10

For a good discussion of the latter, see Hawthorne (2002). For instance, Keynes (1921) proposed that e is irrelevant to p (given background knowledge K) if the probability of p on K is the same as the probability of p on K + e, and relevant otherwise. 11

Rational Agnosticism and Degrees of Belief

61

of understanding a subject’s lack of evidence is incomplete in some respects (for instance, it so far counts a rational subject with a great deal of evidence that merely confirms, but does not change, her initial credence as having no evidence), it is a fine start. It can capture an important way of understanding a subject’s lack of relevant evidence about whether the 2018 Olympics will be in South Korea, whether amber is heterogeneous in composition, whether someone who played on the Michigan football team in the late 1800s became chairman of the Iowa Democratic Party, or whether electric potential has only magnitude and not direction. My claim is that when a subject is in these sorts of impoverished evidential circumstances with respect to an ordinary contingent p, she is epistemically permitted to suspend judgment about p (the reader can assume that the absence of evidence is as transparent as can be to the subject). The dictum to respect one’s evidence is fundamental for traditional normative epistemology. The absence of evidence norm is among the most minimal ways for a subject to respect her evidence. It says just (and very roughly) that when she has none, she may suspend. When your evidence gives you no guidance as to whether an ordinary contingent proposition p is true or false, the norm says that you are thereby permitted to suspend judgment about p. The alternative is to say that (at least in some cases) despite having no evidence relevant to whether that sort of proposition is true or false, suspension of judgment is epistemically prohibited. It is extremely difficult to see how that could be right. In fact it is hard to think of evidential circumstances more appropriate for suspension about these sorts of propositions than these kinds of absolutely impoverished ones. If you are going to have some attitude towards an ordinary contingent proposition that you understand, but about which you have absolutely no evidence either for or against, you cannot be going wrong by suspending judgment. These are exactly the sorts of circumstances suspension of judgment is for.12

12 Does this mean that belief is not permitted in these sorts of evidential circumstances? In general, I do not think that claims about the epistemic permissibility of suspension of judgment should be thought to entail or imply or be easily translated into claims about the epistemic impermissibility of belief. The possibility of a sort of “weak permissivism” should be open: there may well be evidential circumstances that permit either belief or suspension of judgment. That is, given some evidential circumstances with respect to p, it may be epistemically permissible to either (say) believe p or suspend about p (see White (2005) for some related discussion and potential pitfalls though). But is belief epistemically permissible in the sort of cases I am interested in here? In some cases I will argue that it is not, but otherwise I want to remain as neutral as I can on the issue. Can a subject both believe and suspend at once? I think that once we think of belief and suspension as independent attitudes we should think that the combination of attitudes is possible at a time. I take it though that we’ll want to say that a subject who has both attitudes at once is at least sometimes in a rationally conflicted state (much like the subject who believes both p and ¬p at a single time). Of course much more would need to be said about the nature of the objects of the agnostic attitude to get clear about this rational incompatibility. See Salmon (1995) for a good discussion of cases in which there may be no rational failing in both believing p (or ¬p) and suspending about p at a time. I am going to try to skirt these sorts

62

Jane Friedman

This is not to say though that these are the only sorts of evidential circumstances in which suspension of judgment is epistemically permitted. That is, it is not only in absolutely impoverished evidential circumstances that subjects are permitted to suspend. For instance, subjects who have some evidence, but (say) not nearly enough to fully settle the question of whether p is true or false, are also epistemically permitted to suspend judgment. As are subjects with massive bodies of evidence that are (roughly) evenly balanced (as just discussed). Sextus may have thought that suspension was always rationally permissible.13 This is fertile territory. For now I want to try to stay focused on absolutely impoverished evidential circumstances since they make for the tidiest versions of some of the arguments to come. Here is how I will proceed. In section 2 I will argue that thinking about cases in which subjects lack evidence shows that suspension about long conjunctions and disjunctions as well as the individual conjuncts/disjuncts is epistemically permitted. This will leave the Straightforward Reduction in a perilous position. In section 3 I will buttress and extend these conclusions by thinking about priors. I will show that the Straightforward Reduction really cannot succeed and that suspension of judgment about p, ¬p cannot (just) be a matter of having standard credences for p, ¬p (no matter which). In section 4 I will suggest a way to extend the arguments from the earlier sections. And in section 5 I will tie things up and look further forward.

2.

CONJUNCTIONS AND DISJUNCTIONS

Say S is given a drawing of a snowflake and is told that real snowflakes from different locations will be collected. She is asked to consider, for each collected snowflake, whether that snowflake has the same structure or shape as the one in her drawing. S will not see any of the collected snowflakes. Let a1 be the proposition that the first snowflake is a match, a2 the proposition that the second snowflake is a match, and so on through to an (for some finite n). This S does not know very much about snowflakes, in fact she’s never seen any before (she’s from a small island in the tropics). She knows roughly what they are—that they are bits of ice that fall from the sky when it is cold—but other than that, she has no evidence at all about their shapes: she has no idea what sorts of shapes they can take, how many different shapes they come in, the frequency with which a given shape occurs, or occurs with other shapes, and so on. There is simply nothing in her body of total evidence that bears on whether the collected flakes match her flake drawing. of issues in the paper. The state of both believing p (or ¬p) and suspending about p, ¬p at a time will be treated as an irrational state here. 13 Sextus claimed that we ought to be aiming for a kind of doxastic tranquility which can be arrived at by suspending judgment on all matters. See Empiricus (1933), Book I for his famous discussion of these issues. Of course, exactly what Sextus thought we did or ought to believe (rather than suspend about) is a matter of some controversy. See Burnyeat and Frede (1997) for some pillars of the debate.

Rational Agnosticism and Degrees of Belief

63

Take a1 − a3 . Say S considers each in turn. Is it epistemically permissible that she suspend judgment about each of these propositions individually? These are ordinary contingent propositions and we are stipulating that S has no evidence either for or against any of them. From this perspective, as we have seen, suspension is epistemically permissible. When she wonders whether a given flake is a match and recognizes her complete ignorance, she is permitted to suspend. What about the conjunction of these propositions? Or the disjunction? Can S rationally suspend about those alongside suspending about a1 − a3 individually? It looks that way. If we’re to imagine that S has no evidence at all that is relevant to whether each of a1 − a3 is true or false, and that she has no other evidence relevant to whether the conjunction or disjunction of those propositions is true or false, then it looks as though she also has no evidence relevant to whether that conjunction and disjunction are true or false. Given this, suspension of judgment still looks epistemically permissible. From the perspective of her evidence she is as ignorant about the conjunction and disjunction as she is about the individual conjuncts/disjuncts. But this means that we want to be able to say that S can suspend about a1 − a3 and about the conjunction and/or disjunction of those propositions at the same time.14 It looks as though the reasoning in the last paragraph carries over to longer conjunctions and disjunctions as well. For instance, imagine that on top of a1 − a3 S also considers a4 − a10 . The very same considerations about the epistemic permissibility of suspending about each individual ai (1 ≤ i ≤ 10) apply. Again, if we assume that S has no evidence relevant to each and no other evidence relevant to conjunctions and disjunctions of these propositions, then she still has no evidence relevant to whether those conjunctions and disjunctions are true or false. Again, suspension about each of a1 − a10 , as well as conjunctions and disjunctions of those are all epistemically permitted at a time. And the same goes for much longer conjunctions/disjunctions even. She might consider whether the first 100 flakes are a match, or the first 1,000, and so on. Our reasoning about a1 − a3 carries over. With no evidence for or against any individual conjunct/disjunct, and no additional evidence that bears just on the conjunctions/disjunctions, she has no evidence for or against those conjunctions/disjunctions either. In this sense, she has no idea about whether all or any of those flakes are a match. So now it looks as though it can be epistemically permissible for S to suspend judgment about very long conjunctions and disjunctions as well as each of their conjuncts/disjuncts.15 14 I have argued that suspending about each conjunct/disjunct is epistemically permitted and that so is suspending about the conjunction and disjunction, but is it epistemically permissible to suspend about all of those at once? I don’t see why not. S has no evidence either for or against any of these, their combinations are neither contradictory nor tautologous, and so at this point we have no reason to enact a ban on this combination of suspendings. In general, this looks like a perfectly epistemically appropriate (and commendable) combination of attitudes to have. 15 There’s nothing special about this snowflake example either. We could replace the a s with i other propositions. For example, p1 : the peace lily is a member of the Araceae family, p2 : the

64

Jane Friedman

This means trouble for the Straightforward Reduction. The easiest way to see this is to start with a plausible credence-theoretic account of agnosticism (for the Straightforward Reductionist) and see why it fails. The Straightforward Reductionist is going to try to find some subinterval of [0, 1] that is such that suspending judgment about p, ¬p is just a matter of having standard credences for p, ¬p in that subinterval. Let’s call this special subinterval the SJ-subinterval. Having credences for p, ¬p in the SJ-subinterval is necessary and sufficient for being in a state of suspended judgment about p, ¬p for the Straightforward Reductionist. Presumably, the first-pass suggestion is that the SJ-subinterval should be a roughly middling subinterval of [0, 1]. Earlier I said that the agnostic attitude is an attitude of neutrality or indecision about the truth of p, and a middling p-credence surely looks like a good candidate for capturing that sort of neutrality or indecision. Moreover, the Straightforward Reductionist is going to have to make room for belief and disbelief in the formalist’s taxonomy and presumably those will take up the high and low subintervals of [0, 1] (respectively). Let’s start with the suggestion that the SJ-subinterval is [1/3, 2/3] (nothing hangs on this specific precisification of ‘middling’; in a moment we’ll see why others fail as well). (Mid1 ) S suspends about p at t iff Cs (p) ∈ [1/3, 2/3] at t. But the Straightforward Reductionist has to say that (Mid1 ) is false. It is false because it renders impermissible combinations of suspendings that as we’ve just seen need not be impermissible. To see this we need only focus on a1 − a3 . Given that a1 , a2 , a3 are probabilistically independent, Cs (a1 ∧ a2 ∧ a3 ) ought to be equal to Cs (a1 )Cs (a2 )Cs (a3 ).16 But if each of Cs (a1 ), Cs (a2 ) magnolia is a member of the Araceae family, p3 : duckweed is a member of the Araceae family. Or, q1 : that star is a Population I star; q2 : that star is a Population I star; q3 : that star is a Population I star. Or, r1 : A has the DMD gene; r2 : B has the DMD gene; r3 : C has the DMD gene. Or, s1 : A went to the party; s2 : B went to the party; s3 : C went to the party. And so on. In each case we’re to imagine that the relevant subjects don’t have any relevant evidence. 16 Worry. If S ought to treat a − a as probabilistically independent doesn’t that mean that 1 3 she needs to have some information about their dependence relations and doesn’t that amount to her having some evidence? I’m not convinced of either. That is, it may well be true that she ought to treat them as independent even if she has no information about dependence relations (she has to assume something about dependence relations to have all of the relevant credences), and it isn’t clear that her having this information amounts to her having evidence in the sense at issue here. It is worth pointing out though that even if she did have that information and we did count it as evidence, this doesn’t significantly change the impact of the case. Suspension about the individual ai s as well as their conjunctions and disjunctions still looks epistemically permissible. More importantly though, I want to make clear that a version of this argument goes through no matter what we say that S ought or is permitted to assume about dependence relations. S is not only permitted to suspend about the conjunction of a1 − a3 but any conjunction that for each ai has either it or its negation as a conjunct, e.g. that flakes 1 and 2 are matches, but not flake 3, that flake 1 is not a match, but flakes 2 and 3 are, and so on. Each such conjunction is a possible outcome of this flake-matching experiment, and S is epistemically permitted to suspend about which of those outcomes obtain for the very same reasons she is permitted to suspend about whether the outcome in which all flakes match obtains. But if there are n ai s then there are 2n such outcomes/conjunctions (when n < 2 these

Rational Agnosticism and Degrees of Belief

65

and Cs (a3 ) are middling1 (in [1/3, 2/3]), then Cs (a1 )Cs (a2 )Cs (a3 ) cannot be middling1 , and so Cs (a1 ∧ a2 ∧ a3 ) ought not to be middling1 .17 And Cs (a1 ∨ a2 ∨ a3 ) ought to be equal to Cs (a1 ) + Cs (a2 ) + Cs (a3 ) − Cs (a1 ∧ a2 ) − Cs (a1 ∧ a3 ) − Cs (a2 ∧ a3 ) + Cs (a1 ∧ a2 ∧ a3 ). But if each of Cs (a1 ), Cs (a2 ) and Cs (a3 ) are middling1 , then Cs (a1 ) + Cs (a2 ) + Cs (a3 ) − Cs (a1 ∧ a2 ) − Cs (a1 ∧ a3 ) − Cs (a2 ∧ a3 ) + Cs (a1 ∧ a2 ∧ a3 ) cannot be middling1 , and Cs (a1 ∨ a2 ∨ a3 ) ought not to be middling1 .18 If S’s credences for a1 − a3 are middling1 at t, then she is not permitted to have a middling1 credence for the conjunction or for the disjunction of those propositions at t. If (Mid1 ) is true then S can only suspend about each of a1 − a3 as well as their conjunction or disjunction at t by having credences she is not permitted to have at t. (Mid1 ) renders the relevant combinations of suspendings epistemically impermissible. But we just saw that these combinations of attitudes are permissible. (Mid1 ) is false. In general, if we assume that credences are standard, then if (Mid1 ) is true it is never epistemically permissible for S to suspend judgment about each of (at least) three probabilistically independent propositions as well as the conjunction of those three propositions, and it is never epistemically permissible for S to suspend judgment about each of (at least) three probabilistically independent propositions as well as the disjunction of those three propositions. This does not look like a good result.19 Either way, we have now confirmed that it is false: this combination of suspendings can indeed be epistemically permissible. So, the Straightforward Reductionist cannot say that the SJ-subinterval is a middling1 subinterval of [0, 1]. But we also know that this is only the tip of the iceberg. Even if we focus on just a1 − a10 and say that Cs (ai ) = 0.5,  then Cs ( 10 i=1 ai ) ought to be 0.0009765625. And even this is still just the tip of the iceberg. Given the sort of reasoning from the snowflake case, it  doesn’t look as though there is any n such that suspending about ( ni=1 ai ) alongside each individual conjunct is epistemically impermissible (and the same goes, mutatis mutandis for the disjunctions). If this is right then take any x, y such that 0 < x ≤ y < 1, and say that the SJ-subintervals [x, y]. No matter what x and y are we can extend our snowflake example so that there are conjunctions and disjunctions such that it is epistemically permissible for S to suspend about those conjunctions and disjunctions as well as their outcomes are obviously not equivalent to conjunctions of distinct atomic propositions). So long as n > 1, whatever we say that S ought to or is permitted to assume about dependence relations, she will not be permitted to have middling1 credences for all of these conjunctions. In general, for any n ai s, probabilistic coherence alone requires S to have credence 1/2n or less for at least one conjunction that for each ai has either ai or ¬ai as a conjunct. But if n > 1, 1/2n ∈ / [1/3, 2/3]. More on this to come in the next section. If x ≤ 2/3, then x3 ≤ 8/27. And 8/27 < 1/3. If x, y, z ≥ 1/3, then [x + y + z − xy − xz − yz − xyz] ≥ 19/27. And 19/27 > 2/3. 19 In fact, the situation is worse than this even. If credences are standard, then it is never epistemically permissible that S be agnostic about just two probabilistically independent propositions as well as a conjunction and a disjunction of those propositions (I leave the details to the reader). This cannot be right. 17 18

66

Jane Friedman

individual conjuncts/disjuncts at t, but it is not epistemically permissible for her to have credences for all of these in [x, y] at t. The only proper subinterval of [0, 1] that won’t be susceptible to this sort of counter-example will be the open unit interval (0, 1).20 If S’s credences for all of the ai s are in (0, 1) then her credences for the conjunctions and disjunctions ought to be as well.21 So far then it looks as though the Straightforward Reductionist has to say that the SJ-subinterval is at least (0, 1).22 There might be some temptation to say that S shouldn’t have any credence at all in these snowflake propositions. After all, she’s completely ignorant about snowflake shapes. But we know that it is epistemically permissible to suspend about the relevant propositions. If it’s permissible to suspend about p but not permissible to have a standard credence for p, then suspending about p cannot (just) be a matter of having a standard credence for p. In general, if it’s possible for S to suspend about these snowflake propositions (which it is) but to not have credences for them, then any position that aims to reduce suspension to having a credence in some range (standard or otherwise) is in some trouble. These sorts of “credence gaps” are a real worry for the Straightforward Reduction then. If the ai s don’t generate credence gaps we can try to think of other propositions and circumstances that might. A subject might be so utterly in the dark about what the price of copper will be in 100 years, or whether in Minuscule 545 iota adscript occurs up to Luke 1:77, then ceases, or whether the Hill 50 Gold Mine was Australia’s most profitable mine between 1955 and 1961 that he ought to simply refuse to have any degrees of belief for the relevant propositions. But it is epistemically permissible that he suspend judgment about the relevant propositions even if we think he can’t or shouldn’t assign credences to them. No matter how confused or ignorant one is about how much credence to give a proposition in a case, so long as one can grasp the proposition in that case, suspension looks epistemically appropriate.23 20 Of course, there are other options left, i.e. [0, 1), and (0, 1]. I am simply assuming that the best case scenario for  the Straightforward Reductionist is the “shortest” interval, n n i.e. (0,1). 21 S’s credence for ( n a ) ought to be C(a ) and her credence for ( i i i=1 i=1 i=1 ai ) ought to be   |I|+1 I⊆{1,...,n} (−1) i∈I C(ai ). But for any n, these will remain in (0, 1) so long as her credences for the individual ai s do. 22 The Lockean (about belief) is someone who thinks that believing p is just a matter of having a sufficiently high degree of belief for p (see, e.g. Foley (1992)). Let’s say that the Standard Lockean is someone who thinks this about belief and wants to maintain that credences are standard. This Standard Lockean might try to dig in his heels here and insist that suspension is not permissible in some of my cases since sufficiently high credence just is belief. My argument might be deployed as the modus tollens to his modus ponens: if credences are standard, then since suspension of judgment about p is epistemically permissible in these cases, believing p cannot just be a matter of having a (sufficiently) high standard credence for p. But this isn’t just a stand-off. The relevant propositions are ordinary contingent propositions for which the subject has no relevant evidence, and as such, suspension of judgment is epistemically permissible. The Standard Lockean then owes us some story not only about why suspension about the relevant sort of p should be prohibited when one has no evidence either for or against p, but also one about why believing an ordinary contingent proposition should be permitted on absolutely no evidence. 23 See Hájek (2003) for a good discussion of probability gaps and their potential prevalence.

Rational Agnosticism and Degrees of Belief

67

So far it looks as though the Straightforward Reduction has to have the entire open unit interval dedicated to suspension of judgment, leaving 1 for belief and 0 for disbelief. I take it this is a bad result. We want to be able to say that subjects can (rationally) believe p with a p-credence less than 1. Certainly if credences are closely related to betting dispositions in the ways that they are standardly thought to be, that is, as identical to dispositions to bet or as explanations of such dispositions, then there is next to nothing that we believe to degree 1. Even if we wanted to keep credences and bets apart, we should still be troubled by the association of belief with credence 1. At least the following is true of credence 1: it cannot be moved by conditionalization. A p-credence of 1 ought to remain 1 no matter what evidence comes in, given the standard way of updating a credence function on new evidence. It is effectively rational to hold on to a belief with credence 1 come what may. Perhaps there are some beliefs like this, but that class is extremely limited. Whatever notion of belief the Straightforward Reduction is left with here, it is not our everyday notion: there is next to nothing we believe by the lights of the Straightforward Reduction.24 Those less worried about the identification of belief with credence 1 and disbelief with credence 0, can take only momentary comfort: in the next section I’ll show why the Straightforward Reduction doesn’t even have this option.

3.

PRIORS

I am going to assume that the Straightforward Reductionist is a Bayesian, at least in the following additional sense: she thinks that the rational updating of a subject’s credence for p upon receiving new evidence e ought to be determined in part as a function of her prior credence for p—her credence for p “before” she received e. Usually this prior credence is the result of updating on past evidence (i.e. it was the posterior credence in some other update), 24 Some people have tried to argue that the belief-making threshold is shifty: that it can vary according to the circumstances of the subject, or different contexts of utterance. We can find suggestions like this in Hájek (1998), Weatherson (2005), Ganson (2008), Sturgeon (2008), Fantl and McGrath (2009) and a related one in Hawthorne and Bovens (1999). These accounts give up on the Straightforward Reduction strictly so-called since they say that belief and suspension will have to supervene on not just degrees of belief, but degrees of belief plus some other sorts of facts. Nonetheless, some of these views might be thought to respect the spirit of the Straightforward Reduction if not the letter, in that they can say that the only doxastic state that one needs to be in to believe or suspend is to have some relevant (standard) degree of belief. As such, I want to make clear that nothing shifty is going on in any of my cases. Standard mechanisms that make belief harder to come by are raised stakes or the making salient of the possibility of error. Absolutely nothing is at stake for the subjects in my cases. Even when stakes are as low as can be, subjects can rationally suspend with any degree of belief in (0, 1) (assuming, of course, that credences are standard). And the same goes for the possibility of error. No mention of such a possibility is or need be made anywhere to make it that subjects can rationally suspend with very high and very low degrees of belief. Fix these other sorts of facts any way you like, we will still be able to get the result that the SJ-subinterval must be at least (0, 1).

68

Jane Friedman

but this need not be the case. The Bayesian makes room for an absolutely prior p-credence: the subject’s credence for p that is prior to the receipt of any evidence for or against p at all. This sort of primordial or original prior credence is sometimes called an ur-prior credence. There is nothing particularly mysterious about ur-priors. In order to have our opinions rationally informed by evidence, we have to start with some prior, uninformed state of opinion. We can think of one’s credence for p at any time t as the one that results from conditionalizing this ur-prior for p on one’s total evidence at t. One’s ur-prior p-credence is the credence for p one has in the complete absence of evidence: it is an a priori state of opinion about p; it is a degree of belief assigned to p a priori. In this section I’ll argue—by thinking about ur-priors—that the Straightforward Reductionist must make the SJ-subinterval [0, 1]. First, I’ll say why, if credences are standard, it can be epistemically rational to have ur-priors anywhere in [0, 1]. After that I’ll draw out the implication of that fact for the Straightforward Reduction. Assume credences are standard and imagine a subject S assigning ur-priors to the propositions in various finite partitions Pi (i = 1, 2, . . . , n), where P1 is a one-celled partition, P2 a two-celled partition, and so on. These cells are propositions. What should a rational S’s ur-prior credences over these partitions look like? We need only be concerned with the minimal demands that her credence for each cell/proposition be in [0, 1] and that her credences over Pi – for any i – sum to 1.25 These demands guarantee that however a rational S distributes her credences over the cells of the Pi s, so long as there are sufficiently many of them, her credences for most of those propositions will have to be very low (and her credences for their negations very high).26 As i increases, a rational S will have to have credences for the cells that approach 0 (and credences for their negations that approach 1). Moreover, if we move to thinking about partitions with infinitely many cells, and keep in place the Straightforward Reductionist’s assumption that credences are standard, we can see that a rational S will have to be able to

25 If S’s credences for the cells of these partitions are each in [0, 1] but don’t sum to 1, then those credences will fail to be countably additive or be in conflict with the normalization norm (or both). 26 I take it that this is fairly obvious. If a rational S’s credences are uniformly distributed over these partitions, then it is very easy to see: her credence for each p ∈ Pi will be 1/i, and so as i goes up, her credences for each p ∈ Pi will go down. But I have not demanded credal uniformity here. If rational S distributes her credences non-uniformly over Pi where i is very high, then not all of her credences for the propositions in Pi need be low. For instance, she may have a credence of 0.999 for some p ∈ P1010 . But given that her credences for the propositions in P1010 ought to sum to 1, her credences for the remaining 1010 − 1 propositions will have to sum to 0.001. Any way she disperses that 0.001 across those remaining propositions, her credences for those individual propositions will have to be very low. If i is sufficiently high, even if rational S’s credence for some p ∈ Pi needn’t be extremely low, her credence for most of those propositions will have to be. In general, even without uniformity, for any Pi , a rational S’s credence for at least one p ∈ Pi will be at most 1/i.

Rational Agnosticism and Degrees of Belief

69

have ur-priors that are 0 or 1 as well. Say Q is such an infinite partition. If Q is merely countably infinite, then a rational subject can have a credence greater than 0 for each q ∈ Q, but her credences over Q cannot be uniform. When the number of possibilities is countably infinite, a rational subject with standard credences must favor some possibilities over others.27 But when Q is uncountable, if credences are standard, then S will be rationally required to have credence 0 for uncountably many members of Q (and credence 1 for each of their negations). If S’s credences over an uncountable Q are to sum to 1, then she can only have credence greater than 0 for countably many q ∈ Q. But this means that a rational S will have credence 0 for uncountably many q ∈ Q (and so credence 1 for their negations).28 What impact should this have for the Straightforward Reduction? In section 2 I argued that the Straightforward Reductionist has to say that the SJsubinterval is at least (0, 1). In fact, let’s imagine a Straightforward Reductionist who says just that: suspending about p, ¬p is just a matter of having credences for p, ¬p in (0, 1), believing p just a matter of having credence 1 for p, and disbelieving p just a matter of having credence 0 for p. I have already said a little bit about why this is an unpalatable view, but now we have a new line on how it goes wrong. Think about our uncountable partition Q, and rational S distributing her credences over Q a priori. Let’s say Q is a partition of ordinary contingent propositions. The Straightforward Reductionist I just described will have to say that a rational S disbelieves uncountably many q ∈ Q. He has to say that S can suspend about every q ∈ Q only by having probabilistically incoherent credences. So he has to say that suspending about every q ∈ Q is epistemically impermissible. But each q ∈ Q is an ordinary

27 A uniform distribution over a countably infinite partition that is countably additive will violate the normalization axiom. Some (e.g. de Finetti (1970), famously) have worried about the claim that a rational subject must have a skewed credence function in these sort of cases. See Williamson (1999) for a good discussion. 28 Here is an informal and (hopefully) intuitive way to see this (adapted from Williamson (2007), p. 173; see Hájek (2003), p. 281–2 as well). First, I am assuming that the Straightforward Reductionist must avoid credence gaps for the reasons already discussed. Now, say Q is uncountable, and imagine a probability distribution (Pr) over Q. Now think of subsets of Q, Qi as follows. Q1 is the subset of Q whose members have probability 1/1 or greater, Q2 the subset of Q whose members have probability 1/2 or greater, Q3 the subset of Q whose members have probability 1/3 or greater, and so on. For each i, Qi has finitely many members. In particular, each can have at most i members or else the probability of the disjunction of its members would exceed 1. Now take any real number x such that 0 < x ≤ 1. If Pr(q) = x then q is in at least one of the Qi s. For instance, if Pr(q) = 0.001, then q is in any subset of Q, Qi such that i ≤ 1/1000. And the same goes (mutatis mutandis, of course) for any x in (0, 1]. But that means that any q ∈ Q that gets assigned a real number in (0, 1] by Pr is Qi for some i. Now take the union of these Qi s. This is a union of countably many finite sets. But a union of countably many finite sets is itself a countable set. This means that every q ∈ Q such that Pr(q) > 0 is in this countable subset of Q. But Q is uncountable which leaves uncountably many q ∈ Q that can only get probability 0. This means that a countably additive probability distribution over an uncountable partition will have to assign 0 to uncountably many cells of that partition. Given that we are assuming that S’s credence function ought to be a (standard) probability function this means that S ought to have credence 0 for uncountably many q ∈ Q (and credence 1 for their negations).

70

Jane Friedman

contingent proposition about which S has absolutely no evidence, and given that, suspension about every q ∈ Q should be epistemically permissible. This means that the suggestion that the SJ-subinterval is (0, 1) renders rationally impermissible combinations of suspendings that are not rationally impermissible. To avoid this result it looks as though the Straightforward Reductionist will have to say that the SJ-subinterval is [0, 1]. Let me flesh out and buttress this argument. First, let me say a bit more in defence of the claim that suspension about these ordinary contingent partition propositions is permissible. Remember, in these cases, S’s credences are had in the compete absence of evidence; her credences are ur-prior credences. This means that we are to think of them as opinions she has completely a priori. But now we can buttress the claim that suspension about these propositions is epistemically permissible as follows: it is not epistemically permissible that S believe or disbelieve an ordinary contingent proposition a priori, and so with respect to these partitions of ordinary contingent propositions, suspension is not only a rationally permissible doxastic attitude to have towards those partition propositions, it is the uniquely epistemically permissible attitude (from the traditionalist’s taxonomy) to have. Let’s flesh this out further. Let’s first focus on a single relevant (countable) partition A. We can think of A as an ordinary contingent question, in this case the question: how many birds are there in France? This question can be thought of as a partition with each possible complete answer (that there are no birds in France, that there is exactly one bird in France, that there are exactly two birds in France, and so on) as a cell of that partition. These answers are mutually exclusive and exhaustive. Now we can imagine S distributing her credences over A a priori. For any a ∈ A, it is not epistemically permissible that S believe a. I take it that this is obvious. It is not epistemically permissible for S to believe that there is some specific number of birds in France a priori. But what about disbelief? Is it also epistemically impermissible that she believe any ¬a a priori? On the one hand there is an obvious and important symmetry between belief and disbelief in this case. Any given a ∈ A is an ordinary contingent proposition and so is its negation. It is standard to claim that it is not epistemically permissible to believe those sorts of propositions a priori. On the other hand, if credences are standard, then S’s credences for most of the propositions in A ought to be extremely low, and from that perspective it can look as though there is also an important asymmetry between the epistemic permissibility of believing some a and the epistemic permissibility of believing some ¬a. If it is epistemically permissible that S be extremely confident that some ¬a is true a priori, should it really be impermissible that she believe it a priori? While I admit that there is something compelling about this line of thought, there is near consensus that ordinary contingent propositions should not be believed a priori. There have been a few arguments over the years that some contingent truths are knowable a priori. Kripke (1980) famously argued that someone who introduces the term ‘one meter’ as a rigid designator for the

Rational Agnosticism and Degrees of Belief

71

length of a particular stick s at time t can know a priori that the length of s at t is one meter. These “superficially contingent” propositions are such that the subject has some actual semantic guarantee of their truth, and so it looks as though she is permitted to believe them a priori.29 Some have even argued that there may well be some deeply contingent propositions that we can be permitted to believe a priori.30 But these too form a limited class (e.g. there is a believer), and the as do not fall into it. This is all to say that there is almost no precedent for claiming that propositions like the ¬a s could be rationally believed a priori.31 Insisting otherwise here would at the very least amount to a significant re-drawing of the bounds of the a priori, and without serious argument for that re-drawing, we should disregard the relevant asymmetry between any a and ¬a here. In general then we should conclude that neither belief nor disbelief towards any a ∈ A is epistemically permissible. A is a countable question or partition, but everything I have said about it will carry over to uncountable ones as well. S might contemplate questions with uncountably many answers: what the president’s credence that it will rain tomorrow is, or exactly how long the tail of the oldest cat in the world is, or what the exact landing point of the winning dart at the 1983 BDO World Championships was, and so on. Everything I have said about A applies to these partitions as well. No matter how many cells make up these partitions, so long as they are partitions of ordinary contingent propositions neither belief nor disbelief is epistemically permitted. This leaves suspension as the uniquely epistemically permissible attitude from the traditionalist’s taxonomy for S to have towards these propositions. I don’t think that the permissibility of suspension in these sorts of cases is up for grabs. I have argued that belief and disbelief are not epistemically permissible attitudes to have towards these propositions given S’s evidential circumstances. S is epistemically prohibited from believing ordinary contingent propositions a priori. But these considerations do not extend to her suspending about these propositions. In fact, suspension looks like exactly the right attitude to have if she is going to have one (and there is no special reason to demand that she have none). If she’s considering a priori whether apples are sweeter than spinach or how long it takes beavers to build dams or

29 I am simply assuming that we can replace ‘can know’ or ‘is in the position to know’ in these cases with ‘is epistemically permitted to believe’ without incident. 30 Again, see Hawthorne (2002), and for some recent responses and discussion see Avnur (forthcoming) and Turri (forthcoming). 31 Almost. Someone like Tyler Burge in Burge (1993) might want to say that we can come to know ordinary contingent propositions a priori if we come to know them via certain kinds of testimony. Of course, my cases here do not involve testimony. See Christensen and Kornblith (1997) for a good discussion of Burge’s view. We might also find a line of thought that can get us to the conclusion that there are a priori justified beliefs in ordinary contingent propositions in Douven (2008). These arguments rely on claims about our “epistemic goal” that seem to me highly contentious though. In fact, if anything, Douven’s arguments make clear part of what is wrong with these teleological claims.

72

Jane Friedman

where B. B King was born, then it is perfectly reasonable for her to suspend judgment.32 Now we can be clear about exactly what this all means for the Straightforward Reduction. The Straightforward Reductionist wants to isolate an SJsubinterval—a subinterval of [0, 1] such that suspending about p, ¬p is just a matter of having credences for p, ¬p in that subinterval. We have already seen why anything short of (0, 1) won’t work: picking any subinterval shorter than that will force the Straightforward Reductionist to say that some epistemically permissible combinations of suspendings are not epistemically permissible (e.g. long conjunctions and each of their conjuncts). This section confirms that result, but also extends it. It confirms it as follows. Say the Straighforward Reductionist sets the SJsubinterval to some interval [x, y] (or (x, y)) shorter than (0, 1). We can now find a finite partition of ordinary contingent propositions such that S is permitted to suspend judgment about all of the propositions in that partition, but such that her credences for at least some of those propositions ought to be less than x and her credences for their negations greater than y. But this means that the cases in this section again show that if the Straightforward Reductionist sets the SJ-subinterval to anything shorter than (0, 1) she will render epistemically impermissible combinations of suspendings that are not epistemically impermissible. But I hope it is also clear now that (0, 1) suffers in much the same way. If the SJ-subinterval is (0, 1), then it is not rationally permissible that S suspend about the ordinary contingent propositions in some infinite partitions. But suspending about all of those propositions is epistemically permissible. Moreover, it looks as though the Straightforward Reductionist who makes the SJ-subinterval (0, 1) (or shorter) will have to say that a rational S believes many of those ordinary contingent propositions a priori. But a rational S won’t believe ordinary contingent propositions a priori. If the SJ-subinterval is (0, 1) (or shorter), then what look like perfectly rational ur-prior credence distributions over these infinite partitions are rendered irrational since they will have the result that the relevant subject believes ordinary contingent propositions a priori. So the SJ-subinterval cannot be (0, 1) either. In fact, it cannot be any proper subinterval of [0, 1]. The only option left for the Straightforward Reductionist is to make the SJ-subinterval [0, 1] itself. The Straightforward Reduction has the result that the SJ-subinterval is [0, 1]. But this is not an acceptable result for the Straightforward Reduction. It says that suspending about p is just a matter of having any degree of belief for p at all. But if suspending about p is a matter of having any standard p-credence at all, then there is no hope of reducing believing p and disbelieving p to having standard p-credences: the whole unit 32 Although I have argued that belief and disbelief are not epistemically permitted in these cases, it is worth making clear that denying those claims does not secure the claim that suspension of judgment is not epistemically permissible. That claim would need an independent defence, whatever one said about the permissibility of belief and disbelief.

Rational Agnosticism and Degrees of Belief

73

interval has to be devoted to suspension of judgment. The Straightforward Reduction cannot succeed since once suspending about p is just a matter of having a standard credence for p, believing p cannot be a matter of having a standard credence for p, and disbelieving p cannot be a matter of having a standard credence for p. Moreover, the conclusion that S suspends about p, ¬p iff S has (any) standard credences for p, ¬p is not a palatable conclusion even for someone not concerned with reducing believing p and disbelieving p to having standard credences for p (and ¬p). We are not agnostic about p simply in virtue of having a credence for p. This would not only make it that believing p and disbelieving p are not reducible to standard credences for p but that it is irrational to either believe p or disbelieve p when one has a standard credence for p. If anyone with a standard credence for p is agnostic about p, then on the assumption that believing p and suspending about p or disbelieving p and suspending about p are irrational combinations of attitudes, no one with a standard credence for p is permitted to believe p or disbelieve p; that is, no one is permitted to be in a state in which they both have standard credences for p, ¬p and either believe p or disbelieve p.33 We should conclude not only that the Straightforward Reduction fails, but also that suspension about p is not (just) a matter of having a standard credence for p.

4.

B E Y O N D T H E S T R A I G H T F O R WA R D R E D U C T I O N

The Straightforward Reduction fails. Believing p, disbelieving p, and suspending about p cannot all just be matters of having standard credences for p. The endeavour falls apart once we try to reduce suspension of judgment about p to a standard credence for p. More generally, we should conclude that suspending judgment about p is not (just) a matter of having a standard credence for p. Do my arguments here leave room for a position like Standard Lockeanism though? The Lockean is someone who thinks that believing p is just a matter of having a sufficiently high degree of belief for p, and the Standard Lockean is someone who thinks this about belief and wants to maintain that credences are standard. The Standard Lockean won’t be bothered if being in a state of suspended judgment about p, ¬p cannot be reduced to having standard credences for p, ¬p, but he does want to maintain that credences are standard and that (dis)believing p is reducible to having a standard credence 33 Although given our commonplace understanding of belief and degrees of belief these are quite bad results, there is obviously some space for a position that bites these bullets. This position could reduce suspending about p to having a standard credence for p (any at all), but claim that belief and disbelief are attitudes that a rational subject has exactly when she commits to p or ¬p in some distinct way (some way that doesn’t at all involve degrees of belief). This would mean claiming (among other things) not just that believing p is something different from having credence 1 for p, but that it is rationally incompatible with having credence 1 for p. Obviously, we’d need to hear more about what belief is on this view before it could become a serious option.

74

Jane Friedman

for p. I want to make clear that my arguments here do not leave room for this Standard Lockean. In this paper I have described a number of cases in which suspending judgment about p is epistemically permissible. Giving up on reducing suspending about p to a standard credence does nothing to change this. The Standard Lockean (and anyone else) will still need to make room for rational suspension of judgment in those cases. But this will not be possible if the Standard Lockean insists that rational subjects in the sorts of cases I have been describing have standard credences. If they do still have standard credences then they will have very high/low credences. Given this, the Standard Lockean will have to say that they are believers and disbelievers, and so cannot be rationally agnostic. This Lockean has to say that these subjects are not permitted to suspend in the complete absence of evidence and that they are permitted to believe ordinary contingent propositions a priori. Put slightly differently, giving up on reducing suspension about p to having a standard credence for p but maintaining that subjects in the relevant cases have standard credences, while respecting the norms for suspension and a priori believing I’ve discussed here, will have the result that suspension about p is rationally compatible with any standard credence for p (even if it is not reducible to having a standard credence). But if suspending about p can be rationally compatible with having any standard credence then no standard credence can amount to belief since believing p and suspending about p are not rationally compatible. So the Standard Lockean must do more than just give up on reducing suspending about p to having some relevant standard credence for p. In fact, his only option is to claim that subjects in the cases I’ve described have no credences at all. While this might look plausible in some of those cases, it cannot be right for all of them. That would amount to claiming that rational subjects largely have no absolutely prior credences. If subjects need priors to update their degrees of belief on new evidence, then insisting that rational subjects have no absolutely prior degrees of belief, threatens to leave them with no posterior credences either, that is, threatens to leave rational subjects with no credences for anything at all. Obviously this is not a real option for the Standard Lockean, and so we must conclude that Standard Lockeanism is false. Anyone keen on reducing belief and disbelief to high and low credences (respectively), who wants to say that subjects in my cases do have degrees of belief, will have to allow for non-standard degrees of belief in those cases. In fact, some (not just those keen on reducing belief to credence) argue that standard credences are not epistemically appropriate in some of the sorts of cases I have described. For instance, some claim that we need to use infinitesimal degrees of belief to accurately capture a subject’s state of opinion when those opinions range over uncountably many possibilities.34 34 See (e.g.) Lewis (1980) and Skyrms (1980) for the suggestion, and see Hájek (2003) for a good discussion of some of its drawbacks (both philosophical and mathematical).

Rational Agnosticism and Degrees of Belief

75

Making a credence function a hyperreal-valued function might help to escape the conclusion that suspending about p is rationally compatible with having a p-credence of 0 or 1, but the allowance will do nothing to avoid the result that suspension is rationally compatible with all real numbers (and now perhaps infinitesimals as well) between 0 and 1. Another non-standard approach might be more helpful in this regard though. Some claim that in the absence of evidence, or in a state of complete ignorance one ought to have vague or imprecise (mushy, interval) credences. These are degrees of belief that are measured with or represented by sets of real numbers (e.g. [0.2, 0.8]) rather than single real numbers.35,36 Can the Lockean (not the Standard Lockean) say that subjects in the cases I’ve described have imprecise credences? Perhaps, however, this solution does not come all that easily.37 Even in the cases I’ve described, the claim that rational subjects cannot have standard credences is a significant one. Someone who wanted to avoid the conclusion that suspension is rationally compatible with most any precise degree of belief by turning to imprecise credences will have to then say that a rational subject mostly does not have absolutely prior precise credences that are bound by the standard axioms of the probability calculus. This isn’t just the claim that some rational subjects sometimes have or can have non-standard credences. It’s the claim that it is largely rationally impermissible to have standard ur-priors (at least for ordinary contingent propositions). While many Bayesians are comfortable with imprecise credences it isn’t clear that everyone will be happy with this strong claim about priors (certainly the ultra-subjective Bayesian won’t be, but it is easy to imagine complaints from the more objectively minded as well). The much bigger problem though is that this “solution” is not a general solution at all. It may help with the cases I have discussed so far, but that is 35 Typically this is achieved by using sets of credence functions rather than a single function to represent a subject’s total doxastic standing at a time. We can start with a set of credence functions R and the relevant set of propositions F that is the domain of the functions in R. We can then say that each C(·) ∈ R ought to be a probability function, and that a set of real numbers in the unit interval is assigned to the propositions in F. If any two functions in R disagree on their p-assignments (so that the resulting set is a non-singleton set), we can think of the resulting p-credence as imprecise (vague, mushy, etc.). It is also standard (although not required) to demand that R be convex and so demand that the sets assigned to the propositions in F be intervals. 36 For instance, Joyce (2005) claims that we should not try to capture states of “ambiguous or incomplete” (see p. 167 for more detail about what these amount to for him) evidence using a single credence function. He argues that picking any single credence function amounts to pretending to have information that one does not possess. He claims that when one’s evidence is ambiguous or incomplete it is compatible with many distribution of objective probability over the hypotheses, so by distributing credences in any one way over them one ignores a vast number of possibilities that are consistent with one’s evidence. We can find the suggestion that suspension of judgment be captured with credal imprecision in van Fraassen (1989) and Sturgeon (2010). 37 For some general worries about imprecise credences and the sorts of states they represent, see Elga (2010) and White (2010).

76

Jane Friedman

not enough to guarantee that suspension of judgment about p is not rationally compatible with most any standard credence for p. That is the conclusion that the Lockean (and others like him) is trying to avoid. In particular, his goal is to have it that suspension about p is not compatible with high and low standard credences for p. While arguing that S ought not have standard credences in the complete absence of evidence might help to block one path to the conclusion that suspension about p is compatible with high and low p-credences, it does nothing to block other very nearby paths (nor further away ones, of course). Let’s stay focused on questions about the epistemic permissibility of suspending with high or low credence. As I mentioned at the outset, the absence of evidence norm is just one norm among many for epistemically rational suspension. Even if we say that it needn’t license suspension about p with high/low credence for p, that does not mean that some other (closely related) norm won’t. It is not at all clear that suspension about p is going to be epistemically prohibited in cases in which it is relatively uncontroversial that p-credences ought to be precise and either very high or very low. Genuinely making the case that such a combination of attitudes is epistemically permissible is beyond the scope of this paper. But I want to end this section by saying a little bit in defence of the claim. When chances are known it is fairly widely agreed that subjects should set their credences to match those known chances.38 Let’s imagine a slight variation on our French birds case. Say France is birdless at t1 , but at t3 God will give the French some birds. At t1 S assigns her ur-priors to propositions of the form ‘the number of birds in France at t3 will be x’ (where x ∈ N). At t1 she has absolutely no evidence that bears on how many French birds there will be at t3 . Given the arguments so far we should say that suspension about these French bird propositions is epistemically permissible at t1 . But now say that at t2 God tells S that at t3 there will be between 1 and 108 birds in France, and that the number of French birds will be decided by a random process such that each number in the relevant range has the same probability as any other (maybe God will roll his 100,000,000-sided die). Say B = {b1 , b2 , . . . , b100,000,000 } is the partition of French bird number propositions (where b1 is the proposition that there will be exactly one bird in France at t3 , and so on). It is fairly widely agreed that at t2 S ought to have the same, precise degree of belief for each bi ∈ B, that is, for every b ∈ B, Cs (b) ought to be 1/108 . The challenge now is to say just why this sort of information should make suspension of judgment about the number of birds there will be in France at t3 epistemically impermissible. S has more information than before about the number of French birds at t3 , and so we may not be able to lean as much on worries about the contingent a priori, but the new information that she 38 The canonical account of the rationality of this sort of chance-credence relation comes in Lewis (1980). Lewis proposes that the correct relation between chance and credence is given by his Principal Principle, which states roughly that credences ought to track known chances.

Rational Agnosticism and Degrees of Belief

77

has is still incredibly minimal. While she knows how many possibilities there are, and that each is equally probable, she knows nothing at all that makes any remaining possible answer to the question of how many birds there will be in France at t3 stand out from any other. That is, with respect to the question of the number of birds there will be in France at t3 , the evidence she has does not discriminate between the remaining possible answers. Why should suspension of judgment be epistemically impermissible when one’s evidence does nothing to support any one possible remaining outcome over any other?39 We can make the worry acute as follows. Think about the change in view that S undergoes when she learns that the possible number of birds in France at t3 is limited to within a certain range and will be determined by a chance process. Let’s call this new evidence e. e is evidence in favor of the propositions in B. It is evidence that one of those propositions is true, and that the French bird propositions not in that set are false. Here is one way of fleshing this thought out. Before acquiring e let’s say that S had credences for all of the French bird propositions, (for each n ∈ N there is one such proposition). We can assume that S is rational and updates her credences as she ought. Given that, when S learns e her credence for any bk (that the number of birds in France at t3 will be k) for every k greater than 108 will drop to 0 (as will her credence that there will be no birds in France at t3 ). What should we say about her confidence in the remaining 108 French bird propositions? Well, surely her confidence in at least some of them will increase upon learning e and dropping her credence in those French bird propositions incompatible with e. Even if one wants to insist that S’s confidence is not to be measured with standard credences before acquiring e, we want to be able to say that rational S is more confident about at least some of the remaining propositions after discovering e than she was before discovering e. Her outcome space is significantly smaller and more stable and so her confidence in at least some of the remaining propositions should have increased. However we cash out this increase in confidence (it won’t be straightforward if we’re comparing a non-standard credence to a standard one), the point is that we want to be able to somehow say that e makes it epistemically permissible that S be more confident about at least some of the remaining answers than she was before receiving that evidence. e is evidence in favor of those propositions. Let’s say that b5 is one of the propositions about which S becomes more confident upon receiving e. We know that upon receiving e (at t2 ), S credence for each b ∈ B ought to be 1/108 . Anyone keen on avoiding the conclusion

39 In fact we can think of this as an extension of one of the oldest norms for suspension of judgment. In Outlines of Pyrrhonism Sextus says, “the term ‘suspension’ is derived from the fact of the mind being held up or ‘suspended’ so that it neither affirms nor denies anything owing to the equipollence of the matters in question” (Empiricus (1933), PH I, 196). When evidence supports all outcomes equally it is equipollent. Notice, the cases in which subjects have no evidence for or against any outcome are also cases in which they have equipollent evidence.

78

Jane Friedman

that suspension of judgment is rationally compatible with high/low credences will have to say that suspension about b5 (or any other b ∈ B) is no longer permissible at t2 . But is that plausible? Obviously believing b5 (or any other b ∈ B) is not epistemically permissible at t2 . So what can we say about S’s doxastic options if suspension is impermissible (and so is belief)? There are two possibilities: either e permits disbelief (belief in ¬b5 ) or it does not. If we say that disbelief is also epistemically prohibited, then we are left having to say that in acquiring good new evidence, evidence that grounds greater confidence in b5 , and a clear and precise credence for that proposition, S is no longer permitted to have any attitude from the traditionalist’s taxonomy towards b5 . This can’t be right. Moreover, this option obviously will not sit well with the Lockean who wants to identify disbelief with low credence. But if we say that disbelief is permitted, then we’ll have to say that in acquiring good new evidence for b5 , evidence that grounds greater confidence in b5 (but not belief), the uniquely permissible attitude for S to have towards b5 becomes disbelief. We will have to say that in acquiring evidence that counts in favor of b5 and grounds increased confidence in that proposition (but not belief), S will move from circumstances in which suspension is permitted to ones in which only disbelief is. But this can’t be right either. Say I am suspending about whether my national team will win some Olympic event. I have no evidence either way. Then I learn that a member of one of the the opposing teams has a minor injury. This is evidence that counts in favor of my team winning; it grounds increased confidence that my team will win, but cannot ground a belief that they will win (the injury is minor, there are still plenty of other teams in the competition). It would be absurd to claim that I am now no longer permitted to suspend judgment about whether my team will win, but only permitted to believe that my team won’t win: I’ve just acquired more evidence that they will win; I am rationally more confident than before that they will win. But the same is true when it comes to S, e, and b5 . Given that at t2 S has more evidence that there will be exactly five birds in France at t3 than she did at t1 , it would be bizarre to say that at t2 her evidence demands she move from suspension to disbelief about that proposition. It simply does not look as though this new evidence can bar suspension of judgment; suspension should still be epistemically permissible. If this is right then we get the result that suspension about p is rationally compatible with very high and very low credence for p. This is just the start of a new argument for the rational compatibility of high/low standard credences for p with suspension about p. It extends the sort of general worries that have emerged in this paper about minimal evidence justifying maximal credence alongside suspension of judgment, but it requires further thinking about the conditions under which suspension of judgment is epistemically permissible (among other things).40 And of course making the 40 It is also worth making clear that genuinely making good on this line will leave the prospect of identifying suspension about p with, or reducing suspension about p to, having an imprecise p-credence extremely dim.

Rational Agnosticism and Degrees of Belief

79

case that suspension about p is rationally compatible with high/low credences genuinely compelling would involve much more discussion of the roles of credences, belief, and suspension of judgment in cognition, inquiry, and action more generally.

5.

CONCLUDING REMARKS

Where does this leave us? I have argued that the Straightforward Reduction fails. Given that suspending judgment about ordinary contingent propositions is epistemically permissible in the absence of evidence, the demand that suspending about p be a matter of having a standard credence for p has the result that suspending about p is a matter of having any standard credence for p. If that’s right then believing p cannot be a matter of having a standard p-credence and disbelieving p cannot be a matter of having a standard p-credence. However, the aim of the Straightforward Reduction is to reduce all three attitudes from the traditionalist’s doxastic taxonomy to some relevant degrees of belief. But it looks as though the only way to reduce suspension about p, ¬p to some standard credences for p, ¬p is to make it that believing p and disbelieving p can no longer be so reduced. So the Straightforward Reduction cannot succeed. Furthermore, the result that suspending about p is a matter of having any standard credence for p at all is not acceptable even for someone not concerned with reducing believing p and disbelieving p to standard credences for p. It looks false that anyone with a standard credence for p—any at all—is agnostic about p. Moreover, it would mean not only that (dis)believing p could not just be a matter of having a standard credence for p, but that (dis)believing p couldn’t even be rationally compatible with having a standard credence for p. We get this last result since both (dis)believing p and suspending about p is not a rational combination of attitudes at a time, and so if anyone with any credences at all for p, ¬p at t is suspending about p, ¬p at t, then no one with credences for p, ¬p at t can also rationally believe p or ¬p at t. We must conclude that suspending about p, ¬p is not (just) a matter of having standard credences for p, ¬p. I have also argued that anyone who wants to try to avoid the result that suspension of judgment about p is (rationally) compatible with high/low credence for p will have to say that subjects in the cases I discussed in sections 2 and 3 have non-standard credences in those cases. I ended though by saying a little bit about why—even if subjects in those cases do have non-standard credences—the conclusion that suspension of judgment about p is genuinely (rationally) compatible with very high and very low credences for p might prove difficult to escape. If it does prove unavoidable, then common versions of Lockeanism are false. If we can rationally suspend about p despite having (say) a very high degree of belief for p, then believing p must be more than or different from merely having a high degree of belief for p.

80

Jane Friedman

What about our two taxonomies? Even if the Straightforward Reduction fails, there are plenty of more sophisticated ways we might try to reduce the traditionalist’s doxastic attitudes to degrees of belief. For instance, perhaps the difference between believing p and suspending about p does not lie in one’s unconditional p-credences but elsewhere in one’s credence distribution: in some of one’s conditional p-credences or even one’s credences for propositions other than p (e.g. some have tied suspending about p to having higherorder beliefs about one’s first-order epistemic standing). We might also try starting with a different, non-Kolmogorov axiomatization. And I think that reductions in the other direction (i.e. reducing degrees of belief to traditional doxastic attitudes) are more promising than is often acknowledged. Each of these suggestions has some plausibility but faces difficulties, and it remains to be seen whether any can succeed. For now we should conclude that the Straightforward Reduction fails, and that suspending about p is not (just) a matter of having a standard credence for p. And we should be worried about whether suspending about p isn’t genuinely rationally compatible with most any precise degree of belief for p.

REFERENCES

Avnur, Y. (2011). Hawthorne on the Deeply Contingent A Priori. Philosophy and Phenomenological Research 83(1), 174–83. Bergmann, M. (2005). Defeaters and Higher-Level Requirements. The Philosophical Quarterly 55, 419–36. Burge, T. (1993). Content Preservation. Philosophical Review 102(4), 457–88. Burnyeat, M. and M. Frede (1997). The Original Sceptics: A Controversy. Indianapolis: Hackett Publishing Co. Christensen, D. and H. Kornblith (1997). Testimony, Memory and the Limits of the a Priori. Philosophical Studies 86(1), 1–20. de Finetti, B. (1970). Theory of Probability, Volume 1. New York: Wiley. Douven, B. I. (2008). The Lottery Paradox and Our Epistemic Goal. Pacific Philosophical Quarterly 89(2), 204–25. Elga, A. (2010). Subjective Probabilities Should Be Sharp. Philosophers’ Imprint 10(05), 19, 1–11. Empiricus, S. (1933). Outlines of Pyrrhonism. Loeb Classical Library. Cambridge, MA: Harvard University Press. Fantl, J. and M. McGrath (2009). Knowledge in an Uncertain World. New York: Oxford University Press. Foley, R. (1992). The Epistemology of Belief and the Epistemology of Degrees of Belief. American Philosophical Quarterly 29, 111–24. Friedman, J. (2013). Suspended Judgment. Philosophical Studies 162(2), 165–81. Ganson, D. (2008). Evidentialism and Pragmatic Constraints on Outright Belief. Philosophical Studies 139(3), 441–58. Hájek, A. (1998). Agnosticism meets Bayesianism. Analysis 58, 199–206.

Rational Agnosticism and Degrees of Belief

81

(2003). What Conditional Probability Could Not Be. Synthese 137(3), 273–323. Hawthorne, J. (2002). Deeply Contingent A Priori Knowledge. Philosophy and Phenomenological Research 65(2), 247–69. Hawthorne, J. and L. Bovens (1999). The Preface, the Lottery, and the Logic of Belief. Mind 108(430), 241–64. Joyce, J. M. (2005). How Probabilities Reflect Evidence. Philosophical Perspectives 19, 153–78. Keynes, J. M. (1921). A Treatise on Probability. London: Macmillan. Kripke, S. A. (1980). Naming and Necessity. Cambridge: Harvard University Press. Lewis, D. (1980). A Subjectivist’s Guide to Objective Chance. In Philosophical Papers: Volume II. Oxford: Oxford University Press. Monton, B. (1998). Bayesian Agnosticism and Constructive Empiricism. Analysis 58, 207–12. Salmon, N. (1995). Being of Two Minds: Belief with Doubt. Noûs 29, 1–20. Skyrms, B. (1980). Causal Necessity: A Pragmatic Investigation of the Necessity of Laws. New Haven: Yale University Press. Sturgeon, S. (2008). Reason and the Grain of Belief. Noûs 42, 139–65. (2010). Confidence and Coarse-Grained Attitudes. In Oxford Studies in Epistemology Volume 3. Oxford: Oxford University Press. Turri, J. (2011). Contingent A Priori Knowledge. Philosophy and Phenomenological Research 83(2), 327–44. van Fraassen, B. C. (1989). Laws and Symmetry. Oxford: Clarendon Press. van Fraassen, B. C. (1998). The Agnostic Subtly Probabilified. Analysis 58, 212–20. Weatherson, B. (2005). Can We Do Without Pragmatic Encroachment? Philosophical Perspectives 19(1), 417–43. White, R. (2005). Epistemic Permissiveness. Philosophical Perspectives 19(1), 445–59. (2010). Evidential Symmetry and Mushy Credence. In Oxford Studies in Epistemology Volume 3. Oxford: Oxford University Press. Williamson, J. (1999). Countable Additivity and Subjective Probability. British Journal for the Philosophy of Science 50(3), 401–16. Williamson, T. (2007). How Probable is an Infinite Sequence of Heads? Analysis 67(3), 173–80.

4. Probability and Prodigality∗ Daniel Greco

1.

INTRODUCTION

There seem to be systematic relations between knowledge and rational belief on the one hand, and rational decision on the other. If you know that it will rain, then if you are rational you will carry an umbrella. The more reason you have to believe that a bridge is unstable, the more reason you have not to walk across it. Insofar as we are interested in both epistemology and practical rationality, we will want some explanation of how they bear on one another. One approach that seems to promise a powerful and unified explanation of these relations is that of Bayesian decision theory. According to Bayesian decision theory, agents ought to maximize expected utility. The relevant notion of expected utility, however, is defined relative to a probability function, and this seems like a promising place to try to forge a connection between decision theory and epistemology. A natural thought is that the probability functions relevant to rational choice are somehow sensitive to evidence. Suppose I am betting on the outcome of a coin toss, and the coin to be tossed is, unbeknownst to me, double-headed. If I have no reason to believe that it is double-headed, and in fact my evidence suggests that it is a fair coin, then I would plausibly be rationally required to accept a low-stakes bet at 1:2 odds on tails. This shows that the notion of probability relevant to rational choice is not objective physical probability; even though the objective physical probability of my winning the bet was zero, my decision was rationally required. Orthodox subjective Bayesians take examples like the one above to show that the notion of probability relevant to rational choice is not objective chance, but instead subjective doxastic probability.1 While this suggestion is an improvement over the idea that objective probability is the sort of probability that is relevant to rational choice, it still faces serious obstacles. Suppose I am completely convinced that I am the target of a Martian conspiracy. Acting on this belief, I purchase large quantities of tinfoil, so as to fashion hats that will render me invulnerable to Martian mind-scrambling rays. Even if my beliefs are probabilistically coherent, there is a perfectly natural, intuitive

∗ For helpful comments and discussion, thanks to Alan Hájek, Agustin Rayo, Robert Stalnaker, Roger White, two anonymous referees for Oxford Studies in Epistemology as well as the editors thereof, and audiences at MIT and the ANU. 1 See e.g. Howson and Urbach (1996).

Probability and Prodigality

83

sense of “irrational” in which these beliefs and the actions based on them are irrational.2 If we want to hold on to the idea that buying large quantities of tinfoil would be irrational in some sense, and we want to use decision theory to provide an account of rational choiceworthiness in this sense, then we need an evidence-sensitive notion of probability which is identical neither to objective physical probability, nor to subjective doxastic probability. Given such a notion of probability, we can then hold that agents are rational when they act so as to maximize expected utility relative to that probability function. Let us preliminarily assume that there is such a notion of probability, and let us call it “epistemic probability.”3 In this paper, I will argue against a view about the connection between epistemic probability and knowledge. According to this view, what an agent knows has epistemic probability 1 for that agent. There are a number of ways one might be led to such a view. One way involves reflecting on the impropriety of “concessive knowledge attributions” such as the following:4 (1) I know that I will always be too short to play in the NBA, but it is possible that I will have a dramatic mid-life growth spurt and be recruited by the Bulls. (2) I know that I will never be able to afford a private jet, but there is some chance that I will win the $1,000,000,000 jackpot in the MegaBucks lottery. There are many potential explanations for the impropriety of such knowledge attributions, but some explanations will involve the claim that what an agent knows has epistemic probability 1 for that agent.5 Suppose we take the impropriety of such attributions to indicate that they are false.6 If “chance” in (2)

2 We might want to draw a distinction here between rationality and reasonability—perhaps a subject acts rationally just in case she acts in ways that make sense given her beliefs and desires, and she acts reasonably if she acts rationally and she has sensible beliefs and desires. When I buy tinfoil, then, we might say that my actions are rational, but not reasonable. While some writers use “rational” in a broad enough sense that it would probably cover both sorts of evaluation—see Gibbard (1990, p. 7) on “rational” as carrying “a kind of direct and flavorless endorsement”—this is an important distinction that deserves to be marked somehow, whether with the words “rational” and “reasonable” or some other way. Once we make this distinction, the position in the text amounts to the position that decision theory should shed light not just on questions about rationality, but also on questions about reasonability. See Broome (1995, chap. 5) for some persuasive arguments to the effect that decision theory cannot ignore reasonability. 3 Drawing a threefold distinction between objective probability, subjective probability, and epistemic probability is standard in the epistemological literature—see Mellor (2005). 4 The phrase “concessive knowledge attribution” is from Rysiew (2001). 5 I will often drop “for an agent” and “epistemic,” and will just refer to this claim as the claim that knowledge has probability 1. Since epistemic probabilities are always agent-relative, and I won’t be talking about non-epistemic probability in the remainder of this paper except when I explicitly indicate otherwise, such omissions shouldn’t create confusion. 6 That such attributions are false is by no means uncontroversial. Rysiew (2001) holds that they are true, but inappropriate to assert on pragmatic grounds.

84

Daniel Greco

refers to epistemic probability, or if a proposition’s being possible in the sense of (1) requires that it have non-zero epistemic probability, then the claim that knowledge has epistemic probability 1 could explain why claims like (1) and (2) must be false. Alternatively, we might accept that knowledge has epistemic probability 1 on other grounds. If we hold that an agent’s epistemic probabilities are obtained by conditionalizing some prior probability function on the conjunction of all the propositions the agent knows—as Williamson (2000) argues— then it is an immediate consequence that any proposition an agent knows must have epistemic probability 1 for that agent.7 In the remainder of this paper I won’t be concerned with how we might motivate the claim that knowledge has epistemic probability 1, but instead with the consequences of this claim for our views about practical rationality. In §2, I will argue that the view that knowledge has probability 1 leads to unacceptable conclusions about rational choice. In §3, I will consider and reject a response from Timothy Williamson, according to which the blame for these conclusions should not be laid on the link between knowledge and probability 1, but instead on expected utility maximization; according to this response, we should reject decision theory as an all-purpose guide to rational choice, rather than rejecting the association between knowledge and probability 1. In §4, I will consider responses that try to reconcile decision theory with the claim that knowledge has probability 1 by appealing to contextualism or subject sensitive invariantism (SSI) about knowledge.8 I will argue that while such views may avoid the problem identified in §2, they do so only by generating other uancceptable conclusions about rational choice. Lastly, in §5, I will consider a response on behalf of the sensitive invariantist that will lead into more general issues about idealization in epistemology and decision theory.

2.

THE PRODIGALITY PROBLEM

If skepticism is false, then we know quite a lot. If skepticism is false and knowledge has epistemic probability 1, then quite a lot of propositions have epistemic probability 1. It is easy to see why we might worry about this— orthodox decision theory famously does some odd things when probability 1

7 Williamson uses the term “evidential probability” rather than “epistemic probability,” but I have followed standard usage and gone with the latter. This may mark an important distinciton, since as will become clear later in the paper, Williamson doesn’t think that his notion of evidential probability links up with decision theory in the way I have said epistemic probability should. 8 See Cohen (1986), DeRose (1995), and Lewis (1999), for a sampling of contextualist epistemological views. See Hawthorne (2004) and Stanley (2005) for defenses of subject-sensitive invariantism. The term “subject-sensitive invariantism” is from Derose (2004).

Probability and Prodigality

85

is too easy to attain.9 The desire for a model of probabilistic updating in which updating doesn’t involve assigning any new propositions probability 1 was the main motivation behind Jeffrey’s (1965) generalization of strict conditionalization. Williamson (2000, chapter 10) stresses the problems associated with the fact that in orthodox decision theory, once a proposition has probability 1, it retains that probability for all time. This problem, along with many others associated with allowing propositions to have maximal probability, can be avoided by taking conditional probabilty as primitive and allowing for conditionalizing on probability zero events.10 However, the problem I will focus on in this paper cannot be solved by such methods. The problem I will focus on is that, when combined with other plausible claims about knowledge, the claim that knowledge has probability 1 leads to implausible claims about rational choice. The following example will bring out the worry. Suppose I read in the most recent edition of the Enyclopedia Britannica that the 2010 estimate of the population of Canada is just over thirty-four million.11 Suppose that this is true, and that the actual population of Canada is quite close to the estimate. Given plausible, non-skeptical assumptions about knowledge, in such a situation I know that the population of Canada is between thirty and forty million. Now suppose I am offered a bet by a credible bookie that will pay one cent if the population of Canada is between thirty and forty million, and which will bankrupt me if Canada has a population of fewer than thirty or greater than forty million. Given more controversial assumptions about knowledge to the effect that whether one knows that P doesn’t depend on the practical significance of P (assumptions which will be relaxed in later sections of this paper), I can still know that Canada’s population lies between thirty and forty million, even after being offered such a bet. Intuitively, however, it would be irrational for me to accept this bet— given the pay-offs, I shouldn’t do so, even though it is extremely probable in light of my evidence that I will win if I take it. The view that knowledge has probability 1, however, cannot account for this fact about rational choiceworthiness, at least when combined with the assumptions about knowledge mentioned above. The argument that I ought to accept the bet if knowledge has probability 1 and the aforementioned assumptions about knowledge hold is straightforward. While I spell it out in detail in a footnote, the basic idea is that if I know that the population of Canada is in the relevant bounds, and that I will win the bet if the population is in the relevant bounds, then the epistemic

9 Hájek (Draft) considers ten apparent problems with liberal assignments of maximal probability, though he argues that they can each be circumvented. Easwaran (Draft) takes a similar line. 10 This is the strategy favored by Hájek (Draft) and Easwaran (Draft). Williamson (2000, chapter 10) opts for a different route. 11 Full disclosure: I actually read it in Wikipedia.

86

Daniel Greco

probability that I will win if I accept the bet is 1. And if this is the case, then the expected utility of accepting the bet is equal to the utility of winning the bet: the disutility of losing the bet doesn’t affect its expected utility at all.12 Call this apparent result—that the defender of the view that knowledge has probability 1 is committed to the view that one is rationally required to accept high-stakes bets on propositions one knows no matter how unfavorable the odds—the prodigality problem. Perhaps it seems odd to call this a special problem for the view that knowledge has epistemic probability 1—orthodox subjective Bayesian agents could easily conditionalize on claims about Canada’s population and claims about the behavior of bookies in ways that would lead to their being committed to betting at any odds on claims about Canada’s population. But as mentioned earlier, the orthodox subjective Bayesian isn’t trying to explain constraints on rational choice that go beyond formal coherence constraints—she doesn’t think I am irrational if I blow the bank buying tinfoil to prevent Martian mind-scrambling, so long as such actions maximize expected utility relative to my subjective doxastic probability function. If we think there are further constraints, and we want to use decision theory to shed some light on them, then the orthodox subjective Bayesian isn’t a very helpful companion in guilt for the defender of the view that knowledge has epistemic probability 1. Before considering responses to the prodigality problem, it is important to be clear on exactly what is alleged to be problematic. The worry isn’t that, if knowledge has probability 1 and some further assumptions hold, there are situations where we ought to take high-stakes bets at extremely unfavorable odds. Plausibly, there are at least possible cases where I should bet my life against a penny, if the proposition that I am betting on is probable enough. It is hard to imagine any attractive theory of rational choice that wouldn’t have this consequence—no matter how bad the potential downside of a bet, and how small the potential upside, if we hold these fixed, then we should be 12 Let P represent my epistemic probability function, let H be the proposition that the population of Canada is between thirty and forty million, let A be the proposition that I accept the bet, let W be the proposition that I win the bet, and let L be the proposition that I lose the bet. Lastly, let U be my utility function. I assume evidential decision theory for simplicity.

1. Expected utility of (A) = P(W|A)U(W) + P(L|A)U(L) (by the definition of expected utility, together with assumptions that P(W ∨ L|A) = 1, and P(W ∧ L) = 0.) 2. P(W ↔ H|A) = 1 (Because I know that the bookie is telling the truth about the conditions of the bet) 3. P(W|A) = P(H|A) (from 2) 4. P(H) = 1 (Because I know H) 5. P(H|A) = 1 (from 4) 6. P(W|A) = 1 (From 3 and 5) 7. P(L|W) = 0 8. P(L|A) = 0 (From 6 and 7) 9. Expected utility of (A) = U(W) (From 1, 6, and 7) Note that the expected utility of accepting the bet depends only on the utility of winning, and not at all on the disutility of losing.

Probability and Prodigality

87

able to adjust the probabilities so that taking it is the rational thing to do, even without letting the proposition one is betting on have maximal probability.13 Rather, the problem is that in everyday cases, once we hold fixed certain aspects of the situation—for example, that I recently read the Encyclopedia Britannica, that I don’t have any defeating evidence for the claim that the population of Canada is in the relevant bounds, etc.—I ought to take certain high-stakes bets no matter how unfavorable the odds are. Intuitively, there ought to be some level of disutility of my losing the bet such that it wouldn’t be rational for me to take the bet. We can stomach the thought that this level is very high—and so we can stomach the thought that I ought to bet on very unfavorable odds—so long as such a level exists; what we cannot stomach is the idea that there is no way of making the downside bad enough that it is no longer rational for me to bet. One potential line of response to the problem is a thoroughgoing skepticism about knowledge, according to which we never know very much at all; for instance, one might hold that when it comes to contingent propositions, the only ones we can know (if any at all) concern our present mental states. Such a position would avoid being committed to the view that we ought to accept bets at any odds, since while claims about one’s present mental states would have epistemic probability 1, claims about the population of Canada, or the betting behavior of bookies, will not. While this line of response would avoid the problem, it would do so at considerable cost—it is a highly revisionary epistemological view. If the claim that knowledge has probability 1 is only plausible if we know very little, then it is less plausible than we might have hoped. In the remainder of the paper, I will consider two lines of response to the prodigality problem that defenders of a link between knowledge and probability 1 might offer. The first response is due to Williamson (2000, 2005b, 2009). As I interpret him, he maintains that while we should hold onto the idea of epistemic probabilities as an epistemological notion—roughly, they represent the degree to which one’s evidence supports a hypothesis—we should reject the idea that they will combine in systematic ways with utilities to govern rational choice in the manner specified by decision theory. So we can agree that everyday propositions often have epistemic probability 1 in some sense without committing ourselves to any absurd conclusions about radical choice. Since this response gives up on decision theory as a guide to understanding choiceworthiness, it would be disappointing if we were forced to it. I will argue that we are not. The second response involves allowing that while we may know propositions about, for example, the population of Canada, before we are offered high-stakes bets on them, things change once we are offered such bets.14 So we 13

Unless, perhaps, infinite disutilities are at stake. Exactly how this response works will differ depending on whether we are discussing the sensitive invariantist, or contextualist version of this response. 14

88

Daniel Greco

can hold on to the link between probability 1 and knowledge, and the idea that decision theory is a guide to rational choiceworthiness, without committing ourselves to endorsing prodigal betting behavior.15 I will argue that while this response avoids the prodigality problem, it faces other serious objections.16

3.

I S P R O D I G A L I T Y E V E RY O N E ’ S P R O B L E M ?

It may seem as if no particular view about the connections between epistemic probability and knowledge is necessary to generate the prodigality problem. The assumption that rational agents maximize expected utility may seem to generate the problem all by itself. I take this to be the idea motivating the following comments of Williamson’s: We should question the association between evidential probability 1 and absolute certainty. For subjective Bayesians, probability 1 is the highest possible degree of belief, which presumably is absolute certainty. If one’s credence in P is 1, one should be willing to accept a bet on which one gains a penny if P is true and is tortured horribly to death if P is false. Few propositions pass that test. Surely complex logical truths do not, even though the probabiltiy axioms assign them probability 1. (2000, p. 213)

In replying to commentators on Knowledge and its Limits, Williamson (2005b, p. 478) reiterates the point, claiming that “no decision theory based on expected utility, calculated according to the standard axioms and definitions of mathematical probability theory, will be everywhere consistent with what pretheoretic common sense predicts a sensible person would do.” His defense of this claim appeals to another example involving complex logical truths. If Williamson really is right that the assumption that it is rational to maximize expected utility is enough to commit us to prodigality all on its own, then rejecting this assumption may be our best option. Still, a defender of decision theory who still wanted to press the prodigality problem against the view that knowledge has probability 1 might bite the bullet here, and hold that while betting the farm on contingent (but known) truths is irrational, betting at any odds on logical truths is rationally required.17 After all, rejecting the link between expected utility maximization and choiceworthiness is a serious price to pay—once we do so, we lose out on

15 While this response is inspired by Hawthorne (2004) and Stanley (2005), I don’t claim that they would endorse it. Hawthorne (2004, p. 137), in particular, seems to want to distance himself from broadly decision-theoretic approaches to rational choice, so I suspect that he would endorse the spirit of Williamson’s response to the prodigality problem. 16 There are other possible responses that I take seriously, but which I won’t consider here. See the discussion of “objection 7” in Hawthorne and Stanley (2008). Also, see “Unreasonable Knowledge,” by Maria Lasonen-Aarnio (2010). Lasonen-Aarnio might argue that while knowledge has probability 1 and acting so as to maximize expected utility is rational, such action may nevertheless be unreasonable, in a sense that she explicates. 17 As I understand Kaplan (2009, p. 136, note 22), this is his position.

Probability and Prodigality

89

a promising candidate explanation of the systematic connections between evidence and rational action. While Williamson’s remarks may suggest that it is just the identification between probability 1 and rational willingness to bet at any odds that we must reject, it is not clear whether we can maintain any general connection between expected utility and rational choiceworthiness once this link is severed. For instance, if we accept that 0.5 epistemic probability is associated with rational willingness to bet at even odds, must we think that the relationship between choiceworthiness and expected utility changes at some point between probability 0.5 and probability 1, or perhaps that there is a discontinuity at probability 1? Neither of these options is attractive. The above objection doesn’t require that epistemic probabilities be identified with rational betting odds. While it is plausible that this link holds ceteris paribus, we should allow that it is severed when, for example, agents have aversions to betting, or have other desires that bear on the attractiveness of accepting a bet independently of its potential pay-off, as Ramsey (1931) notes. When ceteris is paribus, however, the link should hold, and this is all that is necessary to support the objection to the view that knowledge has probability 1. Still, perhaps we should regard decision theory as applicable at best only in special cases, and should instead adopt a conception of practical rationality that takes something like Aristotelian practical syllogism as the paradigm of rational practical deliberation, and which shunts probabilsitic notions to the sidelines; we might hold some version of the view that it would be rational for an agent to ϕ when she knows the premises of a valid deductive argument whose conclusion is that she should ϕ. For instance, we might follow Williamson (2005a) and hold that one often does know propositions when one is offered bets on them at extremely unfavorable odds, but avoid prodigality by holding that when one is making high stakes bets that depend on whether P, it is not enough to know that P in order to reasonably rely on P in one’s practical reasoning. Instead, as the stakes go up, one must have increasingly many iterations of knowledge that P for it to be reasonable to act as if P—one must know that one knows, or know that one knows that one knows, and so on. Since one will generally not have arbitrarily many iterations of knowledge, in cases of very high-stakes bets on a proposition P, one will typically not have enough iterations of knowledge in order to reasonably act on the assumption that P. In the absence of some account of how the magnitude of the stakes in a situation fixes some level of iterations of knowledge that one must have in order for it to be reasonable to act as if some proposition is true, this approach is less elegant than Bayesian decision theory. There isn’t anything analogous to the notion of expected utility to give a systematic story about how epistemic and practical factors combine to generate facts about choiceworthiness. While I take this to be an unhappy position that gives up on much of the promise of decision theory, if decision theory commits us to prodigality, then we might

90

Daniel Greco

reasonably retreat to something like the Williamsonian position sketched above. 3.1. Prodigality and Bridge Principles Luckily, we can avoid prodigality without rejecting decision theory. Suppose we accept decision theory and the probabilistic framework associated with it—in particular, we accept that logical truths have epistemic probability 1.18 Nothing immediately follows about what bets we must make. Suppose I am offered a bet that is purported to pay off just in case it will rain tomorrow or it’s not the case that it will rain tomorrow. Call the proposition that it will rain tomorrow or it’s not the case that it will rain tomorrow—a logical truth if there ever was one—“Truth.” Let Accept be the proposition that I accept the bet on Truth. Let Win be the proposition that I win the bet. For it to be rational for me to accept the bet no matter the odds, it is not enough that Truth has epistemic probability 1. What is also required is that the following probability be maximal: P ((Win ↔ Truth) | Accept). Equivalently, it is required that P (Accept ∧ Truth ∧ ∼Win) = 0. Put in the language of certainty, it isn’t enough that I be certain that it’s either going to rain or not in order for me to bet at any odds—I must also be certain that betting will have a favorable outcome if and only if it is either going to rain or not. But this claim—a claim about the result of handing some money to a bookie—is exactly the sort of contingent claim that a defender of decision theory can consistently be quite hesitant to assign probability 1. Put more generally, logical truths on their own have no implications about which actions are more promising than others.19 Assigning probability 1 to logical claims only warrants prodigal betting behavior if bridge claims linking

18 This is a bit quick. There have been attempts to model logical ignorance in a Bayesian framework: see especially Garber (1982). However, while Garber’s approach does allow for logically equivalent empirical hypotheses to be assigned different probabilities, it doesn’t allow purely logical truths to be assigned sub-maximal probabilities. Generally, I am sympathetic to Williamson’s claim that “one could try to construct a non-standard probability calculus in which truth-functional tautologies can have probability less than 1, but such modifications tend to make huge sacrifices in mathematical power for tiny gains in psychological realism” (Williamson, 2005b, p. 478). However, later in this section I briefly discuss a different approach to the problem of logical omniscience, which doesn’t involve assigning logical truths submaximal probabilities. 19 I’m making some assumptions here about what counts as an action for the purposes of decision theory. In particular, I’m assuming that actions are the sorts of things one can in some sense “directly” perform, so that [accept a bet that will pay $1 if P ∨ ∼ P] will not count as an action, but [utter “I’ll take the bet”], or perhaps [form an intention to utter “I’ll take the bet”] will. See Lewis (1981, p. 7), Joyce (1999, p. 57), and Hedden (2012) for views along these lines. Without this assumption, logical truths would have implications about which actions are more promising than others—[accept a bet that will pay $1 if P ∨ ∼ P] would have to be more promising than [accept a bet that will pay $1 if P ∧ ∼ P]. I suspect my arguments could be reworked without this assumption, but matters would become more complicated.

Probability and Prodigality

91

logical truths with empirical propositions also have probability 1. But nothing in the probability calculus, and nothing in decision theory, requires that these bridge claims should have maximal probability. The view that knowledge has probability 1, on the other hand does require that bridge claims often have maximal probability, at least when it is married to views about knowledge according to which one can know everyday propositions even while being offered arbitrarily unfavorable bets. So while decision theory alone does not lead to the prodigality problem, decision theory, the claim that knowledge has probability 1, and some plausible assumptions about knowledge, together do. Before considering some ways of trying to reconcile the knowledge/ probability 1 link with decision theory by adopting alternative theories about knowledge, I will consider two objections to my claim that decision theory alone doesn’t lead to prodigality.

3.1.1. Objection 1: Even Contingent Claims Must Have Probability 1 My defense of decision theory from prodigality relied on the claim that while decision theory requires logical truths to have probability 1, it doesn’t require bridge principles linking claims about logical truths to claims about betting outcomes to have probability 1, as such principles are not themselves logical truths. It might be objected, however, that this merely postpones the problem—in some cases, decision theory requires that even propositions that are contingent, non-logical truths (or even falsehoods), have probability 1; in particular, this is so in cases where the event space over which probabilites are divided is infinite. Williamson (2007) argues that if a coin is to be flipped infinitely many times, the probability of any particular outcome (e.g. the coin landing heads every time) is zero, even though that such an outcome should occur is logically possible. Assuming his argument is sound, must a decision theorist bet at arbitrarily unfavorable odds that a coin flipped infinitely many times will not land heads every time? Perhaps, but not obviously. By the argument earlier in this section, this will not lead to prodigality, as long as crucial bridge claims do not have probability 1. And even if considerations about, for example, zero measure sets, force us to grant that some contingent a posteriori claims have epistemic probability 1, such considerations do not force us to grant that the relevant bridge claims must have probability 1. For instance, in a case where someone is offered the chance to bet on the outcome of an infinite sequence of coin flips, nothing in decision theory forces the epistemic probabilities of bridge claims linking the bookie’s behavior with the outcome of the coin tosses to get probability 1. Still, I am willing to grant that one could design an infinitary example in which the epistemic probabilities of contingent claims about some bookie’s behavior are forced all the way to 1. In such a case, decision theory would tell us to bet on any odds against a possible (albeit probability zero) outcome. Is this an unwelcome result?

92

Daniel Greco

Not obviously. Betting at arbitrarily unfavorable odds in ordinary, humdrum cases involving topics like the population of Canada is clearly irrational. But how one ought to behave in complex cases crucially involving infinities is far from apparent—my intuitions don’t speak strongly about such cases, and if they did, I would regard them with skepticism, since we have independent evidence that intuitions aren’t very reliable when it comes to infinity. The defender of decision theory, then, may adopt the attitude recommended by David Lewis concerning certain putative infinitary counterexamples to analyses of causation: I do not worry about these far-fetched cases. They go against what we take to be the ways of this world; they violate the presuppositions of our habits of thought; it would be no surprise if our common-sense judgments about them therefore went astray—spoils to the victor! (Lewis, 1986c)

The defender of decision theory can go on to add that if the war is fought on other ground, then decision theory is likely to win, given the powerful and systematic explanations it can provide of so many phenomena concerning rational choice. 3.1.2. Objection 2: Focusing on Bridge Principles Misidentifies the Problem I have argued that even if logical truths have epistemic probability 1, decision theory doesn’t recommend that one bet on them at arbitrarily unfavorable odds. But it may seem that this addresses the letter of the prodigality problem without addressing the spirit. One might object that the reason we oughtn’t to bet on logical truths at arbitrarily unfavorable odds isn’t that we can’t be sure whether the bookie will pay up, but is instead that we can’t be certain in the logical truths themselves.20 If the epistemic probability of a proposition is supposed to reflect something like the intuitive notion of how well supported that proposition is in light of our evidence, then according to this objection it simply isn’t true that logical truths have maximal epistemic probability. While I grant that this objection has intuitive force, much of that force is dissipated, I think, once we realize just how wide a range of actions we can agree are irrational consistent with maintaining that logical truths have epistemic probability 1. For instance, in the face of disagreement, we needn’t hold that expected utility maximizing agents ought to confidently assert the sentences they take to express those truths. Claims about which logical propositions are expressed by which sentences are empirical claims that need not have epistemic probability 1. Forceful disagreement from one’s logical superiors is, plausibly, exactly the sort of phenomenon that should lead one to revise one’s confidence that one hasn’t misunderstood the meaning of the sentence one initially took to be necessarily true. Once we have a clear picture of how an expected-utility maximizing agent will conduct herself if logical truths have 20

See, e.g. Christensen (2004).

Probability and Prodigality

93

epistemic probability 1, but sub-maximal epistemic probabilities are associated with biconditionals linking logical truths with propositions about the assertibility of sentences, the outcomes of bets, and so on, we may be inclined to think that such an agent’s conduct as a whole seems perfectly reasonable and undogmatic. And if that is our reaction, perhaps we needn’t insist that there is nevertheless some residual problem associated with holding that logical truths have maximal epistemic probability. More generally, there are reasons completely independent of decision theory for thinking that uncertainty concerning logical truths should be treated differently from uncertainty concerning empirical propositions. If we follow Stalnaker (1984) and Lewis (1986a) in modeling beliefs with sets of possible worlds, then we will already want some special treatment of logical uncertainty, since logical truths are true in all possible worlds, and so on a crude version of the possible worlds account, they are automatically believed by all agents. To take one example already alluded to, we might treat some cases of ignorance about logical truths as metalinguistic ignorance; nothing in decision theory requires metalinguistic propositions to receive maximal probability.21 Prodigality does not automatically threaten all decision theorists. We can avoid prodigality while retaining decision theory, so long as we don’t accept that knowledge has probability 1. In the next section, however, I will consider responses that try to reconcile the claim that knowledge has probability 1 with decision theory, while avoiding prodigality.

4.

C O N T E X T UA L I S M , S E N S I T I V E I N VA R I A N T I S M , A N D R AT I O N A L C H O I C E

In my presentation of the prodigality problem, I assumed a view about knowledge according to which anti-skeptical insensitive invariantism is correct. Anti-skeptical insensitive invariantism is called “anti-skeptical” because it says that it is possible to know everyday propositions about matters like the population of Canada, “insensitive” because it says that whether we know isn’t sensitive to practical stakes, and “invariantist” because it says that which propositions we express by uttering sentences containing “knows” doesn’t vary from context to context.22 In particular, I assumed that it is typically the case that when you know something, you will continue to know it when you are offered the chance to bet on it at very unfavorable odds. But many epistemologists reject this assumption, along with insensitive invariantism more 21 Whether such a strategy can be successfully carried out is a controversial matter. See Field (1986) for convincing arguments to the effect that solving the problem of logical omniscience requires more than just recognizing the possibility of metalinguistic ignorance. My own view is that if we understand logical ignorance as a combination of metalinguistic ignorance and fragmentation (in the sense discussed by Lewis (1982) and Stalnaker (1991, 1999)), the problems Field raises look more tractable. 22 See Hawthorne (2004) and Stanley (2005) for some helpful taxonomizing with respect to these terminological issues.

94

Daniel Greco

generally.23 It might seem that views about knowledge other than insensitive invariantism would be well positioned to reconcile the claim that knowledge has probability 1 with decision theory. In this section I will discuss two such strategies for achieving a reconcilliation—one contextualist, one sensitive invariantist. I will argue that neither strategy succeeds, though the bulk of my discussion will focus on sensitive invariantism. 4.1. Contextualism and Choiceworthiness According to contextualists, which proposition one expresses by uttering a sentence of the form “S knows that P” depends on features of one’s context, often including how practically important the parties to one’s conversation take the truth of P to be.24 If we accept a link between knowledge and probability 1, then epistemic probability will be similarly context-sensitive. This might seem to give us the resources we need to handle the prodigality problem. In particular, contextualism seems to allow us to hold that within a context, what counts as “known” also counts as having “epistemic probability” 1, while also holding that the following two sorts of claims are both typically true: 1. Claims made in ordinary contexts to the effect that people know various things. 2. Claims made in contexts in which bets at extremely unfavorable odds have been offered to the effect that people oughtn’t accept such bets, even though they are bets on claims that we would ordinarily take ourselves to know. The contextualist can accept the latter claims because she needn’t hold that propositions which we “know” in ordinary contexts can be truly described as having “epistemic probability” 1 in contexts in which extremely high stakes bets are under consideration. Not all versions of contextualism are able to allow that epistemic probability inherits the context-sensitivity of knowledge. Some natural ways of developing contextualist theories of knowledge involve taking epistemic probability as insensitive to context, and holding that whether a proposition is known depends on how features of context interact with epistemic probabilities. For instance, Stewart Cohen (1988) defends a contextualist version of the relevant alternatives theory of knowledge; on his view epistemic probability is taken as fixed independently of context, and how probable an alternative has to be before it counts as relevant is determined by context. More simply, we might treat epistemic probability as fixed independently of context, and hold that propositions are known only if their probability exceeds some contextually 23 24

I won’t be questioning the anti-skeptical part of anti-skeptical insensitive invariantism. See e.g. DeRose’s (1992) discussion of what have come to be called “bank cases.”

Probability and Prodigality

95

determined threshold—in contexts in which skeptical scenarios are salient or high-stakes bets have been offered, the threshold might be quite high, while in normal contexts it would be lower. While this sort of strategy isn’t the only one available to the contextualist, it is a natural one, and it is incompatible with the claim that knowledge has probability 1. Let us assume, however, that there are versions of contextualism that are consistent with the claim that knowledge has probability 1.25 If epistemic probability is context sensitive, then it is hard to see how rational choice worthiness could fail to be context sensitive as well. While it is possible to come up with examples in which context insensitive notions can be analyzed in terms of context sensitive ones, the examples are unusual, and seem disanalogous to the present case.26 But once we admit that rational choiceworthiness is context sensitive, we run into problems. Rational choiceworthiness is a concept that has roles to play in both third-person evaluation of agents’ choices, and first-person deliberation about which option to choose. Both of these roles are ill-suited to be played by a context-sensitive concept. Let us first consider the third-person, evaluative role of the concept of rational choiceworthiness. If we are contextualists about rational choiceworthiness, then we can use the notion of rational choiceworthiness to assess an agent S’s action, but whether we assess S positively or negatively will depend on our context. If we are in a context in which “knowledge” is hard(easy) to have, we will treat S as “knowing” less(more), and will evaluate her actions relative to the associated epistemic probability function. This would seem to allow irrelevant factors to influence our evaluations, at least if we allow that features of context that determine the extension of “knows” typically appealed to by contextualists—for example, which possibilities are salient in the conversation—in fact do determine the extension.27 If I am trying to decide whether some agent S made a good decision, it shouldn’t matter what possibilities are salient in my conversational context; I am asking a question to which the only relevant factors concern S’s situation, not my own. Perhaps the contextualist can find some way to finesse the above worries— they are sufficiently similar to standard objections to contextualism that we might expect that standard responses will serve.28 Where the contextualist 25 David Lewis (1999) defends a version of contextualism that, as far as I can tell, doesn’t require that epistemic probability be insensitive to context. 26 For instance, we might (pointlessly) analyze “even number” as “large number divisible by two, or non-large number divisible by two.” “Large” is context sensitive, but even though a context-sensitive term appeared in our analysans, it doesn’t make for context-sensitivity in the analysandum (at least if we assme that “large” and “non-large” partition the numbers in every context). In any context, the analysis will give the same results as the simpler analysis of even number as “number divisible by two.” But it seems unlikely that anything like this is going on in the case of a knowledge/probability 1 link, contextualism about knowledge, and choiceworthiness. Thanks to Alan Hájek for the example. 27 See e.g. Lewis (1999). 28 See DeRose (1992, fn. 10).

96

Daniel Greco

has special difficulty is in accounting for the first-person, deliberative role of the concept of rational choiceworthiness. If we are contextualists about knowledge and rational choiceworthiness, we will recognize that there is in fact a family of context insensitive notions of rational choiceworthiness—there is one for each set of possible contextually fixed standards for knowledge, none of which seem to be privileged in any interesting sense. This makes it hard to see how to use the notion of choiceworthiness in deliberation. I could try to perform the action that is truly describable as “choiceworthy” in my context. But there is nothing special about my context—if my conversational partners and I were to start considering more (or fewer) possibilities, a different action might be truly describable as “choiceworthy,” even though nothing that is (intuitively) practically relevant would have changed. If there is nothing practically privileged about either context, or either extension of “choiceworthy,” then it is hard to see why I should prefer to act in the way that is truly describable as “choiceworthy” in my context. If all that was at stake was how to use language (e.g. whether to describe someone as “knowing”) then we might be perfectly happy to allow that considering more (or fewer) possibilities would change our (merely linguistic) behavior. But in cases where there is much of practical import at stake, letting our decisions be sensitive to features of our conversational context seems perverse. The above considerations suggest that a contextualist view about knowledge attributions that emphasizes features of the conversational context of the person making the knowledge attributions—as traditional contextualist views have—won’t extend happily to a contextualist view about choiceworthiness. But accepting a link between knowledge and probability 1, a decision theoretic account of choiceworthiness, and contextualism about knowledge requires accepting a contextualist view about choiceworthiness. Therefore, traditional contextualist views are ill-suited to reconcile decision theory with a link between knowledge and probability 1.29 I believe this points to a more general tension in the theory of knowledge. One desideratum for a theory of knowledge is that it entail (or at least be consistent with) the claim that knowledge attributions are typically true.30 Developing a theory of knowledge on which knowledge attributions are typically true can seem to require holding that the truth of knowledge attributions is somehow sensitive to facts concerning which possibilities the parties in some conversation have mentioned, or are considering.31 Another 29 I don’t rule out that non-traditional versions of contextualism—in particular, versions that reject David Lewis’ “Rule of Attention” (1999)—might be able to avoid the difficulties I’ve been discussing. See Williams (2001). 30 DeRose (1992) writes that “the obvious attraction of contextualism . . . is that it seems to have the result that very many of the knowledge attributions and denials uttered by speakers of English are true—more than any form of invariantism can allow for.” 31 Perhaps the conversation of the knowledge attributer, perhaps the conversation of the subject of the knowledge attribution, perhaps the conversation of the person assessing the truth of the knowledge attribution.

Probability and Prodigality

97

desideratum for a theory of knowledge is that a theory of knowledge should explain the links between knowledge and rational action. But developing a theory of knowledge on which there is a close connection between knowledge and action—for instance, developing a view according to which knowledge has probability 1—seems to require holding that the truth of knowledge attributions not be sensitive to these facts about conversational contexts, since such facts are typically of no practical relevance: they don’t systematically bear on how it is rational to act. Perhaps this isn’t a deep tension—there may be good reasons to reject one of the putative desiderata mentioned above, or the desiderata may ultimately be jointly satisfiable.32 In the case of contextualism, however, the tension is real; the features of contextualism that (at least seem to) allow it to accommodate the idea that knowledge attributions are typically true are the very same features that make trouble when we try to accommodate claims about links between knowledge and rational action. We have already seen that contextualists have difficulty accommodating a knowledge/probability 1 link. Subject sensitive invariantism, however, emphasizes the respects in which the truth of knowledge attributions is (supposedly) sensitive, not to the conversational context of the attributer, but to the practical situation of the subject who is said to know. Such a view might seem better placed than contextualism to accommodate links between knowledge and rational action, and it is to this view that we now turn. 4.2. SSI and Dutch books Unlike contextualists, sensitive invariantists hold that which proposition one expresses by uttering a sentence of the form “S knows that P” is the same across different contexts. However, they hold that the truth of the proposition expressed is sensitive to features not appealed to by traditional epistemological views. Sensitive invariantists hold that whether a subject knows some proposition depends not only on “truth-conducive factors,” but also on features of a subject’s practical situation.33 For instance, if I say “Susie knows that the coin is biased,” traditional views might have it that the truth of my utterance is sensitive to facts such as whether Susie has observed a distribution of heads and tails that would be unlikely if the coin were fair, whether Susie has received testimony to the effect that the coin is biased, whether Susie has subjected the coin to various tests, and so on. On traditional views, my utterance will not be sensitive to facts concerning how important it is for Susie whether the coin is biased. On sensitive invariantist views, both 32 Williamson (2005a) argues that any theory of knowledge will have to attribute a great deal of metalinguistic error to speakers, and that it isn’t obvious that contextualism and sensitive invariantism attribute substantially fewer or less significant errors than insensitive invariantism. 33 The phrase “truth-conducive factors” is from Stanley (2005).

98

Daniel Greco

sorts of facts are relevant. Even if Susie has seen the coin land heads 600 out of 1,000 times, and has received reliable testimony that it is biased, she might not know that it is biased if the proposition that the coin is biased is one on which a great deal of practical importance rests. For instance, a sensitive invariantist would hold that if Susie has been offered a bet at extremely unfavorable odds on the proposition that the coin is biased, it is much harder for her to know that the coin is biased than it is if there is less of practical importance at stake. Sensitive invariantism might seem well placed to reconcile decision theoretic accounts of choiceworthiness with the claim that knowledge has probability 1. Unlike insensitive invariantism, sensitive invariantism doesn’t seem to face the prodigality problem—a sensitive invariantist can agree that while we ordinarily know a great deal, we needn’t accept high-stakes bets at arbitrarily unfavorable odds, since being offered such bets destroys our knowledge; once we have been offered such bets, the claims we are betting on will not have probability 1. Unlike contextualism, sensitive invariantism seems to allow choiceworthiness to depend only on factors that are intuitively relevant—by emphasizing facts about a subject’s practical situation, rather than facts about conversational salience, sensitive invariantism can make sense of why choiceworthiness is a good standard for evaluating a third party’s actions, and a good thing to aim at in deliberation.34 One obstacle to sensitive invariantist strategies for saving the claim that knowledge has probability 1 is similar to one mentioned earlier for contextualism—some ways of implementing the sensitive invariantist proposal involve taking epistemic probability as fixed, and holding that whether one knows depends on the interaction of epistemic probability with practical factors. Fantl and McGrath (2009) employ a strategy along these lines, as does Stanley (2005, p. 87).35 Because Stanley ultimately accepts the claim that knowledge has probability 1, however, he takes this presentation to be merely a convenient oversimplification, and not the official version of his view. In what follows, I won’t focus on problems like this one—I assume there are ways of developing sensitive invariantism that don’t require an insensitive notion of epistemic probability—and will instead argue that sensitive invari-

34 This may be a bit too charitable to the sensitive invariantist. The sensitive invariantist who holds that knowledge has probability 1 could hold that facts about conversational salience are relevant to the truth of knowledge attributions, and differ from the contextualist only in thinking that the relevant salience facts concern what is salient in the subject’s context, rather than the attributer’s context. If she takes this route, then she will have just as much trouble making sense of the deliberative role of the concept of choiceworthiness as will the contextualist. That is, the tension I pointed to at the end of the last section will be just as much of a difficulty for the sensitive invariantist as for the contextualist. 35 Another example involves Weatherson (2012), who argues that knowledge is sensitive, not to the stakes of the bets one faces, but to the odds of those bets. As I understand his position, it requires that facts about subject’s epistemic probabilities are explanatorily prior to facts about what they know; his theory explains what you know in terms of facts about what your epistemic probabilities are, together with facts about what practical options are open to you. Like Fantl and McGrath’s view, Weatherson’s requires the failure of the knowledge-probability 1 link.

Probability and Prodigality

99

antism leads to implausible claims about rational choice when combined with the claim that knowledge has probability 1. Suppose we have access to a device that randomly generates a natural number between 1 and 3, inclusive, with each number having an equal objective chance of being generated. We set the device working, and before looking to see what number is generated, we bet on its outcome. I offer you a bet at 1:2 odds that it generated the number 1. If we are non-skeptics, we should allow that you can know that the device really did generate a number between 1 and 3, and that each number had an equal chance of being generated. And if we accept that knowledge has probability 1, we should also agree that given that you know these things, they have probability 1, so evaluated relative to your epistemic probability function, the bet is fair. Suppose you take it, betting $1 against my $2. Now I make another offer with the following conditions. I will give you a penny if it generated a 1, 2, or 3, but I will take your life savings if it did not generate any of these numbers. You should not take this bet, and SSI has a say about why. Now it is much more important to you whether the machine did generate a 1, 2, or 3, and raising the stakes of betting on this proposition is exactly the sort of thing that can destroy one’s knowledge according to SSI. So the subject sensitive invariantist can grant that while earlier you knew that the machine had generated a 1, 2, or 3, now that this question has become one of great practical importance, you don’t (perhaps possibilities that previously had epistemic probability zero, e.g. that the device malfunctioned, now have positive probability). However, now that my second offer has been made and your knowledge that the device will generate one of the first three natural numbers has been destroyed, your earlier bet doesn’t look quite as good as it did before.36 The probability that the device generated the number one is now less than onethird since there is a chance that it didn’t generate any number at all.37 I offer you the chance to cancel your earlier bet, for a small fee, and you take it, as canceling the bet now has positive expected utility for you in light of your new probabiltiy function. All in all, without my having any information that you lack, I have extracted a small fee from you for nothing in return, by offering you a series of deals each of which you regarded as fair at the time.38

36 One mechanism by which this could work is that my offering you the bet could straightforwardly provide you with evidence that the machine sometimes generates numbers other than 1, 2, or 3. Often the fact that somebody is willing to bet on a claim is some evidence about its truth. But we may assume that nothing like that is going on in this case. 37 I assume that the relative probabilities of the individual numbers being generated haven’t changed. While nothing I’ve said so far forces that assumption, and on some ways of fleshing out the case we might reject it, I assume that the example can be set up in such a way that it is a reasonable assumption to make. 38 A defender of SSI might object that while accepting the first bet is rationally permissible given your utility function, it is also permissible for you to reject it. If Dutch book worries only indicate irrationality when there is a series of bets each of which is rationally mandatory given one’s utility function, but which together guarantee a sure loss, then my initial example doesn’t

100

Daniel Greco

The problem that the above example illustrates is this: if the probabilities that govern an agent’s choices are sensitive to the stakes of the bets she’s offered, then the agent is predictably manipulable by a bookie with the ability to offer high-stakes bets. Since this kind of predictable manipulability is irrational, SSI together with the claim that knowledge has probability 1 entails false conclusions about rational choice.39 It shouldn’t be surprising that the defender of SSI is committed to recommending that agents accept Dutch books if she also accepts that knowledge has probability 1. According to SSI together with the claim that knowledge has probability 1, epistemic probabilities change when the stakes go up, but this change does not occur through conditionalization. But there are familiar arguments to the effect that agents who update the probabilities that guide their actions by rules other than conditionalization may be vulnerable to Dutch books; the argument I’ve given in this section can be seen as pointing to yet another example in which using an update rule other than conditionalization can get one into trouble.40 Should the defender of SSI and the knowledge/probability 1 link bite this bullet and endorse accepting Dutch books, or does she have a way out? One potential option for the sensitive invariantist who wants to save the claim that knowledge has probability 1 is to employ a strategy similar to the one that I used in §3.1 to argue that decision theory alone isn’t enough to generate prodigality. The defender of SSI can claim that when the second bet is offered, it is not just one’s knowledge that the machine generated a 1, 2, or 3 that is destroyed, but also one’s knowledge of the bridge principles necessary to make canceling the initial bet rationally required. The defender of SSI could claim that after the second bet has been offered, one not only no longer knows that the machine generated a 1, 2, or 3, but one also fails to know that one will lose the first bet just in case the machine failed to generate a 1, 2, or 3. At the very least, nothing in the machinery of SSI seems to rule out this response. And this response allows for the possibility that canceling the first bet isn’t rationally required. While the letter of the initial problem may have been avoided, a serious difficulty remains. Being offered the second bet should have no effect on one’s willingness to hold onto the first bet, and our theory of rational choice should explain why. While SSI may not be straightforwardly committed to endorsing the acceptance of Dutch books in cases like these, it lacks the resources to explain why the attractiveness of holding onto the first bet should

threaten SSI. But we could have set up the example so that the bets were mandatory rather than merely permissible, at the cost of some elegance. 39 There is a vast literature on Dutch book arguments, and it is far from uncontroversial exactly what sort of rational shortcoming is indicated by susceptibility to a Dutch book. That susceptibility typically indicates some sort of rational shortcoming, however, is relatively less controversial. See Christensen (1996). 40 See Teller (1973), who credits David Lewis.

Probability and Prodigality

101

have nothing to do with whether or not one has been offered the second. Rather than offering such an explanation, SSI seems to suggest that one’s willingness to stick to earlier bets should systematically depend on which bets one is offered later; according to decision theory together with claims linking knowledge and probability, one’s actions should depend on what one knows, and according to SSI, what one knows depends on which bets one has been offered. What we would like to say is that the epistemic probabilities of the device having generated various outcomes are unaffected by the fact that the second offer has been made, but SSI doesn’t seem to let us say this, at least once it is combined with claims linking knowledge and probability.41 This points to a more general problem involved in combining SSI with decision theory and the claim that knowledge has probability 1, which I discuss in the next section. 4.3. SSI, Probabilities, and Utilities Part of the appeal of the decision theoretic apparatus is that it lets us separate questions about the values of outcomes from questions about how likely it is that various outcomes will come about. Once we have separated considerations concerning utility from considerations concerning probability, we can isolate the contributions that each of these factors make to determining rational choiceworthiness. For example, a course of action might come to be more choiceworthy either because one of the outcomes it might lead to becomes more valuable, or because the probability that it will lead to a favorable outcome might increase (while the value of each possible outcome remains constant). When we marry the claim that knowlede has probability 1 to SSI, however, probabilities and utilities become interdependent in a way that makes it impossible for us to isolate the roles that they play in determining facts about rational choiceworthiness. For instance, suppose, in a variant on the example from the previous section, I first offer you the chance to bet a dollar at even odds on the claim that the device generated either a 1, a 2, or a 3. Obviously, you ought to take the bet. Next, I make the knowledge-destroying offer—I offer you a bet 41 One possibility I have not considered is that the defender of SSI might modify her theory and stop treating knowledge as a relation between a subject and a proposition, but instead as a relation between a subject, a proposition, and an action. That is, the defender of SSI might allow that one can know that P for the purpose of performing some actions, but not for the purpose of performing others. She could then hold that for the purposes of deciding whether or not to accept the first bet, one knows that the machine will generate a 1, 2, or 3, but not for the purpose of deciding whether or not to accept the second bet. She could also hold that being offered new bets typically doesn’t change whether or not you know something for the purpose of accepting bets you have already been offered. While this move would avoid the problems discussed above, it is relatively radical. More importantly, like SSI, it faces a problem which I discuss in §4.3. Thanks to Agustín Rayo (who argues for a view along similar lines in response to a different problem in his 2011 paper, “A Puzzle about Ineffable Propositions”) for suggesting this strategy on behalf of the sensitive invariantist.

102

Daniel Greco

on the same proposition, but with the potential pay-off of a penny, and the potential downside of losing your life savings. You ought not take this new bet. How is this bet different from the earlier one? According to SSI, two things have changed—the disutility of one of the potential outcomes of the bet has increased (since there is now the potential downside of losing your life savings, rather than just $1), and the probability that the favorable potential outcome of the bet will be realized has decreased (since there is no longer a probability of 1 that the favorable outcome will be realized). But we can’t separate out these changes—they are inextricably linked. I take it that this is a drawback of the combination of SSI with the claim that knowledge has probability 1. Intuitively, questions about probabilities and questions about utilities are distinct, and we should be able to separate factors responsible for changing probabilities from those responsible for changing utlities. Even if we could live with the phenomenon noted in the previous section, accepting SSI and the claim that knowledge has probability 1 would make decision theory much less well suited to offering an illuminating explanation of how facts about one’s evidence and facts about value come together to rationalize choices.42

5.

I D E A L I Z AT I O N

In this last section I will consider a potential response on behalf of the sensitive invariantist that will place the debate over whether knowledge has probability 1 in the context of a broader debate about the role of idealization in our theorizing about epistemology and practical rationality. Keeping track of small probabilities is hard. If I am trying to decide whether to take the train or the bus, even if there is some tiny probability that the bus will be hijacked by terrorists, or that the train will derail, it is much easier (and more sensible) to ignore these probabilities in my deliberations than it is to factor each of them in. We might see sensitive invariantism combined with the claim that knowledge has probability 1 as a suggestion for how to ignore

42 Hájek (2006) offers a similar criticism of a certain response to the Saint Petersbug paradox. According to this response, while the utilities a rational agent assigns to various outcomes can increase without bound, the expected utilities assigned to gambles must always be finite; on this accout, one can never rationally take oneself to be playing a Saint Petersburg game. He argues that this creates an illegitimate dependence between probabilities and utilities—someone who changes her mind about the value of an outcome may be directly forced to revise her confidence that it will occur downwards. While in some cases indirect revision of this sort is rational, it is only rational when there is some sort of evidential link between probabilities and utilities, e.g. if you find out that a state lottery is offering a very large jackpot, you may reasonably revise your confidence that you will win downwards, given that you know that the state will only offer lotteries that raise revenues. What wouldn’t be rational would be to revise one’s probabilities as a result of revising one’s utilities directly, in the absence of such an evidential link. This is what the response to the Saint Petersburg paradox considered by Hájek amounts to, and it is also what seems to be recommended by SSI together with the claim that knowledge has probability 1.

Probability and Prodigality

103

them—except in contexts where (prior to considering them) it seems likely that considering them might change our minds about what to do, we should treat ourselves as knowing that they won’t obtain, and should assign them probability 0 when we are calculating expected utilities. After all, considering such possibilities costs brainpower, and there are usually better things we could be doing with our time. If we understand things this way, then it will look as if views on which propositions we ordinarily take ourselves to know have high but nonmaximal epistemic probabilities are views that idealize away from computational costs, while views like sensitive invariantism (when combined with the claim that knowledge has probability 1) eschew this idealization.43 This would allow the sensitive invariantist to explain why—as it came out in §4.2—agents that act in accordance with her recommendations are predictably manipulable; they are predictably manipulable by agents who have more computational resources than them. It is not surprising that agents who have the time and computational resources to consider more possibilities than you can will be able to come up with strategies for offering you bets that will leave you predictably worse off—if they rely on the fact that you are ignoring certain possibilities (while they are taking them into account), and then offer you bets that exploit your oversight, they can put you at a disadvantage. Understood this way, arguments against the claim that knowledge has probability 1 are only attractive insofar as we are ignoring actual agents’ limitations—a realistic, useful theory of practical rationality will accept that knowledge has probability 1. This take on the debate underestimates the resources available to the decision theorist who allows that known propositions have high but sub-maximal probability—such a theorist needn’t be understood as idealizing away from the computational costs of various courses of deliberation. In the remainder of this paper I will suggest a way for the opponent of the view that knowledge has probability 1 to find some room in her story for computational costs. Just as we can use decision theory to evaluate actions like taking the train rather than the bus, we can use decision theory to evaluate long-term deliberative strategies such as taking into account a very wide range of possibilities when deliberating rather than only considering a narrower set.44 Even if we allow that propositions we ordinarily take ourselves to know don’t have

43 Of course, there is another perspective from which it looks like it is the sensitive invariantist who is idealizing—we might say that she is the one who is idealizing by treating small probabilities as if they were zero. This looks a lot like standard ways of idealizing, e.g. treating small frictional forces as if they were non-existent. 44 We might argue that one can’t “directly” bring about that one pursues a long-term strategy, and conclude from this that we shouldn’t use decision theory to evaluate long-term strategies (perhaps because one accepts some of the arguments alluded to in footnote 19). If the reader is worried about this issue, she can instead take the present section to be pointing out that we can use decision theory to evaluate, not only decisions about what bodily actions to perform, but also decisions about how to deliberate.

104

Daniel Greco

maximal probability, we can hold that the long-term strategy of considering possibilities in which those propositions are false (and attempting to think about probabilities of those possibilities in deciding what to do) will have submaximal expected utility. This is because factoring such possibilities into our deliberations costs time and brainpower, and usually doesn’t have any benefit (since it usually won’t change what we do). This allows for an interesting sort of case—it might be that an action ϕ fails to maximize expected utility, even though engaging in a strategy for deliberating that leads to one’s ϕ-ing does maximize expected utility. In those rare cases where which action maximizes expected utility does depend on what happens in far-out possibilities where propositions ordinarily taken to be known are false, it may be that the strategy of ignoring possibilities like those ones itself maximizes expected utility, even though the decision we go on to make about what to do (after having ignored those possibilities) will not maximize expected utility. In fact, the example discussed in the previous section is plausibly such a case. The opponent of the claim that knowledge has probability 1 can hold that throughout the example, the proposition that the device generated a 1, 2, or 3 has high but sub-maximal probability for the subject facing the bets. When no high-stakes bets have been offered, however, it may be rational for the subject to ignore the possibility that this proposition is false in conducting her deliberations, and treat the proposition as if it had probability 1. After all, there is some cost associated with thinking about such possibilities, and in situations in which it is unlikely that considering them will provide any significant benefit, the expected utility of factoring such possibilities into one’s deliberations might be less than the expected utility of ignoring them and rounding high probabilities to 1. Because of this, the decision to accept the first bet—the bet that pays off at 1:2 odds if the machine generated a 1—is a decision that is the upshot of a rational decision about how to deliberate, but which is not itself a rational decision; the expected utility of accepting the bet will be slightly negative if the probability that the machine generated a 1 is slightly less than one-third, as it plausibly is if the probability that it generated any number at all is slightly less than 1.45 Similar cases are familiar from the literature in game theory. It might be that binding oneself to a certain strategy maximizes expected utility, even though actually acting on that strategy in particular cases does not maximize expected utility. For instance, binding oneself to the strategy of always ignoring threats

45 We can also use this framework to say plausible things about the decision to reconsider the first bet. Once the second bet has been offered, it clearly pays to consider the outlandish possibilities in which the machine generates no number at all. And as long as one is already devoting time and computational resources to considering those possibilities for the purpose of determining whether to accept the second bet, one may as well consider (at little extra cost) how their existence bears on the question of whether to cancel the first bet.

Probability and Prodigality

105

might be a good idea because it makes people less likely to threaten you, even though in some particular cases you would be better off complying with a threat—the costs of the threat’s being carried out might be quite high and the effects on one’s reputation as someone who gives into threats could be small or non-existent.46 While such cases aren’t entirely uncontroversial, many philosophers hold that, even if some strategy is generally a good one to follow (and even if it would make sense, ex ante, to bind oneself to acting in accordance with that strategy in all cases), if acting in accordance with it in some particular case would predictably have bad effects on net, it would be irrational to do so.47 Once we acknowledge the possibility of such cases, the view that propositions we ordinarily take ourselves to know have high but sub-maximal probability looks a good deal better. Such a view can give an appropriately nuanced treatment of cases in which agents who ignore far-fetched scenarios in their deliberations make irrational decisions and are thereby predictably manipulable. The view can positively evaluate such agents by holding that they are rational to have a policy of ignoring such scenarios when engaging in deliberation. However, the view can also negatively evaluate such agents by holding that the actions they ultimately perform, having ignored such scenarios, are not rationally choiceworthy. This combination is more attractive than the verdicts of the view on which both SSI and the knowledge/probability 1 link hold; that view seems to entail that such agents are rationally unimpeachable. Ultimately, denying that knowledge has probability 1 and holding that propositions we ordinarily take ourselves to know in fact have high but sub-maximal probability needn’t lead to an uninteresting, unrealistic view that is irrelevant to the concerns of computationally limited agents. To the contrary, such a theory can explain how it is rational to cope with our computational limitations (by positively evaluating the strategy of ignoring farout possibilities), while also acknowledging that rational decisions sometimes beget irrational ones, as in cases where ignoring such possibilities makes us predictably manipulable.

REFERENCES

Broome, J. (1995). Weighing Goods: Equality, Uncertainty and Time. Blackwell, Oxford. Christensen, D. (1996). Dutch-Book Arguments Depragmatized: Epistemic Consistency for Partial Believers. Journal of Philosophy, 93(9):450–79. (2004). Putting Logic in its Place. Oxford University Press, Oxford.

46

See Schelling (1960, 1966). See Parft (1984) and Kelly (2002, 2004). For a contrary view, see Gauthier (1986). The way I have put things in the text presupposes that some version of consequentialism is correct, but this is inessential. 47

106

Daniel Greco

Cohen, S. (1986). Knowledge and Context. Journal of Philosophy, 83(10):574–83. (1988). How to be a Fallibilist. Philosophical Perspectives, 2:91–123. DeRose, K. (1992). Contextualism and Knowledge Attributions. Philosophy and Phenomenological Research, 52(4):913–29. (1995). Solving the Skeptical Problem. Philosophical Review, 104(1):1–52. (2004). The Problem with Subject-Sensitive Invariantism. Philosophy and Phenomenological Research, 68(2):346–50. Easwaran, K. (Draft). Regularity and Hyperreal Credences. Fantl, J. and McGrath, M. (2009). Knowledge in an Uncertain World. Oxford University Press, Oxford. Field, H. (1986). Stalnaker on Intentionality: On Robert Stalnaker’s Inquiry. Pacific Philosophical Quarterly, 67:98–112. Garber, D. (1982). Old Evidence and Logical Omniscience in Bayesian Confirmation Theory. In Earman, J., editor, Testing Scientific Theories. University of Minnesota Press, Minneapolis. Gauthier, D. P. (1986). Morals by Agreement. Oxford University Press, Oxford. Gibbard, A. (1990). Wise Choices, Apt Feelings. Harvard University Press, Cambridge MA. Hájek, A. (2006). In Memory of Richard Je_rey: Some Reminiscences and Some Reflections on The Logic of Decision. Philosophy of Science, 73(5):947–58. (Draft). Staying Regular. Hawthorne, J. (2004). Knowledge and Lotteries. Oxford University Press, Oxford. and Stanley, J. (2008). Knowledge and Action. Journal of Philosophy, 105(10): 571–90. Hedden, B. (2012). Options and the Subjective Ought. Philosophical Studies, 158(2):343–360. Howson, C. and Urbach, P. (1996). Scientific Reasoning: The Bayesian Approach. Open Court, La Salle. IL/ Jeffrey, R. (1965). The Logic of Decision. University of Chicago Press, Chicago. Joyce, J. (1999). The Foundations of Causal Decision Theory. Cambridge University Press, Cambridge. Kaplan, M. (2009). Williamson’s Casual Approach to Probabilism. In Greenough, P. and Pritchard, D., editors, Williamson on Knowledge. Oxford University Press, Oxford. Kelly, T. (2002). The Rationality of Belief and Some Other Propositional Attitudes. Philosophical Studies, 110:163–96. ˆ (2004). Sunk Costs, Rationality, and Acting for the Sake of the Past. Nous, 38(1):60–85. Lasonen-Aarnio, M. (2010). Unreasonable Knowledge. Philosophical Perspectives, 24(1):1–21. Lewis, D. (1981). Causal Decision Theory. Australasian Journal of Philosophy, 59(1): 5–30. ˆ 16(3):431–41. (1982). Logic for Equivocators. Nous, (1986a). On the Plurality of Worlds. Blackwell, London. (1986b). Philosophical Papers, volume II. Oxford University Press, Oxford. (1986c). Postscripts to ‘Causation’. In Lewis 1986b. (1999). Elusive Knowledge. In Lewis, Papers in Metaphysics and Epistemology. Cambridge University Press, Cambridge.

Probability and Prodigality

107

Mellor, D. H. (2005). Probability: A Philosophical Introduction. Routledge, London. Parfit, D. A. (1984). Reasons and Persons. Oxford University Press, Oxford. Ramsey, F. (1931). Truth and Probability. In Braithwaite, R. B., editor, The Foundations of Mathematics and Other Logical Essays. Harcourt, Brace, and Co., New York. Rayo, A. (2011). A Puzzle about Ineffable Propositions. Australasian Journal of Philosophy, 89(2): 289–295. Rysiew, P. (2001). The Context-Sensitivity of Knowledge Attributions. Nous, 35(4):477–514. Schelling, T. (1960). The Strategy of Conict. Harvard University Press, Cambridge, MA. (1966). Arms and Inuence. Yale University Press, New Haven. Stalnaker, R. (1984). Inquiry. The MIT Press, Cambridge, MA. (1991). The Problem of Logical Omniscience, I. Synthese, 89(3):425–440. (1999). The Problem of Logical Omniscience II. In Stalnaker, R., Context and Content. Oxford University Press, Oxford. Stanley, J. (2005). Knowledge and Practical Interests. Oxford University Press, Oxford. Teller, P. (1973). Conditionalization and Observation. Synthese, 26(2):218–58. Weatherson, B. (2012). Defending Interest-Relative Invariantim. Logos and Episteme. 3:117–122. Williams, M. (2001). Contextualism, Externalism and Epistemic Standards. Philosophical Studies, 103(1):1–23. Williamson, T. (2000). Knowledge and its Limits. Oxford University Press, Oxford. (2005a). Contextualism, Subject-Sensitive Invariantism and Knowledge of Knowledge. Philosophical Quarterly, 55(219):213–35. (2005b). Replies to Commentators. Philosophy and Phenomenological Research, 70(2):468–91. (2007). How Probable is an Infinite Sequence of Heads? Analysis, 67(3):173–80. Williamson, T. (2009). Reply to Mark Kaplan. In Greenough, P. and Pritchard, D., editors, Williamson on Knowledge. Oxford University Press.

5. Essence and Natural Kinds: When Science Meets Preschooler Intuition1 Sarah-Jane Leslie

INTRODUCTION

It is common practice in philosophy to “rely on intuitions” in the course of an argument, or sometimes simply to establish a conclusion. One question that is therefore important to settle is: what is the source of these intuitions? Correspondingly: what is their epistemological status? Philosophical discussion often proceeds as though these intuitions stem from insight into the nature of things—as though they are born of rational reflection and judicious discernment. If these intuitions do not have some such status, then their role in philosophical theorizing rapidly becomes suspect. We would not, for example, wish to place philosophical weight on intuitions that are in effect the unreflective articulation of inchoate cognitive biases. Developmental psychology has discovered a range of belief sets that emerge in the first few years of life, and which plausibly go beyond the evidence to which the child has had access in that time period. In such cases, it is reasonable to suppose that the belief sets do not derive solely from the child’s rational reflection on her evidence, but rather show something about the way human beings are fundamentally disposed to see the world. (In some cases, the deep-seated dispositions are also shared with non-human animals.) There are many explanations of why we may be fundamentally disposed to see the world in a particular way, only one of which is that metaphysically or scientifically speaking, the world actually is that way. Another explanation may be that it is simply useful and practical to see the world that way—and this may be so even if it misleads us with respect to the metaphysical and scientific structure of reality. One particular way of carving up the world may be, say, efficient from the information processing point of view, without reflecting much about the fundamental nature of reality. Suppose, then, we find that a particular set of philosophical intuitions closely resembles such an early-emerging implicit belief set. This certainly does not establish the falsity of the intuitions, but it should give us reason to subject them to further scrutiny. This is because such a set of intuitions

1 Preparation of this article was supported by NSF grant BCS-1226942. I am indebted to Andrei Cimpian, Michael Devitt, Susan Gelman, Sally Haslanger, Elizabeth Harman, Mark Johnston, Paul Needham, Michael Weisberg, and an anonymous reviewer for helpful discussion, guidance, and invaluable feedback on earlier drafts. All the remaining mistakes are my own.

Essence and Natural Kinds

109

may strike us as completely compelling, and yet in fact reflect no more than a useful but ultimately misleading cognitive bias. Of course it may turn out that this is not the case; there is no argument from early developing to false. The point is purely epistemological: such a discovery should lead us to scrutinize the intuitions, and look for independent and converging evidence for the conclusions they urge.2 The present paper focuses on essentialism about natural kinds as a case study in order to illustrate this more general point. Saul Kripke and Hilary Putnam famously argued that natural kinds have essences, which are discovered by science, and which determine the extensions of our natural kind terms and concepts. This line of thought has been enormously influential in philosophy, and is often taken to have been established beyond doubt. The argument for the conclusion, however, makes critical use of intuitions, and I note that the intuitions are of the sort had by preschool children, and that they are traceable to a deep-seated cognitive outlook, which is often called “psychological essentialism.” Further, if we did not have such a cognitive outlook or implicit belief set—a belief set which is in fact in place by at least age 4—then we would not have the relevant philosophical intuitions. In light of this, I consider the question of whether natural kinds actually have scientifically discovered or discoverable essences, and whether these putative essences could determine the extensions of our terms and concepts as Kripke, Putnam, and many others have supposed. In fact, a number of philosophers of biology and chemistry have argued that biological and chemical kinds do not have such essences, yet these arguments—particularly in the case of chemistry—have not been assimilated by philosophers more generally. The reason for this poor assimilation is, I suggest, that the Kripke/Putnam view is just so intuitive. But this fact, I argue, is due to a deep-seated cognitive bias, rather than to any special insight into the nature of reality. The first half of the paper lays out the theory of psychological essentialism (which I rename quintessentialism for ease of exposition) in comprehensive detail, and summarizes some of the major experimental results that support it. I then consider further experimental evidence that suggests that lay people frequently misunderstand or misconstrue scientific findings as confirming their (quint)essentialist beliefs, when in fact the science does the opposite. Some of the more vivid and readily accessible cases involve quintessentialist beliefs concerning social kinds such as race and gender, and so such examples are also considered throughout the paper. The second half of the paper takes up philosophical essentialism about natural kinds, and argues that quintessentialist beliefs are required for the crucial Kripke/Putnam intuitions. I then review extensive findings in biology 2 The most general question here is the epistemological status of aspects of philosophical methodology, in light of recent psychological findings. Several theorists have recently begun to consider this question; for intriguing further discussion see, e.g. Gendler (2007), Stich (2009), and Weinberg, Nichols, and Stich (2001).

110

Sarah-Jane Leslie

and chemistry and argue—along with a number of philosophers of biology and chemistry—that the findings conflict with Kripke/Putnam-style essentialism. To suppose otherwise is to misconstrue science through the lens of quintessentialism. The intuition that such essences must exist is no more than an expression of a belief set that is firmly in place by the middle of the preschool years, but this belief set does not comport with the complexities of actual science. To drive this last point home, we must first understand the hypothesis of psychological essentialism and its detailed predictions, then recall the Kripke/Putnam view, and then investigate the actual facts as they stand in biology and chemistry. The psychology explains where our kind-essentialist intuitions come from, and the biology and chemistry explains why we should suspect and/or suspend those intuitions.

1.

PSYCHOLOGICAL (QUINT)ESSENTIALISM

1.1. The Quintessentialists Consider an intelligent species—let us dub them the Quintessentialists— whose members implicitly believe that a wide range of entities have within them a substance-like quintessence, which causally grounds their most important, stable, and enduring properties. Not all entities are believed to have quintessences: for example, artificial and manufactured items are rarely believed to have quintessences. However, biological beings, along with certain non-biological substances, are believed to be bearers of quintessence. It must be emphasized that the Quintessentialists rarely, if ever, explicitly entertain thoughts about quintessence as such; rather, their quintessentialist beliefs are tacit or implicit, though they are frequently manifested in a number of explicit ways. Though they are not the only bearers of quintessence, biological individuals are paradigmatic bearers of quintessence. According to the quintessentialist belief set, each such individual has its own particular quintessence, yet there can be considerable similarities (and in the extreme case, potentially qualitative identity) between the quintessences of distinct individuals.3 For example, 3 Sometimes quintessentialism (= psychological essentialism) is characterized as not allowing for individual variation in quintessence—that is, sometimes it is assumed that members of the same natural kind will have the same quintessence rather than highly similar quintessence. I think in the general case, quintessentialism is best articulated as allowing at least for the possibility of individual variation. This is most vivid in the case of individual people; since a wide range of social kinds are quintessentialized, there must be as much variation between people as there are different quintessentialized social kinds for them to belong to. Further, the literature on organ transplantation (discussed in more detail in the section on the transmissibility of quintessence) reflects the belief that a recipient can take on individual characteristics of the donor—for example, heart recipient Claire Sylvia (Sylvia and Novak, 1997) believed that she came to enjoy fried chicken and beer as the result of a transplant from a young man named Tim (see also Inspector, Kutz, and David, 2004; Meyer, Leslie, Gelman, and Stilwell, 2013). In

Essence and Natural Kinds

111

offspring have quintessences that are highly similar to those of their parents. Further, two individuals are generally only able (or disposed) to produce offspring if their quintessences are sufficiently alike. Like begets like, as far as quintessence is concerned, and Quintessentialists are convinced that one’s parents’ quintessence is the main determiner of one’s own quintessence. These sorts of similarities in quintessence are taken to form the basis of real or “natural” kinds, as opposed to merely conventional groupings. (In the case of non-biological individuals, such as certain substances, origins may also be considered important—for example, if two substances are formed in very different ways, then it is reasonable to suppose they have distinct quintessences.) Thus the Quintessentialists believe that there are ways of dividing individuals up into kinds that carve nature at its joints—namely the ones that group according to objective likeness in quintessence. Conversely, they also acknowledges that some groupings are merely nominal, and do not reflect genuine similarity classes of quintessences. However, Quintessentialists firmly believe that many of their own words and concepts map directly on to the real, objective kinds. The Quintessentialists also believe that there are a number of levels or degrees of similarity in quintessence, all of which are real and objective, in effect constituting a taxonomic hierarchy of kinds. At the lowest levels of this taxonomy, there are considerable similarities between the quintessences of members of some distinct kinds, while at the higher levels, there is considerable variation between the quintessences of members of the same kind. Importantly, there is a privileged level of this subjective taxonomy that occupies a “sweet spot” in this trade-off between within-kind variation in quintessence, and cross-kind quintessential distinctness. At this level, individual members of the same kind have only minimal differences in their quintessences, and these quintessences are quite dramatically different from the quintessences had by members of other kinds. (The Quintessentialists’ cognitive psychologists call this taxonomic level “the basic-level.”) The Quintessentialists believe that this privileged taxonomic level is objectively determined, and so there is a privileged way of answering the question of whether a given individual is the same kind of animal (or same kind of plant) as another: namely whether they belong to the same basic-level kind. The notion of a basic-level kind extends beyond the biological realm and into the chemical realm too; samples of substances can also share quintessences, and there is again a privileged level of the taxonomy of substances where the kinds in play maximize intrakind similarity (in the limit case, intra-kind qualitative identity) and inter-kind order to accommodate these sorts of considerations, the possibility should exist for individual variation in quintessence. It should further be emphasized that the model I am introducing here would allow also for the possibility that, in some cases, the nature of the individual’s quintessence would be exhausted by the nature of the quintessence of the (most specific) natural kind to which they belong. (That is, my model is compatible with the limit case where similarity becomes identity.) For example, one might believe that, e.g. mosquitoes do not display variation in their individual quintessences.

112

Sarah-Jane Leslie

dissimilarity in quintessence. Thus the Quintessentialists are strongly inclined to believe there is a privileged way of answering the question of whether one thing is the same substance as another. It must be emphasized that the Quintessentialists believe that these taxonomic categories are determined by objective facts about quintessence. This has several consequences; firstly that the Quintessentialists strongly believe that membership in these kinds is a purely intrinsic matter—a given individual’s quintessence is the sole determiner of its membership in a real kind. Further, they believe that quintessence lends itself to being “carved at its joints”—that is, quintessence does not vary continuously between individuals of different kinds, but rather is objectively distributed in such a way that, especially at the basic-level, members of the same kind have considerable sameness of quintessence, while non-members have distinctly different quintessences. Thus, membership in these kinds ought to be close to an all-ornothing matter; that is, Quintessentialists believe that real kinds should have sharp boundaries. Sometimes they themselves are not able to tell for certain the kind to which an individual belongs, but they believe there should be a correct answer to the question—an answer that is determined by the individual’s quintessence. (However, other beliefs about quintessence—namely that it is transmittable and mixable, as discussed below—compete with this one, creating an inevitable tension.) The single most important feature of quintessences is that they have causal powers. An individual’s quintessence is understood to be the causal root of many/most of that individual’s stable and enduring intrinsic properties. The respects in which members of a real kind are outwardly similar to each other is understood by the Quintessentialists to be the direct upshot of the similarities between their quintessences. Thus, the Quintessentialists consider outward signs of kind-membership to be important, but only in as much as they are indicators of underlying similarities in quintessence. It would be a mistake, though, to think that quintessences only have kind-related causal powers. For example, the Quintessentialists believe that they themselves each have a distinct quintessence, which grounds many of their individually distinguishing properties (in addition, that is, to grounding kind-general properties). For example, individual personality traits are considered to be the upshot of the individual’s quintessence.4 The Quintessentialists further believe that an item’s quintessence causes its properties in a defeasible way—that is, adventitious factors can prevent an individual from manifesting all the properties that the quintessence would otherwise cause it to manifest. This is particularly clear to them when they consider kind-wide quintessential properties that have been altered by clearly external circumstances—for example, some Quintessentialists have 4 Again, not all articulations of quintessentialism (= psychological essentialism) take this line, however it seems to me the best way to account for a number of findings; see footnote 2, and the section below on the transmissibility of quintessence.

Essence and Natural Kinds

113

lost limbs to amputation or accident. It can also happen that an individual’s quintessence is somehow thwarted from ever expressing itself in a particular feature—for example, some Quintessentialists are simply born without limbs. Since quintessence plays such a central role in beliefs about kindmembership, and since quintessence causally grounds a range of properties, Quintessentialists are ready and able to exploit kind-membership in their reasoning practices. Since natural kinds—especially “basic-level” kinds—are supposed to be groups with highly similar quintessences, Quintessentialists treat membership in a natural kind as a cornerstone of their inductive practices. If one member of a basic-level kind has a plausibly quintessential property, then other members of that kind may well have the property too. (Of course, these inferences are defeasible in a number of ways, given the nature of quintessential properties. They are, nonetheless, a good default starting point.) In this way, knowledge of kind-membership is more important to the Quintessentialists’ inductive inferences than anything else: more important than perceptual similarity between individuals, and more important than their sharing any other properties. When it comes to beliefs about their fellows, the same framework is employed. Quintessentialists manifest a tendency to believe that certain social categories—most prominently race and gender categories—group according to real similarities and distinctions in quintessence, and so reflect objective, non-constructed differences in nature. Just as in the non-social case, quintessentialist beliefs about these categories involve thinking that membership in them is determined by natural, inborn, intrinsic properties of the individual; that members of these groups are highly similar to one another and distinctly different from non-members; and that membership in these groups is more important and informative about the person than his or her appearance, demeanor, and other such readily observable properties. This quintessentialist belief set about the social domain has a number of unfortunate consequences; in particular, social groups whose members are seen to share very similar quintessences often experience significant degrees of social prejudice. When it comes to the nature of quintessence itself, the Quintessentialists do not have very specific beliefs at all. Instead, they have what might be termed a ‘placeholder’ notion of quintessence—they are sure that something fills this role, but they know not what. They are confident that it has a nature and independent existence, however, and they see no reason why they shouldn’t discover more about its nature, especially as their scientific practices mature. However, if one were to push for an elaboration of the metaphysics of quintessence, one might arrive at something like the following: quintessences are substance-like entities in that they occupy—or better, pervade—space-time regions. They can also mix with each other, albeit only under unusual circumstances. Quintessences can permeate objects, including inanimate objects, and thus sometimes be transmitted. Notably, such transmissions almost always require a physical bearer.

114

Sarah-Jane Leslie

In the normal case, an individual’s quintessence pervades its insides; internal parts are regarded as more important for quintessence than external surface parts. In this way, Quintessentialists believe that altering an individual’s insides is more likely to produce a change in the individual’s quintessence than altering an individual’s surface properties. In keeping with their beliefs that real, basic-level kinds reflect highly similar quintessences, they also expect that members of the same basic-level kind will have highly similar insides—even if their external appearances are quite different. Since one’s insides are suffused with one’s quintessence, the possibility of physically transplanting a part of an individual’s insides into another raises the possibility of transferring some of an individual’s quintessence into another. In recent years, the Quintessentialists’ medical technology has evolved to a point where this is more than just a possibility: internal organs can now be transplanted from one Quintessentialist to another. This raises particular anxieties for the Quintessentialists, since they believe that this practice transfers some of the donor’s quintessence to the recipient. The donor’s quintessence retains its causal powers, and as a result, the Quintessentialists believe that the recipient may be irreversibly changed. (The Quintessentialists are often reminded by their medical professionals that there is no scientific evidence to support this claim, but the lay belief persists nonetheless.) Recently, some Quintessentialist scientists have made considerable technological progress on the possibility of transplanting organs from other species. This is most disquieting to the Quintessentialist public. They fear that receiving transplants from other species may alter them, making them more animallike. They feel that mixing the quintessences of different basic-level kinds is profoundly unnatural. While internal parts are the main bearers of quintessence, it strikes the Quintessentialists that a small amount of an individual’s quintessence ‘rubs off’ on things that he or she handles and uses. This belief manifests itself in several ways. For example, Quintessentialists are willing to pay large sums of money for otherwise unremarkable objects, just because they were once owned and used by a celebrated person. They also shrink from items that were handled or used by an infamous and hated person. (This disposition increases as the contact becomes more intimate—e.g. a worn item of clothing is more potent in this respect than a touched pen.) At least to some extent, the Quintessentialists believe that they can be causally affected by quintessence transmitted through an object in this way. To the extent that they hold this belief, it may be weaker than their belief in the potency of organ transplantation—which is, of course, reasonable given their beliefs about the relationship between internal parts and quintessence. There is, nonetheless, some degree of belief in the causal potency of quintessence transmitted even in such an indirect way. It must be emphasized again that these beliefs in quintessence are rarely explicitly formulated, articulated, or entertained by the Quintessentialists. Rather, the belief structure sketched out here is a tacit framework that guides

Essence and Natural Kinds

115

their thinking. Thus, to determine whether an individual is a Quintessentialist, it does not do to ask them directly—rather, one must use a number of indirect measures. Further, the Quintessentialists are not taught to have the beliefs they do. These beliefs are not a mere cultural artifact: Quintessentialists from a wide variety of culturally independent groups exhibit the belief structure sketched above. And tellingly, Quintessentialist children manifest this way of thinking from a very young age. The role of culture is often to fill out more specific forms of the implicit beliefs that are characteristic of the early-developing quintessentialist syndrome.

1.2. Psychological (Quint)Essentialism Psychological essentialism (henceforth psychological quintessentialism) holds that we are the Quintessentialists described above. This well-confirmed psychological hypothesis attributes to us an implicit and early-developing belief in quintessence—in a substance-like entity possessed by some individuals and some forms of stuff, a substance which pervades their insides and causally grounds many of their stable and persisting properties. Membership in a natural kind is taken to be determined by objective features of one’s quintessence—namely the features that are shared by other members of the kind. It is crucial to understand the hypothesis of psychological quintessentialism correctly. It does not involve the claim that to be a Quintessentialist is to subscribe to all the beliefs listed above; rather quintessentialism is better thought of as a syndrome, which can be manifested in a variety of default implicit beliefs and ways of interpreting one’s world. Thus, one might be a Quintessentialist without holding each belief mentioned above. Compare, for example, the symptoms of the syndrome that is depression. Such symptoms include having suicidal thoughts; however it would be a mistake to suppose that one simply could not be depressed unless one has suicidal thoughts. Moreover, default implicit beliefs can be suppressed by explicit learning of contrary facts. For example, a well-informed adult may be less likely to explicitly hold the relevant beliefs about, say, organ transplantation. But this is fully compatible with being a Quintessentialist; that is, exhibiting the psychological syndrome that is manifested in the default beliefs mentioned above. It would also be a mistake to suppose that introspection alone could deliver the result that one is not a Quintessentialist. Introspection at best provides access to explicitly held beliefs; however, it is unhelpful in assessing beliefs that are largely implicit. Sometimes, one belief may be held implicitly, while a contradictory belief may be held explicitly. A dramatic illustration of this phenomenon concerns prejudiced attitudes as revealed by tests such as the Implicit Association Test (IAT) (e.g., Greenwald and Banaji, 1995). Many individuals who honestly explicitly disavow sexist and racist (and other prejudiced) beliefs nonetheless may implicitly hold such attitudes—even dedicated

116

Sarah-Jane Leslie

feminists may implicitly associate, for example, women with nurturance but men with intelligence. Further, implicit attitudes have been found to better predict a range of behaviors as compared to explicit/introspection-based measures of attitudes (Greenwald, Poehlman, Uhlmann, and Banaji, 2009; for excellent further philosophical discussion of these points and related ones, see Gendler, 2008a, 2008b, 2011). Implicit attitudes and beliefs are not somehow of lesser significance than explicit ones—only a limited range of the beliefs and attitudes that shape our behavior are available to introspection. And again, the explicit rejection of a given belief does not entail that that belief is not held implicitly. So quintessentialism is a syndrome, which encourages one or another cluster or range of default implicit beliefs drawn from the inventory provided in the previous section. These beliefs are not readily available to introspection, and they can persist in the face of known evidence to the contrary. The combination of these three observations means that a certain tempting response to the arguments given in this paper would be mistaken. I have in mind the following protestation, crudely characterized: “I know that it is scientifically impossible for a heart transplant to alter one’s personality, therefore I am not a Quintessentialist; however I still have the Kripke/Putnam intuitions, therefore it cannot be the case that Kripke/Putnam intuitions are due to quintessentialism!” This line of objection, along with its more subtle variants, turns on a failure to understand the hypothesis of psychological quintessentialism, and the significance of implicit beliefs more generally. 1.2.1. Some Crucial Findings Despite the fact that the beliefs in question are largely implicit, and so are somewhat more difficult to discern than our explicit beliefs, the last two decades or so have produced a wealth of empirical evidence suggesting we are, indeed, Quintessentialists. In the interests of space, I will not attempt a complete review here, but will rather highlight some of the main findings. (For an extremely thorough review of the data available through 2003, see Gelman, 2003.) There is a considerable amount of evidence that supports the claim that there is a privileged level in our psychological taxonomies—namely the so-called “basic-level” (Rosch, 1978; see also Coley, Medin, and Atran, 1997; Coley, Medin, Proffitt, Lynch, and Atran, 1999; Gelman, 2003; and many others). Members of the same basic-level kind are perceived to have quintessences that are highly similar, and also highly distinctive, in that members of other basic-level kinds have notably different quintessences (Gelman, 2003). Accordingly, in inductive reasoning tasks, both children and adults very often generalize properties to other members of the same basic-level kind (e.g. tigers), but not to more inclusive kinds (e.g. mammals); nor do they limit their generalizations to less inclusive kinds (e.g. Bengal tigers) (Coley et al., 1997; Gelman and O’Reilly, 1988; Waxman, Lynch, Casey, and Baer, 1997).

Essence and Natural Kinds

117

Information about membership in a basic-level kind has a powerful impact on people’s inductive inferences from a very early age. For example, even twoyear-olds prefer to base their inductive inferences on information about kind membership (as communicated via language) than on perceptual similarity (Gelman and Coley, 1990)—despite the general importance of perceptual similarity (especially shape) in guiding young children’s inferences. Further, preschool children distinguish between the sorts of properties that can be generalized along kind boundaries from those that cannot: children do not generalize transient or accidental properties such having fallen on the floor that morning (Gelman and Coley, 1990; Gelman, Collman, and Maccoby, 1986; Gelman and Markman, 1986, 1987). The latter are paradigmatically the sorts of properties that are independent of one’s quintessence. It is also worth noting that the results of many of the studies cited above depend on even very young children being willing to accept that an individual can appear for all the world like a member of one kind, yet really belong to another kind. Further, it is kind membership that is understood to be most important when making inferences about deep, shared properties. Notably, these findings extend to substance kinds, with preschoolers understanding that something can look more like coal than gold yet still be gold, and also basing their inferences about projectible, non-obvious properties on this shared kind membership (Gelman and Markman, 1986). Young children also have definite views on the causal roles of nature vs nurture. For example, preschoolers believe that an infant creature with kangaroo parents will grow up have a pouch and hop, even if it is raised exclusively with goats (Gelman and Wellman, 1991; see also Heyman and Gelman, 2000; Solomon and Johnson, 2000; Springer and Keil, 1989). Young children evidence the same set of beliefs when it comes to gender: for example, that a male baby raised on an island populated only by females will grow up to display boy-typical behavior and preferences (e.g. preferring to play with trucks rather than dolls; Taylor, 1996; Taylor, Rhodes and Gelman, 2009). Preschoolers also display at least some nature-over-nurture beliefs in the case of race, expecting that a child born to black parents will grow up to be black (Hirschfeld, 1996), though in general quintessentialist beliefs about race develop later than quintessentialist beliefs about species and gender (e.g. Rhodes and Gelman, 2009a). More generally, a number of studies have documented beliefs in children from very different cultural backgrounds about the relative power of nature vs nurture, for example Menominee Indian children, Yucatec Mayan children, and urban Brazilian children (Atran, Medin, Lynch, Vapnarsky, Ek’ and Sousa, 2001; Sousa, Atran, and Medin, 2002; Waxman, Medin and Ross, 2007). These findings are exactly what one would expect if one’s subjects were Quintessentialists. Children and adults also tend to ‘intensify the boundaries’ of natural kinds—that is, they judge that there should be sharp, exclusive boundaries between natural kinds (at the same taxonomic level). For example,

118

Sarah-Jane Leslie

Diesendruck and Gelman (1999) found that adults almost always judged that a given animal was definitely a member of a kind or definitely not a member of a kind; their judgments of how typical an exemplar of the kind the individual was, however, was more graded. The same pattern did not hold for artifact kinds, however, reflecting that this pattern is specific to natural kinds. Rhodes and Gelman (2009b) documented similar beliefs in young children. Further, even if adults are themselves uncertain as to how to classify an animal, they judge that there is a single correct answer which an expert could determine (Luhman and Coley, 2000). Young children also judge that membership in both animal and gender kinds reflect correct, objective, non-conventional facts (Kalish, 1998; Rhodes and Gelman, 2009a). Children at least as young as four understand that animate individuals have very different insides from inanimate individuals (e.g. Gelman, 1987; Gelman and O’Reilly, 1988; Gelman and Wellman, 1991; Gottfried and Gelman, 2005; Simons and Keil, 1995). They also understand that the internal parts of animate creatures have special importance when it comes to determining both an individual’s behavior and its kind-membership (Gelman and Wellman, 1991). Some recent work on infants suggests that the importance of insides is recognized from a very young age. Newman, Herrmann, Wynn, and Keil (2008) found that fourteen-month-olds expected that animals with the same internal features (e.g. visible red stomachs), rather than external features (e.g. blue hats), would display the same sorts of self-generated motion. However, when the motion of the target did not appear to be self-generated, these expectations did not arise. Thus it appears that, when it comes to apparently animate, self-generated patterns of motion, even infants expect that internal parts will be better predictors than external parts. Additional studies have found that both children and adults understand that insides are of considerable importance in determining membership in both biological and chemical kinds (e.g. Keil, 1989; Lizotte and Gelman, 1999; Newman and Keil, 2008). For example, Keil (1989) reports a classic series of experiments that show children’s increasingly sophisticated understanding of the importance of insides in determining kind-membership, with even very young children judging that, for example, a raccoon that has had its fur shaved and dyed so as to make it look just like a skunk is nonetheless still a raccoon. Interestingly, recent evidence suggests that rhesus monkeys also privilege insides over outward appearance in determining kind-membership (Phillips, Shankar, and Santos, 2010; see also Phillips and Santos, 2007). The majority of studies mentioned so far were conducted with participants primarily from North American communities. However, recent work has discovered similar dispositions among other cultural groups. As of now, essentialist beliefs have been documented in at least the following communities: Indians (Mahalingam, 2003; Meyer, Leslie, Gelman, and Stilwell, 2013), Brazilians (Diesendruck, 2001; Sousa et al., 2002), the Menominee (Waxman et al., 2007), the Vezo in Madagascar (Astuti, Solomon, and Carey, 2004), Yucatec Mayans (Atran et al., 2001), the Yoruba in Nigeria (Walker, 1999), and

Essence and Natural Kinds

119

the Torguud of Mongolia (Gil-White, 2001). Thus, while of course more work is needed, the available evidence suggests that quintessentialist thinking is not a local phenomenon, but rather is a pervasive aspect of human psychology. 1.2.2. The Transmissibility of Quintessence One question that arises at this point is why quintessence should be modeled as something substance-like, transmissible, and mixable; the above data do not obviously require this to be so. Consider, however, Sir Arthur Conan Doyle’s Adventure of the Creeping Man. In this story, Sherlock Holmes is called to investigate some alarming and mysterious goings-on at the house of an ageing professor. Through a characteristically brilliant string of ‘deductions’, Holmes figures out what is happening: in an effort to recapture his youthful vitality, the professor has been injecting himself with a serum extracted from the glands of monkeys. This has caused the professor to take on several monkey-like characteristics: he has become increasingly aggressive and his knuckles have become thickened and hairy; he has developed superhuman climbing abilities; and at times he adopts the gait of a monkey. This story—however much at odds with modern scientific understanding— reflects a basic quintessentialist way of thinking. A serum extracted from the glands of a monkey confers a portion of the monkey’s quintessence, complete with its causal powers, thus making its recipient take on the characteristics of the monkey. The tale reflects some of the most specific characteristics of quintessentialist thinking: the notion that the quintessence is causally potent, lies in the internal parts of the individual, and is substance-like in that it can be physically removed, relocated, and mixed with the recipient’s own quintessence to produce a hybrid mix of outward properties. (Does this explain the demand for bulls’ testicles among those who desire a certain ‘outward’ property?) Can such beliefs be systematically studied and documented? In an early study, Johnson (1990) found evidence that young children do indeed believe that such hybrid mixtures of quintessence are possible; for example, they believe that a heart transplant from a mean person would make the recipient become meaner. Meredith Meyer, Susan Gelman, and I replicated and extended these findings, and found they held up even when children were asked to consider characteristics that are not culturally associated with hearts, for example being smart (Meyer, Gelman, and Leslie, submitted). In collaboration with Sarah Stilwell, we also recently conducted a series of experiments to test for such beliefs in adults. In particular, we wished to assess whether people would endorse the possibility of a transplant or transfusion recipient’s personality changing to become more like the donor’s. We asked participants to consider donors from a range of social categories and two animal categories. We studied adult participants in both the United States and India, and have found that a sizeable portion of participants from both communities judge that the recipient’s personality may change to become more like the

120

Sarah-Jane Leslie

donor’s as a result of receiving a transplant or transfusion (Meyer, Leslie, Gelman, and Stilwell, 2013).5 Are transplants and transfusions the only means by which quintessence may be transmitted? A quick perusal of auction catalogs will show that people are willing to pay large sums of money for rather ordinary objects, if those objects once had contact with a famous or admired person. Hood and Bloom (2008) and Frazier and Gelman (2009) found that even young children place higher value on such objects, as compared to qualitatively identical duplicate objects. The converse of valuing objects that have been in contact with admired persons is shrinking from items that have been in contact with reviled individuals; this phenomenon has been studied and documented at length by Carol Nemeroff, Paul Rozin, and their colleagues (see Rozin and Nemeroff (2002) for a review). For example, in one famous study, Nemeroff and Rozin (1994) asked participants to rate how they would feel about wearing a sweater that had come into contact with various people, desirable and undesirable, including an enemy of theirs, and a person that the participants judged to be evil. (Unsurprisingly, Adolf Hitler was the favored choice for the latter category.) Participants were strongly opposed to wearing a sweater worn by either a personal enemy or by Hitler; in fact they rated a sweater that had fallen in dog feces as more desirable to wear than Hitler’s sweater. Notice that such results are not accounted for by the simple observation that we happen to value or care about historical properties. There are many historical properties of paintings and sweaters and keepsakes and the like that we do not care about, such as having once been located in Europe or twice having been scanned by an x-ray machine. The crucial thing to be explained is the quasi-infection-based character of the “authenticity and contamination” findings, and a natural hypothesis is that it is an implicit belief in transmissible quintessence which accounts for this. As with transplants and transfusions, the clearest way to determine whether the “authenticity and contamination” findings are evidence of a belief in a transmissible quintessence would be to see whether people believe that wear5 Such beliefs are also attested to in a number of (non-experimental) surveys of attitudes towards transplants (Basch, 1973; Belk, 1990;Hayward and Madill, 2003; Inspector et al., 2004; Sanner, 2001a, 2001b). For example, Inspector et al (2004) found that a third of heart-transplant recipients believed they had in fact taken on characteristics of the donor. Sanner (2001a, b) interviewed people about their attitudes towards various organ transplants, and she reports participants’ explicit formulations of ‘Creeping Man’ concerns: ‘I would perhaps look more piggish with a pig’s kidney.’. ‘Would I become half a pig, if I got an organ from a pig?’ ‘What if I would start grunting?’ (2001a, p. 22). A number of participants also expressed more general discomfort at the idea of ‘mixing’ species in this way: ‘I feel instinctively that it’s wrong to mix different species, it would go wrong.’ ‘My body would let me know that an animal organ didn’t fit. It’s contrary to nature.’ ‘It’s unnatural to move body parts between species’ (2001b, p. 1495). And quite decisively: ‘ The whole pig nature just feels like a big no’ (2001b, p.1496).

Essence and Natural Kinds

121

ing an item of clothing might cause a person to become more like the original wearer, even if the person in question does not know about the item’s history. For example, suppose one is about to stand trial, and one learns that the judge is unwittingly wearing an item of Stalin’s clothing, worn by him on and off over several years. Might one become anxious that the judge will be less just as a result? This phenomenon has not been extensively studied; however, one experiment suggests that such beliefs can be found. In particular, Johnson and Jacobs (2001) found that elementary school children and adults alike believed that wearing Mr. Rogers’ sweater would make a person friendlier, even if the wearer did not know about the prior owner. This finding suggests that, just as with transplants and transfusions, the notion of transmissible quintessence is at play here. The history of close-range adoration of relics purportedly from sainted persons may also be worthy of examination. 1.3. Quintessentialism and Science Some quintessentialist beliefs—such as the beliefs that receiving a blood transfusion or that wearing Mr. Roger’s sweater can alter one’s personality—are quite clearly at odds with the deliverances of science. Many other quintessentialist beliefs are, I would argue, similarly at odds with science, although on the surface they do not seem so. In particular, scientific findings, particularly in biology, are often misunderstood by the general population in a specifically quintessentialist manner. That is, people inappropriately map scientific concepts onto their pre-existing quintessentialist beliefs, and then consider those beliefs to be scientifically underwritten. (This may be part of a more general phenomenon, whereby people adopt culturally available ways of elaborating and articulating the same core, quintessentialist beliefs (e.g. Waxman et al., 2007).) While most educated adults have some familiarity with the concepts of genes and of DNA, misunderstandings abound, many of which at least indirectly suggest that people may simply have mapped the concepts onto their implicit notion of quintessence. One illustrative pair of anecdotal examples can be drawn from the geopolitical sphere: on 13 March 2000, Vladimir Putin reportedly asserted that “central power is in Russia’s genes.”6 Apparently President G. W. Bush concurred; at a White House news conference held on 17 October 2007, he wondered “whether or not it’s possible to reprogram the kind of basic Russian DNA, which is a centralized authority.”7 There is, of course, no such thing as specifically Russian DNA; however a quintessentialist view of Russians as a social kind entails belief in a particular Russian quintessence. There is also, of course, no scientific evidence to suggest that genetics per se would specially predispose a person or a group of people to 6 Oleg Shchedrov, “Central Power in Russia’s Genes, Putin Says,” Reuters (Moscow), 13 March 2000, from Johnson’s Russia List, 13 March 2000. 7 “Russia’s DNA,” Washington Post editorial, 19 October 2007.

122

Sarah-Jane Leslie

accept centralized authority, yet quintessentialist thinking is highly compatible with such a belief.8 As the above example illustrates, “DNA” and “genes” are often invoked in people’s conceptions of quintessentialized social groups. A number of social groups, such as groups demarcated by race and gender, are often highly quintessentialized; that is, people believe that the members of those social groups have highly similar quintessences to one another, and further that nonmembers have quite distinct quintessences—in effect, these groups are treated as social analogs of basic-level kinds. Lay beliefs about race and genetics often show precisely this pattern. For example, the majority of adults in the United States agree with the following statement: “Two people from the same race will always be more genetically similar to each other than two people from different races” (Jayaratne, 2001). Such a belief is wholly at odds with contemporary scientific thinking about genetic variability. Instead, the genetic variability within a racial group is just as high as the degree of variability across racial groups (e.g. Graves 2001; Lewontin, Rose, and Kamin, 1984; Templeton, 1999). In general, contemporary genetics is far more concerned with understanding how genetic differences among individuals can explain the phenotypic differences among those individuals, rather than attempting to find genetic explanations for putative group differences. This point has been poorly assimilated, however. As an illustration, when asked to consider possession of the trait being nurturing, Cole, Jayaratne, Cecchi, Feldbaum, and Petty (2007) found that people thought that genetics could better explain (perceived) gender differences than individual differences within genders. That is, contrary to the actual deliverances of genetics, they thought that group-level genetic explanations were more applicable than individual-level ones.9 Social groups aside, there are numerous other examples of quintessentialist misinterpretations of genetics. For example, Lanie, Jayaratne, Sheldon, Kardia, and Petty (2004) note that people with a family history of a heritable disease sometimes think that they can’t get the disease if they don’t look like the family members who have it. This is a natural belief to have against the backdrop of quintessentialism: one is susceptible to a heritable disease only in as much as one shares quintessence with the relevant relative, and similarity of appearance is an excellent guide to the extent to which quintessence is shared. People also confidently attribute traits, abilities, and dispositions to their genes in the absence of any scientific evidence supporting the idea that those characteristics are genetically based. Lanie et al. report that, even though few respondents could give a remotely adequate characterization of what genes even are, “it is interesting that elsewhere in the interview about three quarters of respondents had no trouble giving an example of at least one nonphysical, 8

I am extremely grateful to Peter Godrey-Smith for drawing my attention to these examples. Interestingly, Keller (2005) found that belief in “genetic determinism”—that one’s genes are a powerful determiner of one’s traits and character—correlated with social prejudice, and that priming people to think about genetics further increased their level of prejudice. 9

Essence and Natural Kinds

123

nonmedical ‘genetic’ characteristic that ran in their families, even if there was no scientific research in support of their beliefs” (2004, p. 311). Such examples included characteristics like being good at home repairs (and perhaps also could have included embracing centralized authority). As another illustration, consider that on the quintessentialist outlook, a quintessence has the power to cause its bearer to have certain characteristics, more or less independently of the bearer’s environment. And so relatedly, the outlook encourages the belief that if a certain characteristic does not completely spring from the bearer’s quintessence, then it must be determined by the environment. Thus, quintessence and environment are seen as largely mutually exclusive influences when it comes to determining the causal source of a trait.10 In contrast, modern genetics holds that phenotypic traits arise from complex interactions between an individual’s genes and its environment. Genetics and environment are not in any sense mutually exclusive as causes, rather they operate in tandem. Nevertheless, lay understanding of genetics often implements the quintessentialist model, in which genetic causes and environmental causes are seen as exclusive of one another. For example, in an extensive study, Jayaratne, Gelman, Feldbaum, Sheldon, Petty, and Kardia (2009) found numerous negative correlations, and no positive correlations, between participants’ endorsements of genetic and environmental factors in explaining the source of a range of traits. In recent years, a number of researchers have argued that quintessentialism is responsible for both resistance to and misunderstanding of evolutionary theory (e.g. Gelman and Rhodes, 2012; Hull, 1965; Mayr, 1982, 1988, 1991; Samarapungavan and Wiers, 1997; Shtulman, 2006; Shtulman and Schulz, 2008). The main culprit is the quintessentialist belief that within-species variability is a limited and uninteresting phenomenon, in comparison to withinspecies similarity. That is, quintessentialism places considerable emphasis on the similarities between members of a given species (or more accurately, a given basic-level kind, since these are not always one and the same), and de-emphasizes the differences between them. (This is, of course, why quintessentialism encourages such rich inductive inferences over these basiclevel kinds.) To correctly understand evolutionary theory, however, one must appreciate that there is a considerable degree of variability within a species, since this variability is precisely what allows the evolutionary process to occur. Not only does quintessentialism make people resistant to accepting evolutionary theory—for example, Samarapungavan and Wiers (1997) found a sizeable portion of Dutch third- and fifth-graders believe the species to be ‘eternal and unchanging’—but it further means that even educated adults often wholly misunderstand what the theory claims. Shtulman (2006) found 10 Of course this is not intended as an absolute and universal quintessentialist principle, but rather a general guiding one—for example, there could still be cases in which the environment prevents a quintessential property from being had, an obvious example being a tiger that has lost its tail.

124

Sarah-Jane Leslie

that the majority of the talented group of students enrolled in the Harvard Summer School held “transformational” rather than “variational” conceptions of evolutionary theory. The correct variational conception of evolution holds that random mutations occur all the time, and some of these confer an adaptive benefit on their possessors, and so render them more likely to successfully reproduce and pass along the mutation to the next generation. The transformational conception of evolution, in contrast, holds that, if a trait is beneficial to the species, then over generations the species as a whole is likely to acquire the trait. Thus, on the transformational view, the quintessentialist emphasis on within-species commonalities is preserved—what happens is that (somehow) over time the common quintessence of the species alters so as to produce more beneficial traits. Perhaps unsurprisingly, people who hold this transformational view of evolution are less likely to believe in evolution as scientific fact. After all, what they understand it to be saying is not in fact true. As emerges at some length below, biology’s understanding of species and other taxa is deeply at odds with the quintessentialist mindset. The relationship between one’s genotype and one’s species membership is complex, probabilistic, and highly dependent on extrinstic factors; there is no such thing as the ‘species’ genotype. Just as the systematic misconstruals of evolutionary theory illustrate, educated adults may appear to have a working understanding of biology, while in fact their understanding is rife with quintessentialist confusion. For example, consider the following remark by the brilliant semanticists Hans Kamp and Barbara Partee: “the vast majority of natural kind terms are sharp in the strict sense of being determinately true or false of everything that is found in the real world. For instance, to belong to a particular biological species an individual must have the DNA of that species; and almost without exception this is a property which an individual organism either definitely has or else definitely lacks” (1995, p. 175).11 How far does this kind of quintesssentialist thinking extend, and has it left its mark on philosophy?

2.

PHILOSOPHICAL ESSENTIALISM AND PSYCHOLOGICAL QUINTESSENTIALISM

2.1. Natural Kinds and Philosophical Essentialism Hilary Putnam and Saul Kripke famously argued for a version of philosophical essentialism as applied to natural kinds, according to which there are scientifically discoverable necessary and sufficient conditions for belonging to a natural kind. In particular, these necessary and sufficient conditions consist in a specification of the kind’s hidden underlying structure. This underlying 11 I am grateful to Susan Gelman for directing my attention to this passage. She and Marjorie Rhodes use this passage to illustrate a very similar point in Gelman and Rhodes (2012).

Essence and Natural Kinds

125

structure is discoverable by science, and plays a causal/explanatory role in the determination of the kind’s perceptible, or manifest, properties. Both Kripke and Putnam further argue that the extensions of natural kind terms are determined by sameness of underlying essence—rather than by, for example, sameness of manifest properties.12 2.1.1. The Twin-Earth Thought Experiment Perhaps the best-known illustration of the point is Putnam’s Twin-Earth thought experiment. We are asked to imagine a planet that is qualitatively identical to Earth, except that where Earth contains H2 O, Twin-Earth contains a superficially indistinguishable substance whose complex chemical formula is abbreviated as “XYZ.” Thus, on Twin-Earth, XYZ fills the oceans and lakes, and falls from the sky as rain; Twin-Earth people drink XYZ to quench their thirst, bathe in it, use to make soups, and so forth. Putnam then asks us to imagine that, in 1950, some inhabitants of Earth set out on a spaceship and reach Twin-Earth. When they arrive there, they are astonished at the similarities, including what they initially believe to be the abundance of water on the planet. Crucially, though, we are encouraged to have the intuition that when the Earth’s astronauts say “there is water in the lakes on Twin Earth,” what they say is false. If they go on to perform chemical tests on the liquid that fills the oceans and lakes on Twin Earth, they will realize they were just wrong to have called the substance “water.” This is because the extension of our term “water” picks out all and only that which has the underlying chemical composition H2 O. Alternatively, imagine an inhabitant of Earth, Oscar, and his Twin-Earthian doppelganger, Twin-Oscar. Putnam argues that the word “water” as used by each has different extensions—for Oscar “water” applies to H2 O, for TwinOscar “water” applies to XYZ. Of course, Oscar and Twin-Oscar may have different beliefs about water and twin-water respectively: namely, Oscar may believe that water is composed of H2 O, while Twin-Oscar may believe that twin-water is composed of XYZ. However, Putnam then asks us to ‘roll the clock back’ to 1750, before anything was known about the chemical 12 It must be acknowledged that Hilary Putnam came to revise his view on the matter, and allow that, e.g. our interests also figure in the determination of the extension of natural kind terms (e.g. Putnam, 1992). This revised view has not, however, been nearly so influential as his original view, and in fact comparatively few philosophers are even aware that Putnam changed his view (Hacking, 2007a). The discussion in this half of the paper is directed towards the view— articulated in “The Meaning of ‘Meaning,” at least as it is widely interpreted—that has been so influential, and which is so often characterized as Kripke/Putnam essentialism. For further discussion, see Hacking (2007a); nothing in this paper disagrees with Hacking on the historical observations, however I will continue to speak of the “Kripke/Putnam view,” since, as Hacking himself observes, this is how the view is widely understood. There is a crucial further point. It is one thing to qualify the view in the light of isolated counterexamples and quite another to hope that this will deal with the full systematic range of examples of natural kind terms drawn from biology and chemistry. What follows is intended to strongly suggest that we need much more than the concession that our interests play a role.

126

Sarah-Jane Leslie

composition of water. Putnam maintains that “water” still has a different extension in the mouths of Oscar and Twin-Oscar—even though we may stipulate now that neither Oscar nor Twin-Oscar (nor any local experts) have any beliefs about the nature of the chemical composition of these substances. If we agree with these intuitions, we must suppose that two people can be in duplicate inner psychological states, while their terms have different extensions, due to the nature of the local natural kinds. Putnam famously concludes that “meanings just ain’t in the head!” (1975, p. 227). In what follows below, I will argue that the intuitions associated with Putnam’s highly influential thought experiment are quintessentialist intuitions. That is, only beings who are (at least in some respects) Quintessentialists would speak a language whose terms behave as Putnam suggests (or at least share Putnam’s intuitions that the terms would behave in the relevant ways). If our psychology was relevantly different, then we would not have TwinEarth intuitions—even if the world does indeed conform metaphysically to the Kripke/Putnam essentialist view of it. I will then review evidence that the world does not in fact conform to Kripke/Putnam essentialism, and argue that the relevant intuitions derive solely from our quintessentialist outlook. As one might put it, Twin-Earth intuitions are driven by what’s ‘in the head’ rather than by what’s in the world.

2.2. Kripke/Putnam Essentialism and Quintessentialism In the wake of this compelling thought-experiment, Putnam proposes an account of natural kind terms. He suggests that natural kind terms can be given “ostensive definitions”—that is, one may provide a definition for a natural kind term by ostending an instance of the kind, and indicating that the term applies to anything that is the same as the instance in important respects. For example, we could define the term “water” by pointing to a glass of it and noting that “water” applies to anything that bears the “same liquid relation” to the ostended sample: The logic of natural-kind terms like “water” is a complicated matter, but the following is a sketch of an answer. Suppose I point to a glass of water and say “this liquid is water” . . . My ‘ostensive definition’ of water has the following empirical presupposition: that the body of liquid I am pointing to bears a certain sameness relation (say, x is the same liquid as y . . . ) to most of the stuff I and other speakers in my linguistic community have on other occasions called “water” (1975, p. 224–5).

And more generally: “One can give [someone seeking to learn a natural kind term] a so-called ostensive definition—‘this (liquid) is water’; ‘this (animal) is a tiger’; ‘this (fruit) is a lemon’ ” (1975, p. 229). Kripke makes similar remarks about ostension in Naming and Necessity (e.g. p. 135). The crucial thought common to Kripke and Putnam is that underlying essential features determine the extension of “same liquid” and “same substance”.

Essence and Natural Kinds

127

Nathan Salmon (1979) convincingly argues that one would be mistaken to suppose that the essentialist metaphysical doctrine is entailed by the above semantic picture. Rather, Salmon argues, the advocates of the semantic picture make essentialist assumptions from the outset. Indeed, Kripke himself explicitly denies that he ever thought to derive conclusions about the metaphysical status of essentialism from his theory of reference (1980, preface). I am in complete agreement with Salmon that the essentialist metaphysics is presupposed by the semantic analysis, rather than entailed by it. (I argue in the final sections of the paper, however, that the detailed presuppositions are ill-founded.) Here, I would note another implicit assumption in Putnam’s argument: namely that the communities of speakers that he describes are, like his readers, Quintessentialists, at least to some extent. Absent such an assumption, the argument would fail. (Unlike his metaphysical assumptions, however, I believe that this assumption is completely correct.) Imagine a community of speakers—let us call them the Phenomenalists— whose psychology is notably different from the Quintessentialists, especially when it comes to beliefs about the natures of certain kinds and individuals. The Phenomenalists staunchly deny that anything of interest lies below the surface—it would never occur to them to consider two individuals to be essentially similar if they differed in their obvious surface properties. The Phenomenalists thus form their categories and concepts on the basis of perceptual appearance and other such readily-accessible qualities. (In this way, the Phenomenalists take Oscar Wilde’s remark “only the most superficial people do not judge by appearances” to be words to live by.) It should be clear that an ostensive definition of the sort Putnam discusses will yield a term with a very different extension for the Phenomenalists. Putnam writes, “Suppose I point to a glass of water and say ‘this is water,’ in order to teach someone the word ‘water”’ (1975, p. 230). If one attempts to teach a Phenomenalist a term in such a way, the Phenomenalist will acquire a term that he will take to have a very different extension than our term “water” does, at least on Putnam’s account of how that term functions. For example, it will strike the Phenomenalist as undeniable that XYZ falls in the extension of this term: it looks like it, smells like it, tastes like it, and so on—and what else matters to the Phenomenalist? Of course, if we are simply imagining an isolated Phenomenalist attempting to learn English, we might just dismiss his reaction as a semantic error, especially if we believe in public languages whose meanings are determined by community-wide dispositions to use terms in certain ways. The response to this is obvious: let us imagine an entire community of Phenomenalists, whose numbers and practices are sufficient to determine a public language of their own. Even if one believes that languages are public and are thus not determined by individual psychological dispositions, there can be no denying that the psychological dispositions of the community of speakers determine which public language they are collectively speaking. If we do indeed speak

128

Sarah-Jane Leslie

a language whose term “water” does not include XYZ in its extension, this is because we are psychologically very different from the Phenomenalists. It might be objected that the Phenomenalists’ mindset is at odds with science. That is, haven’t we discovered that water has an essence (namely H2 O), and so wouldn’t it be somehow unscientific of the Phenomenalists not to revise the extension of the term after making such a discovery? Even if we grant the (dubious) scientific claim about the essence of water, it should be clear that this objection has no bite. First of all, no speaker is under an obligation to revise her use of a term in face of such scientific discoveries unless her community already had a (quint)essentialist view of the term’s extension. If a term in a language has its extension determined by manifest appearance, then scientific discoveries are not relevant to determining its extension (unless of course they bear on facts about manifest appearance). Second, Putnam himself is quite explicit that his claims about meaning do not depend on what is known at a time about the relevant underlying essences. Oscar and TwinOscar’s words have different extensions in 1750; similarly, “χρυσ óς” (gold) as used by Archimedes had its extension restricted to substances composed of the element with atomic number 79—even if Archimedes himself (let us suppose) was unable to tell this substance apart from, say, iron pyrite. It is not the scientific discoveries per se that fix the extension of the terms: it is the dispositions of the community of speakers to intend to use their natural kind terms to carve nature at its (quint)essential joints, whatever those may be. Science is here no more than a post-hoc guide as to the determinate semantic consequences of such dispositions. If we consider a Quintessentialist community instead of a Phenomenalist one, it is evident that the speakers are likely to speak (or at least believe that they speak) a language of the sort that Putnam describes. If one teaches the word “water” to a Quintessentialist via ostention, then she will readily suppose the word’s extension is determined by similarities in underlying quintessence. That is, she will be prepared to limit the use of the term “water” to just those quantities of substance whose quintessences are relevantly similar to that of the ostended substance—and she will be happy to admit that there may be cases where a given quantity of stuff looks just like the original, and yet differs in its quintessence enough that the term “water” will not apply to it. Similarly, if she is taught “tiger” in the same way, she will suppose that the term applies to just those individuals whose quintessences are appropriately similar to the demonstrated individual. And importantly, since their quintessentialist view of the world preceded modern scientific discoveries, Quintessentialists would have understood these terms in this way throughout their entire history, just as Putnam supposes that we do. The Quintessentialists are even able to resolve (or at least appear to resolve) a lingering ambiguity that haunts these ostensive definitions (often referred to as “the qua problem;” for example Devitt and Sterelny, 1987; Dupré, 1981, 1993). There are, for example, a host of non-tigers whose quintessences are in many respects similar to the quintessence of our demonstrated tiger: namely,

Essence and Natural Kinds

129

lions, panthers, cougars, etc. There are also sub-groups of tigers whose quintessences differ in some important and systematic respects: Bengal tigers, Sumatran tigers, Siberian tigers. Let us suppose the demonstrated tiger was a Bengal tiger. How does the Quintessentialist determine that Siberian tigers but not lions have quintessences that are “relevantly similar” to the originally labeled animal? The answer is that she employs a helpful assumption: that novel terms name basic-level categories.13 This assumption fits very well with the quintessentialist outlook: members of basic-level kinds are taken to have maximally similar quintessences, modulo their also having highly distinctive quintessences relative to other kinds. By settling on the basic-level kind, the Quintessentialist thereby adopts a strategy that maximizes the information an application of a natural kind term to an individual conveys about how the individual’s quintessence is similar to and different from other individuals’ quintessences. As a consequence, the Quintessentialists favor using this strategy when learning terms via ostension. Since the Quintessentialists are so accustomed to (unconsciously) applying this strategy, they sometimes leave it wholly implicit in their philosophical works, as they assume that their readers will be Quintessentialists themselves. Put another way, the qua problem can seem not to be that deep precisely because we are Quintessentialists, who privilege basic-level kinds. If we are indeed Quintessentialists, it is easy to see why our intuitions concerning Twin-Earth and the like accord with Putnam’s. Were we, say, Phenomenalists instead, the thought-experiment and its conclusions would gain no traction with us. It is thus at least a necessary condition for Putnam’s account to be successful that we think like Quintessentialists in the relevant respects. Is it also sufficient, or does the world have to cooperate? Do we have Twin-Earth intuitions even when the kind in question has no essence? More generally: might we have the intuition that members of a particular kind must share an essence even if in fact they do not? If so, perhaps not only are the specifically Twin-Earth intuitions driven by quintessentialism, but so too is the bedrock idea that the kinds that correspond to our natural kind terms group individuals by their essences. 2.2.1. The Tragic Mulatto Before considering whether the kinds that Putnam and Kripke discuss actually do group their members by essence, it is worth recalling a nineteenth and twentieth-century North American literary and cinematic trope, the Tragic

13 A separate question is how a Quintessentialist figures out which categories are basic level. There is a wealth of interesting work on this issues (e.g. Murphy, 2002; Rosch, 1973, 1975, 1978; Rosch, Mervis, Gray, Johnson, and Boyes-Braem, 1976). Shape is an important (though defeasible) guide. It is important to separate this empirical question from the point that a Quintessentialist, who can somehow distinguish basic-level from non-basic-level kinds, even when the kinds in question are novel, can use this ability to fix on a meaning for a term introduced via Kripke/Putnam style ostension.

130

Sarah-Jane Leslie

Mulatto (or Tragic Mulatta, as the trope is sometimes styled). The trope typically involves a young woman who “looks and acts like a white person,” and in many cases believes herself to be white. Everyone around her unquestioningly believes that she is white, and she rises to an enviable position in society and is sought after by eligible young white men. Then the ‘terrible’ news is discovered: the woman is not, in fact, white, but rather has mixed racial ancestry. The discovery of one black ancestor spells her ruination. As the literary trope has it, a ‘single drop’ of ‘black blood’ suffices to make one black; she loses everything overnight, is rejected by friends and lovers, and cast out of white society. In some instances, the tale ends with her being sold into slavery. This depressing trope was very popular, and clearly was not in any way difficult for its readers to comprehend. Let us consider its structure then: it requires the reader to suppose than a person can appear in every observable respect to be white, and yet in fact not be white, but rather black. This is something that would be incomprehensible to the Phenomenalist about race, yet is immediately comprehensible to the racial Quintessentialist. Since quintessence only defeasibly causes its bearers to have their characteristic observable properties, the possibility exists for a member of a kind to have the relevant quintessence, and yet not share any of the characteristic properties of the kind. Conversely, an individual may have the characteristic superficial qualities associated with a kind, and yet in fact have the quintessence of another kind. (Compare Gelman and Markman’s (1986) study, where preschoolers readily accepted that an individual could be a member of chemical or biological kind, despite better resembling members of a different kind.) It is this quintessentialist understanding of race that the Tragic Mulatto trope exploits. The Tragic Mulatto trope has much in common with the Twin-Earth thought experiment; someone whose mindset allowed them to accept the Tragic Mulatto tale hook, line, and sinker would surely also have Twin-Earth intuitions about race. If there can be one “tragic mulatto,” why not a whole planet? Why not a Twin-Earth populated by individuals who look and act “like white people,” but are in fact, like the Tragic Mulatto, really black? Just as the Tragic Mulatto’s suitors incorrectly apply the term “white” to her, a visitor to this Twin-Earth would incorrectly apply the term “white” to its inhabitants, and so on, so forth. The same conceptual structure allows for both the Twin-Earth example and the Tragic Mulatto trope. Both hinge on the idea that something can appear in every respect to belong to a kind, but in fact fail to, since the individual lacks the relevant essence. There is, of course, no such thing as black or white essence. Racial groups are social constructs: they are not based on any biologically real essences. Genetic variation within racial groups is just as high as it is across racial groups—there is no such thing as “black DNA,” and there is most certainly nothing intrinsic that would set apart a young woman as truly being black, despite having pale skin and other phenotypic features associated with white

Essence and Natural Kinds

131

people.14 In this case the world does not vindicate the belief in hidden essence, but racial groups have historically been—and continue to be—highly quintessentialized. These quintessentialist beliefs suffice for the analog of the Twin-Earth intuitions in the case of race. This suggests that quintessentialism may be not only necessary, but also sufficient for the relevant intuitions. 2.3. Do Biological Kinds have Essences? One outstanding question at this point is whether the kinds that Putnam and Kripke actually discuss turn out to have essences, or whether—as in the race case—both our Twin-Earth style intuitions and our confident belief in kind essence are simply driven by unjustified quintessentialist beliefs. For a kind to “have an essence” in the Kripke/Putnam sense, there must be hidden underlying features that are necessary and sufficient for kind membership— features of the sort that science investigates and discovers. Moreover, for a natural kind term to have the sort of extension that Kripke and Putnam suppose, it is necessary that any two members of the relevant kind (including any two quantities of the relevant substance) be similar in some such scientifically discoverable and essential respect. Putnam is very explicit that having chemical composition H2 O is the relevant essential respect in the case of water, and also that having the appropriate genetic code is the relevant essential respect in the case of lemons. He writes: [A] critic has maintained that the predominant sense of, say, “lemon” is the one in which anything with. . . the superficial characteristics of a lemon is a lemon. The same critic has suggested that having the hidden structure—the genetic code—of a lemon is necessary to being a lemon only when “lemon” is used as a term of science. Both of these contentions seem to me to rest on a misunderstanding . . . The sense in which literally anything with the superficial characteristics of a lemon is necessarily a lemon . . . is extremely deviant . . . At the same time the sense in which to be a lemon something has to have the genetic code of a lemon is not the same 14 Some researchers have recently claimed that race does, in fact, have a biological reality, in particular that racial divisions can be at least approximately cashed out in terms of continentbased ancestral breeding populations, which affect to some extent the nature and frequencies of some alleles (see, e.g. Kitcher (2007) for sympathetic discussion). Such findings, even generously interpreted, do not amount to “racial essences” or “black DNA” or the like, as Kitcher makes very clear. For example, the claims in question are statistical/probabilistic in nature, and quite weak ones at that: only .0005% of human genetic variation is even putatively explained by membership in these groups (Maglo, 2011). It is also far from clear that these findings hold up to scrutiny (as opposed, perhaps, to simply being expressions of quintessentialism); there are a number of reasons to be skeptical regarding them—reasons which range from concerns about the statistical methods employed to objections to the theoretical interpretations bestowed (for an excellent review, see Maglo (2011), and sources cited therein). Further, the Tragic Mulatto is specifically characterized by having largely white ancestry but a “single drop of black blood,” which suffices to make her black. Nothing in these statistically based genomic analyses would deliver this result (in fact one criticism of them is that they tend to abstract away from cases of mixed ancestry, thus potentially making the findings appear more clean-cut than they in fact are).

132

Sarah-Jane Leslie

as the technical sense (if there is one, which I doubt). The technical sense, I take it, would be one in which “lemon” was synonymous with a description which specified the genetic code. But when we said . . . that to be water something has to be H2 O we did not mean . . . that the speaker has to know this . . . similarly, even though the predominant sense of “lemon” is one in which to be a lemon something has to have the genetic code of a lemon (I believe), it does not follow that “lemon” is synonymous with a description which specifies the genetic code explicitly or otherwise (1975, pp. 239–240, original emphasis).

Kripke makes similar remarks concerning the “internal structure” of tigers (1980, pp.120–1), and later speaks of “scientific discoveries of species essence” (p. 138). Does the world cooperate with these claims, or are these quintessentialist misinterpretations of science? That is, are Kripke and Putnam simply reporting scientific facts in their discussions, or are the discussions fueled instead by unfounded quintessentialist convictions that natural kinds must have essences? Might the very claim that natural kinds have scientifically discovered essences be no more than an articulation of inchoate quintessentialist intuition? 2.3.1. Species and Sex; Genes and Essence On the question of whether species have these sorts of essences, there is a degree of consensus among philosophers of biology (and indeed biologists) that is almost unprecedented in philosophy at large (e.g. Dupré, 1981, 1993; Ghiselin, 1987; Hull, 1965; Laporte, 1997, 2004; Mayr, 1982, 1988, 1991; Okasha, 2002; Sterelny and Griffiths, 1999, and many others.). There is no such thing as “lemon DNA,” no common genetic code that makes for membership in the kind Panthera tigris.15 To a first approximation, most biologists subscribe to the notion that a species is delineated by the boundaries of an ecological niche, or by the boundaries of a reproductive community.16 Within such bounds, considerable genetic variation is possible, and conversely, it is possible for there to be less genetic variation across such boundaries. As a concrete illustration, consider the cutthroat trout (Salmo clarki). This species has a number of subspecies, and there is considerable genetic divergence between these subspecies. In a study of seven subspecies of cutthroat trout, zoologists Fred Allendorf and Robb Leary report that: A highly variable pattern of genetic divergence exists among the seven subspecies . . . Very little genetic divergence exists among Colorado, Snake River, greenback, and Yellowstone cutthroat trout. Nei’s genetic distances between these subspecies are

15 In what follows, I will limit my discussion to the species-level in the taxonomy, since this is, I think, in keeping with Kripke’s and Putnam’s intentions. 16 These distinct ways of characterizing species are considered in more detail in the section headed “Extrinsic Species-Essence and Plenitude”. .

Essence and Natural Kinds

133

typical of those reported for conspecific populations in a diversity of freshwater and anadromous fishes (Avise 1974; Avise & Smith 1977; Buth & Burr 1978; Loudenslager & Gall 1980; Buth et al. 1984). In contrast, substantial biochemical genetic divergence exists between coastal, Lahontan, and westslope cutthroat trout and between these fishes and the other four subspecies. These genetic distances are truly exceptional for conspecific populations; they are similar to or larger than values observed between many species of fish (Johnson 1975; Avise & Ayala 1976; Buth & Burr 1978; Phelps & Allendorf 1983; Yates, Lewis, & Hatch 1984). Surprisingly, the coastal, Lahontan, and westslope cutthroat trout are as similar or more similar to the rainbow trout than they are to the other subspecies. (Allendorf and Leary, 1988, pp. 172–3, emphasis added)

Thus a member of one species (e.g. a member of the Lahontan subspecies) may have more genetically in common with a member of another species (namely rainbow trout) than with a member of its own species (e.g. a member of the Snake River subspecies). While such findings may not be the norm in biology, this is not a wholly isolated occurrence; for example, high degrees of intraspecies genetic divergence can be found among various other Salmonids as well (Pennell and Barton, 1996). As we shall see, these are not just weird cases: biology systematically confounds our quintessentialist convictions. As against this, recently Michael Devitt (2008) has criticized the consensus in the philosophy of biology, and indeed in biology and zoology, and argued for the existence of intrinsic, microstructral species essences of the Kripke/Putnam variety. It is instructive to consider Devitt’s arguments, since he articulates a view that is, I think, widely held in philosophy (outside of philosophy of biology). He argues that biologists and zoologists are committed to positing such essences—even if they assert otherwise—on pain of explanatory inadequacy. Devitt writes: There has to be something about the very nature of the group—a group that appears to be a species or taxon of some other sort—that, given its environment, determines the truth of the generalization [e.g. that Indian rhinos have one horn and African rhinos have two horns]. That something is an intrinsic underlying, probably largely genetic, property that is part of the essence of the group. Indeed, what else could it be? Some intrinsic underlying property of each Indian rhino causes it, in its environment, to grow just one horn. A different such property of each African rhino causes it, in its environment, to grow two horns. The intrinsic difference explains the physiological difference. If we put together each intrinsic underlying property that similarly explains a similar generalization about a species, then we have the intrinsic part of its essence. (2008, p. 355)

And summarizing: I have presented a positive argument for Intrinsic Biological Essentialism. We might sum it up: structural explanations in biology demand that kinds have essential intrinsic properties. (2008, p. 365, original emphasis)

Devitt’s argument is intuitively appealing, but not, I think, ultimately successful (see Barker (2010) and Ereshefsky (2010) for further discussion). First of all,

134

Sarah-Jane Leslie

the argument is undermined by the fact that members of the same species often do not share the same phenotypic traits; conspecificity is compatible with a great deal of variation in phenotype at a time, and even more dramatically over time. That is, it is possible for a species to radically change its phenotypic features over time without a speciation event occurring. In the extreme case, members of a species at t1 might not have any non-trivial phenotypic properties in common with the members of the species at t2. However, in what follows I will set aside this particular concern, and consider only cases in which a particular phenotypic trait actually is shared by members of a species. Devitt is, of course, indisputably correct that each particular African rhino has some intrinsic features that, in combination with the environment, are causally responsible for that individual’s having horns. This does not entail, however, that those very same intrinsic features are also responsible for other African rhinos’ having horns. Whether this is so is a substantive empirical hypothesis, not one whose truth can be intuited in advance. For a simple illustration of the general point, let us revert to a well-known chemical case, that of jade. There are two very different chemical compounds that both fall under our term ‘jade’: jadeite (NaAl(SiO3 )2 ) and nephrite (Ca2 (MgFe)5 Si8 O22 (OH)2 ). Substances composed of these two respective chemical compounds have very similar observable properties (in fact they are far more similar than many conspecific plants and animals). Thus a sample of jadeite and a sample of nephrite will have many properties in common (in fact they will be indistinguishable to all but the most experienced artisans). It is also true that the observable properties of the sample of jadeite are determined by its intrinsic chemical structure in conjunction with the environment, and similarly for the sample of nephrite. Yet there will be no common intrinsic chemical structure that explains the shared features of the two samples of jade. (There will, of course, be the non-explanatory disjunctive property of being composed of either NaAl(SiO3 )2 or Ca2 (MgFe)5 Si8 O22 (OH)2 .However, it is important to see that disjunctive properties cannot play the explanatory role that Devitt has in mind, or else the whole enterprise is trivialized. For example, let us suppose with Devitt that there is a common intrinsic property had by tigers that explains why they are striped. Let us also suppose that there is a different common property that explains why canna lilies are striped. If disjunctive properties are allowed to figure as common intrinsic explanatory properties in Devitt’s sense, then there will be a further shared intrinsic property that explains why this tiger and this lily both have stripes. If disjunctive properties are countenanced in this endeavor, then shared properties become far too cheap to be of interest. Certainly, it would not then be a biological hypothesis that a common property explains why Indian rhinos have one horn—it would simply be a familiar point about the logic of disjunction.) In the case of jade, we have selected these two chemically different but manifestly similar substances to figure in some of our statues and jewelry; but natural selection can do something similar when it comes to phenotypic features. Consider, for example, the property of having typical external human female

Essence and Natural Kinds

135

genitalia—the property that is overwhelmingly used to classify humans, particularly newborn infants, as female.17 What is the genetic basis for having such a property—what intrinsic genetic basis explains the possession of such a property (as Devitt might put it)? This is a question that many educated adults believe they know how to answer: it is having two X chromosomes, they say! Possession of two X chromosomes (or more accurately, possession of at least one X chromosome and no Y chromosome), it is thought, is both necessary and sufficient for being a human female, and this further explains the possession of typical female characteristics. The distinctive essential feature is thus not only thought to be necessary and sufficient for having typical external female genitalia, but it is also explanatory in the sense that Devitt seeks. This proposal does not in fact withstand scrutiny. While the majority of females do not have Y chromosomes, there are some that do; a small but substantial portion of the population have 46, XY karyotypes, but have typical external female genitalia.18 There are numerous ways in which this situation can arise. For example, various mutations can lead to androgen insensitivity, meaning that that cell receptors do not respond to androgenic hormones, so cells are unaffected by the presence of these hormones. Alternatively, there can be any number of changes to the biochemical processes involved in producing these androgens: for example, there may be a range of alterations affecting the synthesis of testosterone from cholesterol, or there may be changes in enzyme levels that prevent testosterone from being broken down into dihydratestosterone. For each of the determinate ways in which this developmental pathway can be altered, there are a range of different genetic mutations, or combinations of mutations, that may be responsible. Further, this is far from an exhaustive list of the ways in which a 46, XY karyotype may be associated with typical external female genitalia (Kolon, 2008).19

17 This discussion in this section is concerned with the biological categories of sex (e.g. male/female), not with the social categories of gender (e.g. man/woman). The two are, of course, often confused, in accordance with our tendency to quintessentialize gender to a very high degree. While it is often pointed out that the latter has no intrinsic essential bases, the corresponding point is less frequently made in the case of the former. The discussion here should illustrate that even the seemingly simple categories of sex are in fact highly complex and resistant to delineation in any simple essentialist terms. 18 This discussion abstracts away from cases of individuals with ambiguous genitalia, which can (though need not) result from many of the conditions described here, and from other conditions besides. 19 As a further level of complication, consider the phenomena of mosaicism and chimerism. These are conditions in which the cells within a given individual have different genetic makeups. For example an individual with mosaicism/chimerism might have some cells with karyotype 46, XX and others with 46, XY, or some with 47, XXY and others with 46, XY, or some with SRY mutations and others without, and so on. The effect on the phenotype depends on a range of factors, for example the relative proportion of the different cells, their location and degree of distribution around the body, and so on so forth. For example, if someone has a mix of 46, XX and 46, XY cells (which is rare but possible), they may have typical male genitalia, typical female genitalia, or ambiguous genitalia, depending among other factors on the proportion and location of the two cell types.

136

Sarah-Jane Leslie

To return to Devitt’s argument, suppose we have a number of newborn females, all with typical external genitalia, and suppose we seek to explain why their genitals are formed as they are. For each individual infant, there will be a genetic component to the explanation (as well as an environmental component), but this genetic component need not be the same for each infant. There may be considerable overlap in explanation (the majority of individuals with female genitalia have 46, XX karyotypes, and—crucially—lack the factors that lead such a karyotype to be associated with typical male genitalia), but in the case of some individuals, the explanation will go by a rather different route. One cannot infer a universal and essential intrinsic genetic component on the basis of shared morphological characteristics, even within a given sex of a given species. Of course, Quintessentialists that we are, we find such inferences appealing to the point of being almost irresistible. As a result, one might be tempted to rescue the idea by ‘setting aside’ cases of 46, XY females, so as to preserve the idea that there is a genetic essence shared by all and only females that explains their typical morphological features. (I want to be absolutely clear that I am in no way attributing such a response to Michael Devitt, however.) There is no question that the individuals in question have the relevant morphological feature—typical external female genitalia—so if the explanation of why females have this feature is to make reference to a genetic essence shared by all females, the only option is to insist that these individuals are not, in fact, female. That is, if we insist that only 46, XX individuals are female, then we can rescue the idea that the kind human female has an essence, and further that this essence explains the typical morphological features of the category. The real females are the 46, XX females, the others are aberrations, freaks, not really female. I say this is no more than quintessentialism in its most unjustified and pernicious form. Biology does not provide neat, essentialized categories of this sort, however much quintessentialism demands them.20 There are a number of different biological routes to the phenotype, and the natural world does not label one or another of these as the genuine, true, real, or normatively right one.21 Sadly, a number of female athletes have had their careers and even their lives ruined by the quintessentialist way of thinking, which is blind to this natural fact. Further, the sort of phenomenon illustrated here is widespread in the biological world. In general, within a given species, individuals who share a common phenotypic feature need not share intrinsic microstructural bases that gave rise to the feature. One simple illustration of this phenomenon 20 For a number of years, 46, XY females were referred to in the clinical literature as “male pseudohermaphrodites,” a label which is no longer in use. Even in the context of medicine, quintessentialism can leave its traces, but it is hard to maintain in the face of a real investigation of nature. 21 Ah, but the variations described in this section tend to be incompatible with reproduction— is this not biology’s way of signaling that these individuals are not really female? Not more so here than with any genetically based condition that results in infertility.

Essence and Natural Kinds

137

involves what are known as phenocopies—individuals who display environmentally induced, non-hereditary phenotypic features which are identical to genetically induced, hereditary features displayed by other individuals. This can occur within a particular species—for example, the Common Rabbit (Oryctolagus cuniculus) frequently sports black fur. This trait—black fur—can arise relatively straightforwardly from a given rabbit’s genetic make-up— that is, as an inherited trait that manifests itself across various environments. The trait can also occur, however, among the Himalayan breed under certain environmental conditions. Himalayan rabbits, when raised in moderate temperatures, have white body fur with black tails, noses, and ears; if they are raised in cold temperatures, however, they develop wholly black fur. The genetic, developmental and biochemical pathways to black fur differ significantly between the two cases (Baum et al., 2010; Sawin, 1932); there is no common intrinsic explanation for the possession of black fur by the members of O. cuniculus. (And conversely, there need be no genetic difference between a Himalayan rabbit with black fur and one with mostly white fur.) Perhaps one might be tempted to think that this example is misplaced because having black fur is not a species-wide characteristic of rabbits, or perhaps because appeal was made to a particular breed of rabbit (which, one might be tempted to think, would be such that its members share an intrinsic microstructural basis for the trait). This is not so. Consider, for example, that having three toes on the hind feet is a characteristic property of guinea pigs (Cavia porcellus). Possession of this phenotypic property is due to a flexible interaction-effect between a number of factors, both genetic and non-genetic— that is, there is a generous range of pathways, all of which lead to having three hind toes. Individual guinea pigs can differ significantly from each other with respect to these factors and yet each have three toes, as was demonstrated in a series of classic experiments conducted by Sewall Wright (1934). Alternatively, consider the salamander Ambystoma talpoideum. A. talpoideum can undergo either metamorphosis from its larval stage to its adult stage, or else can undergo pedomorphosis, that is, it becomes sexually mature while retaining its larval characteristics. A general characteristic property of A. talpoideum is that it tends to undergo pedomorphosis if it is raised in a favorable aquatic environment. However, the genetic basis for this property has been found to differ between two populations of A. talpoideum that are separated by as little as 15 km (Harris, Semtlisch, Wilbur, and Fauth, 1990). The developmental pathway leading to metamorphosis can be disrupted in any number of ways (e.g. by increased prolactin secretion, by blockage of thyrotropin-releasing hormone secretion, by reduced sensitivity to thyroxine, or by other means), so the differences in the genetic basis of pedomorphism likely translate into significant differences in the biochemical processes undergone by individuals from the two populations (Harris et al., 1990; WestEberhard, 2003). Further, the two populations of salamanders are such that the females typically produce approximately eighteen eggs each; however, once

138

Sarah-Jane Leslie

more the genetic basis for this property differs between the two populations (Harris et al., 1990). Such examples abound. Phenotypic traits are the upshot of complex biochemical processes controlled in most cases by a variety of genes. Differences in the genetic level need not translate into differences in the biochemical processes, and even differences in the biochemical processes need not translate into differences in phenotypic traits (see, e.g. Zuckerkandl and Villet (1988) for detailed discussion of the latter point; for example, in some cases there may be an important trade-off between chemical affinity and chemical concentration). More generally, biologists and geneticists often make use of the notion of canalization—the idea that the development of a phenotypic trait, particularly one that is important to survival, can arise despite variation in genes, environment, and the resulting developmental pathways. Canalization of a trait insures that the trait is stably expressed in the face of underlying genetic variation (e.g. Gibson and Wagner, 2000; Waddington, 1942).22 To put the point in what is perhaps more familiar terminology, we might say that phenotypic traits often exploit a certain multiple realizability at the microstructural level. Notice that it would not be an adequate defense of Devitt’s idea to say that the human female, the trout, the rabbit, the guinea pig, and the salamander may be the less common sort of case, and that phenocopying and canalization may not be universal phenomena. Whether or not this is true, it is not the sheer number of such cases, but the way in which they subvert the idea that species as such must meet the relevant (quint?)essentialist standards for being a respectable natural kind. The general theoretical point is this: Every macroscopic phenotypic property depends on a massive number of biochemical reactions, originating with the genes themselves but continuing along the entire developmental pathway, at each point potentially subject to environmental influences, influences from other genes, and so on. If there was a one-to-one correspondence between sameness of macroscopic phenotypic properties and sameness of biochemical pathway, then—simply put—there would be considerably more phenotypic variation than there in fact is. The idea that one may generally infer shared ‘underlying’ features from a shared phenotypic feature just does not pass empirical muster. Nature is, fortunately, far more robust. 2.3.2. Extrinsic Species—Essence and Plenitude If species do not have intrinsic essences, if there are no general microstructural genetic facts of the sort that Devitt, Putnam and (more offhandedly) Kripke posit as making for species membership then does this mean that species have no essences whatsoever? Recently, several philosophers of biology have argued that species can indeed be said to have essences, albeit extrinsic ones. 22 Consider, for example, the wide range of developmental pathways that can all lead to typical external female genitalia, as discussed.

Essence and Natural Kinds

139

For example, Samir Okasha (2002) and Joseph LaPorte (1997, 2004) and others have argued that species do indeed have essences, in the sense that there are necessary and sufficient conditions for belonging to a given species. These necessary and sufficient conditions are relational, however, and do not yield the result that species-membership is an essential property of the individual (e.g. it will not be an essential or even necessary property of Socrates that he be a member of Homo sapiens, contra Wiggins (1980)). Nonetheless, if such species essences are to be found, then there would at least be a determinate same animal relation, of the sort that Kripke and Putnam appeal to. (This result would be arrived at only on the basis of some charitable reconstruction—in particular we would need to supplement the discussion of ostensive definition with the notion that the same animal relation in fact amounts to the same species relation, which is in effect to grant a solution to the qua problem (Devitt and Sterelny, 1987). Since basic-level kinds do not map neatly onto species, or in some cases onto any taxon (Dupré, 1993), this is already quite a generous amendment to the view.) Broadly speaking, on most of the popular accounts, to be a member of a given species, for example Panthera tigris, is to be part of a particular chunk of the genealogical nexus, which begins with a particular speciation event, and ends with either another speciation event, or with extinction. A more specific version of the species concept is the phylogenetic (or cladistic) species concept. LaPorte offers the following as an elaboration of the phylogenetic species concept: “a species name like ‘Panthera tigris’ is to be defined something like as follows: ‘Panthera tigris = df the lineage descending from ancestral population P and terminating in speciation or extinction,’ P being . . . an appropriate population in the lineage that gave rise to today’s tigers.” (2004, p. 54) Of course we are still owed an account of what a speciation event consists in, as many philosophers of biology have noted (e.g. Dupré, 1993; Kitcher, 1984; LaPorte, 2004; Okasha, 2002; Sterelny and Griffiths, 1999). There are two dominant lines of thought here: that speciation events occur when populations become reproductively isolated from each other, and that such events occur when populations come to occupy distinct ecological niches. Both lines of thought are frequently appealed to by biologists and zoologists, and each plays an important theoretical and explanatory role in at least some contexts of biological inquiry. Before considering the significance of having multiple scientifically useful ways of characterizing speciation events, it is worth briefly noting that, whatever details one embraces, species membership is not an intrinsic matter (see, e.g., Dupré, 1981, 1993; Kitcher, 1984; LaPorte, 1997, 2004; Okasha, 2002; Sober, 1980; Sterelny and Griffiths, 1999). For example, the dominant accounts of the species concepts deliver the result that, if two populations are geographically separated in such a way as to prevent interbreeding, and result in the two populations occupying distinct ecological niches, then these populations constitute distinct species (at least on many tenable construals of these speciation conditions). No reference is here made to the intrinsic

140

Sarah-Jane Leslie

properties of the members of the species; thus the possibility remains that two individuals with identical intrinsic properties could belong to distinct species (as a result of belonging to such geographically separated populations). As another (somewhat simplified) illustration, consider the case of an individual with significant genetic mutations that render the individual incapable of either interbreeding with his parents’ population or occupying their ecological niche. If there are other similar individuals around to form a distinct interbreeding population or co-occupy a distinct ecological niche, then a speciation event may be underway, and the individual—sometimes termed a “hopeful monster”—may count among the first population of a new species. If no such fellows are to be found, however, the individual will simply count as an atypical and infertile member of his parents’ species.23 Thus, even if we grant that we may be able to specify necessary and sufficient conditions for membership in a species, we are very far indeed from the sort of essentialism Kripke and Putnam had in mind (LaPorte, 2004; Okasha, 2002). By now it should occasion no surprise to note that the pursuit of extrinsic essential conditions for species membership quickly runs into its own complications. As the above discussion indicates, the phylogenetic species concept can be supplemented in at least two useful and explanatory ways—that is, one can characterize speciation events as arising when two populations cease to interbreed, or when two populations come to occupy distinct ecological roles. These two characterizations of speciation events can lead to competing results: for example, it is possible for two populations to occupy different ecological niches and yet still interbreed, and so on. Further, the phylogenetic species concept (with its premium on shared ancestry) is not even the only concept on the table; there is, for example, the biological species concept (BSC), which takes interbreeding to be the sole determiner of species membership. These two different species concepts may deliver different results, even if the phylogenetic species concept is supplemented by the interbreeding approach to the speciation question. As an illustration, suppose a species splits into two (i.e. non-interbreeding) species S1 and S2 at time t1. Then later at time t2, S1 and S2 cease to be reproductively isolated from each other, and form an interbreeding population. The phylogenetic species concept will still entail that there are two distinct species at t2 (distinguished by their different ancestry going back to time t1), whereas the BSC in its most straightforward form would posit a single species at t2. A further species concept takes occupancy of an ecological niche to be the sole determiner of species membership, rather than simply a determiner of speciation, and so on so forth.24 23 Devitt (2008) writes as though the standard species concepts are not in conflict with his Intrinsic Biological Essentialism. The fact that the standard species concepts allow for such cases suggests otherwise, however (Barker, 2010; Ereshefsky, 2010). Thus Devitt’s proposal is highly revisionary. 24 Devitt (2008) correctly points out that these species concepts fall short of telling us what it is to be a tiger—that is, while they may tell us what it is for a population to be a species, or what it is for two organisms to be conspecific, they do not tell us what it is for an organism

Essence and Natural Kinds

141

Notably, despite the fact that these competing concepts deliver different results about whether two individuals are conspecific or not, and also about how many species should be recognized, and so on, they all have a claim to being scientifically useful. This has led a number of prominent philosophers of biology, beginning with Philip Kitcher (1984) and John Dupré (1981, 1993) to argue for pluralism concerning species concepts (see also Ereshefsky, 1998; Holter, 2009; Rosenberg, 1994; Stanford, 1995, and others).25 The extent of the pluralism is not limited to the range of species concepts laid out above either; for each of the seemingly particular species concepts, there are any number of ways of spelling out the details. Kitcher notes this in a recent paper: I proposed [in 1984] that there were many different species concepts, appropriate for different purposes of inquiry Both Dupré and I, however, tended to think in terms of manageable pluralism, or limited promiscuity; for my part, I took the Biological Species Concept to be one among a number of contenders. The real trouble, however, is that the Biological Species Concept itself allows for indefinitely many ways of development, depending on how one approaches the notions of population and of reproductive isolation. (2007, pp. 300–301)

The arguments of Kitcher and his colleagues on this point seem to me completely convincing: if we seek to characterize species—to give necessary and sufficient conditions for membership in a species—then we will need to countenance a plenitude of sets of such conditions, each of which is scientifically useful and appropriate. Thus even the most charitable attempts to reconstruct essentialism about species face the problem of there being too many candidate to be a tiger (or a lion or a zebra, and so on). Some qualification is in order here: we cannot produce any such purely qualitative specification of the essence of a kind like the tiger or the lion. However, if we are allowed to directly refer to particular individuals—e.g. a particular founding population—then we can provide such necessary and sufficient conditions (e.g. to be a tiger is to be descended from this ancestral population prior to any further speciation events occurring among the population’s descendents). Devitt’s point here is correct, though I disagree with his interpretation to the effect that this indicates a significant hole in biology/philosophy of biology. An alternative interpretation would simply be that, in light of the best science and philosophy, we have found that fullblooded, purely qualitative necessary and sufficient conditions are only to be found at the level of what it is to be a species, or what it is to be conspecific, not at the level of what it is to be a tiger. Of course, since tigers constitute a species, this places some necessary conditions on being a tiger, though they will fall short of being also sufficient conditions. This state of affairs parallels a relatively common view in the discussion of philosophical essentialism: many philosophers hold that we can provide qualitative necessary and sufficient conditions for, e.g. being a person (where this may be distinct from being a member of the species Homo sapiens; compare the literature on personal identity), and further that Socrates is essentially a person. However, they do not hold that there are qualitative necessary and sufficient conditions for being Socrates (though there may be non-qualitative ones, e.g. coming from this sperm and this egg). The situation in philosophy of biology does not seem worryingly different. 25 There are a number of points of disagreement and discussion in the literature, for example whether pluralism is compatible with realism about species, and I will not attempt to reconstruct all the subtleties here.

142

Sarah-Jane Leslie

species essences for us to be able to say: here is the essence of the kind Panthera tigris—the set of properties that all and only actual and possible members of the kind possess. Biological science has not uncovered such hidden conditions that properly govern our use of the term “tiger;” biology does not issue in a privileged same animal relation. At best it has uncovered a multitude of distinct sets of conditions and relations, each with equal claim.26

2.4. Chemical Kinds and Essence Kripke/Putnam essentialism about biological kinds looks to be untenable, even if one attempts to reconstruct it in the most charitable of ways. But what of chemical kinds, such as water and gold? Surely here the world cooperates with our quintessentialist intuitions. After all, don’t we know the essences of water and gold to be H2 O and the element with atomic number 79 respectively? More generally, has not chemistry actually discovered a privileged same substance relation (namely being either the same element or the same compound)—one which could be pressed into duty in the following sort of manner: Suppose I point to a glass of water and say ‘this liquid is called water’ . . . My “ostensive definition” has the following empirical presupposition: that the body of liquid I am pointing to bears a certain sameness relation (say, x is the same liquid as y, or x is the sameL as y) to most of the stuff I and other speakers in my linguistic community have on other occasions called ‘water’ . . . The key point is that the relation sameL is a theoretical relation: whether something is or is not the same liquid as this may take an indeterminate amount of scientific investigation to determine. Moreover, even if a “definite” answer has been obtained either through scientific investigation or through the application of some “common sense” test, the answer is defeasible: future investigation might reverse even the most ‘certain’ example. (Putnam, 1975, pp. 224–5)27

26 I believe that this is anyway a widespread consequence of philosophical essentialism in its most general form—i.e. the thesis that there are two ways in which an item may have a property, namely essentially or accidentally. Put very crudely, where one might suppose there is a single item with n properties, essentialism has difficulty avoiding the consequence that there is something on the order of the cardinality of the power of set of n items instead, differing only in which of the n properties are had essentially vs accidentally. While a number of refinements are in order, essentialism without plenitude on a vast scale is, I think, an untenable position. For the details and the arguments, see Leslie (2011), and Leslie and Johnston (in preparation). 27 Here and elsewhere, Putnam speaks of a “same liquid” relation. The appeal to chemistry is more plausible, however, if this is characterized as a “same substance” relation instead. For example, the dominant use of the term “water” in chemistry is a phase-neutral one, so that water can also occur in solid and gas phases. Nothing in the criticisms of essentialism that follow hang on this adjustment; if anything, this adjustment makes the case for essentialism more plausible. For example, the oft-made claim that “water is identical to H2 O” is clearly false as stated if “water” is a phase-specific term. For then, by parity of reasoning and the transitivity of identity, the absurdity that water is identical to ice is derivable. See Johnston (1997). In what follows, I will use the term “water” in a phase-neutral sense, unless otherwise specified.

Essence and Natural Kinds

143

The idea is elaborated in the following manner: [A] liquid with the superficial properties of “water” but a different microstructure isn’t really water . . . Suppose, now, that I discover the microstructure of water—that water is H2 O. At this point I will be able to say that the stuff on Twin Earth that I earlier mistook for water isn’t really water. (Putnam, 1975, pp. 232–3)

As this passage makes clear, Putnam believes that the same substance relation is determined by ‘microstructural’ properties, and further that ‘H2 O’ specifies such a microstructure. Also emphasized in these last two passages is the notion that these microstructural properties—these determiners of the same substance relation—are discovered. The same theme is also found in Kripke, who makes it clear that we may not only discover, contrary to what we believed, that two substances are distinct, but also that two substances are in fact one and the same: [I]f this substance [H2 O] can take another form—such as the polywater allegedly discovered in the Soviet Union with very different identifying marks from that of what we now call water—it is a form of water because it is the same substance, even though it doesn’t have the appearance by which we originally identified water. (1980, pp. 128–9).

And more generally: [S]uppose some items (let the set of them be I) are discovered and are believed to belong to a new kind K. Suppose later it is discovered that the items in I are indeed of a single kind; however, they belong to a previously known kind, L. Observational error led to the false initial belief that the items in I possessed some characteristic C excluding them from L. (1980, p. 136, emphasis added).

The basic idea behind this essentialist model for chemical kinds is clear: there is a privileged relation same substance as that science discovers, and which determines the extension of our natural kind terms. Further, this scientific discovery has already been made, at least in the case of water and gold, which have been found to be identical to H2 O and the element with atomic number 79 respectively. It may now seem reasonable to infer from these examples and others that, for chemical kinds, the same substance relation can be characterized as a) same atomic number in the case of elements, b) same chemical formula in the case of compounds. Such ‘microstructural’ essences are the determiners of the extension of chemical kind terms, and macroscopic manifest features are at best a guide to these true essential features. We can now pose the question: does science actually deliver such a privileged same substance relation, or do we simply have once again the quintessentialist intuition that science must do so? That is, does the Kripke/Putnam discussion simply retell the scientific facts, or is the science once again distorted, perhaps by quintessentialist convictions? In this section I will argue that the relevant sciences deliver no such privileged same substance relations. The idea that same atomic number/same

144

Sarah-Jane Leslie

chemical formula captures this relation simply will not do, especially in the case of compounds; nor is there a unique privileged substitute relation. At best, science delivers a number of candidate same substance relations, many of which rely on the macroscopic and the manifest rather than the microstructural. Finally, I discuss some examples that suggest that the chemical kinds of thought and talk may not line up especially neatly with the kinds of chemistry indicating, once again that the extensions and truth conditions of our thought and talk are governed more by our false quintessentialist mindset than by any “scientifically discoverable” essentialism, contrary to what philosophers often suppose.28 2.4.1. Essentialism and Compounds Let us begin with the case of water, which we are told, is “identical” to H2 O. One initial observation that has not been properly assimilated is that “H2 O” is not in fact a microstructural description (Needham, 2000, 2002, 2011; van Brakel, 1986). It is simply a compositional formula, specifying the proportion of hydrogen to oxygen that is to be found in the substance. It is not a specification of the molecular structure of water; in fact, it can apply to phases of water that have no molecular structure such as ice X, as noted below. Moreover, historically speaking compositional formulas were introduced and widely used at a time before atomism in chemistry was generally accepted; in this sense, compositional formulas do not even involve a commitment to the existence of atoms or molecules, the actual constituents of microstructure (Needham, 2000). Further, since compositional formulas only specify the proportion of elements in a compound, they do not even distinguish between structural isomers, which are chemically distinct compounds composed of the same proportions of elements. For example, the compositional formula “C2 H6 O” applies equally to ethyl alcohol and dimethyl ether, but these two compounds are distinct along any number of chemically important dimensions. As it happens, water has no structural isomers, however many, many compounds do. (And many compounds have many, many structural isomers. For example, in the case of large organic molecules—even ones that consist only of carbon and hydrogen—the number of distinct compounds corresponding to a given compositional formula can number in the millions and greater (Smith, 2011).) A somewhat better candidate for capturing the ‘microstructure of water’ would be a structural formula—a formula which would specify to some approximation how the relevant molecule is structured, for example by indicating that the two hydrogen atoms are each bonded to the central oxygen atom. There are numerous ways of representing structural formulas, many of them

28 This next section draws on the excellent work of a number of philosophers of chemistry, including Michael Weisberg, Joseph LaPorte, Jaap van Brakel, and especially on the exemplary work of Paul Needham, whose papers should be mandatory reading for philosophers working on natural kinds. The presentation and many of the illustrations are my own, however.

Essence and Natural Kinds

145

graphic in nature, so as to illustrate the geometry of the bonding patterns. For simplicity, I will use ‘H–O–H’ as an abbreviation for a more informative structural description of the molecule. The real complications arise when we learn that only in the gas phase does water primarily consist of such H–O–H molecules. That is, the gas phase of water can be reasonably modeled (though with some idealization) by supposing that there are a large number of discrete, separate H–O–H molecules, but this is not true for the liquid phase, nor for the many ice phases. For example, there are at least fifteen different forms of ice that have been observed experimentally, each formed under different combinations of temperature and pressure. (And more are theoretically projected, including a metallic phase of ice!) Interestingly, one form of ice, ice X, has no molecular structure at all; its microstructural arrangement is such that there is no distinction between the intra-molecular bonds and the inter-molecular bonds. Rather ice X is an atomic solid, composed of hydrogen and oxygen atoms arranged in a particular lattice structure. It does not contain any molecules to be described, so the question of how to describe its molecules does not arise. It still contains hydrogen and oxygen in a 2:1 ratio, however, and so falls under the compositional formula “H2 O,” despite lacking any molecular structure. In the liquid phase, there is molecular structure, but unlike the gas phase, only a sub-portion of the molecules are H–O–H molecules. For example, some of the H–O–H molecules dissociate into H + and OH– ions. Some of these ions attach to H–O–H molecules to form complex ions, for example H3 O + and (H2 O)OH– (since the complexity of the molecules is rapidly increasing, I will use condensed structural formulas such as these). Other molecules bond together to form polymers (i.e. chains of repeating structural units) of arbitrary length: two H–O–H molecules will combine to form a (H2 O)2 molecule, which can combine with another H–O–H molecule to form (H2 O)3 , and so on and so forth. The polymers and the ions can further combine, so in a given sample of liquid water there may be molecules of the form (H2 O)n H + and (H2 O)n OH– for any reasonable n. The patterns of disassociation and bonding in liquid water happen continuously—what is a polymer one moment may disassociate into smaller parts the next, and so on (see Needham, 2000, 2011 for more discussion). This is the beginning (and just the beginning!) of a serious description of the microstructure of water. The idea that “H2 O” constitutes a description of water’s microstructure may have its roots in the incorrect idea that the liquid and solid phases of water (and other substances) are basically like the (still somewhat idealized conception of the) gas phase—which we may say is composed by H–O–H molecules. That is, the idea may persist that the liquid and solid phases have the same microstructural composition, only with the molecules staying closer together in these phases. In fact the important manifest properties of water—for example its relatively high boiling point, which ensures it is liquid at standard temperatures and pressures and therefore drinkable; the fact that ice is less dense than liquid water, which means

146

Sarah-Jane Leslie

fish can survive the freezing of lakes by swimming lower; and so on—are due specifically to that fact that liquid and solid water do not simply consist of H–O–H molecules, but rather a variety of dynamically related complex molecules. In light of this complexity, what are we to make of the very idea of a microstructural essence for water and other chemical kinds? As a general rule, a compositional formula like “H2 O” does not provide a good candidate for capturing the essence of a substance. For example, recall that the compositional formula “C2 H6 O” applies equally to ethyl alcohol (condensed structural formula “C2 H5 OH”) and dimethyl ether (condensed structural formula “(CH3 )2 O)”, but these two compounds are quite dissimilar (and certainly considered distinct by chemists), and we should not want the result that they are somehow to be said to share a microstrucural essence, and so be the same substance. In general, a compositional formula does not specify a unique chemical kind; in fact, in some cases the same compositional formula applies to millions of distinct chemical kinds (Smith, 2011). But on the other hand, relying on a structural formula to specify the composition of the constituent molecules will also be problematic as general strategy, as the case of water illustrates—if we attempt to insist on something to the effect that water is essentially composed of H–O–H molecules, then we will derive the result that water exists only in the gaseous phase. And certainly, no description of molecules will suffice to characterize ice X, since it does not contain molecules.

2.4.2. The ‘Same Substance’ Relation Perhaps the solution will be to examine the microstructure of water and other chemical kinds in even greater detail, and hope that such examination will uncover the elusive essence of chemical kinds. Returning to the case of liquid water (which, remember, was supposed to be the easy case, the poster child for Kripke/Putnam essentialism!), we can begin by noting that the rates and patterns of polymer formation and ionic disassociation are sensitive to the temperature and pressure. Thus, if we take seriously the idea of deriving detailed microstructural descriptions, water at 4 degrees C will have a considerably different microstructure than water at 98 degrees C. Any microstructural specification of the essence of water would have to take account of this variation without delivering the result that these are different substances. As Needham puts it, In view of the sensitivity of the distribution of particular determinate forms of the many determinable features to prevailing conditions, the actual description of the relevant microscopic structure—which must be sufficiently relaxed to accommodate all these variations without including too much—would be a truly daunting task . . . But first and foremost, it should be asked why this vast range of microscopic structures is associated with one and the same substance kind, rather than a genus of related substances. (2000, p. 20)

Essence and Natural Kinds

147

Given the extensive variation at the microscopic level, is water even a single substance? And more generally: what criteria, if any, does chemistry give us for determining how many different substances are present in a given sample— or put differently, does chemistry give us a way of determinately answering Putnam’s question ‘is x the same substance as y’? We have seen that neither sameness of compositional formula nor sameness of molecular structure provides a satisfying answer to the question (because of the possibility of isomers and the possibility of massive variation on the microstructural level respectively). As I understand Needham’s work, the real answer is two-fold: 1) such criteria for sameness are far more readily found at the macroscopic level than at the microscopic level, and 2) there is nonetheless no unique and privileged way of answering the question at either level. Put in a slightly simplified way, the point is this: When it comes to manifest kinds, chemists typically find unexpected complexity and variability in the constituents of the examples of the kinds in question, and they are fully prepared to note this complexity and variation without any surprise, because they treat the manifest and macroscopic features as partly determinative of what it is to be, for example, water or gold—contrary to what Kripke and Putnam envisaged. Chemistry’s partial reliance on and respect for the macroscopic and the manifest is at odds with the Kripke/Putnam project.29 Consider their repeated reference to common microstructure as wholly determinative of the extension of our natural kind terms. Consider, for example, Kripke in the following passage: what I would have wanted to do would have been to discuss the part about gold being a metal. This, however, is complicated because first, I don’t know too much about chemistry. Investigating this a few days ago in just a couple of references, I found in a more phenomenological account of metals the statement that it’s very difficult to say what a metal is. (It talks about malleability, ductility, and the like, but none of these exactly work.) On the other hand, something about the periodic table gave a description of elements as metals in terms of their valency properties. This may make some people think right away that there are really two concepts of metal operating here, a phenomenological one and a scientific one which then replaces it. This I reject. (1980, pp. 117–8)

The thought is that the valency properties or “something like that” is what makes things deserve the name of metals, rather than the macroscopic features of malleability, ductility and the like. In fact, there are at least two scientific

29 Putnam, especially in his later work (but also in “The Meaning of ‘Meaning”’ p. 239), does not seem to take such a consistently hard line on this, and allows instead that our interests also matter when determining whether one thing counts as the same substance as another. See Hacking (2007a) for a helpful discussion. Again, here and throughout the paper, I use “the Kripke/Putnam project” to denote the one that is rooted in Naming and Necessity and “The Meaning of ‘Meaning” ’ as it is widely interpreted—that is, the view that has been so historically influential in philosophy.

148

Sarah-Jane Leslie

concepts of metal in play in chemistry—one denoting a set of elements in the periodic table, the other applying when a solid of any atomic constitution displays certain macroscopic properties—as is illustrated by the hypothesis that there exists a metallic phase of ice. This latter is characterized in macroscopic terms, but is nonetheless an important scientific notion, despite Kripke’s suggestion to the contrary. This dichotomy between the “scientific” and the “phenomenological”—especially if the latter includes notions such as malleability and ductility—is artificial and cannot be sustained. Indeed, there is an interesting further question as to whether the elements that are called metals are so-called because they typically form metallic solids at normal temperatures and pressures, where what counts as a metallic solid is “phenomenologically” determined, that is determined by such properties as malleability, ductility, and the like. Returning to the question of the same substance relation, one promising place to look is what is known as Gibbs’ phase rule, which relates 1) the number of independent substances in a closed system (thus offering a way to ‘count substances’), 2) the number of phases in the system (e.g. solid, liquid, etc.), and 3) the number of ‘intensive variables’ (e.g. temperature, pressure), which can be varied independently. The phase rule does deliver the result that water is a single substance.30 However, it must be understood that Gibbs’ rule counts 30

The phase rule is as follows: c – f + 2 = V: (where c is the number of independent substances, f is the number of phases, and V is the number of intensive variables) If V ≥ 0, then the system will be able to be in equilibrium under at least some conditions. Suppose c = 1—that is, that we have but a single substance in the system—then the phase rule allows us to make empirical predictions about how that substance will behave with respect to its different phases, as a function of temperature and pressure. For example, if we have but a single substance in the closed system, Gibbs’ phase rule lets us derive that there will be a unique combination of temperature and pressure at which three phases of the substance will be at equilibrium—that is, there will be a specific temperature and a specific pressure at which, e.g. solid, liquid, and gas phases can all co-exist in equilibrium (V = 0, so no intensive variables can be varied; this is sometimes known as the triple point). The phase rule also lets us predict that, if c = 1, there will be a range of temperature and pressure pairs at which two phases (e.g. solid and liquid) can co-exist in equilibrium, but also that temperature and pressure cannot be varied independently. However, if again c = 1, but only one phase is present, the temperature and pressure can each vary independently. To return to the question of whether water is a single substance or not, the answer is that, according to Gibbs’ phase rule, it is. The behavior of water fits this thermodynamic profile of a single substance. As Needham illustrates: [S]uppose the two-phase quantity of water [liquid and gas] is confined at fixed temperature to a closed container fitted with a piston. Any attempt to decrease the pressure by raising the piston and increasing the volume available to the water will fail (as long as two phases remain), because the liquid phase will decrease in volume as matter goes into the gas phase to maintain the pressure. Similarly, attempting to increase the pressure by decreasing the volume will be thwarted by the volume of the gas decreasing as matter goes into the liquid phase to maintain the pressure. Continuing the processes of increasing or decreasing the volume will eventually lead to a single phase being formed (gas in the first case, liquid in the second), which is bivariant, so that pressure and temperature can vary independently. This behaviour is in accordance with the phase rule for a system for which the number of independent substances, c, is one (ms, p. 6). This macroscopic criterion for individuating substances thus delivers the result that water is a single substance.

Essence and Natural Kinds

149

independent substances. If multiple substances are related to each other within the system by a reversible chemical reaction, the phase rule will count fewer substances than seems correct. Consider, for example, what happens to a system that contains H2 O heated to around 2000K (Here I follow Needham, ms.). The H2 O begins to dissociate into H2 and O2 molecules, which in turn recombine into H2 O molecules. The phase rule, along with empirical observations about the behavior of the system, delivers the result that there is only one substance in the system, despite the fact that in general H2 , O2 , and H2 O are not the same chemical kind. What constitutes a distinct independent substance in a system is not in general the same as what constitutes a distinct substance simpliciter. The fact that the phase rule deems water to be a single substance could thus be interpreted in the following way: water is a conglomeration of distinct substances, but because they dissociate and recombine as they do, the distinct substances are not independent. An alternative way of distinguishing substances is to consider the ‘entropy of mixing’. Suppose we have two separate containers of gas sitting side by side, and we then remove the divider between them. If the two gases truly consist of the same substance, there will be no increase in entropy as a result of allowing them to combine. If, however, they are distinct substances then— even if they are highly similar—the entropy of the system (a measurable quantity) will increase (Needham, 2000). This turns out to be an exceedingly fine-grained test. For example, in addition to structural isomers (e.g. ethyl alcohol and dimethyl ether), there are also spin isomers, for example orthoand para-hydrogen. A molecule of ortho-hydrogen is one in which the nuclear spins of the two H atoms are aligned parallel; a molecule of para-hydrogen is one in which the nuclear spins are anti-parallel. What is termed hydrogen gas and denoted by the formula “H2 ” normally consists of a mixture of these two spin isomers (at a ratio of about 1:3 of para-hydrogen to ortho-hydrogen). The two spin isomers of hydrogen can be separated in a laboratory situation, however. Further, if a container of ortho-hydrogen is allowed to mix with a container of para-hydrogen, then the entropy of the system will increase; by this test, there are thus (at least) two distinct substances in a normal sample of hydrogen. There are also spin isomers of water, ortho- and para-water, which are distinguished in the same way, depending on whether the spins of the hydrogen nuclei are aligned or not in the H–O–H molecule (and in H–O–H links in the polymer chains, and so on). Just as with hydrogen, normal samples of water contain both spin isomers, in a ratio of 1:3 (para to ortho). For various reasons, ortho- and para-water have proved more difficult to separate than ortho- and para-hydrogen, however several groups of researchers have recently succeeded in doing so (Kravchuk, Reznikov, Tichonov, Avidor, Meir, Bekkerman, and Alexandrovich, 2011; Tikhonov & Volkov, 2002). By the entropy of mixing test, ortho- and para-water count as distinct substances, suggesting that normal samples of water are in fact mixtures of two distinct substances. This disagrees with the results of the phase rule test, however, which suggests that water comprises a single substance.

150

Sarah-Jane Leslie

So, are ortho- and para-water really distinct substances? The answer, in some sense, is clearly “yes.” For example, efforts to separate them are clearly not fool’s errands—attempts to separate a single thing into identical components. On the other hand, it is a distinction that is not generally made—water is normally treated as a single substance in chemistry. Tikhonov and Volkov, the first researchers to separate these spin isomers, found that the separated ortho- and para-water took 55 minutes and 26 minutes respectively to settle back into the normal 3:1 ratio when kept in a liquid state (far longer than they had expected), and estimated that ortho- and para-ice frozen at -18 degrees C would take several months to return to the normal ratio. They also note that “[t]he OP [ortho/para] separation procedure realized is quite straightforward and may occur in nature—in soil, atmosphere, living organisms, and cosmic objects. The scope and the role of this phenomenon are yet to be studied” (p. 2363). It would thus not be totally beyond the realm of scientific possibility (though it may stretch the bounds) to imagine a version of the Twin-Earth thought experiment that compared ortho- with para-water. Would the visitors from Ortho-Earth be making an error when they first applied the word “water” to the substance on Para-Earth? Do these spin isomers bear the “same substance” relation to each other, or not? By some tests, yes (certainly both are H2 O); by other tests, no. There are scientifically useful ways of dividing up substances according to which they are the same substance, but there are other scientifically useful ways of dividing up substances according to which they are distinct. Nor are we awaiting the results of further investigation to settle which way of dividing up substances is the scientifically real one. A similar example which is often discussed in the literature concerns isotopic variation (LaPorte, 2004; Needham, 2008; Weisberg, 2005). Two atoms are isotopes if they share the same atomic number (number of protons in the nucleus) but differ in mass due to the number of neutrons in their nuclei. Elements are standardly defined in chemistry by their atomic numbers, thus two distinct isotopes are nonetheless examples of the same element. Hydrogen atoms, for example, usually have a single proton and no neutron in their nuclei (protium); however, there are two naturally occurring stable hydrogen isotopes, one with a single neutron (deuterium), the other with two neutrons (tritium). Thus, protium, deuterium, tritium all fall under the same element kind (though of course they can obviously be distinguished from each other). Isotopic variation does not tend to affect the qualitative nature of the chemical interactions—if the most common variant of an element enters into a given chemical reaction, so will its isotopes. Sometimes isotopic variation can affect the speed of a chemical reaction, but much of the time the effect is negligible. Thus the addition, as it were, of a neutron—in contrast to the addition, say, of a proton—has little effect on an atom’s chemical behavior. In this way, identifying elements with their atomic numbers regardless of isotopic variation is a very reasonable and useful way to classify elements and the compounds containing them.

Essence and Natural Kinds

151

In the case of hydrogen, however, isotopic variation has non-trivial effects— due in part to the fact that hydrogen atoms are so small (e.g. deuterium has roughly twice the atomic weight of protium). Consider, then, “heavy water,” which has deuterium in place of protium. Is heavy water the same as regular water? By the most standard ways of categorizing compounds—which is insensitive to isotopic variation—they would count as the same substance. (Though of course by other ways of categorizing, they would be distinct—for example, there is an entropy of mixing associated with combining isotopic variants. Nothing in the observation that elements group isotopic variants together implies this is somehow a distinction without a difference.) Imagine, then, a miniature version of Twin-Earth, in which a lake that contains heavy water is newly discovered, and let us pose the question of whether the substance in it is the same substance as the water encountered thus far. It should be clear from the foregoing that chemistry does not deliver a univocal answer to the question. Let us consider instead, then, how the members of population at large would be inclined to answer the question. While this paper in general cautions against reliance on intuition in philosophy, this would seem one occasion where we might nonetheless confidently “intuit” how this would go: the substance in this newly discovered lake would most certainly not be judged to be the same substance that comes out of our taps, etc. (LaPorte, 2004; Needham, 2008; Weisberg, 2005). This is for the straightforward reason that heavy water, when drunk in suitable quantities, causes sterility, neurological problems, and death. If approximately 25 percent of one’s daily intake of protium oxide is replaced by deuterium oxide, one rapidly becomes sterile. If higher proportions of deuterium oxide are consumed, this soon leads to “acute neurological symptoms, liver hyperplasia, anemia, other symptoms, and eventually death” (Kushner, Baker, and Dunstall, 1998, p. 81). These effects, broadly speaking, are due to the fact that deuterium in heavy water exchanges with protium in organic molecules in the body. The bond between carbon and deuterium is considerably stronger than the bond between carbon and protium, which slows and otherwise disrupts a number of biochemical processes (Kushner et al., 1998). I submit that we would not consult with chemists before concluding this substance was most certainly not the same substance as our water. Suppose an overenthusiastic philosopher should try to argue as follows: “in general isotopic variation has very little impact on chemical properties, and so elements are identified by their atomic numbers, and compounds by the elements in them, and so it has been discovered that the substance in this new lake is in fact the very same substance as the one we are used to drinking!” An apt response to the philosopher here might be: “well then, drink up!” The moral of this particular tale is that there are occasions on which chemistry does not deliver a univocal answer to the “same substance” question, yet the significance of the manifest properties of the substance nonetheless do provide such an answer. We have seen thus far that chemistry does not provide a privileged same substance relation. Further, in the case of compounds, we also saw that there

152

Sarah-Jane Leslie

is no generally effective way of specifying the essence of the compound (even abstracting away from isotopism and spin isomerism)—simply specifying the composition of the compound in terms of its constituent elements is too broad (since it does not distinguish structural isomers), yet specifying the structure of the component molecules cuts too finely (since, e.g. water in its liquid phase is composed of many, many different types of molecules, while in its ice X phase it is not composed of molecules at all). 2.4.3. Essentialism and the Elements What, though, of the elements? Certainly, the elements are more hospitable to essentialism than the compounds: in the context of chemistry, elements are quite clearly identified by their atomic numbers. (Again, though, as the case of isotopes illustrates, this does not clearly constitute a privileged same substance relation, as opposed to a same element relation.) Of course, for the Kripke/Putnam thesis to be correct, it is not enough that science provide the essences of kinds, it also has to be the case that these essences determine the extension of the relevant terms/concepts. Thus, an interesting question to explore is the extent to which the natural language terms/concepts for elements have their extensions determined simply by the relevant atomic number. Consider, for example, Kripke’s assertion that “a material object is (pure) gold if and only if the only element contained therein is that with atomic number 79” (1980, p. 138). Since the word “gold” is used in scientific contexts as the English name for the element Au, there is a sense in which something like this biconditional is unassailable as a necessary truth—this is the sense in which something is copernicum if and only if it has atomic number 112. The term “copernicum” is not a term of ordinary non-technical language; copernicum is not a manifest kind—rather it specifically names a certain element: the one with atomic number 112. That “gold” can have such a usage also is not something I would dispute. But Kripke intends to make a far more substantive claim: in fact he refers not to elements, but to material objects made of gold, and thus intends his statement to apply to the manifest kind gold—the kind which has played such an important role in our economic history. Gold’s historical and economic significance depends, of course, on its manifest properties—for example, its lack of chemical reactivity make it very stable and free from corrosion, and its luster and general appearance make it desirable as jewelry and a store of wealth, while its malleability/ductility make it relatively easy for artisans and currency minters to work with. But the manifest properties of an elemental substance depend on more than simply its atomic constituents: they also depend on how those constituents are arranged. When the atoms of a single element bond in different structural patterns, this is known as allotropy. As it happens, Au does not exhibit allotropy at normal temperatures and pressures, but rather forms only what is known as a face-centered cubic lattice, which accounts for many of the substance’s

Essence and Natural Kinds

153

important manifest properties, such as its ductility and malleability. It is not, however, necessary that Au atoms bond this way; for example, at very high pressures, Au atoms form instead a hexagonal close-packed lattice. The resulting substance is far less malleable and ductile than the face-centered cubic lattice of normal temperatures and pressures. If such allotropes of Au existed at normal temperatures and pressures, would we be so quick to assert that being composed of Au atoms is necessary and sufficient for falling under the manifest kind gold, despite lacking these economically significant properties? Or consider the element tin (Sn), which has in addition to its familiar shiny, metallic, malleable form, a non-metallic allotrope (stable below 13.2 degrees C), which is dull, brittle, and powdery. What if Au displayed such allotropic variation? Would such a dull, brittle, and powdery substance fall under the manifest kind gold? One version of this question is a linguistic question: whether the word “gold” would be applied to such a counterfactual substance. A further question is whether this stuff would be considered the same substance as regular gold—for example, whether it would be assimilated into the same economic role as regular gold. Whatever the linguistic outcome would be in such circumstances, the economic outcome seems quite easy to predict: this counterfactual non-metallic Au would not play the same economic role as regular gold. (Of course if there were a cheap and easy way to convert the one to the other, that would change the economic role of the former, but only in a way that is parasitic on the role of the latter.) What of the linguistic question? My view here is that the linguistic question is less probative than it is often taken to be, in large part because—like most linguists who study lexical semantics— I believe that polysemy is a widespread and important phenomenon. That is, a single phonological form, for example /gold/, may be associated with a range of related (sometime very closely related) but nonetheless distinct meanings. (Polysemy differs from lexical ambiguity, since in the latter case, the meanings are typically not at all related.) Thus, in the counterfactual circumstances we are considering, the possibility that the phonological form /gold/ might come to apply to the non-metallic allotrope would not show that the two were taken to belong to the same manifest kind. In this respect, economic behavior is more telling—since it requires one to “to put your money where your mouth is,” as it were. A real-world example of this form concerns the allotropes of carbon. Carbon has many allotropes, including graphite and diamond. It is an undeniable point that we do not consider graphite and diamond to belong to the same manifest kind, and the discovery that both consist of the 6th element of the periodic table does not alter this one bit (Johnston, 1997; LaPorte, 2004). It was a surprising chemical discovery that diamond contained only the humble carbon atom, but this did not somehow undermine diamond’s economic role. In the case of the 6th element, distinct phonological forms (at least in English) are used for each of the allotropes, and a further one (/carbon/) for the element itself. Had history gone differently, so that this 6th element

154

Sarah-Jane Leslie

was first discovered in the context of analyzing diamond, it may have been that this 6th element would have been named “diamond” also, paralleling the situation with gold. The surprising discovery would then have been that lowly graphite contained “diamond” atoms. This is a counterfactual history in which “diamond” is more polysemous than it actually is, not one in which the manifest kind diamond was discovered to include just anything composed of the 6th element. For another illustration of polysemy as it applies to terms which have a use in which they pick out elements, let us consider the phonological form /sodium/. As background information, this phonological form applies to the 11th element of the periodic table. Pure macroscopic quantities of this 11th element can take the form of a metal, which is so reactive that it almost never occurs naturally, and must be covered with paraffin when stored in a laboratory. The phonological form /sodium/ also applies to this dangerous manifest kind. More commonly, however, the 11th element is found as a component of an ionic compound. Table salt is such a compound: it is composed of ions of sodium and chlorine. Sodium ions are typically ingested in the form of table salt, and play a number of extremely important roles in human physiology. for example sodium ions—in conjunction with potassium ions— help maintain blood pressure. The phonological form /sodium/ is again used without qualification in everyday discourse to pick out these ions. Consider, then, a doctor’s injunction to increase one’s sodium intake, as a means of combating low blood pressure. The doctor here is not encouraging her patient to consume a highly reactive and toxic metal, though of course there is a perfectly standard use of /sodium/ that applies exactly to such a substance. Rather, the doctor is suggesting that her patient consume more sodium ions, probably in the form of table salt. (Note that for the halogens the two senses are distinguished phonologically: e.g. chlorine vs chloride, where the latter refers only to ions.) In fact there are at least three related but distinct uses of /sodium/ in play here: the use that concerns ions as in “increase your intake of sodium”, the use that concerns a metallic manifest kind as in “sodium is an explosive metal,” and the use that concerns an element as in “sodium is the element with atomic number 11.” It would however be a mistake—and in this case, a deadly one—to suppose that philosophical musings on chemistry establish that one of the latter two uses employs the real sense of “sodium.” (Suppose someone makes this mistake and interprets his doctor’s advice on increasing his sodium intake as the injunction to ingest a lump of sodium metal. One would not fancy the chances of a malpractice lawsuit pursued on the grounds that “sodium” is univocal and therefore applies to the lump of metal.) Similar remarks apply to “oxygen.” “Oxygen” can be used to pick out the 8th atomic element, and in this sense both O2 gas and O3 gas are composed of oxygen. The manifest kind oxygen—the one we must continually inhale to stay alive—includes only the former gas. The latter gas, ozone, is in an excellent sense simply not oxygen. (Inhaling ozone in sufficient concentrations

Essence and Natural Kinds

155

rapidly leads to respiratory distress and eventually death.) When a paramedic declares that a patient needs oxygen, it is clear what she means, and what she means does not include ozone. 2.4.4. Beyond Chemistry: Quintessence and Chemical Kinds When two manifest kinds are distinguished on the basis of their observable, macroscopic properties, could chemistry (or any other science) ever actually discover that they are one and the same? Of course we might discover they are closely related in some respect, and we may introduce a term that applies to both. Since polysemy is a widespread feature of the lexicon, it is possible that this umbrella term will be phonologically the same as the term for one of the two kinds. But if the two kinds genuinely differ in their important manifest properties, what discovery could lead us to discount this difference? For one thing, even if we allow a distinction between manifest properties and “scientifically significant” ones, the former do not float free of the latter— if there is a difference at the manifest level, there will be a difference at the “scientifically significant” level, and who is to say that difference is somehow unimportant? Consider, for example, the fact that chemists generally use the term “water” in a phase-neutral sense. This does not mean that chemistry somehow obliges us to discount differences in phase—on the contrary, chemists are at times very interested in phase differences. What then of the term “water”? It is clearly used in a phase-neutral sense in most scientific contexts (such as when a chemist writes a paper), but in other contexts in a phase-specific sense (such as when the same chemist requests a glass of water). The term “water” is polysemous in this way. Moreover, some of the best reasons for supposing that water in any sense comprises a single substance depend on its manifest and macroscopic properties, such as its behavior with respect to Gibbs’ phase rule. In contrast to the Kripke/Putnam view, which supposes that when it comes to the determination of reference, the manifest and macroscopic is held wholly hostage to the microstructural and microscopic, the picture that emerges in chemistry itself is one on which macroscopic and manifest properties figure prominently. The majority of examples considered thus far have been cases in which (a naïve and misleading conception of) chemistry might have led us to consider distinct manifest kinds to be one and the same. What then of the converse? Are there examples in which chemistry suggests that there is more than one kind present where there is arguably but one manifest kind? If we take seriously the Kripke/Putnam emphasis on microstructure, then ortho-water and para-water provide such an example. Moreover, Joseph LaPorte (2004) elegantly and persuasively argues that the case of jade also fits this profile, and that the case of jade in China constitutes a historical Twin-Earth scenario that breaks in a way that is the opposite of what the Kripke/Putnam theory would predict.

156

Sarah-Jane Leslie

Recall that there are in fact two chemically distinct compounds with extremely similar manifest properties—nephrite (Ca2 (MgFe)5 Si8 O22 (OH)2 ) and jadeite (NaAl(SiO3 )2 ). For approximately 5,000 years, nephrite played an extremely important economic and cultural role in Chinese society, and was more highly prized than even gold. In 1784, the Qianlong Emperor annexed Northern Burma. The region contained a number of mines that were rich in jadeite, a substance that the Chinese had not previously encountered. Jadeite and nephrite are exceedingly similar in their manifest properties, though not strictly identical; highly skilled artisans could tell the difference, since jadeite is slightly harder and heavier than nephrite. However, both substances share their most significant and prized manifest properties, for example their ability to sustain intricate carving. The Qianlong Emperor loved jade, and welcomed the newfound resources as more of the same substance. Jadeite was seamlessly assimilated into the same economic, artistic, and cultural role as nephrite, and the term “yü” was readily applied to jadeite and nephrite alike (Hacking, 2007b, LaPorte, 2004).31 This example again illustrates the significance of manifest properties in determining what counts as the same substance as something else. One should not, however, be mislead into supposing that manifest properties are all that matters. On this point, I believe that Kripke and Putnam were completely correct; however, I would interpret their insight in terms of our being, at heart, Quintessentialists. We do not believe that looking like a K, feeling like a K, and so on suffices for being a K. To truly be a K requires that one have the quintessence of Ks. On this point, it is instructive to note that, while both jadeite and nephrite have been embraced as true jade, other substances have been deemed false jade, despite having very similar appearances. That is, the manifest kind jade is not simply characterized in terms of superficial descriptive properties—the possibility exists for there to be exceedingly convincing cases of fake jade. (The Internet is replete with advice to buyers on how to avoid being conned in this way by unscrupulous jewelers.) The situation may, I think, be characterized in the following way: we believe that there is a quintessence of jade, which is shared by both jadeite and nephrite, but not by other stones, no matter their appearance. (This leaves open whether we may believe that there are still minor quintessential differences between jadeite and nephrite—perhaps in the way that animal kinds below the basic-level are believed to differ in minor ways that are unimportant for general purposes. Compare Bengal tigers and Siberian tigers.) As the case

31 Might this also be a case of polysemy? To best assess this question, it would be important to examine whether there were any notable economic or cultural distinctions drawn between the two. If none were drawn, then it would be reasonable to suppose that this is not a case of polysemy (at least not for most speakers of the language; for the artisans the situation may be different). I cannot rule out this possibility, but I do not think that chemical differences necessitate it either.

Essence and Natural Kinds

157

of jade suggests, quintessence does not map well onto chemical properties— though of course we may be inclined to suppose it does, just as we do in the biological case. A large difference in chemical properties does not entail a difference in supposed quintessence. I would further propose that the converse holds as, for example, the case of water and heavy water would illustrate: regardless of how chemically similar two substances may be, if one sustains life while the other ends it, these will be believed to have decidedly different quintessences. As a final illustration of quintessentialist thinking, consider the distinction between natural and synthetic diamonds. Current technological advances have made it possible to produce diamonds in the laboratory which are chemically and physically identical to naturally occurring ones. In some cases it is possible for a trained gemologist with enough time and equipment to tell them apart, but even this is not fail-safe. Further, the ways in which they are distinguished often involve detecting imperfections that are more likely to be present in the natural diamonds than in the synthetic ones. It is certainly not possible for a consumer or even for the average jeweler to be able to tell them apart. Nonetheless, consumers are willing to pay far more for natural diamonds and, further, the overwhelming majority of consumers choose the natural ones. These two kinds of diamonds are clearly distinguished in people’s thinking: there are websites which offer advice on how to “get your fiancée to accept a synthetic diamond,” and conversely websites in which people (including, we may suppose, the target fiancées) declare that they will “never settle for a synthetic diamond.” Further, the qualifier “synthetic” is often replaced with a more derogatory term, such as “imitation,” “simulated,” or even “fake,” which strongly suggests that these diamonds are not the same thing as the natural ones, but rather merely mimic their surface and chemical properties while being deeply and underlyingly different. As one jewelry website succinctly puts it, “Some customers may feel content to buy simulated diamonds, knowing that they could never afford a near flawless diamond. Others though, insist upon buying only genuine diamonds, as these are a more natural treasure and very rare, especially when compared to laboratoryproduced items.”32 The quintessentialist intuition here is that man-made diamonds just aren’t the same things as natural ones. They are no more than forgeries, simulacra of the real things, whose natures are shaped deep in the ground under high pressure over many years. No lab can confer this nature on anything, because the nature does not simply reside in the chemical structure. The man-made stones simply do not have the quintessence of diamonds. Reflecting on chemical composition and bonding structure does not silence this complaint 32 . Accessed 24 November 2012.

158

Sarah-Jane Leslie

because our quintessentialist dispositions project differences even when there is no difference in the underlying chemistry.

3.

CONCLUSION

We are Quintessentialists, and we have a tendency to interpret science through the lens of our quintessentialist intuitions. Thus, at first glance, it may seem to us that science confirms our quintessentialist intuitions, when in fact the opposite is true. Closer inspection of the scientific findings shows that— against the very natural and highly intuitive expectations of Kripke and Putnam—there are not good prospects for the discovery of microstructural biological or chemical essences that determine the references of our natural kind terms. There are cases in which members of the same species have quite different DNA and conversely ones in which members of different species have quite similar DNA (recall for example the cutthroat and the rainbow trout). There are cases of sameness of manifest kind with different microstructures (as with jade) and cases of different manifest kinds with the same microstructure (as with natural and synthetic diamonds). Further, neither biology nor chemistry delivers a privileged same species or same substance relation. The highly intuitive character of the Kripke/Putnam picture derives from our quintessentialist mindset; it does not reflect scientific or metaphysical facts about the world. If we did not have something like quintessentialist beliefs then we would not have the relevant Kripke/Putnam intuitions. We do, however, have such beliefs, and we have them despite the fact that the world does not conform to them. Our intuitions here reflect only facts about us, not facts about the deep nature of reality. More generally, this discussion of one of the most intuitive views in philosophy illustrates the urgent need to examine the sources of our “philosophical intuitions.” In particular, it is important to understand whether a set of intuitions may be due to an early-developing and deep-rooted bias to see the world in a particular way. Such a finding does not entail that the intuitions are misleading, but it does suggest that they will be extremely compelling even if they are misleading. Thus, such a finding gives us reason to subject the intuitions and any conclusions derived from them to further scrutiny. If a conclusion is supported solely by such intuitions, this may be good reason to remain skeptical concerning the conclusion. A question remains: what if quintessentialism not only implicitly shapes our basic beliefs but also our semantic intentions; what if in using natural kind terms we often do in fact intend to refer to kinds united by a common quintessence? Science shows there are not in general good candidates to be the common quintessences, so what happens to reference under this assumption? It is either much less determinate than the work of Kripke and Putnam led us to suppose or, if this is different, it is determinate only relative to a quintessentialist fiction that we all tend to share. A full exploration of this question must be left for another occasion.

Essence and Natural Kinds

159

REFERENCES

Allendorf, F. W. and Leary, R. F. 1988. Conservation and distribution of genetic variation in a polytypic species: The cutthroat trout. Conservation Biology, 2, 170–84. Astuti, R., Solomon, G. E. A., and Carey, S. (2004). Constraints on conceptual development. Monographs of the Society for Research in Child Development, 69. Atran, S., Medin, D. L., Lynch, E., Vapnarsky, V., Ucan Ek’, and Sousa, P. (2001). Folkbiology doesn’t come from folkpsychology: Evidence from Yukatec Maya in cross-cultural perspective. Journal of Cognition and Culture, 1, 4–42. Barker, M. (2010). Specious Intrinsicalism. Philosophy of Science, 77(1) 73–91. Basch, S. H. (1973). The intrapsychic integration of a new organ: A clinical study of kidney transplantation. Psychoanalytic Quarterly, 42, 364–84. Baum, P., Schmid, R., Ittrich, C., Rust, W., Fundel-Clemens, K., et al. (2010). Phenocopy—A strategy to qualify chemical compounds during hit-to-lead and/or lead optimization. PLoS ONE, 5, e14272. Belk, R. W. (1990). Me and thee versus mine and thine: How perceptions of the body influence organ donation and transplantation. In J. Shanteau and R. J. Harris (eds.), Organ donation and transplantation: Psychological and behavioral factors. Washington, DC: American Psychological Association, 139–49. Cole, E. R., Jayaratne, T. E., Cecchi, L. A., Feldbaum, M., and Petty, E. M. (2007). Vive la difference? Genetic explanations for perceived gender differences in nurturance. Sex Roles, 57, 211–22. Coley, J. D., Medin, D. L., and Atran, S. (1997). Does rank have its privilege? Inductive inferences within folkbiological taxonomies. Cognition, 64, 73–112. Coley, J. D., Medin, D. L., Proffitt, J. B., Lynch, E., and Atran, S. (1999). Inductive reasoning in folkbiological thought. In D. L. Medin and S. Atran (eds.), Folkbiology. Cambridge, MA: Bradford, 205–32. Devitt, M. (2008). Resurrecting biological essentialism, Philosophy of Science, 75, 344–82. and Kim Sterelny (1987). Language and Reality, Oxford, Basil Blackwell. Diesendruck, G. (2001). Essentialism in Brazilian children’s extensions of animal names. Developmental Psychology, 37, 49–60. and Gelman, S. A. (1999). Domain differences in absolute judgments of category membership: Evidence for an essentialist account of categories. Psychonomic Bulletin and Review, 6, 338–46. Dupré, John (1981). Natural kinds and biological taxa. Philosophical Review, 90, 66–90. (1993). The disorder of things: Metaphysical foundations of the disunity of science. Cambridge, MA: Harvard University Press. Ereshefsky, M. (1998). Species pluralism and anti-realism. Philosophy of Science, 65, 103–20. Ereshefsky, M. (2010). What’s wrong with the new biological essentialism? Philosophy of Science, 77, 674–85. Frazier, B. N. and Gelman, S. A. (2009). Developmental changes in judgments of authentic objects. Cognitive Development, 24, 284–92. Gelman, R. (1987). Cognitive Development: Principles guide learning and contribute to conceptual coherence. Paper presented at the Meeting of the American Psychological Association, Division I, New York.

160

Sarah-Jane Leslie

Gelman, S. A. (2003). The essential child: Origins of essentialism in everyday thought. New York: Oxford University Press. and Coley, J. D. (1990). The importance of knowing a dodo is a bird: Categories and inferences in 2-year-old children. Developmental Psychology, 26, 796–804. Gelman, S.A., Collman, P., and Maccoby, E. E. (1986). Inferring properties from categories versus inferring categories from properties: The case of gender. Child Development, 57, 396–404. Gelman, S. A. and Markman, E. M. (1986). Categories and induction in young children. Cognition, 23, 183–209. (1987). Young children’s inductions from natural kinds: The role of categories and appearances. Child Development, 58, 1532–41. Gelman, S. A. and O’Reilly, A. W. (1988). Children’s inductive inferences within superordinate categories: The role of language and category structure. Child Development, 59, 876–87. Gelman, S. A. and Rhodes, M. (2012). “Two-thousand years of stasis”: How psychological essentialism impedes evolutionary understanding. In K. S. Rosengren, S. Brem, E. M. Evans, and G. Sinatra (eds.), Evolution challenges: Integrating research and practice in teaching and learning about evolution. Oxford: Oxford University Press, 3–24. Gelman, S. A. and Wellman, H. M. (1991). Insides and essences: Early understandings of the nonobvious. Cognition, 38, 213–44. Gendler, T. S. (2007). Philosophical thought experiments, intuitions, and cognitive equilibrium. Midwest Studies in Philosophy, 31, 68–89. (2008a). Alief and Belief. Journal of Philosophy,105, 634–63. (2008b). Alief in action (and reaction). Mind and Language, 23, 552–85. (2011). On the epistemic costs of implicit bias. Philosophical Studies, 156, 33–63. Ghiselin, M. (1987). Species Concepts, Individuality, and Objectivity. Biology and Philosophy, 2, 127–43. Gibson, G. and Wagner, G. (2000). Canalization in evolutionary genetics: A stabilizing theory? BioEssays, 22, 372–80. Gil-White, F. J. (2001) Are ethnic groups biological ”species” to the human brain?: Essentialism in our cognition of some social categories. Current Anthropology, 42, 515–54. Gottfried, G. M. and Gelman, S. A. (2005). Developing domain-specific causalexplanatory frameworks: The role of insides and immanence. Cognitive Development, 20, 137–58. Graves, J. L. (2001). The emperor’s new clothes: Biological theories of race at the millennium. New Brunswick, NJ: Rutgers University Press. Greenwald, A. G. and Banaji, M. R. (1995). Implicit social cognition: Attitudes, selfesteem, and stereotypes. Psychological Review, 102, 8. Greenwald, A. G., Poehlman, T. A., Uhlmann, E., and Banaji, M. R. (2009). Understanding and using the Implicit Association Test: III: Meta-analysis of predictive validity. Journal of Personality and Social Psychology, 97, 17–41. Hacking, I. (2007a). Putnam’s theory of natural kinds and their names is not the same as Kripke’s. Principia, 11, 1–24. (2007b). The contingencies of ambiguity. Analysis, 67, 269–77.

Essence and Natural Kinds

161

Harris, R. N., Semtlisch, R. D., Wilbur, H. M., and Fauth, J. E. (1990). Local variation in the genetic basis of paedomorphosis in the salamander Ambystoma talpoideum. Evolution, 44, 1588–603. Hayward, C. and Madill, A. (2003). The meanings of organ donation: Muslims of Pakistani origin and white English nationals living in North England. Social Science and Medicine, 57, 389–401. Heyman, G. and Gelman, S. A. (2000). Beliefs about the origins of human psychological traits. Developmental Psychology, 36, 663–78. Hirschfeld, L. A. (1996). Race in the making: Cognition, culture, and the child’s construction of human kinds. Cambridge, MA: MIT Press. Holter, B. D. (2009). The ontology of species: A radically pluralistic perspective. Ph.D. thesis, Washington State University. Hood, B. M. and Bloom, P. (2008). Children prefer certain individuals over perfect duplicates. Cognition, 106, 455–62. Hull, D. L. (1965). The effect of essentialism on taxonomy: Two thousand years of stasis. British Journal for the Philosophy of Science, 15, 314–26 and 16, 1–18. Inspector, I., Kutz, Y., and David, D. (2004). Another person’s heart: Magical and rational thinking in the psychological adaptation to heart transplantation. The Israeli Journal of Psychiatry and Related Sciences, 41, 161–73. Jayaratne, T. (2001). National sample of adults’ beliefs about genetic bases to race and gender. Unpublished raw data. Jayaratne, T. E., Gelman, S. A., Feldbaum, M., Sheldon, J. P., Petty, E. M., and Kardia, S. L. R. (2009). The perennial debate: Nature, nurture, or choice? Black and White Americans’ explanations for individual differences. Review of General Psychology, 13, 24–33. Johnson, C. N. (1990). If you had my brain, where would I be? Children’s understanding of the brain and identity. Child Development, 61, 962–72. and Jacobs, M. G. (2001, April). Enchanted objects: How positive connections transform thinking about the very nature of things. Poster presented at the Society for Research in Child Development, Minneapolis, MN. Johnston, M. (1997). Manifest Kinds. Journal of Philosophy, 94, 564–83. Kalish, C. W. (1998). Natural and artificial kinds: Are children realists or relativists about categories? Developmental Psychology, 34, 376–91. Kamp, H. and Partee, B. (1995). Prototype theory and compositionality. Cognition, 57, 129–91. Keil, F. C. (1989). Concepts, kinds, and cognitive development. Cambridge, MA: MIT Press. Keller, J. (2005). In genes we trust: The biological component of psychological essentialism and its relationship to mechanisms of motivated social cognition. Journal of Personality and Social Psychology, 88, 686–702. Kitcher, Philip (1984). “Species”, Philosophy of Science, 51, 308–33. (2007). Does ‘race’ have a future? Philosophy and Public Affairs, 35, 293–317. Kolon, T. F. (2008). Disorders of sexual development. Current Urology Reports, 9, 172–7. Kravchuk, T., Reznikov, M., Tichonov, P., Avidor, N., Meir, Y., Bekkerman, A., and Alexandrovich, G. (2011). A magnetically focused beam of ortho-water. Science, 331, 319–21.

162

Sarah-Jane Leslie

Kripke, S. (1980). Naming and necessity. Cambridge, MA: Harvard University Press. Kushner, D. J., Baker, A., and Dunstall, T. G. (1999). Pharmacological uses and perspectives of heavy water and deuterated compounds. Canadian Journal of Physiological Pharmacology, 77, 79–88. Lanie, A. D., Jayaratne, T. E., Sheldon, J. P., Kardia, S. L. R., and Petty, E. M. (2004). Exploring the public understanding of basic genetic concepts. Journal of Genetic Counseling, 13, 305–20. LaPorte, J. (1997). Essential Membership. Philosophy of Science, 64, 96–112. (2004). Natural kinds and conceptual change. Cambridge: Cambridge University Press. Leslie, S. J. (2011). Essence, Plenitude, and Paradox. Philosophical Perspectives, 25, 277–96. Leslie, S. J. and Johnston, M. (In preparation). Essence and accident: Against limited variety. Lewontin, R. C., Rose, S., and Kamin, L. (1984). Not in our genes: Biology, ideology, and human nature. New York: Pantheon Books. Lizotte, D. J. and Gelman, S. A. (1999). Essentialism in children’s categories. Poster presented at the Cognitive Development Society, Chapel Hill, NC. Luhmann, C. and Coley, J. D. (2000). Domain specific relations between typicality and absolute category membership. Poster presented at Northeastern University College of Arts & Sciences Experiential Education Expo, May 2000. Maglo, K. N. (2011). The case again biological realism about race: From Darwin to the post-genomic era. Perspectives on Science, 19, 361–91. Mahalingam, R. (2003). Essentialism, culture, and power: Representations of social class. Journal of Social Issues, 59, 733–49. Mayr, E. (1982). The growth of biological thought. Cambridge, MA: Harvard University Press. (1988). Toward a new philosophy of biology: Observations of an evolutionist. Cambridge, MA: Belknap Press of Harvard University Press. (1991). One long argument: Charles Darwin and the genesis of modern evolutionary thought. Cambridge, MA: Harvard University Press. Meyer, M., Gelman, S. A., and Leslie, S. J. (Submitted). My heart made me do it: Children’s essentialist beliefs about heart transplants. Meyer, M., Leslie, S. J., Gelman, S. A, and Stilwell, S. M. (2013). Essentialist beliefs about bodily transplants in the United States and India, Cognitive Science, doi: 10.1111/cogs.12023. Murphy, G. (2002). The big book of concepts. Cambridge, MA: MIT Press. Needham, P. (2000). What is Water? Analysis, 60, 13–21. (2002). The discovery that water is H2 O. International Studies in the Philosophy of Science, 16, 205–26. (2008). Is water a mixture?—Bridging the distinction between physical and chemical properties. Studies in History and Philosophy of Science, 39, 66–77. (2011). Microessentialism: What is the argument?, Nous, 45, 1–21. (Manuscript). The phase rule and the notion of substance. Stockholm University. Nemeroff, C. and Rozin, P. (1994). The contagion concept in adult thinking in the United States: Transmission of germs and interpersonal influence. Ethos, 22, 158–86.

Essence and Natural Kinds

163

Newman, G., Herrmann, P., Wynn, K., and Keil, F. C. (2008). Biases towards internal features in infants’ reasoning about objects. Cognition, 107, 420–32. Newman, G. and Keil, F.C. (2008). ‘Where’s the essence?’: Developmental shifts in children’s beliefs about the nature of essential features. Child Development, 79, 1344–56. Okasha, Samir (2002). Darwinian metaphysics: Species and the question of essentialism. Synthese 131, 191–213. Pennell, W. and Barton, B. A. (1996). Principles of Salmonid culture. Netherlands: Elsevier. Phillips, W., Shankar, M., and Santos, L. R. (2010). Essentialism in the absence of language? Evidence from rhesus monkeys (Macaca mulatta). Developmental Science, 13, F1–7. Phillips, W. and Santos, L. R. (2007). Evidence for kind representations in the absence of language: Experiments with rhesus monkeys (Macaca mulatta). Cognition, 102, 455–63. Putnam, H. (1975). “The Meaning of ‘Meaning’.” In H. Putnam, Mind, language, and reality. Cambridge: Cambridge University Press, 215–71. Putnam, H. (1992). Is it necessary that water is H2 O? In L. E. Hahn (ed.), The Philosophy of A. J. Ayer. Peru, Illinois: Open Court, pp. 429–455. Rhodes, M. and Gelman, S. A. (2009a). A developmental examination of the conceptual structure of animal, artifact, and human social categories across two cultural contexts. Cognitive Psychology, 59, 244–74. (2009b). Five-year-olds’ beliefs about the discreetness of category boundaries for animals and artifacts. Psychonomic Bulletin and Review, 16, 920–4. Rosch, E. H. (1973). Natural categories. Cognitive Psychology, 4, 328–50. (1975). Cognitive representation of semantic categories. Journal of Experimental Psychology, 104, 192–233. (1978). Principles of Categorization. In E. Rosch and B. B. Lloyd (eds.), Cognition and Categorization. Hillsdale, NJ: Erlbaum, 27–48. Mervis, C. B., Gray, W. D., Johnson, D. M., Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382–439. Rosenberg, A. (1994). Instrumental biology or the disunity of science. Chicago: University of Chicago Press. Rozin, P. and Nemeroff, C. (2002). Sympathetic magical thinking: The contagion and similarity “heuristics”. In Gilovich, T., Griffin, D., and Kahneman, D. (eds.) Heuristics and biases: The psychology of intuitive judgment. Cambridge: Cambridge University Press, 201–16. Salmon, N. U. (1979). How Not to Derive Essentialism From the Theory of Reference. Journal of Philosophy, 76, 703–25. Samarapungavan, A. and Wiers, R. (1997). Children’s thoughts on the origin of species: A study of explanatory coherence. Cognitive Science, 21, 147–77. Sanner, M. A. (2001a). Exchanging spare parts or becoming a new person? People’s attitudes toward receiving and donating organs. Social Science and Medicine, 52, 1491–9. (2001b). People’s feelings and beliefs about receiving transplants of different origins—questions of life and death, identity, and nature’s border. Clinical Transplantation, 15, 19–27.

164

Sarah-Jane Leslie

Sawin, P. B. (1932). Hereditary variation of the chin-chilla rabbit: In coat and eye color. Journal of Heredity, 23, 39–46. Shtulman, A. (2006). Qualitative differences between naïve and scientific theories of evolution. Cognitive Psychology, 52, 170–94. and Schulz, L. (2008). The relation between essentialist beliefs and evolutionary reasoning. Cognitive Science, 32, 1049–62. Simons, D. J. and Keil, F. C. (1995). An abstract to concrete shift in the development of biological thought: The insides story. Cognition, 56, 129–63. Smith, M. B. (2011). Organic chemistry: An acid-based approach. Boca Raton: Taylor & Francis. Sober, Elliott (1980). “Evolution, population thinking and essentialism”, Philosophy of Science, 47, 350–83. Solomon, G. E. A. and Johnson, S. C. (2000). Conceptual change in the classroom: Teaching young children to understand biological inheritance. British Journal of Developmental Psychology, 18, 81–96. Sousa, P., Atran, S., and Medin, D. L. (2002). Essentialism and folkbiology: Evidence from Brazil. Journal of Cognition and Culture, 2, 195–223. Springer, K. and Keil, F. C. (1989). On the development of biologically specific beliefs: The case of inheritance. Child Development, 60, 687–48. Stanford, P. Kyle (1995). “For pluralism and against realism about species”, Philosophy of Science, 62, 70–91. Sterelny, Kim, and Paul Griffiths (1999). Sex and Death. Chicago: University of Chicago Press. Stich, S. (2009). Reply to Sosa. In M. Bishop and D. Murphy (eds.), Stich and His Critics. Oxford: Blackwell, 228–36. Sylvia, C. and Novak, W. (1997). A Change of Heart: Boston: Little Brown. Taylor, M. G. (1996). The development of children’s beliefs about social and biological aspects of gender differences. Child Development, 67, 1555–71. Rhodes, M. and Gelman, S. A. (2009). Boys will be boys; cows will be cows: Children’s essentialist reasoning about gender categories and animal species. Child Development, 79, 1270–87. Templeton, A. (1999). Human Races: A Genetic and Evolutionary Perspective. American Anthropologist, 100, 632–50. Tikhonov, V. I. and Volkov, A. A. (2002). Separation of water into its ortho and para isomers. Science, 296, 2363. van Brakel, Jaak (1986). The Chemistry of Substances and the Philosophy of Mass Terms. Synthese, 69, 291–324. Waddington, C. H. (1942). Canalization of development and the inheritance of acquired characteristics. Nature, 3811, 563–5. Walker, S. (1999). Culture, domain-specificity, and conceptual change: Natural kind and artifact concepts. British Journal of Developmental Psychology, 17, 203–19. Waxman, S. R., Lynch, E. B., Casey, K. L., and Baer, L. (1997). Setters and samoyeds: The emergence of subordinate level categories as a basis for inductive inference. Developmental Psychology, 33(6), 1074–90. Waxman, S. R., Medin, D. L., and Ross, N. (2007). Folkbiological reasoning from a cross-cultural developmental perspective: Early essentialist notions are shaped by cultural beliefs. Developmental Psychology. 43(2), 294–308.

Essence and Natural Kinds

165

Weinberg, J., Nichols, S., and Stich, S. (2001). Normativity and epistemic intuitions. Philosophical Topics, 29, 429–60. Weisberg, M (2005). Water is Not H2 O. In D. Baird et al. (eds.) Philosophy of Chemistry: Synthesis of a New Discipline. New York: Springer, 337–45. West-Eberhard, M. J. (2003). Developmental plasticity and evolution. New York: Oxford University Press. Wiggins, David (1980). Sameness and substance. Oxford: Blackwell. Wright, S. (1934). The results of crosses between inbred strains of guinea pigs, differing in their number of digits. Genetics, 19, 537–51. Zuckerkandl, E. and Villet, R. (1988). Concentration-affinity equivalence in gene regulation: Convergence of genetic and environmental effects. Proceedings of the National Academy of Science, 85, 4784–8.

6. Easy Knowledge, Transmission Failure, and Empiricism Ram Neta

INTRODUCTION

In this paper I discuss a particular epistemological puzzle that is a version— and, I think (though I shall not argue it here), the most fundamental version— of what has sometimes been called “the problem of easy knowledge.” I’ll begin by spelling out, in section I, what I take the problem to be. Then, in sections 2–4, I’ll argue that recent attempts to address the problem (from Jonathan Weisberg, Michael Titelbaum, and Chris Tucker) all fail. In section 5, I’ll articulate a principle (very similar to one that Crispin Wright has recently defended) that solves the problem. The common objection to this principle— an objection that Wright, for instance, accepts—is that it is inconsistent with a plausible empiricism. I argue that this objection fails: in fact, the principle is fully consistent with any plausible empiricism.

1.

THE PUZZLE CASES

The phrase “the problem of easy knowledge” has been used as a label for various epistemological puzzles. Let me be explicit about the particular puzzle that I’ll be discussing here. The puzzle that I’ll be discussing here is really a puzzle about doxastic justification, that is, a belief’s being justified, and its relation to propositional justification, that is, a person’s being justified in believing something (whether or not she believes it). The reason that I discuss justification instead of knowledge is that the epistemological puzzle that I want to discuss is a puzzle that arises in respect to beliefs that need not be true. And justification, unlike knowledge, does not require truth. The puzzle arises if we assume, as I do, that doxastic justification is closed under undefeated competent deduction, where by “competent deduction,” I mean a deduction that one makes from premise set P to conclusion C because one understands that C is entailed by P, and where a competent deduction is “undefeated” if one has no reason for believing that it is unsound. For if we assume the closure of justification under undefeated competent deduction, then we may wonder what to say about the competent deductive inferences that take place in the following four hypothetical cases.

Easy Knowledge, Transmission Failure, and Empiricism

167

1.1. Testimony Case You are walking around St Andrews looking for Market Street, and so you ask some passer-by (who is utterly unknown to you) where Market Street is. She tells you it is three blocks north, and that’s the end of your conversation with her. There is, let’s stipulate, nothing unusual about the situation, or about your interlocutor. Now, you competently reason as follows, and thereby arrive, for the first time, at a belief in the proposition (3), a proposition that you had not heretofore doubted or questioned: (1) My interlocutor said that Market Street is three blocks north of here. (2) Market Street is three blocks north of here. (3) On this occasion, my interlocutor told the truth.

1.2. Gibonacci Case You are calculating sums of the first ten elements of various Fibonacci sequences using the procedure of multiplying the seventh element of the sequence by 11. (Let’s call this the “Gibonacci procedure,” for the generalized Fibonacci procedure.) For the sequence: x y x+y x + 2y 2x + 3y 3x + 5y 5x + 8y 8x + 13y 13x + 21y 21x + 34y you calculate the sum of these 10 elements by multiplying the seventh element by 11, and you get: 55x + 88y. Then you competently reason as follows, and thereby arrive, for the first time, at a belief in the proposition (3’), a proposition that you had not heretofore doubted or questioned: (1’) According to the Gibonacci procedure, the sum of the first 10 elements of the Fibonacci sequence whose first 2 elements are x and y is 55x + 88y. (2’) The sum of the first 10 elements of the Fibonacci sequence whose first 2 elements are x and y is 55x + 88y. (3’) The Gibonacci procedure gives the right result.1 1

(3’) states a surprising necessary truth, as just a bit of calculation will show.

168

Ram Neta

1.3. Leaky Memory Case You have heard that some people have unreliable memories, but you have no reason to suspect that you are one of these people. In fact, you have learned (though you can’t now recall just how it is that you learned this) that your own recall is highly reliable, in the sense that, usually, when you seem to recall that p, it is true that p. And so you competently reason as follows, and thereby arrive, for the first time, at a belief in the proposition (3”), a proposition that you had not heretofore doubted or questioned: (1”) I seem to recall that my recall is, for the most part, highly accurate. (2”) My recall is, for the most part, highly accurate. (3”) On this very occasion, my recall is accurate.

1.4. Proof Case You have just done a very careful 200-step mathematical proof of the following form: a+b=c c+d=e e+f=g ... a + b + c + d + . . . = z. The conclusion of the proof states the sum of 201 numbers. Now, having just done the proof and come to believe the conclusion on its basis, you reason as follows, and thereby arrive, for the first time, at a belief in the proposition (3”’), a proposition that you had not heretofore doubted or questioned: (1”’) I have just done a proof that C. (2”’) C. (3”’) If I made any mistake of addition or transcription in this proof, that mistake was compensated for by some other mistake. There is a problem about doxastic justification that is common to each of the four cases above. In each case, the protagonist justifiably believes the first two premises—let’s stipulate that she does. Furthermore, in each case, the protagonist competently deduces the conclusion from those premises— again, we can stipulate that she does. Given the closure of justification under competent deduction, it follows that the protagonist justifiably believes the conclusion of each argument. But, in a situation in which the protagonist’s justification for believing the second premise depends upon her justification for believing the first premise, the protagonist cannot gain justification for believing the conclusion by performing any of the deductive inferences just sketched, no matter how competently she reasons. For example, if what justifies your belief, in Testimony Case, that Market Street is three blocks north

Easy Knowledge, Transmission Failure, and Empiricism

169

of here, is simply that your interlocutor said so, then your belief that your interlocutor told the truth cannot become justified simply by your inferring (3) from (1) and (2). If what justifies your belief, in Gibonacci Case, that the sum of the first ten elements of the Fibonacci sequence whose first two elements are x and y is 55x + 88y, is simply that the Gibonacci procedure gives that result, then your belief that the Gibonacci procedure gives a correct result cannot become justified simply by your inferring (3’) from (1’) and (2’). And so on for the third and fourth cases. In each case, even if the inference leads you to acquire belief in the conclusion for the first time, and even if your belief in the conclusion happens to be somehow or other justified, still, your inference cannot be what makes your belief in the conclusion justified—at least not in those cases in which you have no justification for believing the second premise that does not depend upon your justification for believing the first premise. The problem that we confront here is the problem of explaining why this is so. Should we solve the problem common to these four cases by simply denying its presupposition, namely, that doxastic justification is closed under undefeated competent deduction? Indeed, doesn’t the lottery paradox give us reason for denying such closure? No. What the lottery paradox shows, at best, is that justification is not closed under conjunction when the conjuncts are negatively epistemically relevant to each other (accepting either conjunct makes it rational to be less confident of the truth of the other conjunct). But the premises of the inferences are not negatively epistemically relevant to each other—on the contrary, they are positively epistemically relevant to each other. So the inferences above cannot be assimilated to lottery inferences in which closure has been thought to fail. Do considerations of risk aggregation (as in the preface paradox) give us reason to deny closure, even in cases in which the premises are not negatively epistemically relevant to each other? Perhaps they do, but this is irrelevant: let the premises be as risk-free as you please—indeed, let them be nearly certain—and the deductive inferences above still cannot serve to justify their conclusions. So what’s going on in the four cases above? I’ll critically assess one proposal built around Jonathan Weisberg’s No Feedback principle, another proposal built around Michael Titelbaum’s argument against “No Lose” investigations, a third proposal from Chris Tucker’s account of transmission failure, and then finally endorse and defend a fourth proposal.

2.

WEISBERG ON NO FEEDBACK

Weisberg clearly addresses cases like the Testimony Case above. He begins by postulating a defeater for inductive reasoning. No Feedback. If (i) L1–Ln are inferred from P1–Pm, and (ii) C is inferred from L1–Ln (and possibly some of P1–Pm) by an argument whose justificatory power depends on making C at least x probable, and (iii) P1–Pm do not make C

170

Ram Neta

at least x probable without the help of L1–Ln, then the argument for C is defeated.2

The basic idea of No Feedback is simply (albeit crudely) stated as follows: If a conclusion C isn’t rendered sufficiently probable by some premise set, then inferring C from lemmas which are in turn inferred from the premise set can’t make C any more probable. No Feedback is supposed to govern inductive reasoning. But what about the kind of deductive reasoning that we find in Testimony Case? How does it apply there? According to Weisberg, the basic point still holds for such cases: If I believe that Market Street is three blocks north of here, and my reason for believing that is merely my interlocutor’s testimony that Market Street is three blocks north of here, then we can justifiably infer that my interlocutor told me the truth only if the proposition that my interlocutor told me the truth is rendered sufficiently probable by my interlocutor’s testimony that Market Street is three blocks north of here. Since the proposition that my interlocutor told me the truth is not (let us suppose) rendered sufficiently probable by my interlocutor’s testimony that Market Street is three blocks north of here, we cannot justifiably infer this conclusion in the way done in Testimony Case. Now, you might have a worry about this treatment of Testimony Case. The worry is this: Given that the prior probability of one’s normal-seeming interlocutor telling one the truth is (presumably) very high, and given that conditionalizing on the receipt of any antecedently plausible testimony from the interlocutor would raise this probability at least slightly higher (since the testimony itself is antecedently plausible, and so helps at least slightly to confirm my initial suspicion of my interlocutor’s truthfulness), why should we suppose that the proposition that my interlocutor told me the truth is not rendered sufficiently probable by my interlocutor’s testimony that Market Street is three blocks north of here? The only way that I can see for Weisberg to address this concern (in fact, the way he does address this concern towards the end of his paper) is by claiming that, in order justifiably to infer a conclusion from a premise, the conditional probability of the conclusion on the premise must be significantly higher than the prior probability of the conclusion. Only so can the premise itself serve to justify the conclusion (as opposed to the conclusion’s simply being justified independently of the premise). Notice that this maneuver seems also to help explain what’s wrong with the argument in Gibonacci Case, for the conclusion of that argument has a probability of 1. The conclusion of that argument, namely that the Gibonacci procedure gave a correct result in a particular case, is a priori certain and necessary (however surprising it may seem). Its prior probability is 1, and its conditional probability on anything else (with a non-zero probability) is 1. So if Weisberg attempts to diagnose the problem with the argument in Testimony 2 Jonathan Weisberg, “Bootstrapping in General”, Philosophy and Phenomenological Research (2011).

Easy Knowledge, Transmission Failure, and Empiricism

171

Case by saying that my interlocutor’s testimony does not have enough of an upward impact on the probability of the conclusion, then he will be able to extend that diagnosis easily to explain what’s wrong with the argument in Gibonacci Case. But now there is a problem, for notice that the reason that the conclusion of the argument in Gibonacci case has a probability of 1 is that any purely mathematical truth will have a probability of 1. (Of course this is not to say that any purely mathematical truth will be maximally justified: of course that is not true. But there is no way to assign probabilities to propositions without assigning probability 1 to propositions that are necessary. So, although mathematical truths can be justified to different degrees, these differing degrees of justification cannot be probabilities, that is, cannot comply with the Kolmogorov axioms.) If the problem with the argument in Gibonacci Case is supposed to be that the premises do not raise the probability of its conclusion, then that will be a problem with any other purely mathematical argument as well. In fact, it will be a problem that the argument in Gibonacci Case will share with the following perfectly fine argument: (1’) According to the Gibonacci procedure, the sum of the first 10 elements of the Fibonacci sequence whose first 2 elements are x and y is 55x + 88y. (3’) The Gibonacci procedure gives the right result. (2’) The sum of the first 10 elements of the Fibonacci sequence whose first 2 elements are x and y is 55x + 88y. But clearly, this prediction is false: there is nothing wrong with the argument just stated, and your belief in (2’) can easily by justified by competently deducing (2’) from (1’) and (3’), supposing your beliefs in those premises are justified. So the problem with the argument given in Gibonacci Case is, in fact, not a problem shared by the argument just stated. More generally, it is obvious that not all mathematical arguments suffer from the defect of the argument given in Gibonacci Case. So there must be something wrong with the proposed diagnosis of the argument in Gibonacci Case: the problem with that argument cannot be simply that the premises do not significantly raise the probability of the conclusion. But then what is wrong with the argument given in Gibonacci Case? Weisberg does not give us any guidance here. Of course this is no surprise: probabilistic constraints on good reasoning generally do not offer a lot of help in understanding the epistemology of the a priori. And this is not a problem for Weisberg’s No Feedback principle: notice that the principle is expressly stated to apply to arguments whose justificatory power depends upon making their conclusions sufficiently probable. But purely mathematical arguments are not like that: their justificatory power does not depend upon making their conclusions sufficiently probable, but rather upon making their conclusions sufficiently justified, in some non-probabilistic way. None of this tells against

172

Ram Neta

the No Feedback principle, but it does indicate that, if we want a unified explanation for what goes wrong in all of the four inferences given above (including the one in Gibonacci Case), we will need to look at other, nonprobabilistic constraints on good reasoning. Even if his No Feedback principle is true (and I do not, for present purposes, dispute its truth), it cannot be used to provide such a unified explanation. Let’s see if we can do any better by drawing on Titlelbaum’s discussion of “no lose” investigations.

3.

T I T E L B AU M O N N O L O S E E P I S T E M O L O G Y

Titelbaum argues that no epistemological theory should license what he calls a “no lose” investigation. What is that? Here’s his initial characterization of such investigations: Suppose an agent knows at t1 that between that time and some specific future time t2 she will investigate a particular proposition (which we’ll call p). Her investigation counts as a no-lose investigation just in case the following three conditions are met: (1) p is not justified for the agent at t1 . (2) At t1 the agent knows that -p will not be justified for her at t2 . (3) At t1 the agent knows that if p is true, p will be justified for her at t2 .3

An investigation that fits this profile could appear to have an incoherent combination of features: if the agent knows that the investigation will justify p if p is true, then the agent can deduce from this knowledge that, if the investigation fails to justify p, then p is false. And so long as the agent retains knowledge of this conditional, then, if the agent comes to know that the investigation fails to justify p, the agent can deduce, and so come to know, that p is false. But the agent can know all of this at t1 , and so the agent can know at t1 that, if the investigation fails to justify p at t2 , then (contra supposition 2) it will have justified –p for her at t2 . This means that the agent can know at t1 that the investigation cannot fail to justify p at t2 . In short, the agent knows that the investigation is guaranteed to justify p. But any “investigation” that is guaranteed to justify p is not really an investigation at all: it is impossible for any such investigation to exist. If it is right that investigations having the three properties above are impossible, then this could help to explain what is wrong with the deductive inference in Testimony Case. Suppose that I want to investigate whether my interlocutor told me the truth, and I do so simply by hearing what my interlocutor says and then reasoning in the way described in Testimony Case. Suppose furthermore that I know that, in the course of doing this, I will not gain or lose any information that is evidentially relevant to the question whether my interlocutor told me the truth, and that does not come from the inference itself. In 3 Michael Titelbaum, “Tell Me You Love Me: Bootstrapping, Externalism, and No-Lose Epistemology”, Philosophical Studies (2011).

Easy Knowledge, Transmission Failure, and Empiricism

173

that case, I can know in advance (just given how the inference works) that this procedure cannot end up justifying me in believing that my interlocutor did not tell me the truth. In other words, my procedure has the second mentioned feature of a no-lose investigation. Furthermore, I know in advance (by closure) that, since this inference is valid, the conclusion will be justified if the premises are. Furthermore, I know that my premises are justified. And so (again by closure) I know in advance that the conclusion of the inference will be justified. It follows that I know in advance that if the conclusion of the inference is true then it will end up being justified. In other words, my procedure has the third feature of a no-lose investigation. But, if no-lose investigations are impossible, then it follows that my procedure cannot have the first feature of a no-lose investigation: in other words, it cannot be the case that the conclusion is not justified before I make the inference. And so the conclusion is justified before I make the inference to that conclusion. But if the conclusion is justified before I draw it, then the inference is not what makes the conclusion justified: in short, the conclusion cannot be justified after the inference if it was not already justified before. Thus, if no-lose investigations are impossible, and if justification is closed under obvious entailment, then there must be something very wrong with the inference given in Testimony Case: namely, the inference must be incapable of making the conclusion justified, since the conclusion must be justified before I ever make the inference. And this seems like just the right thing to say about the deductive inference in Testimony Case. Unfortunately, Titelbaum’s account of what is wrong with the inference in Testimony Case cannot appeal to a principle that is quite as simple as what I’ve just offered, for, as he recognizes, not all possible investigations with properties (1)–(3) above are incoherent. Titelbaum asks us to consider “the proposition p ‘There are memory-erasers who want belief in their existence to be justified.’ Suppose that at t1 I have evidence for p but also have a defeater for that evidence (so that I meet the first condition for a no-lose investigation). Suppose further that I know of some specific future time t2 that I’m not going to get any evidence against p between now and then (so that the second condition is met). Finally, suppose that if p is true the memory-erasers will remove the defeater from my memory so that I have justification to believe in them at t2 (thereby meeting the third condition). Under our definition, this example involves a no-lose investigation, yet such arrangements will be possible on any epistemological theory that allows for defeated justification.” To accommodate a variety of such cases, Titelbaum slightly refines his characterization of no-lose investigations so that they require that “p and all the agent’s relevant evidence concerning p are context-insensitive, and . . . the agent knows at t1 that every proposition relevant to p that is justified for her at t1 will also be justified for her at t2 .” As Titelbaum recognizes, this refinement is needed in order to characterize a kind of investigation that should not be licensed by any epistemological theory. But once we refine our characterization of no-lose investigations in this way, the claim that no-lose investigations are impossible can no longer

174

Ram Neta

explain what’s wrong with the argument given in Leaky memory Case. In that case, you do not know, when you begin your inference, that every proposition relevant to (3”) that is justified for you at that time will also be justified for you when you complete your inference, and this is true for the simple reason that memory is leaky: at least some of the currently recalled episodes that now serve to justify your belief in (1”) will “leak” out of your recall between the time that you begin your inference and the time that you complete it, especially if performing the inference takes some significant amount of time. At the very least, you cannot be confident that this leakage does not happen, and so you cannot be confident that the inference in Leaky Memory Case satisfies the profile of a no-lose investigation, as Titelbaum defines it. (Note that what prevents the inference in Leaky Memory Case from being a clear case of a no-lose investigation is not merely the fact that our memory leaks information, but more specifically the fact that our memory leaks information that is evidentially relevant to the justification of the conclusion of that inference.) So, even if Titelbaum is right to claim that no-lose investigations are incoherent, this claim cannot help us to understand what’s wrong with the inference in Leaky Memory Case. We’ll need to find another explanation of what goes wrong with the four deductive inferences with which we began.

4.

T U C K E R O N T R A N S M I S S I O N FA I L U R E

It is extremely plausible to claim that the deductive inferences in the four cases described above all suffer from something like what Crispin Wright has called “transmission failure”.4 But what is the phenomenon of transmission failure? Chris Tucker gives an account of those inferences that he takes to involve such failure.5 According to Tucker, transmission failure is a phenomenon that concerns the spread of doxastic justification from the premises to the conclusion of an inference, and the principle that governs such failure is the following: TFP4: Necessarily, S’s competent deduction P therefore Q fails to transmit (doxastic) justification if and only if S’s belief in P is justified (at least partly) in virtue of being based on either S’s justified belief in Q or a chain of reasoning that employs S’s justified belief in Q as (part of) a link in that chain.6

Tucker makes clear elsewhere that he understands a belief being “based on” another belief to involve the agent inferring the latter from a premise set that includes the former: in short, TFP4 identifies transmission failure with premise circularity. If Tucker’s principle TFP4 were true, it could give us a nice explanation of what’s wrong with the deductive inference in Testimony Case: When you form your belief in (2) on the basis of your belief in (1), what 4 See, e.g. Crispin Wright, “Cogency and Question-Begging: Some Reflections on McKinsey’s Paradox and Putnam’s Proof” in Philosophical Issues (2000). 5 Chris Tucker, “When Transmission Fails,” Philosophical Review (2010). 6 Tucker, 517.

Easy Knowledge, Transmission Failure, and Empiricism

175

you are really doing is forming your belief in (2) on the basis of your belief in both (1) and (3). And so the inference you are making is really of the form (1) and (3), therefore (2), therefore (3). This inference fails to transmit doxastic justification from belief in its second step to belief in its third step for the very simple reason that it is premise circular, and a premise circular inference can never enhance your justification for believing their conclusions. This form of diagnosis seems to work just fine for the deductive inferences in Testimony Case, Gibonacci case, and inductive memory case. But what about the deductive inference in Proof Case? It seems implausible to suppose that the protagonist in Proof Case is really inferring the mathematical conclusion C from a premise set that includes the premise: (3”’) If I made any mistake of addition or transcription in this proof, that mistake was compensated for by some other mistake. Suppose that, for the sake of preserving Tucker’s otherwise very nice account of transmission failure, we say that the inference to C really does include the proposition (3”’) as a premise. But then we can simply consider a case in which the protagonist, having reached conclusion C from a 201-step proof (a proof that includes (3”’) as a premise), now reasons as follows: (1””) I have just done a 201-step proof that C. (2”’) C (3”’) If I made any mistake of addition or transcription in my 201-step proof, that mistake was compensated for by some other mistake. This inference is no better than the others, and yet (3””) cannot be a premise in the inference to C. Or, if it is—in other words, if the inference to C is really a 202-step proof—then we can simply imagine another case in which the protagonist reasons as follows: (1””’) I have just done a 202-step proof that C. (2”’) C (3””’) If I made any mistake of addition or transcription in my 202-step proof, that mistake was compensated for by some other mistake. Unless we are prepared to say that every mathematical proof has infinitely many empirical premises concerning the proof itself, we will have to end this regress somewhere. And wherever it ends, we will then find it possible to construct an inference that clearly suffers from the same problem as the other deductive inferences that we considered in Testimony Case, Gibonacci Case, and Inductive Memory Case, and the fault with which cannot be a matter of premise circularity. In short, if there is a single problem with the various deductive inferences that are our targets here, that problem is not a matter of premise circularity, and it is not explained by Tucker’s TFP4.

176

5.

Ram Neta T H E C O R R E C T A C C O U N T O F T R A N S M I S S I O N FA I L U R E

Neither Weisberg nor Titelbaum nor Tucker gives us the resources necessary to explain what the problem is with the deductive inferences given in the four cases with which we started. But a number of philosophers, beginning with Crispin Wright, have suggested a principle governing the transmission of justification across inference. I believe that a version of such a principle can be used to spell out what’s wrong with all four of these inferences. What I’d like to do now is to show how that works, and then I’ll rebut what I take to be the most widespread source of resistance to such a proposal—namely, that it commits us to accepting some form of a priori justification for believing deeply contingent propositions. What I intend to show is that Wright’s transmission failure proposal can be accepted by empiricists. We empiricists can abjure all a priori justifications for deeply contingent propositions, and still accept a transmission-failure account of what goes wrong in each of our target inferences. So, what’s wrong with our four target inferences? One plausible answer appeals to the following principle: Transmission Failure Necessarily, S’s competent deduction P therefore Q fails to transmit (doxastic) justification if either S is justified in believing that P, or S is justified in believing that P supports Q, at least partly in virtue of S’s justification for believing Q. In other words, the circumstances under which competent deduction cannot enhance your justification for a conclusion include those under which either the premise beliefs or the inference itself are not justified for you independently of the justification that you have for that very conclusion. One difference between this principle (TF) and Tucker’s principle TFP4 is that TF allows that an agent can enjoy justification for believing one proposition in virtue of enjoying justification for believing another proposition, even when the former proposition is not inferred from the latter as a premise, and thus not “based” on the latter in Tucker’s sense. The epistemic dependence of a conclusion upon its premises is not, I claim, the only kind of epistemic dependence there is. How does TF serve to explain what’s wrong with each of our four target inferences? In each of those inferences, one’s sole justification for the second premise is supplied partly by one’s justification for the conclusion, and so, by TF, one cannot acquire a justified belief in the conclusion by inferring that conclusion from a premise set that includes the second premise. For instance, in Testimony Case, one has justification for believing premise (2)—that Market Street is three blocks north of here—in virtue of one’s having justification for believing (whether or not one believes it, and whether or not one bases his belief in premise (2) on it) that one’s interlocutor told one the truth. And so,

Easy Knowledge, Transmission Failure, and Empiricism

177

by Transmission Failure, one’s belief that one’s interlocutor told one the truth cannot become more justified by inferring it from a premise set that includes premise (2). In Gibonacci Case, one has justification for believing premise (2’)—that the sum of the first ten elements of the Fibonacci sequence whose first two elements are x and y is 55x + 88y—in virtue of having justification for believing that the Gibonacci procedure gives the right result. And so, by Transmission Failure, one’s belief that the Gibonacci procedure gives the right result cannot become more justified by inferring it from a premise set that includes premise (2’). In Leaky memory Case, one has justification for believing premise (2”)—that I got a bowling ball as a present on my last birthday—in virtue of having justification for believing that one’s apparent recall is accurate on this occasion. And so, by Transmission Failure, one’s belief that one’s apparent recall is accurate on this occasion cannot become more justified by inferring it from a premise set that includes premise (2”). And finally, in Proof Case, one has justification for believing premise (2”’)— the mathematical proposition C—in virtue of having justification for believing that one either did not make a mistake in the proof, or else the mistake was compensated for by another mistake. And so, by Transmission Failure, one’s belief in that disjunction cannot become more justified by inferring it from a premise set that includes premise (2”’). Transmission Failure mentions the possibility that an inference itself— rather than any of its premises—might be justified only by virtue of one’s having justification for believing the conclusion. Here is an example to illustrate the possibility: On previous days, the majority of my inferences from past to present had true conclusions. Therefore, today, the majority of my inferences from past to present have true conclusions. This inference fails to justify its conclusion, even if its premise is justified, and justified independently of its conclusion being justified. Transmission Failure explains why: even if we can justifiably believe the premise independently of any justification we might have for believing the conclusion, we cannot be justified in drawing the conclusion from the premise independently of any justification we antecedently have for believing the conclusion. (Notice, by the way, that none of the proposals offered by Weisberg, Titelbaum, or Tucker can explain what’s wrong with the inference above.) Transmission Failure explains what’s wrong with each of our four inferences. And if we imagine the cases differently, and in such a way that our justification for believing the second premise does not depend upon our justification for believing the conclusion, then the inferences no longer seem problematic at all.

178

Ram Neta

One might worry that the sufficient condition that Transmission Failure states for the failure of transmission is too weak. Consider, for instance, the following example, changed slightly from Testimony Case: (1) My interlocutor said that Market Street is three blocks north of here. (2) Market Street is three blocks north of here. (1) My interlocutor said that Market Street is three blocks north of here and (2) Market Street is three blocks north of here and (3) On this occasion, my interlocutor told the truth. This last inference suffers from transmission failure as well, but does one have justification for believing (2) in virtue of having justification for believing the conjunction (1) and (2) and (3)? No. What is true, however, is that, under the conditions specified in Testimony Case, one has justification for believing (2) in virtue of having justification for believing (1) and (3). And it’s also true that (1) ND (3) is logically equivalent to (1) and (2) and (3). So, to account for the transmission failure in this case, we need to revise our principle Transmission Failure slightly, as follows: Transmission Failure’ Necessarily, S’s competent deduction P therefore Q fails to transmit (doxastic) justification if either S is justified in believing that P, or S is justified in believing that P supports Q, at least partly in virtue of something E, such that E is what makes S justified in believing Q. Transmission Failure’ also explains what’s wrong with other variants of the inferences considered above, e.g. (1) My interlocutor said that Market Street is three blocks north of here. (2) Market Street is three blocks north of here. (3x) My interlocutor told the truth, and 1 + 1 = 2. Or this: (1) My interlocutor said that Market Street is three blocks north of here. (2) Market Street is three blocks north of here. (3y) My interlocutor told the truth, and Market Street is three blocks north of here. Each of the inferences above suffers from transmission failure, and the principle Transmission Failure’ can explain why: in each case, I am justified in believing the second premise at least partly in virtue of whatever it is that justifies me in believing the conclusion. I should note that Transmission Failure’ is a principle about deductive inference, not about inductive inference. So Transmission Failure’, by itself, cannot tell us what’s wrong with those self-supporting inferences that are inductive, rather than deductive. Sometimes, the problem of easy knowledge

Easy Knowledge, Transmission Failure, and Empiricism

179

is thought to be a problem about self-supporting inferences of any kind. But, as I argue elsewhere,7 we can tell a simple and elegant explanation of what’s epistemologically wrong with self-supporting inductive inferences if we start by giving the present story about what’s epistemologically wrong with selfsupporting deductive inferences. That claim, however, is not one that I have the space to defend in this paper.

6.

HOW TO BE AN EMPIRICIST

Although the present account of what goes wrong in our target deductions seems to deal with the cases that we’ve considered, it is subject to a worry concerning epistemic circularity. The worry starts to come into focus if, in considering Testimony Case, we ask: how does one acquire justification for believing (3)? One can perhaps justifiably infer (3) from one’s justified belief in: (4) When a normal-seeming human adult A confidently reports that p, where p is some proposition that is easily checkable by her listeners, and about which there is no apparent motive to deceive, then, in normal circumstances, A’s report is true. And (5) I am now in normal circumstances. But then what makes one’s belief in the generalization (4) justified? Wright,8 BonJour,9 Burge,10 Cohen,11 White,12 Wedgwood,13 Zalabardo,14 Silins,15 and Hawthorne,16 all say that one is a priori justified in believing (or at least 7

See my forthcoming “Sosa on Epistemic Circularity”. Crispin Wright, “Warrant for Nothing (and Foundations for Free)”, Proceedings of the Aristotelian Society (2004). 9 Laurence BonJour, In Defense of Pure Reason, Cambridge University Press, 1998. 10 Tyler Burge, “Perceptual Entitlement”, Philosophy and Phenomenological Research (2003). Burge speaks not of a priori justification but rather of a priori entitlement for belief in such generalizations as (4). But that is because Burge is concerned to distinguish between warrants that we can articulate and warrants that we cannot, and that is not a distinction that matters for my purposes here. 11 Stewart Cohen, “Bootstrapping, Defeasible Reasoning, and A Priori Justification,” Philosophical Perspectives (2010). 12 Roger White, “Problems for Dogmatism”, Philosophical Studies (2006). 13 Ralph Wedgwood, “A Priori Bootstrapping”, available at . Accessed 12 December 2012. 14 Jose Zalabardo, “Externalism, Skepticism, and the Problem of Easy Knowledge”, Philosophical Review (2005). Since Zalabardo is an externalist, he thinks of this a priori justification as an externalist warrant. 15 Nico Silins, “Basic Justification and the Moorean Response to the Skeptic”, Oxford Studies in Epistemology (2007). 16 John Hawthorne, “Deeply Contingent A Priori Knowledge”, Philosophy and Phenomenological Research (2002). 8

180

Ram Neta

accepting17 ) some contingent generalizations concerning the reliability of at least some of one’s own cognitive sources: if they are right about this, then why not regard (4) as one such contingent generalization? Let’s suppose, for the sake of argument, that they are right. It is consistent with their view, and also quite obviously true, that we can also gain an empirical justification for (4) by, say, doing some sort of empirical testing: perhaps this would involve scouring testimonial corpus data and checking the truth of a large, representative sample of the confident assertions that occur in that corpus, at least when they are on topics that are easily checkable by the conversational participants, and there is no apparent motive for the testifier to lie about them. Suppose that, by performing this empirical procedure, we gain empirical justification for (4). But, after we gain this empirical justification for (4), we then gain an undermining defeater for (what we have just, for the sake of argument, conceded to be) our a priori justification for (4). Precisely what might constitute such a defeater is going to depend upon what sort of a priori justification we are alleged to have for (4). For instance, if this a priori justification is supposed to be reflectively accessible, then an undermining defeater of this a priori justification might consist in some reason to believe that our powers of reflection are likely to mislead us with respect to what we are a priori justified in believing (but not with respect to what we are empirically justified in believing). If, however, this a priori justification is supposed to be something that we gain simply, say, by virtue of the nature or reliability of a certain proper subset of our cognitive faculties, then an undermining defeater might consist in some reason to believe that those particular cognitive faculties do not have the sort of nature or reliability that would grant us these justifications. But, whatever precisely our a priori justification for (4) is supposed to consist in, it is, I assume, something that is subject to undermining defeat, and the defeat of which is consistent with there being no defeater for the aforementioned empirical justification for (4). So let’s suppose that, after gaining empirical justification for (4), you then gain an undermining defeater for your putative a priori justification for (4). This is all possible, and furthermore, it is compossible with your continuing to have empirical justification for (4), and thereby empirical justification for (3). What this shows is that it is possible for us to have exclusively empirical justification for believing (4), and so (3). So this leaves us with a question: if it is possible to have exclusively empirical justification for believing (4), why should we grant, as all the aforementioned philosophers do, that our justification for believing (4) is ever a priori? One reason we might have for thinking that our justification for believing (4) is at least sometimes a priori is that we didn’t actually perform the empir17 Of the authors mentioned, only Wright says that the a priori justification is not a justification for belief, but rather for a distinct attitude that he calls “acceptance.” I take it that Wright thinks that it is an insight of skepticism that our belief justifications bottom out in something that is not a belief justification at all.

Easy Knowledge, Transmission Failure, and Empiricism

181

ical procedure described above. Or . . . did we? Don’t we constantly receive testimony, and notice when it fails to cohere with other evidence we have? Of course, we don’t typically engage in a deliberate process of acquiring testimony and verifying it against information acquired through non-testimonial sources. But, even if we are not engaged in this deliberate process, we are still (for the most part, and without paying much attention) accumulating the kind of empirical evidence that we would end up acquiring in a more targeted way by going through this process: we hear what people say, and we tend to notice when it is corroborated or undermined by other sources of information (testimonial or non-testimonial). But this raises a further question: what gives justification to our beliefs in all of the empirical evidence propositions that constitute our empirical justification for believing (4)? Aren’t some of those evidence beliefs (e.g. beliefs about what assertions have been made, or which of those assertions have been corroborated) themselves going to be justified at least partly on the basis of testimony? And, if they are, then won’t we need to justifiably believe (4) in order justifiably to believe those evidence propositions? So how can these evidence propositions help to constitute an empirical justification for believing (4), if our justified belief in (4) enables us justifiably to believe those evidence propositions? Here we confront a problem of epistemic circularity. If we abjure a priori justification for deeply contingent generalizations like (4), then it seems that we inevitably run into this problem of epistemic circularity. If we presume to derive all testimonial justifications from perceptual justifications, then this only postpones the problem of epistemic circularity: now we face the question how we can be justified in believing that our perceptual faculties themselves are reliable (this is surely something that we are justified in believing, whether or not such meta-justification is necessary for our various first-order perceptual justifications), if not on the basis of a track record argument from instances, and such an argument would itself seem to suffer from Transmission Failure. How can we empiricists—who do abjure a priori justifications for deeply contingent generalizations like (4)—solve this problem? In order to develop my answer to this question, let me begin by making a point about empirical evidence. Whatever exactly our total empirical evidence consists in, it will be a conjunction of propositions, each of which is such that we are empirically justified in believing it.18 And there are going to be many particular propositions in our evidence set which are such that, in order to be justified in believing those particular propositions, we will also have to be justified in believing in the veracity of the source of that particular conjunct’s justification. (Specifically, this will be true at least of those evidence propositions which, like the second premise in each of the four deductive

18 I will not attempt to defend this propositional conception of evidence here; for a defense of it, see my “What Evidence Do You Have?”, British Journal for the Philosophy of Science (2008).

182

Ram Neta

inferences from our initial cases, we are justified in believing on account of the operation of cognitive sources that we can knowledgeably identify.) But how can we be justified in believing in the veracity of that source, without relying exclusively upon the pieces of evidence produced by that source? This can happen only if our total evidence contains pieces of evidence from a variety of different sources, where the deliverances of each corroborate the deliverances of the others. If you have evidence from any one cognitive source of which you are aware, then you also have evidence that does not derive from that same source. So, with these pieces in place, I can finally state my solution to the problem of epistemic circularity as follows: In order to justifiably believe any deeply contingent proposition as a result of the operation of some cognitive source of which I am aware, I must be justified in believing some deeply contingent generalization about the veracity of that particular cognitive source. But in order to be justified in believing some deeply contingent generalization about the veracity of that particular cognitive source, my total empirical evidence must make it highly probable that the source in question is veracious. In order for my total empirical evidence to make it highly probable that the source in question is veracious, my total empirical evidence must contain pieces of evidence that are not from the source in question, and those other pieces of evidence must corroborate the evidence I get from the source in question. But I am justified in believing each of the particular pieces of my total evidence (at least those particular pieces that I am justified in believing in virtue of the operation of some cognitive source of which I am aware) only in virtue of being justified in believing some deeply contingent generalization about the veracity of its source, which I am in turn justified in believing only in virtue of my total evidence. In short, I am justified in believing each proposition in my total evidence only by virtue of being justified in believing some conjunction of evidence propositions. To sum up the present view in a slightly misleading way: I am justified in believing each particular evidence proposition only because I am justified in believing all of them. This last formulation is misleading in two ways. First, it misleadingly suggests that my justification for believing each particular evidence proposition is somehow inferential, as if I infer each evidence proposition from the conjunction of my total evidence. But this is clearly false. As I said above in my reply to Tucker, there is a difference between justifiably inferring p from q, on the one hand, and being justified in believing p partly in virtue of being justified in believing q, on the other. The former is a species of the latter, but the latter is a much broader category. For instance, I typically infer the conclusion of a mathematical proof simply from the mathematical premises of that very proof, and not from any non-mathematical propositions concerning the proof itself, and yet I am justified in believing the conclusion of the proof only in virtue of my being justified in believing some non-mathematical propositions concerning the proof itself (e.g. I was not careless when I did it). Similarly, I do not infer the particular propositions in my evidence set from their conjunction,

Easy Knowledge, Transmission Failure, and Empiricism

183

but this does not mean that I am not justified in believing those evidence propositions in virtue of being justified in believing some conjunction of them. But I also do not want to suggest that there is some specific conjunction of propositions in my evidence set such that my justifiably believing any single proposition in my evidence set depends in any way (inferential or otherwise) upon my being justified in believing that specific conjunction. Rather, what my justifiably believing any single proposition in my evidence set depends upon is there being some conjunction or other of evidence propositions that includes the single proposition in question, such that I am justified in believing that conjunction. So the overall picture is this. My evidence set consists of a conjunction of empirical propositions, each of which I am (non-inferentially) justified in believing, and each of which I am justified in believing partly in virtue of being justified in believing the others (or any others that could equally well constitute an evidence set that included that particular proposition). The justification of my belief in each evidence proposition thus depends on its coherence with the rest of the propositions in my evidence set. By adopting this position, we can solve the problem of easy knowledge by appeal to transmission failure, and we can do so without forsaking empiricism, and without running into any insurmountable problems of epistemic circularity. But have we really avoided the problem of epistemic circularity altogether? I said that we can corroborate the evidence that we get from each source by appealing to other sources that provide some of our other evidence. But we cannot corroborate the whole of our empirical evidence. So what justifies me in believing that my total empirical evidence is not systematically misleading? Answer: what justifies me in believing that my total empirical evidence is not systematically misleading is simply my total empirical evidence. My total empirical evidence justifies me in believing quite a few things about the world, including the following: What evidence someone has is a result of impacts upon their sensory systems. Such impacts are interrelated in such complicated ways that it would be very difficult to make them systematically misleading (i.e. misleading in a way that left no trace in one’s evidence set itself). I have no evidence of the existence of anything that can perform such a difficult task, and a great deal of evidence that no such thing exists. Thus, my total evidence justifies me in believing that my total evidence is not systematically misleading. Perhaps it is misleading here and there, but it is not so misleading as to make it impossible for me to correct it by appeal to my total evidence.

7.

CONCLUSION

So what’s wrong with the four competent deductive inferences in Testimony Case, Gibonacci Case, Leaky Memory Case, and Proof Case? What’s wrong with them is that they fail to transmit justification: what makes your belief in the premises justified is (at least partly) your justified belief in the conclusion,

184

Ram Neta

and so, by Transmission Failure, you cannot become more justified in believing the conclusion by inferring it from those premises. But if you must justifiably believe the conclusion already in each of these cases, how did you get that justified belief? Does it ultimately rest on some a priori justification for believing some contingent proposition about your own cognitive reliability? No: we needn’t suppose that you have any such a priori justified beliefs. All your justified beliefs about your own reliability can be justified empirically, on the basis of your total empirical evidence. Doesn’t this picture generate a vicious epistemic circle? Not necessarily. Even the strictest foundationalist can accept this picture, so long as she posits a single foundationally justified belief, namely, belief in the conjunction of one’s total evidence. It is one’s justified belief in that conjunction that enables one to have other justified beliefs, including justified beliefs in particular conjuncts of that conjunction.19 19 For helpful discussion, I am grateful to Yuval Avnur, Stewart Cohen, Juan Comesana, John Devlin, Daniel Fogal, Richard Fumerton, Anil Gupta, Alex Jackson, Matt Kotzen, Anna-Sara Malmgren, Matthew McGrath, Michael Titelbaum, Chris Tucker, and an anonymous referee for Oxford Studies in Epistemology.

7. Why Philosophy Can Overturn Common Sense1 Susanna Rinard

INTRODUCTION

Many philosophers have a rather limited view of what our discipline could hope to achieve. They think that philosophical arguments cannot rationally overturn our pre-theoretical common sense convictions. Here, for example, is Kit Fine: In this age of post-Moorean modesty, many of us are inclined to doubt that philosophy is in possession of arguments that might genuinely serve to undermine what we ordinarily believe.2

And David Lewis: One comes to philosophy already endowed with a stock of opinions. It is not the business of philosophy either to undermine or justify these preexisting opinions to any great extent, but only to try to discover ways of expanding them into an orderly system.3

Other advocates of positions of this kind include Lycan 2001, Kelly 2005, and Kelly 2008. On the other hand, some philosophers take the opposing view. They present and endorse philosophical arguments against various claims of common sense. For example, Unger 1975 argues that no one knows anything at all. Van Inwagen 1990, Dorr 2002, and others argue that ordinary objects like tables and chairs do not exist. Finally, I’ll include some remarks on this topic by Kant and Hegel: 1 Thanks to Alex Byrne, Andrew Graham, David Gray, Daniel Greco, Bradford Skow, Agustin Rayo, Robert Stalnaker, John Stamm, three anonymous referees for Oxford Studies in Epistemology, and audiences at the MIT MATTI Reading Group and the MIT Epistemology Reading Group. Thanks especially to Miriam Schoenfeld for our many helpful conversations on these topics. Thanks to Andrew Graham for providing the Kant and Hegel quotes. Special thanks to Roger White—this paper has benefited greatly from his many suggestions and comments. Finally, I would also like to thank Thomas Kelly for writing two detailed and fascinating papers on this topic, the reading of which inspired me to write this paper. This material is based upon work supported under a National Science Foundation Graduate Research Fellowship. 2 Fine (2001, 2). 3 Lewis (1973, 88).

186

Susanna Rinard

To appeal to ordinary common sense . . . is one of the subtle discoveries of recent times, whereby the dullest windbag can confidently take on the most profound thinker and hold his own with him.4 Since the man of common sense makes his appeal to feeling, to an oracle within his breast, he is finished and done with anyone who does not agree; he has only to explain that he has nothing more to say to anyone who does not find and feel the same in himself. In other words, he tramples underfoot the roots of humanity.5

As for myself, I wouldn’t go so far as to say that Lewis, Fine, and company are dull windbags. And I am not yet convinced that they have trampled underfoot the roots of humanity. However, my view shares some of the spirit of these remarks. For I do believe, contra Lewis, Fine, etc., that philosophy can overturn common sense. It is the aim of this paper to defend that position. (I will not begin by defining “common sense.” If this concerns you, please see this footnote.6 ) The paper has two distinct and easily separable parts. In part one, I present and endorse a positive argument for the claim that philosophy can overturn common sense. My opponents and I agree that science can overturn common sense. But, I claim, every scientific argument relies on assumptions that are highly theoretical, even philosophical. If a scientific argument against a common sense proposition is to succeed, then its philosophical assumptions must be more worthy of belief than the common sense proposition under attack. But this means that there could be a philosophical argument against common sense, each of whose premises is just as powerful, epistemically, as the scientist’s philosophical assumptions. If the scientific argument can succeed, then so, too, can the purely philosophical argument, and so philosophy is capable of overturning common sense.7 4

Kant (2008, 9). Hegel (1977, 43). 6 I will not take it upon myself here to provide any general definition of “common sense,” and I don’t think this is required for my arguments to go through. My opponents are committed to the view that there is an important distinction between what is common sense and what isn’t, since they believe that this distinction plays an important epistemic role. Since I deny the latter, I am not committed to the former. That said, in the course of arguing against my opponents, there are a few places in which my arguments do rely on a claim about what counts as common sense. In each such case, I think that most if not all of my opponents would agree. For example, in the second half of the paper, I assume that certain propositions about the external world, like “I have hands,” count as common sense. These are taken to be paradigm examples of commonsense propositions by all parties to the debate. Also, in the first half of the paper, I assume that special relativity conflicts with common sense. Although perhaps slightly more contentious, this is also generally accepted. Moreover, see section 3, response to objection 5, for a response to the objection that special relativity does not conflict with common sense. 7 One might wonder whether this conclusion by itself is enough to seriously challenge a broadly Moorean outlook. Shouldn’t the reasonable Moorean admit that there might possibly be one or two cases in which philosophy can overturn common sense? Surely we should interpret the Moorean as claiming at most that, in general, philosophy can’t overturn common sense. I’d like to direct those worried about this issue to the last three paragraphs of section 2. One central aspect of the Moorean outlook, however exactly it is interpreted, is that we are supposed to be 5

Why Philosophy Can Overturn Common Sense

187

In part two, I undermine some of the main reasons philosophers have given for the opposing view that philosophy can’t overturn common sense. First I consider the Moorean idea that common sense propositions are more plausible than philosophical claims. Then I turn to the view, defended in Kelly 2005, that one should retain belief in judgments about particular cases at the expense of general principles that conflict with them. Finally I consider a version of reflective equilibrium, defended in Harman 2003, in which conservatism plays an important role. In each case, I argue that either the view in question is false, or that it fails to provide an independent motivation for the claim that philosophy can’t overturn common sense. The aim of this paper, then, is to argue for and defend the claim that philosophy can overturn common sense. If I am right, then I think this infuses the project of philosophy with new importance and urgency. Almost everything we think, say, and do presupposes some proposition of common sense. If the business of philosophy is, at least in part, to inquire into the truth of these propositions—with the real possibility left open that we may find good reasons for rejecting them—then philosophy is highly relevant to almost every aspect of our daily lives.

1.

I N T R O D U C T I O N T O PA R T O N E : P H I L O S O P H Y CAN O V E R T U R N COMMON SENSE

In this first part of the paper I will provide a positive argument for the claim that philosophy can overturn common sense. In its simplest form, the argument is as follows: (1) Science can overturn common sense. (2) If science can overturn common sense, then so can philosophy. (3) Therefore, philosophy can overturn common sense. This argument is not original to me. Indeed, it is considered, and rejected, by many of my opponents. My main contribution will come in the form of my particular defense of premise (2), for it is this premise that is generally rejected by the advocates of common sense. Premise (1) is widely accepted, and the argument is valid, so I will focus primarily on defending premise (2) (though see section 3, reply to objection 5 for a defense of premise (1)). It will be helpful to begin by considering my opponents’ argument against premise (2). Here are some relevant quotes, starting with one from William Lycan: Common sense beliefs can be corrected, even trashed entirely, by careful empirical investigation and scientific theorizing . . . No purely philosophical premise can ever able to dismiss philosophical arguments for skepticism on the grounds that they conflict with common sense. I claim at the end of section 2 that, if my arguments are accepted, we can’t do this. So I think that my arguments really do take the wind out of the Moorean’s sails. My reply to objection 5 in section 3 is also relevant here.

188

Susanna Rinard

(legitimately) have as strong a claim to our allegiance as can a humble common sense proposition such as Moore’s autobiographical ones. Science can correct common sense; metaphysics and philosophical “intuition” can only throw spitballs.8

Anil Gupta expresses a similar sentiment: Any theory that would wage war against common sense had better come loaded with some powerful ammunition. Philosophy is incapable of providing such ammunition. Empirical sciences are a better source.9

The idea here—which is also found in Kelly 2008—can be summarized as follows: science, unlike philosophy, can appeal to empirical, observational evidence. When science undermines common sense, it does so by appealing to direct observation. When the philosopher attempts to undermine common sense, however, she can appeal only to highly theoretical premises, which are less powerful, epistemically, than observational evidence. So scientific arguments against common sense are more powerful than philosophical argument against common sense. There are many concerns one might have about this argument. First, it is notoriously difficult to distinguish observational and theoretical claims. Second, even supposing one can do so, it is not clear that observational claims really are epistemically stronger than theoretical ones. However, I want to set these worries aside, and focus on what I think is a deeper flaw in this argument. I agree with my opponents that scientific arguments, unlike philosophical arguments, appeal to observational claims. However, I will argue that scientific arguments must also rely on highly theoretical assumptions that are just as far removed from observation as the kinds of claims typically appealed to in philosophical arguments against common sense. Indeed, many of these theoretical scientific assumptions are straightforward examples of typical philosophical claims. An argument is only as strong as its weakest premise. So if a scientific argument is to succeed in undermining common sense, then each of its premises, individually, must be more epistemically powerful than the common sense proposition it targets.10 Since, as I claim, the scientific argument relies crucially on a philosophical assumption, this philosophical assumption must be more powerful than the common sense proposition. But if one philosophical claim can be more powerful than a common sense proposition, then there could be an argument consisting entirely of philosophical claims, each of which is more powerful than the common sense proposition whose negation they entail. If so, then philosophy can overturn common sense. The argument I just sketched appeals to the following claim:

8

Lycan 2001. Gupta (2006, 178). 10 A is more epistemically powerful than B just in case, if forced to choose between A and B, one should retain belief in A and give up belief in B. 9

Why Philosophy Can Overturn Common Sense

189

Science Requires Philosophy (SRP): Scientific arguments against common sense rely crucially on philosophical assumptions.

The next section will be devoted to defending this claim.

2.

SCIENCE REQUIRES PHILOSOPHY (SRP)

I will begin my case for SRP by considering what is perhaps the most widely accepted example of a case in which science has succeeded in overturning common sense: special relativity. Most philosophers—including many who think that philosophy can’t overturn common sense—believe that special relativity is true, and agree that special relativity conflicts with common sense. So if I can show that the scientific argument for special relativity relies crucially on a philosophical assumption, then this, in combination with the fact that an argument is only as strong as its weakest premise, will suffice to show that philosophy can overturn common sense (as explained in more detail in the previous section). We needn’t get too far into the technical details, but some basic knowledge of the case will be helpful. Consider a simultaneity proposition like this one: Joe’s piano recital and Sarah’s baseball game were happening at the same time. Pre-theoretically, we would think that a proposition like this one, if true, is objectively and absolutely true; its truth is not relative to a particular person or thing. “Licorice ice cream is delicious” may be true for Joe but not for Sarah, but the simultaneity proposition, we would normally think, is true absolutely if true at all. However, according to special relativity (SR), this is not the case. SR says that there are many different reference frames— each object has its own—and the very same simultaneity proposition may be true in one reference frame but not another. Moreover, there’s no particular reference frame that has got it right. Each reference frame has the same status; none is more legitimate than the others. So, special relativity conflicts with the common sense idea that simultaneity claims are absolute. Now, there is an alternative scientific hypothesis—the so-called neoLorentzian view—that is empirically equivalent to special relativity but which does not conflict with the common sense idea that simultaneity is absolute. Special relativity and neo-Lorentzianism agree on almost everything. In particular, they agree on all observational propositions—there is no possible experiment that could decide between them. The main difference is that according to neo-Lorentzianism, one of the reference frames is privileged in the sense that it gets the simultaneity facts right. On this view, one particular reference frame is objectively and absolutely correct. So the neo-Lorentzian view vindicates the common sense idea that simultaneity is absolute. Most scientists—and most philosophers—believe that special relativity, rather than the neo-Lorentzian view, is true. They say that the neo-Lorentzian view is unnecessarily complex. It posits an additional metaphysical fact—a fact about which reference frame gets things absolutely correct—that doesn’t

190

Susanna Rinard

make a difference to the empirical predictions of the theory. Special relativity “gets the job done” with less machinery. So if we follow Ockham in thinking that simpler hypotheses should be given more credence than complex ones, we will give more credence to special relativity than to the neo-Lorentzian view. This, in fact, is what most scientists and philosophers do, and for this reason they give up the common sense idea that simultaneity is absolute. With these facts on the table, we can now ask the crucial question: does the argument from special relativity against the absoluteness of simultaneity rely crucially on a philosophical assumption? I think that it does. In particular, it relies on the philosophical assumption that simpler hypotheses should be preferred over complex ones. Anyone who gives up the view that simultaneity is absolute on the basis of special relativity must have a reason for preferring special relativity to the neo-Lorentzian view. The reason standardly given is that special relativity is simpler. Without the further claim that simpler theories should be preferred, we simply don’t have any reason to give up the common sense idea that simultaneity is absolute. So the defender of special relativity must think that a philosophical assumption—the claim that simpler theories should be preferred over complex ones—is more epistemically powerful than a common sense proposition. Here’s another way to make the point. Suppose that this philosophical assumption were not more powerful than the common sense claim. In that case, one should reason as follows: well, the idea that simpler theories are preferable does have some plausibility to it. But if this is true, then we have to prefer special relativity to the neo-Lorentzian view, since it is simpler. But this would force us to give up the common sense idea that simultaneity is absolute. This common sense claim is more powerful than the philosophical assumption that simpler theories are preferable. So, I will retain belief in the common sense claim, give up my preference for simpler theories, and then believe what the empirical evidence forces me to believe, namely, that the neo-Lorentzian view is true. Most philosophers, however, do not reason in this manner. Since they think we should accept special relativity, I conclude that they must think that the philosophical preference for simplicity is more powerful than the common sense notion that simultaneity is absolute. So, we see that the insistence by Kelly 2008, Lycan 2001, and Gupta 2006 that observational evidence is more powerful than philosophical claims, and their pointing out that science, unlike philosophy, appeals to observational evidence, is beside the point. No matter how much observational evidence is appealed to in a scientific argument against common sense, as long as the argument relies crucially on a philosophical assumption, then, if the argument is to succeed, this philosophical assumption must be more powerful than the targeted proposition of common sense. If so, then there could be a successful argument against common sense that relies only on philosophical assumptions; and if so, then philosophy is capable of overturning common sense. The next section of this paper will be devoted to replying to a variety of objections to the argument I’ve just given. But first, before closing this section,

Why Philosophy Can Overturn Common Sense

191

I’ll briefly address the question of how general this phenomenon might be. That is, is the argument for special relativity unique among scientific arguments in its reliance on philosophical assumptions? Or is the phenomenon more widespread? I think there is a general reason to think that the phenomenon is widespread. Scientific arguments against common sense typically proceed by noting that a currently accepted scientific hypothesis is in conflict with common sense. However, scientific hypotheses are generally not logically entailed by the data that support them. Moreover, it is usually only the hypothesis as a whole that conflicts with common sense, rather than the data themselves. This is true in many other commonly cited examples of the overturning of common sense by science: astronomical hypotheses according to which the Earth is round, rather than flat, and according to which the Earth orbits the Sun, rather than vice versa; and the hypothesis that tables, chairs, and other objects are mostly empty space, rather than solid. Since it is only the hypothesis as a whole and not the empirical data by themselves that conflicts with common sense, there will always exist an empirically equivalent competitor hypothesis that vindicates common sense. If so, then a philosophical assumption will be required if the non-common-sensical theory is preferred. Such an assumption will likely be an epistemological principle about theory choice, such as the claim that one hypothesis explains the data better and should for that reason be preferred; or that one hypothesis is simpler, more internally coherent, or better unified with other accepted theories, and that these constitute reasons for preferring it; etc. So, I think that the reliance of science on philosophy is not an isolated phenomenon restricted to a few cases like special relativity, but is rather the norm. Let’s take stock. I have argued that the paradigm example of a successful scientific argument against common sense—the argument for special relativity—relies crucially on a philosophical assumption, namely the assumption that simpler hypotheses should be preferred over complex ones. Anyone who accepts special relativity on the basis of the scientific argument for it is committed to thinking that this philosophical assumption is more epistemically powerful than the common sense idea that simultaneity is absolute. If so, then there could be a successful argument against a common sense proposition that relied only on philosophical assumptions. If each of its premises is at least as powerful as the claim that simpler theories should be preferred, then the philosophical argument against common sense will succeed. One upshot of this is that we can’t dismiss arguments like the argument for external world skepticism just on the grounds that its premises are purely philosophical. Rather, we must carefully consider the status of each of its premises in comparison to the common sense claim that we know we have hands—and, crucially, in comparison to the status of the philosophical assumptions required by the scientific arguments that one accepts. If the premises of the skeptical argument are as powerful as, for example, the claim

192

Susanna Rinard

that simpler theories should be preferred, then the skeptical argument will succeed. It is not my purpose here to deliver a final verdict on the success of the skeptical argument,11 so I will not undertake an in-depth comparison here. However, I will note that, on the face of it, things don’t look at all bad for the skeptic. Take, for example, one of the key premises in one version of the argument for skepticism: the claim that propositions about the way things appear to us—for example, the proposition that it appears as though I have a hand—are evidentially neutral between the hypothesis that things really are the way they appear (I really do have a hand) and the hypothesis that I am a brain in a vat being fed misleading appearances as of a hand. This claim is extremely compelling. How could the appearance of a hand be evidence for one of these hypotheses over the other, when both predict that I would have exactly this appearance? Compare this claim, now, with the claim that we ought to prefer simpler theories over complex ones. While this claim is accepted by many philosophers, it seems to me if anything less obviously correct than the skeptic’s premise just mentioned. If the preference for simplicity is powerful enough to overturn common sense, then it seems to me that the skeptic’s claim is as well. Of course, there is much more that could be said on this topic—for example, one might try to argue in response that the common sense claim that simultaneity is absolute was antecedently less powerful than the claim that I know I have hands, and so it may not suffice for the skeptic’s premises to be as powerful as the scientist’s philosophical assumptions. My point here is just that once we have seen that, as in the case of special relativity, philosophical assumptions can be more powerful than common sense, skeptical arguments and other philosophical attacks on common sense can no longer be dismissed out of hand. A careful and serious investigation into the epistemic status of their premises needs to be undertaken, and, at the outset, it is not at all clear that these premises won’t in the end prove powerful enough to overturn common sense.

3.

OBJECTIONS AND REPLIES

Objection 1: Your argument (says the objector) presupposes that the scientific argument for special relativity relies crucially on the philosophical assumption that simpler theories should be preferred to complex ones. However (says the objector), not all arguments for special relativity rely on this assumption. The following is a perfectly valid argument: (1) [empirical scientific data];

11 I do so in Rinard (ms). I argue that it is not rational to accept the argument for external world skepticism because anyone who does so is ultimately committed to accepting skepticism about complex reasoning on the basis of a complex argument (via skepticism about the past), a position that is self-undermining.

Why Philosophy Can Overturn Common Sense

193

(2) If [empirical scientific data], then special relativity is true; therefore, (3) special relativity is true. Reply: Let’s think more about the status of premise (2) in the objector’s argument. We’ll assume that the person considering the argument is a scientist or philosopher aware of the existence of the neo-Lorentzian view. What reason could such a person have for believing (2)? After all, the empirical data are entailed by both special relativity and neo-Lorentzianism. It seems to me that such a person must think that special relativity has some other feature that neo-Lorentzianism does not have, such that hypotheses with that feature should be given greater credence. But, whatever the feature, the claim that hypotheses with this feature should be given greater credence will be a philosophical, epistemological claim. If one doesn’t believe a philosophical claim of this kind, then one would not believe (2), and the argument wouldn’t go through. So I claim that any argument for special relativity must rely at least implicitly on a philosophical assumption. Objection 2: I concede (says the objector) that the scientific argument for special relativity relies on a philosophical assumption. However, we have more reason to believe the philosophical assumptions that are typically appealed to in scientific arguments against common sense than the philosophical assumptions that typically appear in philosophical arguments against common sense. This is because science has such an impressive track record. Every time we use a laptop or walk over a bridge, we are getting evidence that scientists know what they are doing, and that we should believe whatever theories and philosophical assumptions are endorsed by science. We have no similar reason for believing the assumptions made by philosophers, since philosophy as a discipline does not have a similar track-record. Reply: According to the objector, it is in virtue of science’s long and impressive track-record of success that we should believe the philosophical assumptions that appear in scientific arguments like the argument for special relativity. If the objector is right, this means that in the absence of such a track record, we would have no reason to believe these assumptions. But I don’t think this is right, and I don’t think my opponents would agree with this either. Consider a hypothetical scenario in which human history went quite differently. Let’s suppose that the very first scientist managed to acquire the empirical evidence that actual scientists now take to constitute their reason for believing special relativity. Let’s suppose that after reflecting on this evidence, the first scientist developed the theory of special relativity. She also developed and considered the neo-Lorentzian view, but reasoned that special relativity should be preferred because it was simpler and more elegant. Now, given that this scientist is working in a society that does not have a long track-record of the success of science, according to the objection just given, this scientist has no reason to believe her philosophical assumption that simpler theories are to be preferred. However, I think the scientists and philosophers who believe

194

Susanna Rinard

special relativity today would agree that this first scientist would be entirely justified in assuming that simpler theories should be preferred, and giving up her common sense beliefs on that basis. If so, then a long track-record is not required for science to overturn common sense, and the objection fails. Objection 3: You (says the objector) have argued that anyone who gives up the common sense idea that simultaneity is absolute must do so on the basis of an argument that relies crucially on philosophical assumptions. But you assumed that the person in question was a scientist or philosopher familiar with the neo-Lorentzian view. However, many lay people have given up the absoluteness of simultaneity without having the faintest idea of what neoLorentzianism is, or even what special relativity is, just on the basis of the testimony of scientists or teachers. Surely they didn’t have to rely on any assumptions about the relative merits of simple and complex scientific theories, but it was rational nonetheless for them to give up the common sense idea that simultaneity is absolute. Reply: First, note that my argument does not require that everyone who rationally gives up the absoluteness of simultaneity must rely on philosophical assumptions. It is enough that one could rationally do so on the basis of an argument that does rely on philosophical assumptions, as that alone is sufficient to show that philosophical assumptions can be more powerful than common sense. However, as a matter of fact, I do think the layperson described by the objector is relying, at least implicitly, on some philosophical assumptions, although they may be quite different from the assumptions relied on by the scientist. For example, she is relying on the assumption that testimony is a legitimate way to acquire justified beliefs. This is an epistemological claim. Objection 4: The scientists’ philosophical assumptions are more powerful than the philosophers’ because there is more agreement about them. Reply: As in my reply to objection 2, we can consider a hypothetical scenario in which there is no established scientific community, nor even an established intellectual community. According to the objector, a philosophical assumption is strong enough to overturn common sense only when there is consensus about it. But I think even my opponents would agree that a lone scientist, in ignorance of what anyone else thinks of the idea that simpler theories are preferable, could, if in possession of the right empirical evidence, rationally come to believe special relativity by relying in part on philosophical claims. This shows that it is not in virtue of the consensus in the scientific community that philosophical assumptions can be more powerful than common sense. At this point, the objector may concede that consensus is not required for one to be justified in the assumption that simpler theories should be preferred to complex ones. However, the objector may then claim that the presence of significant disagreement would suffice to undermine one’s justification in this philosophical assumption. Moreover, says the objector, since

Why Philosophy Can Overturn Common Sense

195

there is significant disagreement about the philosophical assumptions that are employed in philosophical arguments against common sense, this undermines any justification we might otherwise have had for accepting these arguments. On this view, philosophy can’t overturn common sense (even though science can) because there is significant disagreement about the philosophical assumptions made in philosophical arguments. Once again, however, I think the objector is committed to some implausible claims about certain hypothetical scenarios. Suppose that, when special relativity was initially presented to the scientific community, there was significant disagreement about whether the theory should be accepted. Many—perhaps most—scientists thought the theory too absurd to take seriously, and so did not give up their belief that simultaneity is absolute. The objector is committed to thinking that, in such a case, the initial proponent of special relativity would not be justified in believing it. However, this does not seem right. It can be rational for scientists (and philosophers) to maintain their views even in the face of significant disagreement.12 Since disagreement would not be sufficient to undermine justification in the philosophical assumptions required by scientific arguments against common sense, it is not sufficient to undermine justification in the premises of philosophical arguments against common sense. Objection 5: I concede (says the objector) that special relativity has overturned our belief that simultaneity is absolute. The claim that simultaneity is absolute may be plausible; however, I think it is not as basic, as robust, or as central to our way of thinking as the kind of common sense beliefs that philosophers attempt to undermine, such as the belief that I know I have hands or the belief that tables exist. Science may be able to undermine widely accepted propositions that are highly plausible—like the proposition that tables are solid, that the Earth is flat, that the Sun orbits the Earth, and that simultaneity is absolute—but even science couldn’t undermine the real “hard core” of common sense that philosophical arguments target. So examples from science like the example of special relativity don’t really go any way towards making

12 If we also assume that those with whom one disagrees are epistemic peers in the relevant sense, then some epistemologists would disagree with this claim; see, for example, Elga (2007) and Christensen (2007). To someone with that kind of view—someone who denies that it can be rational for a scientist to believe a proposition in the face of significant peer disagreement— I offer the following, alternative response to this objection. According to this objector, it is only because of the widespread disagreement about certain philosophical claims that they are not strong enough to overturn common sense. Thus, according to the objector, it’s entirely possible that a philosopher working in isolation could rationally give up common sense beliefs on the basis of philosophical arguments. If so, then the reason why, according to the objector, philosophy can’t actually undermine common sense is just due to an accidental, contingent fact about the existence of disagreement in the field. But this does not seem to be in the spirit of their view. I would think that many of my Moorean opponents would want to insist that even the lone philosopher should not give up common sense beliefs on the basis of philosophical arguments. If so, then they would not want to endorse objection 4.

196

Susanna Rinard

it more plausible that philosophical arguments like the skeptical argument could succeed. Reply: This objector is objecting to premise (1) of the simple statement of my argument as it appears at the beginning of section 1, which is the premise that science can overturn common sense. According to the objector, there is a “hard core” of common sense that can’t be overturned by any sort of inquiry, either scientific or philosophical, and this “hard core” includes propositions in conflict with external world skepticism and ontological nihilism. Many of my opponents accept premise (1), and so I have simply presupposed it up until this point, for reasons of dialectical effectiveness. However, I think this objector makes a good point, and so I’ll take up the issue here. I don’t want to rest my case on this, but I want to begin by saying that it’s not obvious that special relativity isn’t in conflict with propositions just as central and basic as the proposition that I know I have hands. Consider an ordinary simultaneity proposition like the following: Andrew and I are brushing our teeth at the same time. One might think—indeed, this is my view—that this proposition, as ordinarily understood, could be true only if it is objectively and absolutely true that Andrew and I are brushing our teeth at the same time.13 If so, then it is not merely the abstract and theoreticalsounding proposition that simultaneity is absolute that is in conflict with special relativity; rather, this theory is in conflict with propositions as ordinary, simple, and plausible as any day-to-day simultaneity claim.14 Consider, also, the astronomical hypothesis that the Earth orbits the Sun. One might be inclined to think that if this hypothesis is true, then my bed is not in the same place as it was yesterday. After all, the Earth has moved somewhat, and so the region of space formerly occupied by my bed is now occupied by something else. But the claim that my bed is in the same place it was yesterday seems to me on par with the claim that I know I have hands. This gives us a reason for being skeptical of the objector’s claim that the propositions overturned by science are not part of the “hard core” of common sense that philosophers have attempted to undermine. But I don’t want to rest my case on these claims. Rather, I will argue that there could be successful empirical arguments against the very claims that, according to the objector, are in the “hard core” of common sense, and that these arguments rely crucially on philosophical assumptions. 13 Similarly, although I won’t attempt to argue for this here, one might think that ethical relativism is off the table because, as ordinarily understood, propositions like “Torture is wrong” are true only if true absolutely. On this view, part of what one is asserting when one asserts that torture is wrong is that torture is wrong according to all legitimate perspectives. 14 Special relativity also entails that nothing can go faster than the speed of light. This may also seem to conflict with some very basic common-sensical ideas. For example, suppose I am on a train going almost as fast as the speed of light (but not quite) and I shoot a gun in the direction of the train’s movement whose bullets go half as fast as the speed of light. It seems like the bullet should end up going faster than the speed of light. But according to special relativity, this is impossible. (Thanks to Bradford Skow for suggesting that I include this example.)

Why Philosophy Can Overturn Common Sense

197

Consider, for example, one of the common sense claims that the skeptic aims to undermine: I know I have hands. Epistemologists agree that one could get empirical evidence against this claim. For example, suppose one is told by a reliable source that doctors have found a way to cure cancer by manufacturing a drug that requires some part of a real human hand. People are being asked to donate their hands to the cause, so that enough of this drug can be manufactured to cure everyone of cancer. Those who agree to donate their hands are told they will undergo a surgical procedure under general anesthesia which involves the removal of their hands and the replacement of them with fake hands, which look and feel exactly like real hands, and from which there is no recovery period. (We can imagine that surgery has become quite advanced.) You agree to donate your hands. When you wake up in the hospital bed, you look down at the ends of your arms, and find that what appear to be your hands look and function exactly as they always have, just as you were told they would. In this case, I think you should believe that you do not have real hands (just fake hands), and so give up the common sense belief that you know you have hands. Moreover, in giving up this belief you are (at least implicitly) relying on the epistemological assumption that the fakehand hypothesis is better supported by your evidence than the empirically equivalent conspiracy theory according to which the doctors are all lying to you and your hands were never removed in the first place. Here’s another example in which one could get empirical evidence against the common sense belief that one has hands. Suppose you wake up one morning and find a ticker-tape across the bottom of your visual field. The tape reads: “Your brain has been removed from your body and put in a nutrientfilled vat. Your sensory experiences are being fed to you by our vat-tending scientists.” The tape goes on to make all sorts of predictions about what your sensory experience will be like, and these predictions all come true.15 In such a case, I think one should believe that one is a brain in a vat, and so give up all of the common sense beliefs that are incompatible with that claim. But, once again, in doing so one must rely at least implicitly on an epistemological claim according to which the bain in a vat (BIV) hypothesis is more worthy of belief, given your evidence, than the hypothesis that things really are as they seem, and that the ticker-tape is not reliable. (Perhaps, according to this alternative hypothesis, you are hallucinating the ticker-tape due to some kind of psychological ailment.) It is worth pointing out that the kind of epistemological assumptions featured in these cases are very similar in kind to the types of epistemological assumptions typically appealed to in philosophical arguments for skepticism. For example, the skeptic may employ the premise that your current sensory evidence is neutral between the BIV hypothesis and the hypothesis that things

15 This example is not original to me, but unfortunately I can’t remember where I first heard it.

198

Susanna Rinard

are as they appear. The epistemological assumptions appealed to in the abovedescribed empirical arguments against common sense are claims of a similar kind: claims about which hypotheses about the nature of the real world are best supported by the evidence of your senses. This completes my reply to this objection. I have argued that there could be successful empirical arguments against the very same common sense propositions that philosophical arguments seek to undermine. Moreover, these empirical arguments rely crucially on philosophical assumptions. So, in the relevant sense, science can overturn common sense and premise (1) of the argument remains intact. This concludes part one of the paper. The overall goal of this first part was to argue that philosophy can overturn common sense, since science can. Scientific arguments against common sense, I claim, rely crucially on philosophical assumptions. For example, the argument for special relativity relies crucially on the assumption that simpler hypotheses should be preferred over complex ones. Since an argument is only as strong as its weakest premise, if these scientific arguments succeed, it must be the case that the philosophical assumptions on which they rely are stronger than the common sense propositions they target. But if so, then there could be an argument against common sense whose premises consist entirely of philosophical claims, each of which is stronger than the targeted common sense proposition. So philosophy can overturn common sense, since science can.

4.

I N T R O D U C T I O N T O PA R T T W O

In this second half of the paper I will consider some of the most common motivations for my opponents’ view that philosophy cannot overturn common sense. Each motivation relies crucially on some principle of philosophical methodology. I will consider the following three methodological principles: (1) common sense propositions enjoy greater plausibility than philosophical claims, and thus should be given priority when they conflict (defended most famously by Moore 1962a, 1962b, 1962c); (2) general principles should be rejected when they conflict with a large number of judgments about particular cases (defended in Kelly 2005 as a kind of Mooreanism and also, I think, suggested by certain remarks in Goodman 1983); (3) conservatism—minimizing change—is an important aspect of rational belief revision (defended in Harman 2003 as a version of reflective equilibrium and also suggested by certain remarks in Lewis 1999). In each of these three cases, I will argue either that the methodological principle in question is false, or that it fails to provide a genuinely independent motivation for the idea that philosophy can’t overturn common sense. For example, in section 7 I will present a novel objection to conservatism that, I argue, undermines Harman’s principle of philosophical methodology. I will claim that, on one important notion of epistemic dependence, beliefs

Why Philosophy Can Overturn Common Sense

199

that depend on a belief Q are completely irrelevant to whether belief in Q should be maintained. However, I will show that there is an important class of cases in which principles of methodological conservatism (like Harman’s) have the consequence that whether or not belief in Q should be maintained depends in part on the nature and number of beliefs that depend on Q. I think this brings out a deep problem with the view that conservatism (minimizing overall change) is an important aspect of philosophical methodology. Throughout the discussion of all three methodological principles, I will focus (as do the defenders of these principles) on the case of external world skepticism. The skeptic presents us with a valid argument from plausible philosophical premises for the conclusion that no one knows anything about the external world. For example, the skeptic may argue that all we have to go on in assessing propositions about the external world is the evidence of our senses, that is, propositions about how things appear to us. But, continues the skeptic, this evidence is neutral between the hypothesis that the external world really is as it seems and the hypothesis that one is a brain in a vat being deceived by a mad scientist. So, concludes the skeptic, no one knows whether they really have hands, whether there is a table before them, etc. By presenting her argument, the skeptic has shown us that there is a contradiction between some plausible epistemological principles and our ordinary common sense beliefs about what we know. How should we revise our beliefs in light of this? Should we accept the skeptic’s premises and the radical conclusion that follows from them? Or should we hold on to our pre-theoretic belief that we know many things, and reject one of the skeptic’s premises? Each of the three methodological principles I will discuss yields the verdict (according to its defenders) that we should hold on to our common sense beliefs about the extent of our knowledge, and give up one of the skeptic’s premises. But I will argue that these principles do not succeed in motivating this view.

5.

M O O R E ’ S P L AU S I B I L I T Y P R I N C I P L E

We begin, of course, with G. E. Moore. I’ll follow Lycan 2001 p. 38–39 in characterizing Moore’s view roughly as follows: A common sense proposition is more plausible than the premises in a philosophical argument against it. So, if forced to choose between common sense and a philosophical premise, one should retain belief in common sense and give up the philosophical premise. Call this Moore’s Plausibility Principle. In assessing this view, I think we should first ask what is meant by “plausible.” On one reading, A is more plausible than B just in case A seems pre-theoretically prima facie more worthy of belief than B. If this is what is meant by “plausible,” then I am willing to grant that ordinary common sense may indeed be more plausible than complex abstract philosophical principles.

200

Susanna Rinard

The trouble is that it is simply not the case that, when conflicts arise, we should always retain belief in whichever proposition seems pre-theoretically prima facie more worthy of belief. Sometimes one discovers by reflection that one’s pre-theoretical judgments of comparative worthiness of belief were the reverse of what they should have been. (This point is also made by Conee 2001 and echoed by Kelly 2005.) For example, consider the following two propositions: (A) There are more positive integers than there are positive even integers. (B) Two sets X and Y have the same cardinality just in case there is a one-to-one mapping from the elements of X to the elements of Y. Pre-theoretically, A seems more worthy of belief than B. It just seems obvious that there are more positive integers than positive even integers, and B is a complicated principle that requires some thought before endorsing. So if plausibility is interpreted as suggested above, Moore’s Plausibility Principle says that one should retain belief in A and reject B. However, this is exactly the opposite of what should be done. Reflection should result in a reversal of one’s initial judgment of the comparative worthiness of belief of these two propositions. For all that has been said so far, the skeptical case could be just like this one. Pre-theoretically, it may seem obvious that one knows that one has hands, and the premises of the skeptical argument may seem to be complicated claims that require thought before endorsing. However, after reflecting on the skeptical premises, they can come to seem more and more plausible until it becomes clear that they are undeniable, and the claim to knowledge must be rejected. So, although the skeptic may agree with the Moorean about the pretheoretical, prima facie judgments, neither the skeptic nor the Moorean should accept the principle that these initial judgments determine which of the two propositions it is rational to give up, since reflection can sometimes reveal that one’s initial judgments were the reverse of what they should have been.16 This discussion may suggest an alternative reading of the Plausibility Principle. On the first reading, which of two conflicting propositions you should give up depends on your pre-theoretic judgments about them. On the second version of the principle, which of the two propositions you should give up is not determined by your initial judgments; rather, it is determined by your judgments upon careful reflection. According to this version of the principle, one should give up whichever of the two propositions seems least worthy of belief upon careful reflection. However, this version of the principle does not yield the result the Moorean desires. It makes the normative facts about what one should believe dependent on the contingent psychological facts about what happens to seem most

16 The Monty Hall problem may provide another example of a case in which one’s initial judgments of the comparative worthiness of belief are the reverse of what they should be. In general, psychology experiments like the Wason card selection task and the experiments discussed in the heuristics and biases literature (see for example Kahneman, Tversky, and Slovic 1982) provide another source of counterexamples to Moore’s Plausibility Principle.

Why Philosophy Can Overturn Common Sense

201

worthy of belief upon reflection. As a matter of psychological fact, upon careful reflection, each premise of the skeptical argument seems to the external world skeptic to be more worthy of belief than the common sense view that she has knowledge of the external world. So, according to this version of the plausibility principle, she should give up the common sense belief on the basis of the philosophical argument for skepticism. If so, then philosophical argument can undermine common sense, which is exactly what the Moorean seeks to deny. We may consider a third, more normative understanding of what plausibility amounts to: A is more plausible than B just in case as a matter of fact A is more worthy of belief than B. This does not seem to help much; after all, the skeptic thinks that her premises are more worthy of belief than the common sense claim to knowledge, and so the skeptic will think that this version of the plausibility principle vindicates her position. The Moorean may try to tack on the claim that the skeptic is wrong about this: that, as a matter of fact, it is the common sense claim that is most worthy of belief. Simply to insist that this is the case, however, does not provide us with an independent motivation for the Moorean view. To say that common sense propositions are more worthy of belief than philosophical claims just is to say that, in cases of conflict, we should retain belief in common sense propositions at the expense of philosophical claims, which is just to say that philosophy can’t overturn common sense. What we have here is a restatement of the Moorean view, not an independent motivation for it. I conclude, then, that each version of the plausibility principle is either false, or fails to provide a genuinely independent motivation for the view that philosophy can’t overturn common sense.

6.

G E N E R A L P R I N C I P L E S V S . PA R T I C U L A R C A S E J U D G M E N T S

In his 2005 paper “Moorean Facts and Belief Revision or Can the Skeptic Win?”, Thomas Kelly draws the same conclusion that I did in the previous section: Moore’s plausiblility principle does not successfully make the case that philosophical argumentation can’t overturn common sense propositions like our ordinary claims to knowledge. However, Kelly then goes on to provide an alternative interpretation of Moore that he thinks does succeed in showing this. I think Kelly’s version of Mooreanism is also unsuccessful. Kelly begins by distinguishing two different principles of philosophical methodology: particularism and methodism.17 The particularist and the methodist both begin with some initial judgments about particular cases, and 17 Kelly’s terminology is borrowed from Chisholm’s 1973 book The Problem of the Criterion. Chisholm used the terms slightly differently, to mark the distinction between the epistemologist who builds his theory of knowledge around his initial views about the extent of our knowledge (the particularist) and the epistemologist who builds his theory of knowledge around his initial views about the criteria for knowledge (the methodist).

202

Susanna Rinard

some initial judgments about general principles. The difference between the particularist and the methodist manifests itself when contradictions are discovered between the initial case judgments and general principle judgments. In revising his beliefs in light of such contradictions, the particularist will give more weight to his initial judgments about cases than to his judgments about general principles. So, if just one of his initial case judgments is in conflict with one of his general principle judgments, he will give up the general principle and retain the case judgment. The methodist, on the other hand, will give more weight to the general principle judgment than to the case judgment. So, in a case of conflict between one judgment of each type, the methodist will retain the general principle judgment and revise his case judgment. Kelly recommends that we think of these two methodologies as lying on a spectrum: near the far end of the spectrum is hyper methodism, according to which one should give almost no weight at all to one’s judgments about cases; on the other end is hyper particularism, according to which one should give almost no weight at all to one’s judgements about general principles; and in the middle is the view that judgments at the different levels of generality should be accorded equal weight. Kelly’s discussion of these views presupposes that one point on this spectrum is correct for all cases in which there’s a conflict between one’s case judgments and one’s judgments about principles. It is this presupposition that I will take issue with. But first, let’s get Kelly’s argument on the table. Kelly’s first claim is that the correct principle of philosophical methodology is not very far over on the methodistic side of the spectrum. Kelly leaves open whether it is particularistic, or in the middle, or slightly on the methodistic side, but he claims that any kind of robust methodism is false, as a principle of philosophical methodology. Kelly’s defense of this claim is straightforward: often, we reject a general principle because of a single counterexample. That is, in cases of successful counter examples, we give up a judgment we have about a general principle because it conflicts with a single case judgment. This would not be possible if a principle of robust methodism were applied, since robust methodism assigns substantially more weight to general principles than to case judgments, and so a principle would not be given up just because it conflicts with a single case judgment. Thus, Kelly concludes that since we should sometimes give up a general principle because it conflicts with a single case judgment (as, for example, in the Gettier case), the correct principle of philosophical methodology is either close to the middle of the spectrum, or particularistic. Certainly, claims Kelly, any kind of robust methodism is false. And this result, says Kelly, is all that is needed to mount a Moorean response to the skeptic. Kelly first notes that the skeptical argument relies crucially on a general principle about knowledge or evidence. Moreover, he says, anyone who accepts the skeptical argument will have to give up a very large number of judgments about cases, like the judgment that I know that I have hands,

Why Philosophy Can Overturn Common Sense

203

that I know there is a table before me, etc. When we are confronted with the skeptic’s argument, then, we are forced to choose between a general principle, on the one hand, and very large number of judgments about cases, on the other. According to any methodological principle that gives case judgments and principle judgments roughly equal weight (or more weight to case judgments), the recommended course of action is clear: give up the general principle in favor of the case judgments. If so, then the skeptical argument cannot overturn our common sense beliefs about the extent of our knowledge.18 The only way out of this conclusion, according to Kelly, would be to adopt a position of hyper methodism. (It is this claim that I will disagree with.) Kelly rightly notes that this option is highly undesirable, since a hyper methodist could never give up a general principle in the face of a single counter example. As mentioned briefly above, I disagree with the presupposition implicit in Kelly’s argument that one point on the spectrum of methodologies he describes must be appropriate for every case of conflict between general principles and case judgments. I agree with Kelly that in some cases, a general principle should be rejected because it conflicts with a single case judgment. But, it simply doesn’t follow that one must never retain belief in a general principle that conflicts with many case judgments. The rational response to a conflict of this sort may depend on details of the case that go beyond the different numbers of types of judgments that one has. In some cases, it is rational to give up a general principle because it conflicts with one’s judgments about cases. Kelly provides some examples of this sort. However, there are other cases in which rationality requires one to give up one’s judgments about cases because they conflict with a general principle. I will provide some examples of this sort below. Kelly’s argument commits him to the view that in any case of conflict between many case judgments and a single general principle judgment, one must always reject the general principle judgment. Call this Kelly’s Claim. I don’t think Kelly’s Claim is correct. I will now present two cases in which the rational response to a conflict of this kind is to give up the case judgments and retain the general principle judgement. If I am right about these cases,

18 Nelson Goodman, in chapter three of Fact, Fiction, and Forecast, presents a view concerning inductive skepticism that, I think, is essentially the same as Kelly’s argument against the external world skeptic. According to Goodman, we discover what a term means by finding a general principle about its meaning according to which the cases to which the term is actually applied just are the cases to which the term is properly applied. This is how we define “tree”: we look for a general principle that captures what is in common between all the objects to which we actually apply the term “tree.” Likewise, says Goodman, to define “valid induction,” we should look for whatever it is that is in common between all of the arguments to which we actually apply the term “valid induction.” Clearly, if we follow this procedure, we will never get the result that there are no valid inductions. So, if this procedure is corrent, the inductive skeptic is wrong. The objections I will present against Kelly’s Claim work equally well against this interpretation of Goodman’s view. For example, if we were to apply Goodman’s procedure in figuring out what “justified certainty” means, we would never reach what I will argue is the correct conclusion.

204

Susanna Rinard

then Kelly’s Claim is false, and so Kelly’s argument for Mooreanism will not go through. My first example is taken from the philosophy of probability. My second example is strongly parallel to the skepticism case, and thus is particularly instructive in this context. First, consider someone who tends to commit the gambler’s fallacy. If he sees a fair coin land heads many times in a row, he judges that the coin is more likely to land tails than heads on the next toss; if the coin lands tails many times in a row, he judges that heads is more likely than tails on the next toss; etc. Suppose that this person then takes a philosophy of probability class, and encounters the principle “Independent tosses of a fair coin are equally likely to come up heads.” This principle strikes him as highly plausible. Then, however, it is pointed out to him that if he accepts this principle, he should no longer judge that heads is more likely than tails after a long series of tails (and likewise for the other gambler’s fallacy judgments he is apt to make). According to Kelly’s Claim, the only rational response for this person is to regard the principle as false, since it conflicts with many of his case judgments. But this is clearly wrong: it may well be that the rational response for him is to give up the case judgments. After all, if this person reflects on the principle, he may come to see why it’s true, and come to realize that his previous case judgments were mistaken and confused. In such a case we would want to say that the rational thing for this person to do is to retain belief in the general principle, and let this principle guide his case judgments. But this is exactly what Kelly’s Claim says he must not do. I think there are many examples of this sort: cases in which many of one’s initial case judgments were based on conceptual confusion, which can be cleared up when one reflects on a conflicting (but true) general principle. Many of the other examples from the heuristics and biases literature could be used to make the same point. I’ll now describe my second counterexample to Kelly’s Claim. Consider someone who has never taken a philosophy class, and who has never taken the time to reflect on the epistemic status of ordinary, day-to-day beliefs. Let’s call this person Ordinary. Now imagine that a philosopher goes and has a chat with Ordinary. They are sitting on a sofa in front of a table, on which there lies a book, in plain view, in natural light, etc. Upon chatting with Ordinary, the philosopher discovers that he believes many things, including the following: he believes that there’s a book on the table and he believes that the sun will rise tomorrow. Moreover, the philosopher discovers that Ordinary is certain of these things, and—crucially—that he takes himself to be justified in being certain of them. Having never heard of the fanciful skeptical hypotheses familiar to philosophers, Ordinary takes himself to have definitively ruled out all possibilities in which these propositions are false, and so takes himself to be fully justified and rational in being certain that they are true. Suppose, however, that the philosopher goes on to describe to Ordinary certain skeptical scenarios, involving evil demons who trick people into

Why Philosophy Can Overturn Common Sense

205

believing that there are books in front of them when there really aren’t, counter inductive worlds in which one remarkable day the sun fails to rise, etc. The philosopher goes on to explain to Ordinary that, since such scenarios are possible, many of Ordinary’s judgments of the form “I am justified in being certain that P” conflict with the principle “If there is a possible scenario in which not-P that one cannot definitively rule out, then one is not justified in being certain that P.” Now that he is aware of this contradiction, Ordinary must choose between his belief in this highly plausible general principle and many of his beliefs of the form “I am justified in being certain that P.” How should Ordinary revise his beliefs in this case? I think most epistemologists would agree that he should retain belief in the general principle and give up the case judgments. Of course, most would also say that he should still think that he knows these propositions, or at least that he’s justified in believing them. But he should no longer think that he’s justified in being certain of them, since he no longer takes himself to have ruled out all possible scenarios in which they are false. This is another case in which a large number of one’s initial case judgments were mistaken as a result of conceptual confusion which can be cleared up by reflection. In this case, we might say that Ordinary was lead into these mistakes by a limited imagination: he failed to realize that scenarios like the skeptical scenarios were possible; they had simply never occurred to him. This failure to recognize the possibility of scenarios of a certain sort lead him to misapply his own concept of justified certainty. So this is a case in which Ordinary should, upon realizing this mistake, retain belief in the general principle about certainty and let its proper application guide his case judgments. These two counterexamples to Kelly’s Claim illustrate a general way in which it goes wrong. Kelly’s Claim does not allow for the possibility that one systematically misapplies one’s own concepts due to conceptual confusion.19 But surely this is possible, as these two cases illustrate. It is possible to systematically misapply one’s concept of comparative probability, and it is possible to systematically misapply one’s concept of justified certainty. The skeptic claims that, just as it is possible to systematically misapply these concepts, it is possible to systematically misapply our concept of knowledge. And this, contends the skeptic, is exactly what we have in fact done. In comparing the skeptical case to the two cases just discussed, I think it is especially helpful to focus on the certainty case. This is because the response to the conflict in the certainty case that is widely agreed by epistemologists to be the rational one is highly structurally analogous to what the skeptic takes to be the rational response in the case of knowledge. In both cases, there is some general principle about the conditions under which one stands in a certain kind of epistemic relation to a proposition. In the one case, it is a principle

19 Strictly speaking, Kelly’s Claim does not allow for the possibility that one rationally believes that one has systematically failed to misapply one’s own concept.

206

Susanna Rinard

about when one is justified in being certain of something; in the other case, it is whatever principle about knowledge is employed in the skeptical argument. In both cases, one initially judges, of many of the propositions that one believes, that one does stand in that epistemic relation to them. In fact, it may well be almost the same set of propositions in each case. Most epistemologists would recommend revising one’s belief, about each such proposition, that one is justified in being certain of it. The skeptic recommends revising one’s belief, about each such proposition, that one knows it. The two cases of belief revision are, structurally, almost exactly the same. We have just seen that the certainty case blocks Kelly’s anti-skeptical argument by showing that the methodological principle he relies on—Kelly’s Claim—is not true. In general, it is a consequence of the close structural similarity of the certainty case and the knowledge case that it will be extremely hard for any general principle of philosophical methodology to yield the right result about the certainty case—namely, that one should give up one’s initial case judgments—while also yielding the result that one should retain one’s initial case judgments about what one knows. It’s time to take stock. Kelly endorses the methodological claim that one should always give up a general principle when it conflicts with many case judgments. If this is true, then, when confronted with a conflict between a general philosophical principle and common sense judgments about particular cases, one should give up the philosophical principle and retain belief in the common sense judgments. However, I have argued that Kelly’s methodological principle fails because it does not allow for the possibility that one has systematically misapplied one’s concepts. I used two cases—the gambler’s fallacy case and the certainty case—to illustrate that systematic misapplication of concepts is indeed possible.

7.

C O N S E RVAT I S M A N D R E F L E C T I V E E Q U I L I B R I U M

In this section I will discuss a version of reflective equilibrium that emphasizes minimal change as an important aspect of rational belief revision—not just in philosophy, but belief revision of any kind. I will focus on the version elaborated and defended by Gilbert Harman, but note that Lewis 1999, Pryor 2000, and others also take conservatism to be an important part of proper philosophical methodology. In the section entitled “Philosophical Skepticism” of his paper “Skepticism and Foundations,” Harman describes the basics of his version of reflective equilibrium and argues that it has the result that we should retain belief in the common sense proposition that we know we have hands and give up one of the skeptic’s premises. Harman begins by considering a person who has the usual beliefs about the external world, and about the extent of his knowledge of the external world, and who also believes (implicitly, perhaps) the premises of the skeptical

Why Philosophy Can Overturn Common Sense

207

argument. Suppose this person then discovers the contradiction between his beliefs about the extent of his knowledge and the skeptical premises that he accepts. How should this person revise his beliefs in order to avoid this inconsistency? Harman’s principle says that “the way to resolve such conflicts is by finding the minimal adjustment in S’s beliefs and methods that provides the most simple and coherent upshot.”20 Upon discovering a contradiction in one’s beliefs, one must choose to update to just one of the many possible complete systems of belief that resolve this contradiction in some way. According to Harman’s principle, you should adopt whichever one of these complete systems of belief best balances minimal change, coherence, and simplicity. Now consider the particular case at hand, in which one has discovered a conflict between the skeptic’s premises and one’s beliefs about the extent of one’s knowledge. Retaining the skeptical premises requires giving up a vast number of beliefs about the external world, and about one’s knowledge of it. This would involve a very large change in one’s system of belief. Giving up one of the skeptical premises, however, would require only a very small change in comparison; all of one’s beliefs about the external world, and one’s knowledge of it, would remain intact. Thus, Harman’s principle of belief revision, with its emphasis on minimal change, yields the result that one ought to give up one of the skeptical premises.21 Lewis expresses the same general thought in the following quote from “Elusive Knowledge:” “We are caught between the rock of fallibilism and the whirlpool of skepticism. Both are mad! Yet fallibilism is the less intrusive madness. It demands less frequent corrections of what we want to say. So, if forced to choose, I choose fallibilism.”22 I’ll focus on Harman’s version of the idea, though, since it’s developed in more detail. I will argue that Harman’s principle of rational belief revision is not a good one. First, note that the counterexamples to Kelly’s Claim, discussed in the previous section, also constitute counterexamples to Harman’s principle. In these counterexamples, it is rational to revise a great many of one’s initial case judgments because they conflict with a single general principle. Because there are so many case judgments that must be revised, rejecting the general principle instead would have constituted a more minimal change. So, Harman’s principle shares the flaw identified earlier in Kelly’s principle: it does not allow for the possibility of systematic misapplication of one’s concepts. However, I want to develop and bring out another, perhaps deeper, problem with Harman’s principle of rational belief revision. 20

Harman (2003, 10). In fact, Harman needs to make a few further assumptions in order to reach this conclusion. It must be assumed that giving up the skeptical principle does not result in large losses in simplicity and coherence; and, it must be assumed that the option of retaining the skeptical principle and the claims to knowledge, and giving up the belief that they conflict, would not be acceptable. 22 Lewis (1999, 419). 21

208

Susanna Rinard

As noted above, while Harman takes his principle to correctly describe rational revision of beliefs in philosophy, he does not think it is limited to beliefs in this area. Indeed, Harman takes his principle to apply to all rational belief revision. I will present a counterexample to Harman’s view involving a case of evidence by testimony to bring out the second general problem with Harman’s principle. Ultimately, I will argue that Harman’s principle fails to be appropriately sensitive to the relations of epistemic dependence that obtain between some beliefs. I will claim that the nature and number of beliefs that depend epistemically on a belief Q are irrelevant to whether Q should be given up. However, whether Harman’s principle recommends giving up belief in Q is sometimes determined in part by the nature and number of beliefs that depend on Q. This, I think, is the central problem with Harman’s principle. I’ll start by describing a case that does not itself constitute a counter example to Harman’s principle, but which will help me set up the counter example. Case 1: Imagine that you have lived your entire life in a remote, isolated village, and know nothing about the world beyond the village borders. One day you encounter a visiting stranger, Alice, who tells you she’s spent her life traveling around the world, including a place called “Costa Rica,” and proceeds to tell you about the beautiful birds she saw there. Naturally, you believe what she tells you. Later, however, you meet up with one of your best buddies—Bert. Bert has an uncanny knack for being able to tell when people are lying. Time and again he’s told you that someone was lying about something, and independent investigation proved him right. This time, Bert tells you that Alice was lying—in fact, she has never been to Costa Rica. It is clear that in this situation you should believe Bert. After all, Bert has an excellent track record of detecting lying, and you only just met Alice. Now consider a modified version of this case. Case 2 (below) is exactly like case 1, with the following exception: Case 2: Alice didn’t just tell you a few things about Costa Rica. Rather, she has told you many stories about her travels all around the world. You love listening to these stories, and you have spent most of your evenings listening to them. As a result, you have accumulated a vast and detailed system of beliefs about the world beyond the village border, all on the basis of Alice’s testimony. It is only after this has been going on quite a while that Bert tells you that Alice has been lying the whole time. I think it’s equally clear that you should believe Bert in case 2 as well. After all, as before, Bert has an incredible track record; and, the mere fact that Alice has told you many things does not make any particular one of them more credible. However, Harman’s principle says otherwise. In case 2 (but not case 1) believing Bert would require you to give up a vast number of other beliefs that you have—you would have to give up all your beliefs about the nature of

Why Philosophy Can Overturn Common Sense

209

the world outside your village. This constitutes quite a substantial change in your overall system of belief. Such a substantial change would not be required if you were to believe instead that Bert just happened to be wrong in this particular case. This would be quite a minimal change, overall, since you could retain all the beliefs you formed on the basis of Alice’s testimony. So, Harman’s principle says you should retain belief in everything that Alice told you. But clearly this is not the rational response to hearing what Bert said. You should believe Bert, even though this requires you to give up all of your beliefs about what lies beyond the village border. Before explaining what general problem I think this illustrates with Harman’s view, I will consider and respond to a possible response from a defender of Harman’s principle of belief revision. Harman might respond that in fact, his principle does not have the consequence that I claim it has. I claimed that Harman’s principle has the consequence that, in case 2, you should retain believe in what Alice told you, since this makes for a more minimal change. But Harman’s principle says that minimal change is just one criterion, to be balanced with the other criteria of simplicity and coherence. While there doesn’t seem to be any difference in simplicity between the two belief systems, there may seem to be a difference in coherence. It will be helpful to describe in more detail the two complete belief systems you’re choosing between. I’ve named them BERT and ALICE: BERT: Bert is right that Alice was lying to me; after all, he’s always been right in the past. So, I cannot trust anything that Alice told me, and must give up all the beliefs I had formed on the basis of her testimony. ALICE: Alice has been telling me the truth the entire time. Although Bert is generally right, he was wrong in this case. Harman might claim that ALICE, while a more minimal change than BERT, is less coherent than BERT. After all, on belief system ALICE, you believe that Bert is generally reliable, but you also believe that he was wrong in this case, even though you don’t also believe that there is some particular feature of this case in virtue of which he was likely to make a mistake. Surely this collection of beliefs doesn’t hang together very well, and could be seen as making for a slight incoherence in belief system ALICE. I think this putative problem can be circumvented. A slight modification of ALICE will get rid of the minor incoherence. Consider ALICE*: ALICE*: Alice has been telling me the truth the entire time. Although in general when Bert tells me someone is lying, that person really is lying, I can see that Bert is jealous of all the time I’m spending with Alice, and he told me she is lying only because he hopes it will have the effect of my spending less time with her. I couldn’t explain to someone in words exactly what it is that makes me believe he’s jealous, but I can just tell he is by the way they interact.

210

Susanna Rinard

ALICE* does not have any of the mild incoherence that might have been found in ALICE; and, it is still a much more minimal change than BERT. So, even if Harman’s principle does not yield the result that one should adopt belief system ALICE, it does say that rather than adopting belief system BERT, you should revise your system of beliefs to ALICE*, since that would constitute a much more minimal change, and it is no less simple or coherent. So, for example, consider someone who, prior to hearing Bert’s claim that Alice is lying, did not suspect that Bert was jealous, and who has no good evidence that Bert is jealous. However, this person is loath to make any very serious changes to his belief system, and whenever he encounters evidence against many of his beliefs, he has a tendency to come up with some way of explaining the evidence away, rather than coming to terms with the real implications of the evidence (of course he would not describe himself that way). So, when this person hears Bert’s claim that Alice has been lying, he immediately begins to suspect that there’s something fishy going on, and eventually convinces himself that Bert is jealous of Alice, etc.—in short, he comes to accept belief system ALICE*. According to Harman’s principle, this person is revising his beliefs exactly as he should be. But this is the wrong result. The rational response to Bert’s testimony is not to continue to believe Alice, but rather to give up belief in everything Alice said. Now that I’ve argued that this case does indeed constitute a problem for Harman’s principle, I’ll move on to the diagnosis: what deeper problem with Harman’s principle is illustrated by this counterexample? The core of my diagnosis will be as follows. There is a relation of epistemic dependence that holds between some beliefs (I will say more about epistemic dependence in a moment). I claim that beliefs that depend epistemically on some other belief Q are irrelevant to whether or not Q should be given up. That is, how many beliefs there are that depend on Q, and what the contents of those beliefs are, is irrelevant to whether belief in Q should be maintained. But whether Harman’s principle recommends giving up belief in Q is sometimes determined in part by beliefs that depend on Q. This is the problem with Harman’s principle that is brought out by the counterexample I gave. In order to further explain and defend this diagnosis, I need to give some explanation of the notion of epistemic dependence. There are several formal properties of the dependence relation that can be noted at the outset. First, dependence is relative to an agent; it may be that P depends on Q for one agent but not for another. Second, the dependence relation is never symmetric. If P depends on Q, then Q does not depend on P (relative to the same agent, of course).23

23 Some epistemologists (perhaps some coherentists) may disagree with me about this: they may think that some cases of dependence are symmetric. That’s OK for me, as long as the coherentist is willing to grant that some cases of dependence are asymmetric, and, in particular, that my belief that there are beautiful birds in Costa Rica asymmetrically depends on my belief that Alice has been telling the truth. Instead of claiming that beliefs that depend on Q are never

Why Philosophy Can Overturn Common Sense

211

Now for a more substantive characterization of dependence. The intuitive idea is that P depends on Q for an agent S when S’s belief that P is based (at least in part) on S’s belief that Q, and when Q (perhaps in conjunction with other beliefs that S has) does indeed justify P; P derives its justification (at least in part) from Q. Relations of epistemic dependence are commonplace. My belief that the chicken I just ate had not gone bad depends on my belief that I took the chicken meat out of the freezer and into the fridge only a day or two ago (and not, say, several weeks ago). My belief that the temperature is 84 degrees depends on my belief that my thermometer reads “84.” My belief that the butler did it depends on my belief that the butler’s fingerprints were found at the scene of the crime. Moreover, your belief that there are beautiful birds in Costa Rica depends on your belief that Alice is telling the truth. Harman’s principle says that you should retain your belief that Alice is telling the truth because giving it up would require you to give up your belief that birds in Costa Rica are beautiful, and all other such beliefs. On Harman’s principle, part of what makes it the case that you are justified in holding on to your belief that Alice is telling the truth is that you have many other beliefs that depend on it, and which you would have to give up with it. But that is to make beliefs that depend on Q relevant to whether or not belief in Q should be retained. In particular, it is to make beliefs that depend on your belief that Alice is telling the truth relevant to whether you should continue to believe that Alice is telling the truth. And it is my contention that beliefs that depend on Q are never relevant to whether belief in Q should be retained. Whether you have any beliefs that depend on Q, and if so, what beliefs they are, is irrelevant to whether or not it would be epistemically rational for you to retain belief in Q. The problem with Harman’s principle is that it is not consistent with this fact. Let’s take stock. I have argued that there are two main problems for Harman’s principle of belief revision. The first is that, like Kelly’s Claim, it does not allow for the possibility of systematic misapplication of concepts. The second is that it allows the number and nature of beliefs that depend on Q to be relevant to whether belief in Q should be retained. Thus, Harman’s principle is false, and so, like Moore’s plausibility principle and Kelly’s Claim, it cannot be used to motivate the idea that philosophy can’t overturn common sense.

8.

S U M M A RY A N D C O N C L U D I N G R E M A R K S

It has become popular to think of common sense as an oracle to which the philosopher must always defer. If a philosopher’s theory turns out to conflict with common sense, the philosopher is taken to have overstepped her bounds relevant to whether Q should be given up, I would then re-state my claim as follows: beliefs that depend asymmetrically on Q are never relevant to whether Q should be given up. The argument against Harman goes through just as well this way.

212

Susanna Rinard

and is expected to retreat. It has been my aim in this paper to convince the reader that this conception of philosophy is untenable. Philosophical argument is perfectly capable of undermining our ordinary, pre-theoretic view of the world. In the first half of the paper, I provided a positive argument for my central claim that philosophy can overturn common sense. In particular, I argued that if (as my opponents agree) science can overturn common sense, then so can philosophy. In the second half of the paper, I objected to the main motivations philosophers have had for taking the opposing view, namely that philosophy is not powerful enough to overturn common sense. These motivations turned out to rely on faulty theories of philosophical methodology. One consequence is that we cannot simply dismiss out of hand philosophical arguments—like arguments for skepticism—that target common sense claims. We cannot know in advance that our ordinary beliefs will stand fast in the face of such arguments; only careful and detailed consideration of these arguments can reveal whether or not they succeed. This brings a heightened sense of importance and urgency to philosophical inquiry. Nothing less than our most basic and central beliefs are at stake.

REFERENCES

Chisholm, Roderick (1973) The Problem of the Criterion Milwaukee: Marquette University Press. Christensen, David (2007) “The Epistemology of Disagreement: The Good News” Philosophical Review 116: 187–217. Conee, Earl (2001) “Comments on Bill Lycan’s Moore Against the New Skeptics” Philosophical Studies 103 (1): 55–59. Dorr, Cian (2002) “The Simplicity of Everything,” Princeton University doctoral dissertation. Elga, Adam (2007) “Reflection and Disagreement” Nous 41 (3): 478–502. Fine, Kit (2001) “The Question of Realism” The Philosophers’ Imprint 1 (2): 1–30. Goodman, Nelson (1983) Fact, Fiction, and Forecast Cambridge, Mass: Harvard University Press. Gupta, Anil (2006) Empiricism and Experience Oxford: Oxford University Press Harman, Gilbert (2003) “Skepticism and Foundations,” in Steven Luper (ed.) The Skeptics: Contemporary Essays Aldershot: Ashgate Publishing: 1–11. Hegel, G. W. F. (1977) Phenomenology of Spirit Oxford: Oxford University Press. Kahneman, D., Tversky, A., and Slovic, P. (1982) Judgment Under Uncertainty: Heuristics and Biases New York: Cambridge University Press. Kant, Immanuel (2008) Prolegomena to Any Future Metaphysics New York: Cambridge University Press. Kelly, Thomas (2005) “Moorean Facts and Belief Revision or Can the Skeptic Win?” in John Hawthorne (ed.) Philosophical Perspectives: Epistemology 19: 179–209.

Why Philosophy Can Overturn Common Sense

213

(2008) “Common Sense as Evidence: Against Revisionary Ontology and Skepticism,” in Peter French and Howard Wettstein (eds.) Midwest Studies in Philosophy: Truth and Its Deformities 32: 53–78. Lewis, David (1973) Counterfactuals, Cambridge, Mass: Harvard University Press. (1999) “Elusive Knowledge,” in David Lewis (ed.) Papers in Metaphysics and Epistemology Cambridge: Cambridge University Press, 248–261. Lycan, William (2001) “Moore Against the New Skeptics” Philosophical Studies 103 (1): 35–53. Moore, G. E. (1962) “A Defense of Common Sense,” in G. E. Moore (ed.) Philosophical Papers London: Collier Books. (1962) “Four Forms of Skepticism” in G. E. Moore (ed.) Philosophical Papers London: Collier Books. (1962) “Proof of an External World” in G. E. Moore (ed.) Philosophical Papers London: Collier Books. Pryor, James (2000) “The Skeptic and the Dogmatist” Nous 34: 517–549. Rinard, Susanna (ms) “Reasoning One’s Way out of Skepticism”. Unger, Peter (1975) Ignorance: A Case for Skepticism Oxford: Oxford University Press. Van Inwagen, Peter (1990) Material Beings Ithaca: Cornell University Press.

8. Could Evolution Explain Our Reliability about Logic? Joshua Schechter

1.

INTRODUCTION

Let the “logical propositions” be the logical truths and logical falsehoods. We are reliable about logic in the following sense: The logical propositions that we believe (upon reflection and discussion) are by-and-large true and the logical propositions that we disbelieve (upon reflection and discussion) are by-and-large false.1 This is a striking fact about us, one that stands in need of explanation. But it is not at all clear how to explain it. So we have a puzzle: How is it that our logical beliefs match the logical facts? How is it that we are reliable about logic? This puzzle is akin to the well-known Benacerraf-Field problem for mathematical Platonism.2 According to that argument, mathematical Platonists are unable to explain our reliability about mathematics due to their claim that mathematical objects are abstract. In the absence of some amazing cosmic accident, it is difficult to see how we could have ended up with the correct mathematical beliefs and practices. This provides reason to reject Platonism about mathematics. Logic does not have—or at least does not obviously have—a distinctive ontology. Nevertheless, the challenge of explaining our reliability about logic is also daunting. What gives this reliability challenge its bite is not the ontology of logic but the (apparent) objectivity of logic.3 We can understand the claim that logic is objective as the conjunction of the following three theses:4 First, certain sentences and mental representations express logical truths

1 This characterization of our reliability about logic is broadly analogous to the characterization of our reliability about mathematics in Field (1989). 2 See the introduction and title essay of Field (1989). See Benacerraf (1973) for an important precursor. 3 The same holds true for mathematics. What gives the Benacerraf-Field problem its bite is not the ontology but the objectivity of mathematics. Indeed, there is a pressing challenge for any domain such that we think (i) we are reliable; (ii) the domain is objective; and (iii) our beliefs about the domain are not generated by some kind of perception. 4 There may be several different philosophically interesting notions of objectivity. See Schechter (2010) for defense of the claim that this is the relevant characterization of objectivity in the context of reliability challenges.

Could Evolution Explain Our Reliability about Logic?

215

and logical falsehoods. They are therefore both meaningful and truth-apt.5 Second, the truth of logical truths and the falsity of logical falsehoods do not depend on us. In particular, they do not depend on our thoughts, language, or social practices. Third, even if there is, in some sense, a plenitude of incompatible logical practices, only one—or a small number of them—is correct.6 Given the objectivity of logic, it is not at all clear how to explain our reliability. The intuitive difficulty is this: We have some understanding of how we could have veridical beliefs about non-objective facts. For instance, it is not at all mysterious how Sir Arthur Conan Doyle could be reliable about what is true in the Sherlock Holmes fiction. We also have some understanding of how we could arrive at veridical beliefs about objective facts via perception. But this understanding does not extend to the case of logic. We do not understand how we could be reliable about objective facts that were not learned via some kind of perception. This difficulty is potentially very significant. If we were to come to believe that there is no satisfying explanation of our reliability compatible with the objectivity of logic, this would put pressure on our belief that logic is objective, on our belief that we are reliable about logic, or on our general background views about the world. But giving up on the objectivity of logic—or on our reliability or our general background views—would be devastating to our ordinary ways of thinking. Thus, there is an important explanatory challenge to answer. In this paper, I examine one candidate answer to this challenge. In particular, I pursue the attractive thought that our reliability about logic is to be explained by appeal to evolution by natural selection.7 The account is based on two main ideas. The first is that our reliability about logic is to be explained by appeal to a more basic competence: We are reliable about logic because we are reliable in our deductive reasoning.8 The second is that being a reliable deductive reasoner conferred a heritable survival or reproductive advantage upon our ancestors. We inherited this trait. This explains how we are reliable in our deductive reasoning. In the past, when I have presented this view, it has prompted two radically different responses. Some have claimed that it is obviously correct: We are the 5 To raise a reliability challenge, it suffices that the relevant truth predicate be minimal or deflationary. It is not necessary to appeal to a more robust conception of truth. 6 Notice that the claim that logic is objective does not entail that it is an objective matter whether a truth counts as a logical truth. That is, logicality need not be objective. 7 The idea of explaining the reliability of our reasoning by appeal to evolution has a long history. See, for instance, Darwin (1871), book I, chapter V. Versions of this view can also be found in Spencer, Mach, Avenarius, Boltzmann, Simmel, James, and Dewey, among many others. 8 Some philosophers and psychologists have claimed that we do not reason deductively but rather employ only inductive or abductive patterns of reasoning. I find this view implausible, but cannot argue against it here. If this alternative view were instead adopted, many of the same issues would arise.

216

Joshua Schechter

products of evolution, and since we are reliable about logic, there must be an evolutionary explanation of this fact. Others have claimed that the view is obviously hopeless. As will become clear, both reactions are premature. There are several in principle difficulties facing evolutionary accounts, but there is reason to believe that they can be addressed. The difficulties facing evolutionary accounts are fierce: (i) Even if evolutionary accounts can explain why we employ useful cognitive mechanisms, they seem unable to explain why we employ highly reliable ones; (ii) Even if evolutionary accounts can explain why we reason reliably about a narrow range of simple propositions—those concerning danger, food, reproduction, shelter, and so on—they seem unable to explain our reliability concerning propositions of arbitrary complexity and with arbitrary subject matters; (iii) Even if evolutionary accounts can explain how we came to employ deductive rules of inference that are actually truth-preserving, they seem unable to explain how we came to employ rules that are necessarily truth-preserving; Finally, (iv) since engaging in deductive reasoning does not yield novel information about the world, it is difficult to see how there could be any selective advantage in doing so.9 The purpose of this paper is to sketch an evolutionary explanation of our reliability about logic and to demonstrate how these general difficulties may be addressed. I do not claim that every detail of my evolutionary explanation is correct. I do not even claim that some evolutionary explanation must be correct. Rather, my main claim is that there is no in principle reason to think that evolutionary accounts are incapable of explaining why we are reliable about logic. In particular, the account I sketch provides one plausible answer to the reliability challenge for logic. It demonstrates that there are plausible answers to be had. This defangs the reliability challenge for logic. It defuses the tension between the claim that logic is objective, the claim that we are reliable about logic, and our general background views about the world. This paper will proceed as follows. In the next section, I further develop the reliability challenge for logic. I present what I take to be the crux of the challenge—the challenge of explaining how it is that we have a reliable cognitive mechanism for deductive reasoning. In section three, I discuss the nature of evolutionary explanations. Section four is devoted to sketching an evolutionary explanation of our reliability about logic. I first present an explanation of why it is that we possess logical concepts. I then present an 9 Other alleged difficulties include the following: There is no way to make sense of the distinction between selection for accepting the correct logic whichever that is, and selection for accepting a particular logic that happens to be correct. See Field (1998), page 19. Evolutionary explanations cannot explain why having objective thoughts is within the range of biological options. See Nagel (1986), pages 78–9. The probability that any of our cognitive mechanisms is reliable is low, given naturalism and given that we are the product of evolution, due to the difficulty of providing a naturalistic account of the role of content in causing behavior. See Plantinga (1993), chapter 12.

Could Evolution Explain Our Reliability about Logic?

217

explanation of why it is that we employ reliable deductive rules given that we possess these concepts. In so doing, I address the fourth difficulty listed above. Finally, in section five, I answer the remaining three difficulties for evolutionary accounts of our reliability about logic.

2.

P R E L I M I N A R I E S 10

Before developing the reliability challenge for logic, I should first explain what I mean by “logic.” Logic, as I use the term here, does not concern artificial formal languages. Rather, it concerns propositions that can be expressed in non-technical natural language and believed by ordinary thinkers. Certain propositions are logical truths—for instance, the proposition that every walrus is a walrus. Other propositions are logical falsehoods—for instance, the proposition that some walrus is not a walrus. There are many interesting questions that arise concerning the nature of logic. For my purposes here, I need not presuppose any particular account. In my discussion, however, I will assume that (i) propositions and not sentences are the primary bearers of logical truth and logical falsity; (ii) propositions are fine-grained in the sense that there can be distinct logically equivalent propositions; (iii) whether a proposition is logically true or logically false depends on its logical form; and (iv) logical truths are necessarily true and logical falsehoods are necessarily false, on any reasonable (alethic) kind of necessity. I will also assume that classical logic is at least approximately correct. As a first pass, the reliability challenge for logic is the challenge of explaining why it is that the logical propositions that we believe (upon reflection and discussion) are by-and-large true and the logical propositions that we disbelieve (upon reflection and discussion) are by-and-large false. Yet, this is not the best way to understand the crux of the challenge. This challenge has a straightforward answer. Our reliability about logic can be explained by appeal to a more basic competence: we are reliable about logic because we are reliable in our deductive reasoning.11 Consider some moderately complex logical truth, for instance if both A and if A then B then B, substituting particular sentences for A and B. We believe this proposition, at least upon reflection and discussion. How did we come to believe it? There are many possibilities. For instance, we may have learned it from a logic teacher. Or we may have observed that B and inferred the conditional from it. However, the canonical way one comes to believe this truth is via a chain of reasoning, perhaps one like the following:

10

This section presents some of the main conclusions of Schechter (2010). This claim is broadly analogous to the suggestion that our ability to reason correctly about metaphysical necessity is a byproduct of our more basic ability to reason correctly about counterfactuals. See Hill (2006) and Williamson (2007). 11

218

Joshua Schechter

Suppose both A and if A then B. So A. So if A then B. So B. So if both A and if A then B, then B. More generally, we come to believe logical truths and disbelieve logical falsehoods on the basis of such deductive reasoning. Our logical beliefs are the outputs of deductive reasoning in cases where there are no initial premises. It is plausible that whenever thinkers reason in this way, their reasoning depends on the employment of rules of inference. In particular, it is plausible that deductive reasoning relies on the employment of rules that resemble the rules that appear in natural deduction formulations of logic. On this view, deductive rules include such rules as: From both p and q, infer p; From both p and q, infer q; From p and if p then q, infer q; From q under the supposition p, infer if p then q. There are several reasons to think that the rules that we employ in deductive reasoning are more complex than the simple rules listed above.12 Nevertheless, it is plausible that the rules that we employ are closely related to the standard natural deduction rules, and in what follows, I’ll assume that they are.13 Consider again if both A and if A then B, then B. We believe this proposition because we went through a chain of reasoning like the one displayed above. We ended up with a true belief because the transitions involved in our reasoning were truth-preserving. The transitions were truth-preserving because the deductive rules that governed them are reliable. In general, we are reliable in our logical beliefs because we are reliable in our deductive reasoning. We employ reliable deductive rules.14 Of course, providing this explanation does not fully answer the reliability challenge for logic. It raises a new explanatory demand. Explanation is now needed of how it is that we are reliable in our deductive reasoning. The challenge thus becomes that of explaining the reliability of our cognitive mechanism for deductive reasoning. 12

See Harman (1988; 1995). There is disagreement in the psychological literature over the correct view of deductive reasoning. The view assumed here is closest to that of Rips (1994) and Braine and O’Brien (1998). My discussion would have to be changed if an alternative view were adopted. But in most cases, the changes would be minimal. 14 There are well-known experimental results showing that humans are prone to errors in deductive reasoning. Perhaps the most striking results concerns variants of the Wason selection test. So the claim that we are reliable in our deductive reasoning should not be overstated. But these errors seem largely to be performance errors. The rules built into our deductive competence are reliable. 13

Could Evolution Explain Our Reliability about Logic?

219

What is this challenge? There are two important explanatory questions concerning the reliability of our deductive mechanism. They may be stated as follows: The Operational Question: How does our cognitive mechanism for deductive reasoning work such that it is reliable? The Etiological Question: How is it that we have a cognitive mechanism for deductive reasoning that is reliable?15 To illustrate the distinction between these two questions, it may be helpful to compare a different cognitive mechanism—say, vision. Our visual mechanism is reliable in the sense that it by-and-large produces true beliefs about our environment. This is a striking fact, in need of explanation. Indeed, this fact raises two explanatory questions. First, how does our visual mechanism work such that it reliably produces true beliefs about our environment? Second, how is it that we have a reliable visual mechanism? The answers to these two questions are very different. To answer the reliability challenge for logic, satisfying answers to both the operational and the etiological questions are needed. The operational question has a straightforward answer: Our deductive mechanism works via the employment of rules of inference. This mechanism is reliable because the deductive rules of inference that we employ are necessarily truth-preserving. They are guaranteed to yield true beliefs from true beliefs.16 That’s all that needs to be said. We need not claim that our deductive mechanism somehow tracks the logical facts. The etiological question, however, cannot be answered so easily. To answer this question, an explanation is needed of how it is that we have a reliable cognitive mechanism for deductive reasoning. In particular, an explanation is needed of how it is that we employ reliable deductive rules. There does not seem to be a quick explanation of this fact. This is the crux of the reliability challenge for logic. It is important to recognize that this explanatory demand is not generated by some overly powerful epistemological principle that quickly leads to radical skepticism. For instance, we need not accept the claim that thinkers must possess an explanation of how it is that they have a reliable cognitive method

15 The etiological question can be interpreted in two ways—as concerning ontogeny or as concerning phylogeny. To fully answer the reliability challenge, both questions require answers. But the ontogenetic question is philosophically not very pressing. The development of a reliable deductive mechanism is presumably genetically encoded. The philosophically pressing question concerns phylogeny. 16 This claim must be generalized to handle deductive rules that involve propositional attitudes other than belief. Let’s say that believing a proposition is correct if the proposition is true, disbelieving a proposition is correct if the proposition is false, believing a proposition under a supposition is correct if the proposition is true if the supposition is, and so on. Our deductive rules of inference are reliable in the sense that they necessarily preserve correctness. I leave this generalization implicit in what follows.

220

Joshua Schechter

in order to be justified in employing the method. Rather, the motivating line of thought is simply this: It is a striking fact that we have a reliable cognitive mechanism for deductive reasoning. This is a fact that “cries out” for explanation. Ceteris paribus, it is a cost of a theory if it treats some striking phenomenon as merely accidental or otherwise inexplicable. Thus, it would be very unsatisfying to be forced to claim that it was a brute fact—an amazing cosmic accident—that we employ reliable deductive rules. If that were the only available account, there would be a tension in our overall view of the world. There would be pressure to somehow modify our view. This is why a substantive answer to the etiological question is sorely needed.

3.

E V O L U T I O N A R Y E X P L A N AT I O N S

Perhaps the most natural approach to answering the reliability challenge for logic—that is, for answering the etiological question for deductive reasoning—is to appeal to evolution by natural selection. Evolution by natural selection works as follows: There is initially a population of organisms. Each organism has a genetic endowment, its genotype. Each organism also has phenotypic traits that depend in part on its genotype, and so are heritable. There is variation in the genotypes and phenotypic traits among the organisms in the population. Organisms with certain phenotypic traits tend to be better at survival and reproduction than the rest, given their background environment. Over time, these fitter organisms survive longer and reproduce more frequently than less fit organisms. This yields a change in the gene frequency of the population and a corresponding change in the frequency of phenotypic traits.17 According to the most straightforward kind of evolutionary explanation of why a phenotypic trait came to predominate in a population, possessing the trait enhanced the fitness of the ancestors of the population (in their background environment). In the case of interest, the relevant trait is that of employing reliable deductive rules. On the most straightforward evolutionary explanation of why our population came to employ reliable deductive rules, then, our ancestors were selected for employing reliable deductive rules—a heritable trait—and this explains why it is that we, their descendants, employ reliable deductive rules. An evolutionary account of our reliability about deduction is attractive because it is plausible that employing reliable deductive rules can confer survival and reproductive advantages on an organism. For instance, if an organism believes that a predator is either in this grove of trees or that grove, and then comes to learn that the predator is not in that grove, it is 17 There are biological mechanisms of evolution other than natural selection, such as random drift. I focus on natural selection because such non-selective mechanisms cannot play the central role in a satisfying explanation of our reliability. (Lamarkian inheritance, if it were to exist, could also play such a role.)

Could Evolution Explain Our Reliability about Logic?

221

advantageous for the organism to come to believe that the predator is in this grove. An evolutionary account is also attractive because it fits well with the general scientific picture of our place in the world. It may seem strange to discuss evolutionary explanations of traits such as employing reliable deductive rules. Whether a rule is reliable is not, in some sense, a purely biological fact. It might seem more promising to consider the trait of employing some particular collection of rules. But this line of thought is mistaken. Employing reliable deductive rules is a phenotypic trait like any other. There is no reason to suppose that evolutionary explanations are unable to explain the presence of such traits. Moreover, if we only provided an explanation of how we came to employ some particular collection of deductive rules, we would not answer the reliability challenge for logic. Even though we would have explained why we employ the particular rules, and even though it is a necessary truth that those rules are reliable, we would not have explained why we employ reliable rules. Explanation is not closed under necessary entailment.18 The explanation of our employment of the rules would presumably have nothing to do with their reliability. The reliability of our deductive rules would still seem accidental in a worrisome way. There are several ways in which this simple evolutionary account of our reliability could be modified. For instance, it could be claimed that our ancestors were directly selected, not for employing reliable deductive rules, but for possessing linguistic abilities that required or led to the possession of reliable deductive rules.19 It could be claimed that our ancestors were selected for employing reliable non-deductive learning mechanisms that can be used to acquire reliable deductive rules. It could be claimed that our ancestors were selected to easily adopt reliable deductive rules when taught them by confederates.20 Or it could be claimed that cultural evolution rather than biological evolution was primary. It is worth emphasizing that it is not part of my task here to reject any of these suggestions. My goal is only to sketch one plausible explanation of the reliability of our deductive rules and show how seemingly powerful difficulties for it may be addressed. The main point of this paper is to provide something like a demonstration of possibility (or of plausibility) and not of actuality. Before filling in more of the details of the evolutionary account, it is important that it first be made clear what natural selection can—and cannot— explain. To this end, it is helpful to consider a simple toy example. Suppose that we have very many different sets of fair dice. Let us assume that each set contains five dice and that the sets differ in color. So there are five bright red dice, five faded blue dice, five forest green dice, and so on. Suppose

18

See Schechter (2010), page 447, for an example. See Carruthers (2002a) for arguments that the use of language is central to cognition. 20 This suggestion is related to the Baldwin effect—the idea that there might be selection for organisms that can more easily learn certain abilities and then, only later, for organisms that have those abilities innately. 19

222

Joshua Schechter

that we roll each of the dice simultaneously and select the sets of dice that come up all sixes. That is, we keep the sets of dice that come up all sixes and dispose of the rest. If asked why the current population contains sets of dice that came up all sixes, we can provide an explanation involving selection: There were very many sets of dice and the ones that came up all sixes were selected. In contrast, there is no selective explanation of why some particular set of dice—say the faded blue dice—came up all sixes. That was purely a matter of chance. This toy example involves artificial selection, not natural selection. Yet, the point generalizes. Natural selection can help to explain why organisms with a certain phenotypic trait came to predominate in a population. It cannot explain why a particular organism has the trait in question.21 Instead, the explanation of that fact depends on other processes. Novel phenotypic traits arise in a population on the basis of genetic mutation and recombination. These mutations are due to stray cosmic rays, errors in replication, and other chance events. These traits are then passed on from organisms to their descendants.22 The explanation of how a particular organism came to have a phenotypic trait thus does not involve selection but heredity and extremely chancy events. It is important to be careful here. Natural selection can help to explain why a novel phenotypic trait arose in the population. Previous bouts of selection could have made it very likely that the trait would emerge.23 But what selection cannot do is explain why a particular individual has the trait in question. This feature of selective explanations might provoke the following worry: To fully respond to the reliability challenge, one must provide a satisfying explanation both of why the population primarily includes thinkers who employ reliable deductive rules and of why particular thinkers employ reliable deductive rules. Whatever the prospects are for an evolutionary account to meet the first demand, it is unable to meet the second. And it would be highly unsatisfying to be forced to think that it was merely an accident that each individual thinker employs reliable deductive rules. The claim that it was merely an accident would produce a tension in our overall view of the world. While this worry is arresting, it is ultimately misguided. The primary fact to be explained is a population-level fact—our population primarily includes thinkers that employ reliable deductive rules. To illustrate this point, suppose that there were a vast population of heterogeneous thinkers, each with a different set of inferential rules. Suppose that a few of these thinkers had reliable deductive rules, and that the number of such thinkers was roughly

21

See Sober (1984), section 5.2. I am here ignoring the complexities that arise from the impact of the environment on phenotypic traits and from the existence of sexual reproduction. These complexities do not affect the points to follow. 23 See Neander (1995). Neander also claims that selection can explain why a particular organism has a phenotypic trait. I don’t see why this further claim is true. 22

Could Evolution Explain Our Reliability about Logic?

223

what one would expect if the rules were somehow distributed randomly. If the few reliable individuals had no other striking properties in common, we would not think that their reliability was particularly in need of explanation. Nor would we find it troubling if it turned out to be merely an accident that they were reliable. (Compare: Given that we have many sets of dice, it is not striking that some particular set came up all sixes. What is striking is that every set of dice in our possession came up all sixes.) This suggests that the reliability of a particular individual is not in general a striking fact. What is striking is the reliability of the population at large. There is a complication, however. Consider the first-personal claim that I employ reliable deductive rules. It is intuitive that this is a striking fact. At the very least, this fact seems more in need of explanation than the claim that some arbitrary particular individual is reliable. When I reflect on my own reliability, it is hard to be satisfied with the thought that it was merely a lucky fluke that I am reliable. Since an evolutionary account cannot explain this firstpersonal fact, we again have a worry about the prospects of an evolutionary account. I’m not entirely sure what to make of this concern. There is reason to be cautious here. It is not entirely clear that we should endorse the intuition that it is a striking fact that I am reliable. It is not obvious that my winning the lottery would be any more striking than the fact that some particular person won the lottery. And even if the first-personal fact is a striking fact, it might not be a big cost of a view to claim that this fact came about merely by accident. But it should be granted that there are murky waters here.24 Putting the first-personal case aside, assuming that an evolutionary account of the reliability of our population can be made to work, the fact that it is accidental that particular thinkers employ reliable rules is not worrisome. But can an evolutionary account of the reliability of our population be made to work?

4.

A N E V O L U T I O N A R Y E X P L A N AT I O N O F O U R DEDUCTIVE RELIABILITY

Deductive rules are rules of inference that govern reasoning with logical concepts. One natural strategy for explaining why it is that we employ reliable deductive rules is to divide the task into two parts. The first is to explain why it is that we possess logical concepts. The second is to explain, given that we possess logical concepts, why it is that we employ reliable deductive rules. In this section, I sketch an evolutionary account with this bipartite structure. According to this account, possessing logical concepts and employing reliable 24 In conversation, David Christensen has suggested that although selection cannot explain the first-personal fact that I have a reliable deductive mechanism, it nevertheless raises its probability. While this seems true, it does not assuage the worry. The general principle of theory choice generating the reliability challenge concerns explanation and not probability.

224

Joshua Schechter

deductive rules both conferred survival advantages on our ancestors. This explains why it is that we have these phenotypic traits. 4.1. The Advantages of Possessing Logical Concepts On the proposal under consideration, possessing logical concepts conferred a survival or reproductive advantage on our ancestors. But it is prima facie puzzling what this advantage could be. The main difficulty is as follows: It is tempting to think that the inputs from the world (via perception) to our reasoning faculties are logically simple. It is also tempting to think that our behavior and behavioral dispositions depend only on the logically simple products of our reasoning. Given these two assumptions, it is plausible that being able to hold logically complex beliefs can confer a selective advantage only if such beliefs are important as intermediate steps in our reasoning. But this is apparently ruled out by the fact that logical concepts are conservative in the following sense:25 Adding logical concepts (along with the associated deductive rules) to any conceptual practice involving only logically simple propositions does not license any inferences from logically simple premises to logically simple conclusions that were not already licensed before the addition.26 In response, it might be suggested that possessing conservative concepts can be advantageous in another way—namely by enabling thinkers to reason in a quicker or more efficient manner than they would otherwise be able to.27 But this suggestion does not help for the case of logical concepts. Adding logical concepts (with the associated deductive rules) to a conceptual practice does not speed up derivations of logically simple conclusions from logically simple premises. In answer to this difficulty, I suggest that the two tempting thoughts ought to be rejected. Logically complex beliefs can be inputs to reasoning, delivered by perception or other non-inferential cognitive mechanisms. This is perhaps clearest for conjunctive propositions. Our visual system binds together different features of an object and delivers beliefs in conjunctions, such as the proposition that a given object is both round and red. Beliefs delivered by perception may be logically complex in other ways, too. Consider, for instance, negative 25 Conservativeness (or a related property) is often taken to be a necessary condition on logical constanthood. This proposal stems from the Gentzen-Prawitz tradition of requiring the introduction and elimination rules of a logical concept to appropriately match each other, and from the suggestion in Belnap (1961) that a conceptual role bestows a genuine meaning on a logical constant only if it satisfies a conservativeness requirement. Hacking (1979) demarcates the logical, in part, by appealing to conservativeness. See Dummett (1991) and Tennant (1987) for relevant discussion. 26 In the absence of a detailed list of the deductive rules we employ, a rigorous proof of conservativeness is not possible. But this claim, or a related one, is highly plausible. 27 Compare: Mathematical theories may be useful when added to nominalist physical theories by shortening the derivation of purely nominalist physical consequences from nominalist physical claims. See Field (1980).

Could Evolution Explain Our Reliability about Logic?

225

existential propositions. It seems that just by looking—without any inferential reasoning—I can come to believe that there is no dog in my office.28 Similarly, perhaps thinkers can directly come to believe disjunctive propositions on the basis of perception, such as the proposition that a certain object is either blue or green.29 Perhaps, too, thinkers can directly come to believe general propositions, such as the proposition that every book on the top shelf is blue. Moreover, if beliefs acquired through testimony are not based on inference, arbitrary logically complex beliefs may be inputs to reasoning. On the basis of logically complex inputs to reasoning, thinkers may reason to logically simple conclusions and behave accordingly. Logically complex beliefs can also be important in guiding behavior, and not merely by virtue of their logically simple consequences. For a mundane example, suppose that I am searching for a glass of water. Holding the disjunctive belief that a glass contains either water or vodka should have a different impact on my behavior than a belief in either of the disjuncts. If I were to believe that the glass contains water, I should pick up the glass and abandon my search. If I were to believe that it contains vodka, I should ignore it and continue my search. But if I were to believe the disjunction, I should examine the liquid more closely, perhaps take a sip, and perform other appropriate tests.30 Similarly, holding the negative belief that there is no glass of water nearby should motivate me to search elsewhere. Holding the conditional belief that if there is a glass of water nearby then it is in my office should motivate me to look in that location. In general, possessing logical concepts enables thinkers to represent important information about the world, information of use in guiding behavior. There is a natural picture of the representational abilities provided by the logical concepts. Consider the space of all (metaphysically or epistemically) possible worlds. We can think of each logically simple proposition as picking out a region of this space—namely the worlds at which the proposition is true. The logical concepts enable us to pick out additional regions of this space in our thought. Conjunction enables us to take the intersection of regions. Disjunction enables us to take the union of regions. Negation enables us to take the complement of a region. And so on. Being able to pick out these 28 It might be claimed that my belief that there is no dog in my office is not solely based on perception, but also on a default presumption that what I don’t see isn’t there. This view faces several difficulties. But even if some version of it is correct, the default presumption is itself a logically complex input to reasoning. 29 See Dummett (1991), page 267, for the claim that perceptions can be disjunctive. Dummett’s example is that Hardy may not have been able to tell whether Nelson said, “Kismet, Hardy” or “Kiss me, Hardy”, but may have had a perception with an “irreducibly disjunctive form.” 30 For an analogous example, see Skyrms (1999). In discussing a hypothetical group of logically sophisticated vervet monkeys, Skyrms claims that the optimal evasive action given knowledge that there is either a snake or a leopard nearby may be different both from the optimal action given the knowledge that there is a leopard nearby and from the optimal action given the knowledge that there is a snake nearby.

226

Joshua Schechter

additional regions of modal space helps us to represent information that we acquire from the world and make use of in guiding our behavior. As we will see below, this picture is much too crude. But it provides a useful initial account of the role of the logical concepts in thought. It is interesting to note that it is not necessary for a thinker to possess logical concepts in order to have the representational abilities they provide. For instance, instead of believing of something that it is either blue or green, a thinker could believe that it is blue-or-green, where blue-or-green is a concept that applies to blue things and to green things. Instead of believing that it is not the case there is a chair here, a thinker could believe that it is unchaired here, where un-chaired is a concept that applies to locations lacking chairs. Similar claims hold for more complex logical constructions. Instead of possessing a few general purpose logical concepts, thinkers could possess a bevy of special purpose concepts. However, this does not provide an objection to the claim that possessing logical concepts can be evolutionarily advantageous. It would be computationally very costly for a thinker to dispense with general-purpose logical concepts in favor of a large number of special purpose concepts. Moreover, to have the representational abilities that logical concepts provide, such a thinker would have to add a potentially unbounded number of additional concepts whenever a new concept was acquired. There is an additional proposal about the advantage of possessing logical concepts that merits discussion. This is the idea (inspired by the work of Brandom) that the role of logical concepts is to enable thinkers to “make explicit their implicit inferential commitments.”31 The idea here is that thinkers are disposed or committed to infer in certain ways. Logical concepts provide thinkers with the conceptual resources needed for subjecting these commitments to rational scrutiny. Consider, for instance, the conditional. In believing if A then B, a thinker makes explicit her commitment to inferring B from A.32 This belief can then be assessed by considering reasons for and against it. If the latter are more persuasive than the former, the belief—and the inference—may be rejected. Adapting this view to an evolutionary context, the idea is that possessing logical concepts such as the conditional conferred a survival advantage because they enabled our ancestors to hold up their inferential commitments to rational scrutiny, and thus enabled them to improve their methods of reasoning about the world.33 31 See Brandom (2000, introduction and chapter 1). Brandom’s view concerns logical vocabulary rather than logical concepts. 32 This is reminiscent of the view in Ryle (1950) that conditionals are inference tickets. Brandom sometimes writes that the job of conditionals is to enable speakers to say that certain inferences are acceptable. Presumably, this should be construed in a loose sense. “If A then B” does not say anything about an inference (provided that neither A nor B do so). Rather, it says something about the world. The best way to understand Brandom’s view, I take it, is that in stating “if A then B”, a speaker asserts something about A and B, and in so doing also expresses a commitment to the correctness of the inference from A to B. 33 Consider the following conditional: If Sally is deceiving me, I’ll never believe that she is (because she is so clever). Presumably, a thinker can sensibly believe this conditional with-

Could Evolution Explain Our Reliability about Logic?

227

This suggestion is appealing. It is plausible that an important role of logical concepts is to allow thinkers to better assess their inferential practices and commitments. However, the proposal need not be seen as a competitor to the idea that the function of the logical concepts is to enable thinkers to represent facts about the world that they would not otherwise easily be able to represent. Possessing logical concepts may have conferred a survival advantage both because they improved our ancestors’ representational abilities and because they helped our ancestors better assess their own patterns of reasoning. There is no need to choose between these two suggestions.34

5.

T H E A D VA N TA G E S O F E M P L O Y I N G R E L I A B L E DEDUCTIVE RULES

We see, then, that possessing logical concepts can confer an evolutionary advantage. What about the employment of reliable deductive rules? This question raises an old problem about deduction. The problem is to explain the point of deductive inference.35 Deductive inference does not augment our knowledge of the world. In an intuitive sense, the conclusion of a deductively valid inference does not contain any information not already contained in the premises. This claim has a formal analogue: The set of possible worlds at which the conclusion is true is a superset of the possible worlds at which all of the premises are true. What, then, could the purpose of deductive inference possibly be? How could there be an evolutionary advantage in employing reliable deductive rules? One strategy for answering this question is to appeal to some version of Conceptual Role Semantics. On this view, for a thinker to possess a concept, she must employ certain associated rules of inference or belief-forming methods. Assuming that Conceptual Role Semantics holds true for logical

out being committed to reason from the antecedent to the consequent. So there seems to be a counterexample to this view of conditionals. (This kind of example is originally due to Thomason. See van Fraassen (1980), page 503, for the attribution.) There are several things that can be said in reply. One could claim that a thinker who endorses the conditional does express a commitment to make the corresponding inference but loses this commitment upon coming to believe the antecedent. Alternatively, one could modify the view by claiming that in endorsing a conditional, a thinker expresses a commitment to viewing the corresponding inference as a good inference (but does not necessarily express a commitment to drawing the inference). Finally, it could be claimed that the semantics of conditionals and the selective advantage of possessing the conditional come apart. Although endorsing a conditional does not always go along with a commitment to reasoning from the antecedent to the consequent, it typically does. This is enough to explain why possessing the conditional conferred a survival advantage. Thanks to an anonymous referee for pressing me on this issue. 34 Here are two additional proposals: First, possessing logical concepts is needed to engage in certain kinds of non-deductive reasoning, such as inference to the best explanation. Second, possessing logical concepts such as the conditional enables thinkers to appropriately hedge their beliefs and avoid unnecessary error. The latter suggestion is related to the discussion of the conditional in Boghossian (2003). 35 This problem goes back at least as far as Mill (1843), II.3.1.

228

Joshua Schechter

concepts, it is plausible that the deductive rules we employ are conceptconstituting of the logical concepts we possess. So any advantages there are for possessing logical concepts are also advantages for employing the associated deductive rules. Although this strategy has some appeal, there are reasons for doubt. First, Conceptual Role Semantics is subject to many difficulties, and may well be false.36 Second, the explanation is intuitively the “wrong way around.” On many versions of Conceptual Role Semantics, possessing a concept consists in employing certain rules. So one would expect the evolutionary advantages of possessing a concept to derive from the advantages of employing its associated rules, and not vice-versa. Finally, and more importantly, the strategy is at best able to explain how it is we employ particular deductive rules— those constitutive of the particular logical concepts we possess. It is unable to explain how it is we employ reliable deductive rules. The reliability of the rules plays no obvious role in the explanation.37 The explanation makes it seem to be merely an accident that we came to employ reliable deductive rules.38 There is a better strategy for explaining the evolutionary advantage of employing reliable deductive rules. A preliminary point to make is that information from the world does not come to us all at once in a single package. We acquire it over time and from disparate sources. Deductive reasoning can help us combine new information with old information. It can help us combine information from different sources. It can also help us to reassess old information or to apply old information to new situations. Yet, simply saying this does not suffice to explain the evolutionary advantage of employing reliable deductive rules. When a thinker employs a reliable deductive rule to combine bodies of information or to draw out a consequence of some body of information, the thinker is not acquiring any novel information. All of the relevant information is already present, at least on a coarsegrained understanding of “information.” So what could the evolutionary advantage of employing a deductive rule be? The answer to this question must be this: Employing reliable deductive rules helps thinkers to convert information into a more usable form. To illustrate, suppose again that I am searching for a glass of water. Even if I were to come to believe both that there is a glass of water located either here or over there and that the glass of water is not located here, I would not immediately be motivated to look for it over there. In order for me to be so motivated, I would first have to draw the relevant inference. I would have to come to 36 For some of the difficulties, see Block (2000), Fodor and Lepore (1991), and Williamson (2003). 37 To be fair, on some versions of Conceptual Role Semantics, constitutive rules are required to be truth-preserving. For instance, see Peacocke (1992). However, there are compelling counterexamples to this claim. See Boghossian (2003) and Schechter and Enoch (2006) for discussion. 38 A more plausible suggestion is that the selective advantage of possessing logical concepts required their constitutive rules to be reliable. This suggestion is compatible with the one to follow.

Could Evolution Explain Our Reliability about Logic?

229

explicitly believe that the glass of water is over there. This is so despite the fact that the information that the water is over there is implicit in my original beliefs. (This point is still more striking for cases where the relevant sequence of deductive inferences is very long.) In general, rational behavior depends not only on the information a thinker possesses, but also the manner in which it is represented. This shows that the natural picture of the role of the logical concepts in thought presented above—in terms of picking out regions of modal space—is too crude. Logically equivalent propositions can be importantly different in their relevance to behavior. A deductive consequence of some premise may pick out a larger region of modal space than the premise, but may also be directly relevant to an urgent matter in a way that the premise was not. Part of the role of the logical concepts, then, is to represent information in specific ways. Deductive rules of inference are important because they enable us to manipulate the ways that information is represented. The proposal, then, is this: Employing reliable deductive rules enables thinkers to better make use of the information they possess in guiding their behavior. This is why employing reliable deductive rules conferred an evolutionary advantage on our ancestors.

6.

OBJECTIONS AND REPLIES

We have, then, a (very rough) sketch of an evolutionary explanation of our reliability about logic. We possess logical concepts and employ reliable deductive rules because this conferred an evolutionary advantage on our ancestors. Possessing logical concepts enabled our ancestors to represent important information about the world as well as to assess their own patterns of reasoning. Employing reliable deductive rules enabled our ancestors to better make use of the information in their possession. Our reliability about logic is a sideeffect of our employment of reliable deductive rules. It may be worthwhile to here repeat an earlier point: My discussion is only intended to provide a sketch of an account. To fill out the explanation would require a considerable amount of empirical work. For instance, one might want to look into the following issues: How is it that our cognitive mechanisms are genetically encoded? How is it that natural selection works on the mind? What are the intermediate forms between animals not possessing cognitive mechanisms for deductive reasoning and animals that do possess such mechanisms? Why are humans special in having such powerful cognitive capacities? What is the specific role of each individual logical constant and each individual deductive rule in our thought? What are the costs of having a cognitive mechanism for deductive reasoning? And so on. It is not one of the tasks of this paper to answer these interesting and important empirical questions. The central goal of this paper, rather, is to address the reliability challenge for logic and to thereby ward off the pressure on our

230

Joshua Schechter

belief that logic is objective, on our belief that we are reliable, and on our general background views about the world. This is a task in the epistemology of logic. Providing a plausible sketch of an explanation, even at a high level of generality, is sufficient for achieving this goal. There are, however, several pressing objections that can be raised against evolutionary accounts of our reliability about logic. These objections purport to show that such accounts cannot even in principle explain our reliability about logic. In this section, I discuss the most serious of these objections. I show how they may best be addressed. The objections fall into two categories. The first concerns whether evolutionary accounts can explain the full extent of our reliability. The second concerns whether evolutionary accounts can explain our reliability at all. In responding to these objections, I further develop the evolutionary account. 6.1. Generality, Complexity, and Necessity One difficulty for evolutionary explanations of our reliability is that they seem unable to explain the full extent of our reliability. It is plausible that our ancestors were selected for correctly reasoning only about certain domains. Presumably, what natural selection “cares about” is that our ancestors formed true beliefs about danger, food, mating, shelter, and other topics closely tied to survival and reproduction. Perhaps when sexual selection is taken into account, the list can be somewhat expanded. But even if it can, it is clear that our ancestors were not selected for correctly reasoning about algebraic geometry, the Problem of the Many, the geology of Mars, or how best to make a cheese soufflé. Evolutionary accounts thus seem unable to explain how it is that we employ deductive rules that are reliable for arbitrary subject matters.39 A second difficulty concerns the complexity of propositions. It is plausible that our ancestors were selected for reasoning correctly only with relatively simple propositions. Presumably, there was little survival advantage in reasoning correctly with highly complex propositions. Evolutionary accounts thus seem unable to explain how it is that we employ deductive rules that are reliable when applied to propositions of arbitrary complexity. A third difficulty comes from the reaches of modality. The reliability of our deductive mechanism is not merely a matter of its yielding true beliefs from true beliefs. Our deductive mechanism is reliable in a modally robust sense— the rules we employ necessarily preserve truth. Evolutionary accounts seem 39 A related worry appears in Delbrück (1978), page 353, and is discussed in Sober (1981). Delbrück’s worry concerns the mechanisms used in scientific reasoning. Nagel (1986), page 79, argues against evolutionary explanations of our capacity for objective thought on the grounds that “our capacity to form cosmological and subatomic theories takes us so far from the circumstances in which our ability to think would have had to pass its evolutionary tests . . . .” Field (1989), pages 28–9, argues against evolutionary explanations of our reliability about mathematics on the grounds that only a small portion of mathematics is tested against the world.

Could Evolution Explain Our Reliability about Logic?

231

unable to explain this fact. Success in survival and reproduction depends only on reasoning in ways that are truth-preserving in the actual world, and not on reasoning in ways that would be truth-preserving even in far off possible worlds.40 Indeed, when carefully evaluated, it seems incoherent to claim that our ancestors were selected for employing deductive rules that are necessarily truth-preserving. A necessary condition for there to have been selection for a trait is that the presence of the trait caused the relevant organisms to survive longer or reproduce more frequently. How could it be that a modal property was causally efficacious in this way? In response to the third of these difficulties, one might try to deny that our deductive rules are reliable in a modally robust sense. Perhaps the deductive rules we employ would be truth-preserving if the world were only slightly different, but would not be truth-preserving if the world were radically different.41 However, this response is unattractive. Adopting it would be tantamount to giving up on many of our modal beliefs: If we were to believe that even our deductive rules are not necessarily truth-preserving, we should believe that we are not at all reliable about what is necessarily the case. Moreover, an analogous response cannot be used to answer the first difficulty. It would be extremely difficult to maintain that our deductive rules produce true beliefs when reasoning about predators but not when reasoning about other topics, such as how to get an astronaut to the Moon. Such a view conflicts with the success we have had in many of our endeavors. Such a view is also potentially self-undermining: Any belief that we are unreliable concerning subject matters that were not of immediate evolutionary relevance should undermine our trust in the very reasoning that produced the belief. There is no plausible way to argue that our ancestors were selected for reasoning correctly about arbitrary subject matters, with propositions of arbitrary complexity, and in a way that works in arbitrary possible worlds. The answer to the three difficulties must instead be that these features of our deductive rules are byproducts of the traits for which our ancestors were really selected. Our ancestors were selected for employing deductive rules that are actually truth-preserving for a limited range of propositions. Our ancestors were also selected for employing rules that are computationally efficient, have few storage costs, and the like. A good way to satisfy these constraints is to employ

40 Nozick (2001), page 122, and Stroud (1981) argue that evolutionary accounts make it difficult to see how we could be reliable in our attributions of necessity, since selection only rewards believing truths about the actual world. This objection is different from the one I discuss. I do not here consider the status of attributions of modal properties. 41 A related proposal is put forward by Nozick (1993), page 111, who suggests that the explanation of why logical principles seem self-evident may be that “they are true, even if only contingently, even just ‘true enough’ . . . and that they have held true durably for long enough to leave an imprint upon our evolutionary endowment.” This is reminiscent of the view in Mill (1843), II.6.2, that arithmetic is contingent but appears necessary due to our “early and constant experience”.

232

Joshua Schechter

deductive rules that work generally and necessarily, and this explains why our ancestors came to have such rules. This response might be thought to be somewhat disappointing. What we were after was a satisfying explanation of how we came to have generally and necessarily truth-preserving deductive rules, and the evolutionary explanation on offer claims only that this is a side-effect of something else. But this disappointment should be fleeting. There is nothing inherently problematic about the style of explanation offered. Indeed, evolutionary biologists are familiar with numerous examples in which the most plausible explanation of some trait is that it is a side-effect of what was really selected for.42 What would be unsatisfying is if it were claimed that our employment of reliable deductive rules was a side-effect of a trait that was not directly related to reliability, such as the trait of having large ears. This would make it seem accidental that we came to employ deductive rules that work generally and necessarily. But the explanation under consideration makes no such claim. The trait of employing rules of inference that are truth-preserving in a limited range of cases is closely tied to the trait of employing rules that generally and necessarily preserve truth. The proposal thus provides a satisfying answer to the three difficulties.

6.2. Usefulness and Truth The final difficulty for evolutionary accounts of our reliability is the most general. Namely, even if an evolutionary account can explain why we have useful cognitive mechanisms, it is not clear it can explain why we have cognitive mechanisms that are truth-conducive. A version of this objection has been forcefully presented by Stich.43 Stich directly argues only that natural selection does not guarantee that our cognitive mechanisms generally produce true beliefs (given the appropriate inputs). This conclusion is unobjectionable. But if his arguments can be extended to support the claim that natural selection is unlikely to yield cognitive mechanisms that generally produce true beliefs, this would count against an evolutionary explanation of our reliability. Stich argues for two claims. First, evolution does not always yield the optimal mechanism for a task. That is, it does not always produce a mechanism that provides the greatest contribution to fitness when compared to alternatives. Second, optimal cognitive mechanisms are not always the most

42 Gould and Lewontin (1978) use the word “spandrels” to refer to traits that are not selected for but are side-effects of how an organism develops or is built. 43 See Stich (1990, chapter 3). Stich does not challenge the view that our cognitive mechanisms are the products of evolution, but rather the view that “evolutionary considerations impose interesting limitations on irrationality.” Churchland (1987) argues that there is a significant gap between behaving in a way conducive to survival and having mostly true beliefs.

Could Evolution Explain Our Reliability about Logic?

233

reliable. Features other than truth-conduciveness may be more important to fitness. In support of the first claim, Stich presents several reasons why evolution does not always yield an optimal design: An optimal mechanism may not be biologically possible. It may not be biologically possible for a population to evolve an optimal mechanism given its current characteristics. Random genetic drift may lead to the disappearance of an optimal mechanism from a population or may hamper its spread. The complexities of sexual reproduction may do so, as well. Stich is undoubtedly correct that evolution does not always maximize fitness. However, Stich provides no reason to believe that evolution is unlikely to provide mechanisms that substantially enhance fitness. Moreover, it is difficult to evaluate how well Stich’s considerations apply in the particular case of interest. Not much is known about the evolution of the brain, and more specifically, the evolution of cognitive traits. It is an open question whether any of the factors Stich discusses were of any real importance in the evolution of our reasoning mechanisms. In support of the second claim, Stich argues that there is frequently a trade-off between how truth-conducive a cognitive mechanism is and how economical the mechanism is with respect to the amount of energy, time, and cognitive hardware that it requires. It may be more important to an organism’s survival that a cognitive mechanism be “fast and frugal” than that it be generally accurate. In addition, there may be a trade-off between the number of false positives and the number of false negatives that a cognitive mechanism produces. It may be very important to avoid one of these—for instance, false negatives concerning the existence of lurking predators—at the cost of substantially increasing the other.44 Again, it is undoubtedly correct that the fittest cognitive mechanism is not always the most truth-conducive. Yet, there are general reasons to think that the gap between the two is not as large as Stich’s discussion might lead one to expect. All things being equal, correctly representing the world is conducive to survival: it helps organisms react to threats and achieve important goals.45 It seems likely that the cognitive mechanisms that evolution produces are at least somewhat truth-conducive in a central range of cases. There is also reason to believe that Stich’s considerations carry less force for the particular case of deductive reasoning. The structure of logical entailments is relatively simple. Indeed, cooking up rules that resemble the deductive rules we employ but which are not universally or necessarily truth-preserving would seem to require adding ad hoc conditions or otherwise increasing their complexity. So, for the specific case of deductive reasoning, there seem to be no 44

As Stich notes, these points are originally due to Sober (1981). Stich (1990, chapter 5), argues that it is not at all valuable to have true beliefs. His argument depends on the difficulty of assigning truth conditions to mental states in a privileged way. This is a deep issue in the philosophy of mind and cannot be addressed here. 45

234

Joshua Schechter

additional computational costs in employing fully truth-conducive rules over somewhat-but-not-fully-truth-conducive rules. There may be no cognitive mechanism in the ballpark that is less reliable but more economical. Moreover, deductive reasoning is a general-purpose reasoning mechanism. So there is little pressure for it to minimize false positives at the expense of false negatives, or vice versa. There is no obvious reason for selection to yield any systematic bias for one over the other. There is an additional point to make in response to Stich’s claim. Namely, there is reason to think that there was in fact selection pressure for the employment of deductive rules that are highly truth-conducive in an important range of cases. In general, it can be advantageous for organisms to have cognitive mechanisms that are highly truth-conducive albeit slow or inefficient. In certain circumstances—where energy and time is available, the stakes are high, and the relevant issues are tricky—it can be important not to form beliefs rashly, but rather to have a more considered response. This is a point familiar from everyday life, and it extends to the context of evolutionary fitness. Deductive reasoning has the hallmarks one would expect of a cognitive mechanism that was selected for playing this role. Deductive reasoning is relatively slow, at least compared to instinctive responses or the application of simple heuristics. Its outputs are only indirectly tied to behavior. It is contentneutral, capable of representing and transitioning between a large range of contents. And it is highly reliable. What these features suggest is that our deductive mechanism was designed to get it right when getting it right is very important and time is not of the essence.46 This suggestion fits well with “dual process” theories in psychology, according to which we possess two cognitive systems.47 System 1 is a relatively fast, cognitively undemanding, automatic reasoning system that makes use of special-purpose heuristics. System 2 is a relatively slow, cognitively demanding, general-purpose system for explicit reasoning. Our mechanism for deductive reasoning is a part of System 2. This system is highly reliable, at least in an important range of cases. The claim, then, is that one of the functions of our deductive mechanism requires it to be highly truth-conductive for some range of cases. This raises the question: What is that function? I have proposed that the evolutionary advantage of employing deductive rules is that they enable thinkers to convert information into a form that makes it available for guiding behavior. This goes some way towards answering the question. But the question remains:

46 Papineau (2000) argues for an analogous view about a different cognitive mechanism. He claims that we have a specialized means-end reasoning module that selects the best course of action in the light of the available information when “the stakes are high and time does not press.” 47 See Evans and Over (1996) and Sloman (1996).

Could Evolution Explain Our Reliability about Logic?

235

For what purpose (or purposes) did our ancestors need to represent logically complex information and manipulate it in a highly truth-conducive way? There is one suggestion that I find both plausible and illuminating. Namely, employing highly truth-conducive deductive rules conferred a selective advantage on our ancestors because it helped them to successfully engage in long-term planning.48 There are two types of long-term planning. The first is what might be called “strategic planning.” This is the sort of planning a thinker engages in when constructing a strategy for governing her actions over some stretch of time in order to achieve some desired end. The second is what might be called “contingency planning.” This is the sort of planning a thinker engages in when constructing a plan for handling potential future emergencies or opportunities. Three general features of long-term planning suggest a connection with our employment of reliable deductive rules. First, long-term planning typically relies on many different pieces of information, provided by many different cognitive mechanisms. This suggests that deductive reasoning plays an important role in long-term planning; such reasoning enables thinkers to combine the relevant information and transform it into a more usable form. Second, long-term planning is an important cognitive endeavor, one in which the benefits of success and the costs of failure can be high. We do not always have the time or energy to deliberate about what to do at the time we need to act; long-term planning enables us to ameliorate the impact of such resource constraints. Such planning also enables us to better coordinate our actions (with ourselves and with others); it increases the likelihood that we will achieve important ends.49 This suggests that success in long-term planning provides a significant evolutionary advantage. Finally, when engaging in long-term planning, there is typically little need to make an immediate decision. This suggests that there is little selection pressure against the employment of relatively slow cognitive mechanisms in planning, at least so long as there are compensatory benefits. Taken together, these three features suggest that there was a strong selective pressure for the employment of highly truth-conducive cognitive mechanisms to use in such planning, even if such mechanisms are slower or less efficient than alternatives. The proposal, then, is that our ancestors were selected for employing highly truth-conducive deductive rules in part because it helped them engage in long-term planning.50 Other proposals can also be envisioned. For instance, a currently fashionable view is that many of our cognitive abilities were developed to help with the complexities of living in social groups. On one version of this proposal, our cognitive abilities were developed to help to help compete with conspecifics 48 49 50

I originally owe this suggestion to Paul Boghossian. These claims about the importance of planning are due to Bratman (1987; 2000). Similar remarks can be made about deliberation more generally.

236

Joshua Schechter

for limited resources.51 On a different version of this view, our cognitive abilities were developed to help cooperate with conspecifics in engaging in complicated cooperative endeavors such as foraging.52 Alternatively, it could be that employing highly truth-conducive rules was important for the task of constructing explanations of important phenomena—such as the causes of certain animal traces.53 A still different proposal is that our cognitive abilities are the product of sexual selection. Being able to reason clearly and in a truthconducive way were indicators of the fitness or an organism, and so organisms that were more intelligent were favored as mates.54 There are additional alternatives, too. Many of these views have the resources to explain why we came to employ highly truth-conducive cognitive mechanisms for deductive reasoning. The general moral to draw is not that any particular view (or views) ought to be adopted. Rather, it is that natural selection can sometimes yield highly truth-conducive cognitive mechanisms. Our deductive mechanism is a prime candidate for being such a mechanism.

7.

CONCLUSION

On the proposed account, we are reliable in believing logical truths and disbelieving logical falsehoods because we have a reliable mechanism for deductive reasoning. The explanation of how we came to have a reliable deductive mechanism is an evolutionary one. Our ancestors were selected for possessing logical concepts. Possessing these concepts conferred an evolutionary advantage because it enabled our ancestors to represent important information about the world and because it aided them in assessing their own patterns of reasoning. Our ancestors were also selected for employing deductive rules that are truth-preserving in an important range of cases. Employing such rules enabled our ancestors to convert information into a form that was more useful for guiding their behavior. (According to one proposal, it was advantageous for our ancestors to do this in a highly truth-preserving way because of the importance of long-term planning.) There was an additional selection pressure on our ancestors, namely, to employ rules that are cognitively economical. As a byproduct of these pressures, our ancestors came to employ rules that are generally and necessarily truth-preserving. Employing reliable deductive rules is a heritable trait, and so we came to employ them, too. On this proposal, our reliability about logic is a byproduct of a byproduct. It is a byproduct of the general reliability of our cognitive mechanism for 51

See, for instance, Humphrey (1976) and Flinn, Geary, and Ward (2005). See Sterelny (2003; 2007). 53 Carruthers (2002b) suggests that explanatory reasoning originally emerged to help hunters track animals. 54 See Miller (2000). 52

Could Evolution Explain Our Reliability about Logic?

237

deductive reasoning. This in turn is a byproduct of what was really selected for—an economical cognitive mechanism that is highly reliable in a central range of cases. This proposal provides a satisfying answer to the reliability challenge for logic. It demonstrates that there are plausible answers to be had. Whether or not the precise details of the account are correct, it shows that there is no in principle reason to think that evolutionary accounts are incapable of explaining our reliability about logic. This defuses the tension between the claim that logic is objective, the claim that we are reliable about logic, and our general background views about the world.55

REFERENCES

Belnap, N. (1962) “Tonk, Plonk and Plink”, Analysis, 22: 130–4. Benacerraf, P. (1973) “Mathematical Truth,” The Journal of Philosophy, 70: 661–79. Block, N. (2000) “Semantics, Conceptual Role,” in E. Craig (ed.), Routledge Encyclopedia of Philosophy (London: Routledge). Boghossian, P. (2003) “Blind Reasoning,” Proceedings of the Aristotelian Society, Supplementary Volume, 77: 225–48. Braine, M., and D. O’Brien (eds.) (1998) Mental Logic (Mahweh: Erlbaum). Brandom, R. (2000) Articulating Reasons: An Introduction to Inferentialism (Cambridge, Mass.: Harvard University Press). Bratman, M. (1987) Intention, Plans, and Practical Reason (Cambridge, Mass.: Harvard University Press). (2000) “Reflection, Planning, and Temporally Extended Agency,” Philosophical Review, 109: 35–61. Carruthers, P. (2002a) “The Cognitive Functions of Language,” Behavioral and Brain Science, 25: 657–74. (2002b) “The Roots of Scientific Reasoning: Infancy, Modularity, and the Art of Tracking,” in P. Carruthers, S. Stich, and M. Siegal (eds.), The Cognitive Basis of Science (Cambridge: Cambridge University Press), 73–98. Churchland, P. (1987) “Epistemology in the Age of Neuroscience,” The Journal of Philosophy, 84: 544–53. Darwin, C. (1871) The Descent of Man and Selection in Relation to Sex (New York: D. Appleton and Company). Delbrück, M. (1978) “Mind from Matter?” American Scholar, 47: 339–53. Dummett, M. (1991) The Logical Basis of Metaphysics (Cambridge, Mass.: Harvard University Press). 55 A much earlier version of this paper was presented at an NYU dissertation seminar. I would like to thank the participants for their questions and remarks. I would also like to thank Paul Boghossian, Ray Buchanan, Winston Chiong, David Christensen, Jamie Dreier, David Enoch, Greg Epstein, Dana Evan, Hartry Field, Kit Fine, Don Garrett, Pete Graham, Liz Harman, Jonathan Jenkins Ichikawa, Øystein Linnebo, Anna-Sara Malmgren, Gary Marcus, Christopher Peacocke, Erica Roedder, Katherine Rubin, Katia Samoilova, Karl Schafer, Brad Skow, Declan Smithies, Sharon Street, Michael Strevens, Roger White, and Masahiro Yamada for valuable discussion. I would also like to thank two anonymous referees for their helpful comments.

238

Joshua Schechter

Evans, J. and D. Over (1996) Rationality and Reasoning (East Sussex: Psychology Press). Field, H. (1980) Science without Numbers: A Defense of Nominalism (Princeton: Princeton University Press). (1989) Realism, Mathematics and Modality (Oxford: Basil Blackwell). (1998) “Epistemological Nonfactualism and the A Prioricity of Logic,” Philosophical Studies, 92: 1–24. Flinn, M., D. Geary, and C. Ward (2005) “Ecological Dominance, Social Competition and Coalitionary Arms Races,” Evolution and Human Behavior, 26: 10–46. Fodor, J. and E. Lepore (1991) “Why Meaning (Probably) Isn’t Conceptual Role,” Mind and Language, 6: 328–34. Gould, S. and R. Lewontin (1978) “The Spandrels of San Marco and the Panglossian Paradigm: A Critique of the Adaptationist Programme,” Proceedings of the Royal Society of London, 205: 581–98. Hacking, I. (1979) “What is Logic?,” The Journal of Philosophy, 76: 285–319. Harman, G. (1988) Change in View: Principles of Reasoning (Cambridge, Mass.: The MIT Press). (1995) “Rationality,” in E. Smith and D. Osherson (eds.), Thinking: An Invitation to Cognitive Science, volume 3, second edition (Cambridge, Mass.: The MIT Press), 175–211. Hill, C. (2006) “Modality, Modal Epistemology, and the Metaphysics of Consciousness,” in S. Nichols (ed.), The Architecture of the Imagination: New Essays on Pretense, Possibility, and Fiction (Oxford: Oxford University Press), 205–35. Humphrey, N. (1976) “The Social Function of Intellect,” in P. Bateson and R. Hinde (eds.), Growing Points in Ethology (Cambridge: Cambridge University Press), 303–17. Mill, J. S. (1843) A System of Logic, Ratiocinative and Inductive, volume 1 (London: John W. Parker). Miller, G. (2000) The Mating Mind: How Sexual Choice Shaped the Evolution of Human Nature (New York: Doubleday). Nagel, T. (1986) The View from Nowhere (Oxford: Oxford University Press). Neander, K. (1995) “Pruning the Tree of Life,” British Journal for the Philosophy of Science, 46: 59–80. Nozick, R. (1993) The Nature of Rationality (Princeton: Princeton University Press). (2001) Invariances: The Structure of the Objective World (Cambridge, Mass.: Harvard University Press). Papineau, D. (2000) “The Evolution of Knowledge”. Reprinted in his The Roots of Reason (Oxford: Oxford University Press, 2003), 39–82. Peacocke, C. (1992) A Study of Concepts (Cambridge, Mass.: The MIT Press). Plantinga, A. (1993) Warrant and Proper Function (Oxford: Oxford University Press). Rips, L. (1994) The Psychology of Proof: Deductive Reasoning in Human Thinking (Cambridge, Mass.: The MIT Press). Ryle, G. (1950) “ ‘If,’ ‘So,’ and ‘Because’ ” in M. Black (ed.), Philosophical Analysis (Ithaca: Cornell University Press), 323–40. Schechter, J. (2010) “The Reliability Challenge and the Epistemology of Logic,” Philosophical Perspectives, 24, Epistemology, 437–64. and D. Enoch (2006) “Meaning and Justification: The Case of Modus Ponens,” Noûs, 40: 687–715.

Could Evolution Explain Our Reliability about Logic?

239

Skyrms, B. (1999) “Evolution of Inference,” in T. Kohler and G. Gumerman (eds.), Dynamics in Human and Primate Societies (Oxford: Oxford University Press), 77–88. Sloman, S. (1996) “The Empirical Case for Two Systems of Reasoning,” Psychological Bulletin, 119: 3–22. Sober, E. (1981) “The Evolution of Rationality,” Synthese, 46: 95–120. (1984) The Nature of Selection: Evolutionary Theory in Philosophical Focus (Cambridge, Mass.: The MIT Press). Stich, S. (1990) The Fragmentation of Reason (Cambridge, Mass.: The MIT Press). Sterelny, K. (2003) Thought in a Hostile World: The Evolution of Human Cognition (Oxford: Blackwell Publishers). (2007) “Social Intelligence, Human Intelligence and Niche Construction,” Proceedings of the Royal Society, Series B, 362: 719–30. Stroud, B. (1981) “Evolution and the Necessities of Thought”. Reprinted in his Meaning, Understanding, and Practice: Philosophical Essays (Oxford: Oxford University Press, 2000), 52–66. Tennant, N. (1987) Anti-Realism and Logic (Oxford: Clarendon Press). van Fraassen, B. (1980) “Review of Rational Belief Systems by Brian Ellis,” Canadian Journal of Philosophy, 10: 497–511. Williamson, T. (2003) “Understanding and Inference,” Proceedings of the Aristotelian Society, Supplementary Volume, 77: 249–93. (2007) “Philosophical Knowledge and Knowledge of Counterfactuals,” Grazer Philosophische Studien, 74: 89–123.

9. Can Selection Effects on Experience Influence its Rational Role?∗ Susanna Siegel In a chilling experiment conducted by Keith Payne, when primed with pictures of Blacks, participants more often misclassify a tool (pliers, wrench, or a drill) as a gun, compared with subjects who have been primed with pictures of Whites.1 From this result, we know that whatever psychological state the prime puts the subjects in, it influences their answers on the classification task. But how is their visual experience affected? How do the pliers look to the subjects, when they see them? The experiment does not speak directly to this question. There are several (combinable) options. A first option is cognitive penetration: subjects see the pliers, but nonetheless the pliers look to them like a gun, whereas to subjects given the Whites prime the pliers have their normal appearance. A different option is that the prime generates a selection effect, which can influence either the content of experience, or the role of experience. In selection of features for experience, subjects see the pliers, but due to the prime, they attend only to those features that pliers share with guns, such as being metallic and shiny. Here they have neither a pliers-experience nor a gun-experience, but rather an experience with more impoverished contents, and they end up jumping to the conclusion that it is a gun. A different selection effect is anti-selection of experience for uptake. Here, when participants see the pliers, they look just like pliers, but due to the prime, this experience is ignored when subjects form the belief that the stimulus was a gun. Their pliers-experience is anti-selected for uptake into the usual kind of belief-formation process.2 When subjects form the belief that the stimulus is a gun, their perceptual experience does not play its usual role in the formation ∗ For discussion, thanks to John Bennett, Ned Block, David Braun, Alex Byrne, Bill Child, Brendan Dill, Anya Farennikova, Jane Friedman, Paul Marcucilli, Alyssa Ney, Bernhard Nickel, Keith Payne, James Pryor, Nicholas Silins, participants in a seminar at Harvard, and audiences at the first Sheffield workshop on implicit bias, MIT, NYU, Oxford, Rochester, and Union College, where some of this material was presented. Special thanks to Eric Mandelbaum, Matthew McGrath, Declan Smithies, Scott Sturgeon, and Sebastian Watzl for extensive discussion. 1 Payne 2001. The subjects are non-Black American college students, though similar effects are found for Black American college students. 2 A structurally similar form of anti-selection would stop the process one step later: subjects would form a pliers-belief on the basis of the pliers-experience, but then the pliers-belief would be inferentially quarantined, in something like the way compartmentalized beliefs are. Could a mental state count as a belief, even if it was quarantined from all inferences? I leave this difficult question aside.

Selection Effects on Experience

241

of this belief. Usually, the contents of experience are closely related to the contents of perceptual belief, but not in this scenario, where a gun-belief is formed in response to a pliers-experience. Could cognitive penetration, selection for experience, or anti-selection for uptake explain the role of the Black prime in causing classification errors in the pliers/gun experiment?3 If so, either the resulting perceptual experiences or the resulting perceptual beliefs end up apparently confirming what the subjects expect or fear. These forms of cognitive penetration and selection effects challenge folk theories that construe perception as primarily a means of taking in information about what’s currently in the immediate environment. According to the folk theory, perception is largely the result of initial inputs from the world to the senses, and any intervening perceptual processing disallows it from being influenced by the rest of one’s psychological states, such as what one believes, wants, or fears. Integration with those other psychological states comes after experience, not before.4 In this sense, perception is primarily receptive, and that is why it contrasts so sharply with reasoning. We reason from information that we take ourselves to have already, and sometimes let our reasoning fall prey to our hopes or fears. But perception is different. Perception isn’t subject to reasoning at all—neither good reasoning nor defective reasoning. The idea that experiences are primarily receptive is encouraged by the phenomenology of perception, in which we seem simply to confront the scene before us and take things in as they are.5

3 Cognitive penetration and selection effects are not the only options for explaining Payne’s result, and in fact these hypotheses are better at illustrating types of effects on perception, than they are interpreting this particular experiment. An unpublished follow-up experiment by Payne probing the confidence levels of participants questions the assumption that the participants’ keyboard response reflects what they at that moment take the target shown to be. The follow-up study replicated the main result and then asked participants to rate how confident they were in their responses, and found that confidence levels correlated almost exactly with performance: subjects reported little confidence in answers that turned out to be wrong, and high confidence in answers that turned out to be right. Crucially, in trials with stereotypeconsistent errors, self-reported confidence levels were very low, and subjects reported feeling that their fingers (on the keyboard) could not keep up with the responses they wanted to make. Here participants systematically found themselves indicating an answer they thought was wrong. So perhaps they experience the pliers as pliers and believe they are pliers, but their finger-pressing behavior is not guided by what they believe. Anti-selection for uptake, however, may yet end up explaining Payne’s central result. The follow-up result allows that the pliers-experience is anti-selected for belief, if the subjects have contradictory beliefs about whether the stimulus is a pair of pliers. If we looked ahead to how participants remember what they saw, and found that their memory correlated with their behavior rather than with what they would disown, if asked, at the moment of the experiment, one might find that their behavior was only the first manifestation of exactly the sorts of dispositions we’d take as evidence for believing that a gun was presented. 4 Something like this part of the folk theory is found in Fodor’s idea that perceptual processing leading up to belief is dedicated exclusively to extracting information from those initial inputs, though his theory is directed at perceptual processes that may not be the exclusive determinants of conscious experience. 5 A number of writers have tried to characterize this aspect of perception, including Sturgeon 2000 on scene-immediacy, Martin 2002 on conviction, and Chalmers 2006 on edenic content.

242

Susanna Siegel

The folk theory of perception goes with a folk-epistemology of perception, which in turn assumes a distinction between perceptual experience and perceptual belief.6 Roughly, perceptual beliefs are beliefs about what you’re experiencing, which you form chiefly because you are having the experience. According to folk epistemology, in central cases of perceptual belief, where perceivers are not aware of any reason to discount their experiences, such experiences are the main person-level psychological state that determines the content of perceptual beliefs; and even though those experiences cannot be the product of rational (or irrational) person-level processes, nonetheless they can provide rational support for certain beliefs about what we see, as when we believe that there is mustard in the fridge upon seeing the mustard when we open the fridge door. Putting these two epistemic features together, perception can provide rational support without having any itself, and this winning combination makes it well suited to stop regresses of justification. Most theories of the rational role of experience assume that experience conforms to the folk theory of perception, and are directed at cases in which experiences play the standard role in perceptual belief. The possibilities of cognitive penetration and selection effects on perception create the need to explain the rational role of experiences that don’t conform to the folk theory of perception or don’t play the standard role in perceptual belief. We can see how the challenge to the folk theories unfolds by considering cases in which perception becomes an instrument of confirmation bias. Confirmation bias may be most familiar as a feature of suboptimal searches for evidence in hypothesis-testing, or imbalanced consideration of reasons in connection with forming or maintaining beliefs or fears. In these forms of confirmation bias, one searches for confirming but not for disconfirming evidence for a hypothesis, or one considers reasons that support what one suspects or fears, without considering reasons against it. But a different form of confirmation bias could be found at the level of perception. Here, a subject ends up perceiving, either at the level of experience or belief, what she already suspects or fears to be the case, and is systematically prevented from perceiving counter-evidence, leaving her with apparent confirmation from her perception of her fear or suspicion. This phenomenon can occur in the form of cognitive penetration, and the epistemic consequences in that case are increasingly widely discussed.7 Unlike cognitive penetration, however, the structure of selection effects on perception has barely been examined. We can distinguish between selection effects per se, and selection effects that unfold in a way that leads to confirmation bias. In experiences of objects, some factors 6 For complex forms of dissent to the distinction between experience and belief, see Glüer 2009 and Byrne 2009. 7 Discussion of epistemological consequences of cognitive penetrability of experience can be found in Huemer (2013), Lyons 2011, Siegel (2011 and 2013a), McGrath (2013a and 2013b). For discussion of the epistemological implications of cognitive penetration on perceptual processing generally (not focused exclusively on experience), see Fodor 1983, Churchland 1988, and Raftopolous 2009.

Selection Effects on Experience

243

always determine which objects are being experienced and which features of those objects you focus on—such as where you happen to be looking and why.8 What you believe, want, or fear can easily influence what you are looking at. Likewise, experiences could be anti-selected for uptake due to all sorts of different factors, ranging from load-induced inattentiveness, to antiselection that is at some level explained by its role in maintaining a prior fear, hope, or expectation. Here, the selection effects of interest are those that are generated by prior beliefs (including expectations), hopes, or fears, and that operate as a means of psychologically confirming or strengthening those prior states. These are selection effects that can lead to confirmation bias at the level of perception. My aim is to examine the rational impact of two varieties of selection effects of this kind. First, I’ll present a case in which the same kind of objects are repeatedly selected for experience, where the selection leads to confirmation bias. This type of selection effect is a selection of objects rather than features for experience, but I’ll argue that analogous cases involving selection of features warrant the same treatment. These cases pull us in the direction of allowing that experiences you don’t have can affect the rational status of the experiences you do have. If patterns of experience can be epistemically significant in this way, then ultimately what needs explaining is how the rational role of those experiences can be influenced by their pattern, and by their relationship to the prior psychological states that help select which objects or features are experienced and which are not. I defend the position that patterns of experiences can be epistemically downgraded by certain selection effects. Experiences are epistemically downgraded, if the rational support they offer for believing their contents are reduced or eliminated, relative to a baseline that I’ll assume is the normal case. Normally, seeing a car drive by gives you excellent reason to believe that a car just drove by, and I’ll assume that the experience you have in seeing the car plays a major role providing this rational support.9 Since the perceivers in the cases I consider are unaware of the selection effects and their causal role in confirmation bias, one might think that only externalist theories of rational belief (such as reliabilism) could explain why certain selection effects lead to epistemic downgrade. But I’ll sketch an internalist explanation of epistemic downgrade in these cases. Besides the selection of objects or features for experience, the second variety of selection effect I consider is anti-selection of experience for uptake. Antiselection for uptake creates fragmentation in the mind that is similar in some 8 Here I’ll assume that if you attend to a feature F of an object, then F will characterize the way the object looks to you. This assumption can be taken as a restriction on the type of featureattention at issue. It rules out unconscious attention to features, for example, as well as “feature illusion” in which you attend to an object’s redness (for example) but experience it as blue instead of red. 9 For discussion of the different ways in which experiences could provide rational support for beliefs, see Siegel and Silins 2013.

244

Susanna Siegel

ways to standard cases of compartmentalized beliefs.10 Here the challenge is to explain the rational powers of psychological states that are isolated from one another. In cases of compartmentalized belief, it is controversial how, if at all, beliefs in separate compartments bear rationally on one another. Similarly, in cases of anti-selection for uptake, an account is needed of which other mental states, if any, the anti-selected experiences rationally bear on. I suggest that anti-selected experiences can retain the same rational bearing on mental states that lie outside its “compartment” as it would have in a more unified, less compartmentalized mind, and that this conclusion seems especially plausible in the case where anti-selection for uptake facilitates confirmation bias. The existence of these two varieties of confirmation-bias-inducing selection effects is much less controversial than cognitive penetration. Selection effects are compatible with the idea that all of the psychological processing leading up to perceptual experience is modular. Many experimental results that provide suggestive but not decisive evidence for cognitive penetration are explicable in principle by selection effects.11 But here I set aside controversies about the psychological reality or prevalence of confirmation-bias-inducing selection effects and cognitive penetration. My methodology is to define hypothetical versions of each kind of confirmation bias, discuss its epistemological consequences, and leave the empirical inquiry into whether any of these effects actually happen (and if so, how much) for a separate discussion. Of course, these facts are not epistemologically irrelevant, since we have a special interest in our own epistemic situation, and for that we need to know the empirical facts. But the epistemological significance of those facts can be pursued in a relatively aprioristic way. Once we have answers to the empirical questions about which types of effects on perception we are subject to, we can see what different epistemological theories tell us about our actual epistemic situation. In addition, distinguishing different types of effects on perceptual experience and perceptual belief may help pin down competing interpretations of some experimental results. The discussion proceeds as follows. In section 1, I introduce the main epistemological problem that arises from confirmation-bias-inducing selection of objects or features for experience. In sections 2 and 3, I sketch an internalist way to respond to the problem. I discuss anti-selection for uptake in section 4, and its significance for the epistemology of mutually unintegrated mental states in section 5. Throughout, I use “experience” as shorthand for “perceptual experience,” though I focus on visual experiences.12

10

On compartmentalized belief, see Lewis 1982. Suggestive but indecisive results include Barrick 2002, Balcetis and Dunning 2006, Hansen 2008, Levin and Banaji 2006, Proffitt 2006, among many others. 12 In my terminology, perceptual and visual experiences are phenomenal states, and so can be had in cases of hallucination or illusion, as well as veridical perception. 11

Selection Effects on Experience

1.

245

SELECTION FOR EXPERIENCE AS A MEANS O F C O N F I R M AT I O N B I A S

Selection for experience should be understood to include either (or both) the selection of features and the selection of objects. In Payne’s pliers/gun experiment, the prime pretty clearly isn’t affecting whether one sees the pliers, but it may be influencing how the pliers look when you see them, by influencing which features one attends to. But in general, prior states such as beliefs, desires, or fears could influence not only which features one experiences, but also which objects one experiences. Taken by itself, selection for experience is unremarkable. But when selection for experience operates as a means of confirmation bias, so that one ends up experiencing (roughly speaking) what one antecedently fears or wants or believes, an epistemological problem arises. For instance, suppose you want it to be true that in general, red things are square. Call this the redsquare desire. Unbeknownst to you, your red-square desire affects how you distribute your visual attention when looking at a crowded display, to the extent that you don’t see other red shapes. Although the display includes shapes of many colors, among the red shapes your attention is drawn only to the squares. You see green triangles and orange circles, but overlook the red shapes that aren’t square. A similar selection effect could also hypothetically be induced by a prior belief in the generalization that red things are square. The core of the problem is to assess what kind of rational support the redsquare experiences provide for the generalization that all the red things in the display are square. How might such distributions of a subject’s perceptual experience come about? Other red shapes besides squares could be unconsciously perceived, without representations of them ever becoming conscious. Alternatively, a prior state such as a red-square desire or the belief that in general red things are square might direct attention only to those places where red squares are likely to be found, even if the subject isn’t aware that these places are likely to contain red squares. These examples are unrealistic, but the unrealistic part is limited to the content of the states involved. A more realistic type of example involves stereotypes or prejudices that operate implicitly, and impact behavior in evaluative contexts. For instance, often evaluators assessing candidates for hire harbor implicit negative views of “outgroups,” and focus disproportionately on the negative features of outgroup members when evaluating them. The negative features are not always (or even very often) hallucinated, but a fuller picture of the candidate would show that those features are counterbalanced by positive features. One of the psychological structures that could underlie this behavior is the type of selection effect described here, where the positive features are simply not consciously taken in to begin with.13 This manifes13 Another psychological structure that could underlie the same behavior is that the positive features are perceived, but given less weight in assessment (Steinpries 1999, Goldin and Rouse

246

Susanna Siegel

tation of prejudice is more realistic than our red-square example, but our example lets us isolate one of the epistemic problems raised by this type of phenomenon. What kind of rational support do the red-square experiences provide for the generalization that all the red things in the display are square? On the one hand, you see the red squares in the display perfectly well, without any illusion or hallucination—just as a hiring committee may register the negative features of outgroup candidates without any distortion. On the other hand, the idea that these experiences rationally support the red-square generalization can seem suspicious—just as suspicious as the hiring committee’s judgment might seem that all of the outgroup candidates are relatively weak, when this judgment is made on the basis of their selective focus on those candidates’s weaknesses. This case raises the possibility that facts about what experiences you don’t have can influence the rational role of experiences that you do have. The problem can be sharpened by considering a series of experiences. Imagine a psychological experiment in which you’re given a visually crowded display, such as the kind found in the game Where’s Waldo?. The experimenter gives you a series of three search tasks to find all the targets belonging to a specified type within a challenging time limit.14 First the experimenter tells you to find all the sad faces in the display as quickly as you can, and to let her know when you think you’ve found them all, giving you a short time limit. After finding several sad faces, the time in between searching and finding sad faces lengthens, until you cease to find any more. When you reach the time limit, you are not completely confident that you have found them all, but you report your results to the experimenter. She says: “Well done! You found them all.” Your second task is the same, except with happy faces. Encouraged by your success with the sad faces, you search for the happy faces, and once you stop finding any more, you report your result before the time limit is up, feeling more confident this time. Once again the experimenter tells you that you found all the targets. Your third task is to find the red items in the display. Unbeknownst to you, a prior hope influences how your attention is distributed—the hope that in general, red things are square. Buoyed by your success in the previous search tasks, you report confidently that you have found all the red things. Because of how you distribute your attention, you miss all the red things that weren’t 2000). We might see this phenomenon as a temporary shift in the criterion for what counts as a strong candidate. A similar hypothesis about Payne’s misclassification result would be that when shown the Black prime, subjects jump to the conclusion that the metallic shiny object is a gun, effectively lowering their criterion at the level of judgment for what will count as a gun. Standard shifting could also occur in the transition from subpersonal perceptual processing to experience, if subjects formed a gun-experience in response to the combination of initial sensory inputs and the Black prime. That would be a case of cognitive penetration. 14

Thanks to Eric Mandelbaum for inventing this experiment.

Selection Effects on Experience

247

square. You noticed that all the red things that you found in the display were square, and formed the belief that in this display, all the red things are square. We can call the scenario in the third stage of the experiment the perverse scenario. In the perverse scenario, the prior red-square hope prevents you from consciously perceiving non-square red things. The perverse scenario can be contrasted with a bad-luck scenario. In a bad-luck scenario, you don’t antecedently hope that in general, red things are square and you aren’t subject to any other systematic red-square related bias in your dispositions to attend, but just by chance, you fail to notice the non-square red things in the third stage of the experiment. The sharpened question concerns the rational support provided by the subject’s experiences for the universal generalization that all the red things in the display are square: Distribution Question: Do your red-square experiences provide just as much rational support for believing that all the red things in the display are square, in the perverse and the bad-luck scenarios? If the answer is Yes, then the fact that the red-square experiences in the perverse scenario came about via the red-square hope does not affect how much rational support those experiences provide, and there is no other relevant difference between the two scenarios. Some theories of rational belief may ultimately embrace the Yes answer, on the grounds that the subject is unaware of the role of the red-square hope in selecting red squares for experience. For instance, many internalist theories of justification deem etiological features of experience irrelevant to its rational role, so long as those features are forgotten, or were never within the subject’s awareness, or are otherwise inaccessible to the subject at the time of the experience.15 From this observation, one might conclude that only externalist theories of justification, such as reliabilism, can ground the No answer. But this conclusion misses an internalist option that grounds the No answer in the relations between the experiences and the psychological states that give rise to it—even if the subject is unaware of those relations, and regardless of any considerations about reliability. This option locates some of the features that determine the rational status of experiences in places that are external to the subject’s awareness, but still internal to the subject’s mind. To bring the internalist option into focus, we need to distinguish between two ways in which single-square experiences could provide rational support for the universal generalization that all the red things in the display are square. Incremental Support: Each red-square experience, at the time that you have it, provides incremental rational support for the universal generalization that all the red things in the display are square.

15

See the discussion of the problem of forgotten evidence in Feldman and Conee 2001.

248

Susanna Siegel

Overall Support: The red-square experiences, once you have all of them that you’re going to have, provide rational support for the universal generalization that all the red things in the display are square. From now on I’ll call the generalization “the universal generalization” for short. When the Distribution question asks whether the red-square experiences provide rational support for the universal generalization, we could interpret it as asking about either incremental support or overall support. There is also a different dimension along which the Distribution question can be made more specific. The original question asks about the rational role of the red-square experiences in providing rational support for the universal generalization, without specifying whether the experiences do this all by themselves, or together with other factors that are present in both scenarios. Given your success on the search task (for faces) that precede the search for red things, such a factor might seem to be present in the form of a well-founded belief, based on the experimenter’s feedback, that you are good at these search tasks, or that you are not systematically excluding some of the search targets. To take account of the fact that the red-square experiences might provide rational support of either kind for the universal generalization, in combination with some additional factor X, we can reformulate the distribution question to focus on whether the two scenarios differ in whether they contain any values for X, allowing that it might be a null factor. In case the additional factor might be different for incremental support and overall support we can use different subscripts for X in the two questions: Incremental support question: Is there a factor XI present in the perverse and bad-luck scenarios, such that the combination of each red-square experience with XI provides just as much incremental support for the universal generalization in both scenarios? Overall support question: Is there a factor XO present in the perverse and badluck scenarios, such that the combination of the red-square experiences with XO provides just as much overall support for the universal generalization in both scenarios? The incremental support question faces a controversy over whether, if a proposition ∼e is evidence for a hypothesis ∼h, then e is evidence for h. For instance, let h be the universal generalization, and let e be the proposition that location L does not contain a red non-square. Given that the proposition (∼e) location L in the display contains a red non-square provides evidence for the proposition (∼h) it is not the case that all the red things in the display are square, would the proposition (e) that location L does not contain a red non-square provide evidence for the universal generalization? One way for e to be true is if location L contains a red square. Given that the discovery of red nonsquares would disconfirm the universal generalization, does the discovery of red squares add rational support to it? Kaplan (1996) argues that it does, Popper (1959) argues that it doesn’t. But even if Kaplan’s side is right, and

Selection Effects on Experience

249

each red-square experience incrementally supports the generalization in equal amounts in both scenarios, that support might be insignificantly small. If you weren’t in New York City last night when a murder occurred there, then you didn’t commit the murder. But suppose you were in New York City last night. On Kaplan’s view, the fact that you were there provides some evidence for the hypothesis that you committed the murder, but (as he emphasizes) the amount of evidence it provides is so little that it would need to be supplemented with lots of other evidence to provide any significant support for this hypothesis. Just as what’s relevant in the murder example is a bigger collection of evidence, so too in the red square case, the version of the distribution question we should be asking is a version about a significant amount of rational support, not just any amount. Since if any red-square experiences provide any significant amount of rational support for the universal generalization, all of the red-square experiences taken together do, I’ll interpret the distribution question as asking about overall support from now on. In this context, “rational support from red-square experiences for the generalization” is overall support from those experiences for the generalization. The official Distribution question is the question about overall support. Official Distribution question = Overall support question: Is there a factor XO present in the perverse and bad-luck scenarios, such that the combination of the red-square experiences with XO provides just as much overall support for the universal generalization in both scenarios? My strategy for introducing the internalist grounding the No answer to this question is to argue first that the No answer is compatible with evidentialism, the thesis that the rational beliefs are exactly those that fit the evidence, where a subject’s evidence supervenes on her mental states.16 Evidentialism may seem to be at odds with the No answer. But it isn’t. By seeing how evidentialism can be combined with answering No to this question, an internalist grounding of the No answers comes to light. Why Evidentialism Seems Forced Into Answering Yes If evidentialism is true, then when we ask whether you have rational support for P, what’s relevant is just the evidence for P you have. The fact that there’s some other evidence that bears on P, but which you don’t have, isn’t relevant.17 There may be norms governing the search for evidence—how to do it, where to look, when it is okay to stop, etc. But according to standard versions of evidentialism, these norms are different from norms governing what is rational to believe. How much evidence you have, and the process by which you got it, doesn’t affect what’s reasonable to believe in light of that evidence. 16

Feldman and Conee 2001. For discussion and defense of this idea, see Foley and Fumerton 1982, Kelly 2002, Feldman 2003. 17

250

Susanna Siegel

The same idea applies straightforwardly to experiences, on the assumption that experiences are or provide evidence. If the visual search process had gone differently, for instance because there was no selection effect by the red-square hope of the sort described, then your visual search probably wouldn’t have terminated without your finding some other red shapes. But according to the evidentialist idea as applied to experiences, when we ask whether the redsquare experiences provide incremental rational support for the distribution propositions, we’re asking whether those propositions are supported by the experiences you have. The fact that you could have had some other experiences bearing on them is not relevant. How would the No answers to the Distribution Question come to be at odds with evidentialism? Evidentialism entails that two subjects with exactly the same evidence have the same amount of rational support for the same propositions. The conflict arises when we combine No and evidentialism with an assumption about the evidence present in both scenarios, resulting in the following inconsistent triad: (a) In each scenario, the amount of rational support from red-square experiences you have for the generalization is different. (= the ‘No’ answer) (b) If your evidence for the generalization is the same both times, then your rational support for the generalization is the same both times. (= consequence of Evidentialism) (c) Your evidence for the generalization is the same in both scenarios. (= hypothesis about experiences and search process) What grounds are there for (c)—that your evidence is the same in both scenarios? One might argue: the evidence for the generalization in both cases consists in (i) your experiences, plus (ii) your evidence about the search process, and nothing else. These two factors seem evidentially on par in both scenarios: Same experience → same evidence: Since the red-square experiences in the two scenarios are the same, they contribute the same evidence to the generalization belief in the two scenarios. Rational search termination: The subject has no reason to think that the visual search was terminated too soon, given her success on the previous tasks. Since evidentialism is at odds with answering No only if claim (c) is true, evidentialism can be reconciled with No by denying (c).18 But on what grounds can (c) plausibly be denied? 18 An intermediate response splits the difference between Yes and No, by rejecting thesis (a), but holding that nonetheless, beliefs in the distribution propositions based on the redsquare experiences would be epistemically ill-founded. According to the split-the-difference position, in the perverse scenario, the red-square experiences provide propositional justification for the generalization, but do not support the formation of doxastically justified belief in the generalization.

Selection Effects on Experience

2.

251

HOW EVIDENTIALISM CAN ANSWER NO

Writ large, what seems to force evidentialism into answering Yes is that the red-square experiences seem to provide a rational route to the generalization, when combined with the (apparently justified) belief that you have looked long enough to reach an answer to the question defining the task. In the red square example, the task is to find all the red things, and the supporting belief is based on the experimenter telling you that you succeeded on the previous, similar search tasks. In the hiring case, there is also a search task—to find features of the candidates that are relevant to deciding whether they are a strong candidate. And committee members might have an analogous, apparently justified supporting belief, based on their past success in identifying strong candidates, that they have looked long enough to answer the question. To see why evidentialism is not forced into answering Yes, we need to look more closely at the structure of the red-square experiences that figure in this reasoning. I’m going to make a case for the existence of a temporally extended, compound experience. With the compound experience in the picture, the reasoning that seems to push the evidentialism toward Yes can be defused in two steps. First, the compound experience is the only one that could reasonably be thought to combine with the background beliefs just described, in order to support answering Yes. But second, it is open to the evidentialist to hold that this compound experience is downgraded by the selection effect. This position gives them a way (and apparently the most plausible way) to deny claim (c), even on the assumption that the background beliefs that one has looked long enough are justified. Introducing the evidentialist position involves examining compound experiences (2.1), and then using that notion to defuse the reasoning that seems to force evidentialism to answer Yes (2.2). In section 3, I sketch an argument for the claim that the selection effects epistemically downgrade the compound experiences in each case. Since the argument provides general reason to think the compound experience is downgraded, a fortiori it provides reason for evidentialists to reject claim (c) and answer No to the Distribution question. 2.1. Individual vs Compound Experiences When you look for red things in the display, you have an experience of searching. When you find the red squares, and then keep looking for red things but cease to find any other red shapes, that too is an experience—an experience of searching and not finding any more targets. This option is at odds with a plausible principle linking propositional justification to doxastic justification. According to the linking principle, if a factor X provides propositional justification for a proposition, then if you formed a belief on the basis of X, the belief would not thereby be doxastically unjustified. Without this linking principle, the resulting notion of propositional justification would be watered down a lot, leaving it obscure in what sense X provided S with propositional justification for P in the first place, given that she can’t even use X to form a wellfounded belief in P.

252

Susanna Siegel

These experiences differ phenomenally from the experiences of a red square that punctuates the experience of searching. Those individual experiences can occur in isolation, without occurring in the course of a search for red things. We can therefore distinguish individual red-square experiences (experiences of single red squares) from more complex, temporally extended experiences that involve a sequence of individual red-square experiences. For any putative temporally extended experience, we can ask what makes it stand out as a natural subexperience in the subject’s stream of consciousness— one that is both temporally extended, yet stops short of the entire stream of consciousness, or the entire “ray” of experience that extends forward in time given a temporal starting point. There are doubtless multiple temporally extended experiences that are salient aspects of the subject’s mental life. Call the compound experience the sequence of the single-square experiences, together with the experience of having looked for red things in the display after finding the red squares, and not finding any more red shapes. By hypothesis, the compound experience is temporally extended, starting with the sequence of red-square experiences, and continuing through the experience of looking for red things and not finding any. We can define the compound experience further by answering specific questions about its duration, its relation to memory, its content, and its phenomenal character. Regarding duration: when does the compound experience end? It is natural to think that it ends when the subject stops looking at the display. But what if by the time the subject stops looking at the display, she has forgotten that she saw some (or even all) of the red squares she saw when she had the singlesquare experiences? For some purposes it may be illuminating to focus on a stretch of a subject’s experience, regardless of what she remembers. But here, it is useful to focus on experiences that the subject could explicitly endorse as bases for believing all of the contents of the experience. For that purpose, it is useful to build into the nature of compound experience that the subject retains in memory its earlier parts by the time she has the later parts. Regarding memory: compound experiences satisfy the following assumption. According to the memory assumption, at the end of the compound experience, the subject remembers experiencing each red square that she saw. (The memory assumption is compatible with the subject forgetting about some of the red squares over the course of seeing them, so long as she remembers them at the end). Regarding content: what contents does the compound experience have? Given the memory assumption, it is natural to think that the content of the compound experience includes a conjunction of the contents of all of the individual red-square experiences. Conjoining these contents yields a content of the form: Conjunctive content: this1 red thing is square and this2 red thing is square . . . and this thisN red thing is square

Selection Effects on Experience

253

. . . where N is the number of red things the subject sees. (I am ignoring the differences between “this red thing is square” and “this is a red square,” and assuming that the red things in the display look red to the subject). Factoring in the experiences of searching and not finding any more red things plausibly adds another conjunct: Conjunctive content + : this1 red thing is square and this2 red thing is square . . . and this thisN red thing is square and no other red thing in the display is square. If a subject has an experience with this conjunctive content, then her experience conveys something to the effect that the red things she has seen in the display are all the red things there are. And if it does that, then it seems to include in its contents the universal generalization: Universal generalization: All the red things in the display are square. The considerations suggest that a subject in the red-square experiment could have a compound experience with universally quantified contents. Regarding phenomenal character: one might reasonably wonder whether experiences could have universally quantified contents. Could such content answer to any phenomenal aspect of the experience? If not, then either no experience ever has such contents, or if it does, its link to the subject’s mental life is tenuous. One might think that our experiences can only ever take a stand on how many F’s there are, by representing that there are at least that many F’s. Consider the experience of seeing three pens on a cluttered table. Your experience might represent that there are three pens on the table, but not take a stand on whether there are any other pens among the clutter. If there turned out to be more pens underneath the clutter, the experience would not thereby be inaccurate. If experiences of quantity were like this, then experiences could never represent that the F’s in the relevant domain are all the F’s there are in that domain. And if they can’t do that, then they can’t represent that all the F’s in that domain are G. But some experiences provide natural examples of experiences that represent that the F’s in the relevant domain are all the F’s there are in that domain. For example, suppose you saw three eggs lined up in an egg carton, and had an experience that represents not only that three eggs are in the carton, but also that the other spaces in the carton are empty. If there were (by some amazing illusion) four eggs in the egg carton your experience would be inaccurate. It seems straightforwardly part of the phenomenal character that the eggs in the egg carton are exactly three. Going with that, the experience represents that three eggs are all the eggs in the carton. And if it can do that, then it can represent that all the eggs in the carton share some feature (they are brown, or egg-shaped, or are eggs, for instance).19 I’ll call the phenomenal feature that goes with universally quantified contents ‘phenomenal that’s-allness’. 19 I’ve spoken as if experiences could represent properties such as being an egg or being a pen, but the same points would hold for egg-shaped or pen-shaped volumes.

254

Susanna Siegel

By following on the experience of searching and not finding any more red things, the compound experience has phenomenal that’s-allness. In this respect, it is more like the egg-carton experience than the cluttered-table experience. In a context where one has been searching for a target (such as eggs), an egg-cartony experience has a salience structure that is stable with respect to its target. A salience structure for a subject is a structure in which some objects and properties are experienced by S and others, though present, are not experienced by that subject.20 In a stable salience structure, which items are targets is no longer evolving in the experience. Once the salience structure is stable, you may suddenly notice another target. But when that happens, then (assuming the additional target is experienced as such) the phenomenal that’sallness ceases. When the salience structure is still evolving in the context of a search, so is the structure of experienced affordances—roughly, experienced possibilities of action. If you are in the midst of looking for red things in the display, then as yet unsearched areas of the display are presented to you as places where red things might be found. If the salience structure of your experience of seeing the display was still in flux, you would either still be looking around at the display, or else you experience parts of the display you haven’t seen yet as possibly containing red things.21 Once the phenomenal that’s-allness emerges, these affordances are gone. Putting these observations together, we could say that with respect to the target, the transition into phenomenal that’s-allness can be a transition from a kind of experiential flux, in which the salience structure of targets is still evolving, and a kind of experiential fixity, in which the salience structure of targets is stable. This type of experiential flux and fixity could be seen as a way for an experience to treat the question of which experienced objects or features are targets as open or closed. Treating a question as open or closed can have obvious ethical implications as well as epistemological ones. Abstracting from perceptual experience, it is intuitively possible to prematurely write someone off as unsuitable for a task, just as it is possible to treat it as an open question whether someone is suitable for a task, even when there is already evidence to settle the question one way or another. Social contexts like these can strengthen our understanding of what it is to treat a question as open or closed, by illustrating how these options can make some social relationships into live possibilities while making others socially inaccessible. In this way, in social contexts, transitions between flux and fixity can be mechanisms of social inclusion or exclusion. It is natural to think that these mechanisms sometimes operate in ethically bad

20 Watzl 2010 and 2011 discusses structures of salience, though he defines it in terms of the structure of attention to objects and properties, rather than the structure in which objects and properties are experienced at all. 21 More exactly, you would experience the display as containing parts that are so far unseen by you and that possibly contain more red things.

Selection Effects on Experience

255

(or good) ways by operating in epistemically bad (good) ways. In section 3, I argue that partly by virtue of its transition from flux to fixity, the compound experience is afflicted by an epistemic shortcoming. 2.2. Why Evidentialists Don’t Have to Answer Yes With compound experiences on the table, we can return to the idea that the subject of the experiment seems to be able to combine her knowledge that she succeeded on the first two search tasks with her red-square experience, to get significant rational support for the universal generalization. If she can do that in both the bad-luck and perverse scenario, then by evidentialist lights, claim (c) in the inconsistent triad seems true (the subject has the same evidence in both scenarios for the universal generalization), and so does the answer Yes to the Distribution question (Yes, the subjects in the two scenarios have the same amount of rational support from their experiences for the universal generalization.) A first attempt to challenge this reasoning would be to deny that the subject’s success on the first two search tasks (for faces) gives her reason to think she has succeeded on the third search (for red things). But the subject has no inkling that her red-square hope is impairing her performance on the third tasks, and all three tasks are superficially similar. And the red-square hope has no influence on this belief—its basis is the experimenter’s feedback, together with the experience of looking for targets and eventually not finding any more. So this attempt to challenge the reasoning seems weak. I am going to grant to the proponents of Yes that it fails. A different attempt to challenge the reasoning is to hold that due to the selection effect by the red-square hope, the individual red-square experiences in the perverse scenario do not provide any evidence at all.22 This option entails that the individual red-square experiences would not even rationally support the single-square propositions such as those that could be expressed by this1 is a red square, this2 is a red square, etc., or the plurality proposition that the display contains some red squares. That position is arguably too strong, since on the face of it, the subject is excellently positioned to form justified beliefs with these contents: she veridically sees the red squares, and her experiences result from normal processes. We can even suppose that the process of detecting red squares generates few false positives and few false negatives. Typically, when visual experiences with these features are endorsed, the resulting beliefs are justified. But if individual red-square experiences provide rational support for single red-square propositions, then why can’t these individual experiences be combined with the background knowledge about one’s success on the previous

22 I set aside the question whether this position rules out that self-ascriptions of red-square experiences made in the usual distinctively first-personal way can be rational.

256

Susanna Siegel

search tasks to provide reason for the universal generalization? Here we come to the first step in defusing the reasoning. Suppose you had only individual red-square experiences, without the compound experience. More exactly, suppose you experienced several red squares, but did not have the experience of looking for more red squares in the display and not finding any more. The allotted time ran out before you felt you had looked long enough to complete the task. Merely having a series of individual red-square experiences leaves open that you might experience the display as having large areas in which you hadn’t yet looked. If you had an experience like that, then your background knowledge that you succeeded on the previous search tasks would not give you much reason, if any, to think you have succeeded in finding all the red things. In the previous tasks, you had analogous compound experiences as well—experiences of looking for targets and eventually finding no more. If you lacked that experience on the third task, your background knowledge of previous successes would not go very far in giving you reason to believe the universal generalization. These considerations suggest that the type of experience that the background knowledge needs to combine with to support the universal generalization is something like a compound experience, rather than individual red-square experiences. And this brings us to the second step in defusing the reasoning that seems to force the evidentialist’s hand. While it seems implausible that individual red-square experiences are epistemically downgraded, a case can be made that the compound experience is epistemically downgraded.

3.

WHY THINK COMPOUND EXPERIENCES ARE DOWNGRADED?

Why think the selection effect downgrades the compound experience in perverse scenario? Reliabilists might hope to appeal to reliability considerations.23 A potential difficulty for reliabilist treatment of the red-square example lies in identifying which processes leading from red-square experiences to the belief in the universal generalization have to be reliable, for the belief to count as justified. This is a version of the Generality Problem (Conee and Feldman 1998). In the perverse scenario, the process generating red-square experiences is a highly reliable process for forming beliefs about where the red squares are, a slightly less reliable process for forming beliefs about where the colored shapes are, and an unreliable process for forming beliefs about where the red shapes are. Which of these processes (if any) determines whether the belief that the display contains red squares, red shapes, and colored shapes is justified, and what principled way is there to identify the processes that matter? This problem can be avoided by internalist explanations of the downgrade that start from the idea that an experience or belief can be a conduit of ill23

See Lyons 2011.

Selection Effects on Experience

257

foundedness. An experience or belief is a conduit of ill-foundedness, if any subsequent belief that is formed primarily on its basis would thereby be illfounded. To show that an experience or belief is a conduit of ill-foundedness, it is not necessary to specify which type of unreliable process (if any) makes it ill-founded, and in this way the Generality Problem is avoided. If an experience is a conduit of ill-foundedness, is it epistemically downgraded? It is plausible to think that being a conduit of ill-foundedness and being epistemically downgraded are linked in this way. If any belief formed on the basis of the experience would thereby be ill-founded, then it is hard to see what is left of the idea that the experience provides a significant amount of justification for it.24 If so, then we can argue that the compound experience is downgraded, by arguing that it is a conduit of ill-foundedness. Let’s call a dependence relation whereby a hope, fear, or prejudice that P leads to a belief or experience with content P an ’irrational dependence relation’, if it would make the belief ill-founded. In the case of belief, we can distinguish between the relation of dependence on a hope that makes it ill-founded, and the features that explain why the belief is made illfounded by that dependence relation. And then we can ask: which features of belief explain why depending on a hope in that way makes beliefs conduct ill-foundedness? If a belief’s dependence on a hope makes it conduct illfoundedness because the belief has feature X, and experience E with the same dependence relation on a hope has feature X, then that’s a reason to think that dependence relation makes E conduct ill-foundedness. That is my argumentative strategy for defending the thesis that the compound experience is downgraded, I first compare the compound experience with a belief, B1, that has the same content and depends on a hope in the same way. Then I argue that this dependence relation makes B1 conduct ill-foundedness, by virtue of a feature that beliefs in general share with experiences. To execute this strategy, we need an example of B1. Consider a route to belief via wishful (or fearful) thinking. Here you don’t experience any red things at all, but you form the belief that all the red things in the display are square because you hope or fear that this generalization is true. A different kind of wishful (or fearful) thinking operates by overgeneralization. You see some red squares (as such), and the red-square hope (or fear) makes you jump to the conclusion that all the red things in the display are square. In both of these wishful (or fearful) thinking cases, the red-square hope (or fear) causes the subject to form B1. And in both cases, because of this influence, B1 is a conduit of ill-foundedness, for subsequent beliefs formed primarily on its basis. (Maybe the ill-foundedness of B1 would wash out if the basis of a subsequent belief B2 was big and complex enough and included B1. But it

24 The underlying principle here is the same as the one discussed earlier in rejecting the “split the difference” option for evidentialism (note 18). For more on the link between being a conduit of ill-foundedness and epistemic downgrade, see Siegel 2013a, and McGrath 2013b.

258

Susanna Siegel

doesn’t always wash out. For instance if B2 = thing1 is a red square, if you formed B2 on the basis of B1, B2 would thereby be ill-founded.) In these cases, are the features that make B1 conduct ill-foundedness shared by the compound experience? Let’s call C the universal generalization that all the red things in the display are square. We have already noted a first causal feature that applies to both the direct and the overgeneralization routes to B1: the red-square hope or (fear) causes there to be a state with content C, in a manner characteristic of wishful thinking. The compound experience ends up with content C via the selection effect imposed by the red-square hope, whereas B1 ends up with it via wishful (or fearful) thinking, either directly or by overgeneralization. A second causal feature is shared by the compound experience and by B1 when B1 is formed by overgeneralization. This feature is a transition, caused by the red-square hope (or fear), from a state of flux to a state of fixity. We’ve seen that in the case of the compound experience, the red-square hope causes a transition from experiential flux (when the salience structure is still evolving, and the subject is still looking around the display for red things), to experiential fixity (when the salience structure stabilizes into phenomenal that’s-allness with respect to red things). A similar transition can be found in the overgeneralization route to B1. At the level of belief, the transition between flux and fixity can be thought of as a transition from suspending judgment on whether a proposition P is true, to settling on an answer to that question (an answer other than “I don’t know”). This transition is normatively assessable. In some cases, if one starts out in a state of suspended judgment on what the answer is to the question one is investigating, sometimes it is epistemically appropriate to continue suspending judgment in the face of incoming information, rather than to settle on an answer to that question. When B1 is formed via overgeneralization, in the course of a search for red things in the display, there is a similar transition from flux with respect to which things in the display are red (while you are still looking around the display) to fixity on this question. Just as in the compound experience case, the transition from experiential flux to experiential fixity is caused by the red-square hope, in the case of B1, the analogous transition is caused by the red-square hope. In the compound experience, the red-square hope causes an experience (part of the compound experience) of looking around the display for red things and ceasing to find any more of them. In the case of B1, the red-square hope causes the overgeneralization from the individual red-square experiences. Given the assumption that the compound experience and B1 (in the overgeneralization case) depend on a red-square hope in the same way, we can ask why depending on a red-square hope in that way makes beliefs conduct illfoundedness. Are the features of B1 that explain why the overgeneralization we’ve described leads to ill-foundedness shared with experiences? A first set of reasons to think they are shared is that the features that distinguish belief from experiences don’t explain why irrational dependence relations make beliefs conduct ill-foundedness.

Selection Effects on Experience

259

The major difference between beliefs and experience is that we don’t subject them to the same normative assessment. We consider some beliefs to be unjustified or irrational, and we subject all beliefs to normative assessment. But we rarely, if ever, criticize someone on the grounds that they have an irrational or unjustified experience. We don’t regularly subject experiences to normative assessment. And even if some experiences are downgraded—which could be seen as a form of epistemic normative assessment—it sounds odd to criticize someone for having the experience they have, even if it is downgraded. But on closer examination, these disanalogies are not relevant to finding features that explain why irrational dependence relations ill-found beliefs. When B1 is ill-founded by irrational dependence relations such as wishful thinking, the explanation of what makes it ill-founded is not the fact that B1 is irrational. Rather, the explanation goes around the other way. B1 is irrational, because of the way it depends on wishes or fears. An opponent might reply that beliefs can be ill-founded by as wishful or fearful thinking, because they need to be caused by rational processes to provide justification, and wishful or fearful thinking are not rational processes. In reply, it doesn’t seem true in general that beliefs have to be caused by rational processes to provide justification. For instance, when developmental psychologists invoke “core cognition” to explain behavior of infants, they are naturally interpreted as positing beliefs, such as beliefs that give shape to core concepts of mechanical causation and agency, that constitute our intuitive physics, and that underlie our primitive number systems.25 These beliefs were not ontogenetically acquired at all, a fortiori they are not caused by rational processes. For all that, they may be rational beliefs to hold, and presumably they provide rational support for subsequent beliefs in these domains. If so, then it is not in general true that beliefs need to be caused by rational processes to provide justification. What features of belief explain why they can be ill-founded by irrational dependence relations on wishes, fears, or prejudice? My answer is that beliefs are made to conduct ill-foundedness by irrational dependence relations, by virtue of the fact that belief is a mode of endorsing content. Beliefs are states whose contents subjects endorse, and thereby give shape to their outlook on the world. Whatever else beliefs are, they are endorsements of contents.26 Why think that being an endorsement of content explains why irrational dependence relations make beliefs conduct ill-foundedness? Consider the relationship between ill-founding and well-founding. Schematically, the 25

Carey 2009. What about other belief-like states, such as low credences, the cognitive underpinnings of implicit bias, delusions, or specially circumscribed situations in which subjects act as if P were true? Are these all endorsements of content as well? Low credences can be set aside because the argument requires only that there is a class of beliefs that are endorsements of content. (I am assuming without argument that reductionist approaches that would dismiss our usual notion of belief or replace it with credences cannot be right.) Whether or not the other states are beliefs, they are belief-like to the extent that they involve endorsements of contents. 26

260

Susanna Siegel

reasoning can be put like this: B1 ill-founds B2 only if B1 has feature F. B1 has feature F only if it is an endorsement of content. So B1 ill-founds B2 only if it is an endorsement of content. Less schematically, when a belief B1 ill-founds another belief B2, an epistemically good-making feature of B1 is removed. That feature is: providing rational support for B2. For instance, if B1 were well-founded, and it combined with other states in the right way, then it could provide rational support for B2. But in at least some cases, for B1 to provide rational support for B2, B1 has to be an endorsement of content. For instance, having a very low credence or weak belief in P could not make it reasonable to endorse the disjunction P-or-Q. In general, the relations underlying coherence requirements on belief need a steady attitude all the way through. Putting these points together, in at least some cases, for B1 to illfound B2, B1 has to be an endorsement of content. So being an endorsement of content explains why irrational dependence relations make beliefs conduct illfoundedness. If being an endorsement of content explains why beliefs are illfounded by irrational dependence relations, then this is explained by a feature that is shared by experiences. Experiences are also states whose contents subjects endorse, and thereby give shape to their outlook on the world. By virtue of being endorsements of contents, beliefs and experiences have similar psychological roles. Like beliefs, experiences provide input to the reasoning (including action plans) that we actually go through—whether that reasoning conforms to epistemic and practical norms or not. Experiences and beliefs play similar roles in reasoning (good and poor, epistemic and practical) because are both are modes of endorsing content. Of course sometimes, epistemic rationality dictates that we quarantine experiential endorsement, so that it does not interact in the usual ways with our other beliefs and with our behavior. Other times, rationality dictates that we don’t quarantine the experience—but we do that anyway, refusing to believe our eyes when we should. In both cases, the same is true of belief. Sometimes epistemic rationality dictates that we give up a belief; other times it dictates that we hold on to it. An opponent might object to the hypothesis that being an endorsement of contents explains what makes irrational dependence relations ill-found beliefs, on the grounds that if true, it shows too much. If being an endorsement of content explains why beliefs are made to conduct ill-foundedness by irrational wishful or fearful thinking, then it is natural to think it also explains why beliefs can be made to conduct well-foundedness by good reasoning. But then by parallel reasoning about experience, experiences can also be made to conduct well-foundedness by good reasoning. That result may seem to go too far. The idea that experiences can be justified by person-level, psychological antecedents is at odds with our practices of epistemic normative assessment. Either that part of our practices should be revised, or else the reasoning is too powerful. Two points can be made in reply. First, suppose it followed from the fact that experiences can conduct well-foundedness by being caused by rational

Selection Effects on Experience

261

processes, that experiences had to be caused by a rational process in order to provide justification. That consequence might be worrisome, since it seems false, and seems to rob experience of the potential to stop regresses of justification. But the consequence does not follow. So the result can respect the plausible idea that for experiences to provide justification, they do not need to be the result of good reasoning, or any other rational process operating over person-level psychological states. Second, our conception of experiences as a-rational should be revised in any case, to account for the rational status of a kind of dependence relationship that can hold between experiences, or more exactly between subexperiences (parts of the same overall experiences)—even if these processes do not guarantee that the dependent subexperience conducts well-foundedness. For instance, suppose you experience x as edible, as a result of experiencing it as orange and jutting out from the tree trunk, together with your background belief that mushrooms with those features are edible. Here your experience of the mushroom as edible could be seen as a rational elaboration of the rest of your experience (x is orange and jutting out) combined with your background belief. To take another example, acrophobes standing on high balconies tend to overestimate their distance from the ground, compared with people standing on the same balcony who are not afraid of heights.27 An acrophobe’s fear might lead her to experience a balcony as being at a distance that is dangerous to fall from, and that subexperience might explain why she experiences the distance to the ground as magnitude D + . Here too, there is something like good reasoning that unfolds within the experience: if a balcony is a dangerous height to fall from, then it is at least D + from the ground. The starting point of this reasoning is fear-induced subexperience of danger, but from there it unfolds into another subexperience of distance in a way that seems rational, not in the normative sense, but in the way that rationalizations do. If experiences can stand in rationalizing relations to each other, then the part of our epistemic normative practices on which we prescind from assessing experiences from having rationally assessable causes should be revised. I’ve argued that the compound experience is downgraded by the role of the red-square hope in selecting only square red things for experience, and in producing an experience with the universal generalization as its content. If this conclusion is correct, it challenges the presumption that the etiology of experience is always an a-rational process. The causal history of an experience, and more generally the dependence relations it stands in to hopes, fears, or prejudice, can be rationally assessable, and more specifically they can be irrational. And when it is, it reduces or eliminates the rational support that the experience provides for believing its contents. I now turn to a different kind of selection effect: anti-selection of experiences for uptake into belief.

27

Stefanucci and Proffitt (2009).

262

4.

Susanna Siegel A N T I - S E L E C T I O N F O R U P TA K E T H AT L E A D S T O C O N F I R M AT I O N B I A S

Here is an ordinary phenomenon. You open the fridge to see whether there is any mustard inside. There is a jar of mustard in the fridge, and you see it when you open the door. But instead of forming the belief that there’s mustard in the fridge, you close the door and find yourself as unsure as you were at the start as to whether the fridge contains mustard. You didn’t come to believe and then forget that the mustard is there; instead, you never formed any mustardbelief in the first place, because your mustard-experience—the experience you have when you see the mustard, which we can suppose does not involve any illusion or hallucination—never fed into any belief-formation process at all. Whereas the mustard case as stated doesn’t specify the factors that led your mustard-experience to get passed over in belief-formation, in the kind of antiselection for uptake that leads to confirmation bias, a prior psychological state does this, such as a prior fear, hope, or belief. For instance, if you strongly hope that there is mustard in the fridge, and that hope “freezes” your mustardexperience when you see the nearly empty jar, leading you to form the belief that the mustard jar is full, then that would be the kind of anti-selection for uptake of interest here. To help us discuss this kind of anti-selection for uptake, some terminology is useful. First, let’s call a belief Bp a bypassing belief, just in case Bp is formed or maintained in part by a process that bypasses an experience that would uncontroversially rationally bear on p, if it were selected for uptake—holding constant the rest of what the subject believes. By definition, a subject will either initially form a bypassing belief upon having a bypassed experience, or will form the belief prior to having that experience and maintain it afterward, in which case her state of believing p is a bypassing belief only in the period after her experience is bypassed. I’ll call experience E pertinent to a bypassing belief Bp, just in case E would rationally bear on p, if it were not bypassed. Since the definition of pertinence involves a counterfactual, to assess whether an experience is pertinent to a belief, we need a way of individuating experiences across worlds. There are different ways to individuate experiences, but the way I’ll be individuating them holds constant phenomenal character and level of attentiveness. A simple example of pertinence can be found in typical perceptual demonstrative beliefs. Consider beliefs that would be natural to express by uttering a sentence of the form that is an F, where the speaker is perceiving what she’s intending to refer to in using the demonstrative.28 In tasks that ask subjects to classify certain targets as belonging to a class A or class B, subjects are supposed to form demonstrative beliefs, of each target they see, that it is either (an) A or (a) B. For instance, if asked to classify what they see as a 28 In addition to perceiving what she intends to refer to, the intention and the perception have to be related in the right way. For discussion, see Siegel 2002.

Selection Effects on Experience

263

tool or a gun, or as red or green, subjects are asked to form demonstrative beliefs, of the targets, that they are a tool (or a gun), or red (or green). Here, the experiences of the things demonstrative referred to are pertinent to the demonstrative beliefs. Anti-selecting one’s experience for uptake differs from discounting it. Discounting an experience is a way of acknowledging a defeater for the experience.29 For instance, suppose you open the fridge and have a mustardexperience, but you know that you are liable to hallucinate mustard, so you don’t form the belief that the fridge contains mustard. Here your mustardexperience would be feeding into the process of belief-formation, and discounted in that process. In contrast, if your mustard-experience is antiselected for uptake, it does not feed into beliefs about what the fridge contains at all. The contrast with discounting can be brought out further by comparing different types of irrational perceptual belief. While some cases of discounting give experiences only as much evidential weight as they deserve, other routes to discounting experience seem doomed to irrationality. For instance, consider an anxious subject who has a pervasive, ominous gut feeling of uncertainty that makes her hold all perceptual beliefs less firmly, across the board. The fact that this discounting originates in anxiety seems to make the process of forming beliefs irrational. Or consider a subject who blankly refuses to draw the natural conclusions from her experience, even though she has no background belief that appearance are misleading, nor any (prior) gut feeling of uncertainty. She sees the jar of mustard in the fridge, it looks the way she expected the mustard to look, but she doesn’t form the belief that there is mustard in the fridge. Like the anxious subject’s beliefs that there probably isn’t any mustard in the fridge (or her low credence in the proposition that the fridge contains mustard), this subject’s agnosticism about whether the fridge contains mustard seems irrational. The anxious subject and the resolutely agnostic subject are both crazy. But they are crazy because of the way they respond doxastically to experience, when the experience is pertinent to the belief. In contrast, when experiences are anti-selected for uptake, they have no more than a causal impact on the subject’s forming a belief to which the experience is pertinent. Anti-selection for uptake is a more likely explanation of certain dynamics of belief than discounting. At time 1, you believe that the mustard jar is in the fridge, mostly full. Shortly afterward, at time 2, you open the fridge, and there’s the mustard, just where you knew it would be, but it is visibly empty and looks empty to you. You see the mustard jar, it looks empty, but when you shut the fridge at time 3, you come away believing exactly what you started out believing: that the mustard jar is in the fridge, mostly full. A likely

29 Defeat of experience is introduced by Pollock in his 1974. See also the discussion of undercutting defeat in Pollock 1986.

264

Susanna Siegel

explanation for your behavior—more likely than crazy discounting—is that you failed to integrate the information about the mustard you got from seeing it with your prior beliefs. A similar explanation could apply to an experiment reported by Triesch et al. (2003), who were investigating change-blindness in the context of a sorting task. Participants who are fixating a block while moving it (in a virtual environment) sometimes fail to report changes in height or color, even when the change is relevant to their motor task, such as moving the tall blue blocks to one place and short yellow ones to another. The changes are less frequently missed when the change is relevant to the task, and that link between reporting changes and task-relevance is the central result of the study. But when subjects fail to report the change, it is as if, after the block was first seen, but before it changed color or height, the channel that normally allows information from experience to guide belief and action is shut off. If that’s what happens, then the stretch of one’s perceptual experience after the block changes color is bypassed in the process of maintaining a prior belief. After the un-noticed change in color or height, the experience plays a central role in making it the case that the subject maintains a belief about the block.30

5.

C A N A N T I - S E L E C T E D E X P E R I E N C E S R E TA I N T H E I R R AT I O N A L R O L E ?

Anti-selection of experience for uptake can generate a type of fragmentation in the mind, in which an experience that could potentially impact which beliefs one forms is prevented from having that kind of impact. Beliefs are sometimes compartmentalized in a way that has this same effect. You might believe you made an appointment with X alone at noon, while also believing that you made an appointment with Y alone at noon, without realizing that you made conflicting appointments. Here you fail to believe the straightforward consequence that you made conflicting appointments, as well as failing to believe the contradiction that you both will and will not meet with X alone at noon. In a similar example offered by Lewis (1982), a resident of New Jersey believes that Nassau Street in Princeton runs north-south and is parallel to the train tracks, and also believes that the train tracks run east-west and are parallel to Nassau Street. Here too, the subject fails to believe a straightforward consequence of their beliefs, since they fail to believe the contradiction that Nassau both does and does not run east-west. The metaphor of mental compartments (or mental fragments) is apt for describing all of these cases, because a type of inferential integration among one’s mental states is missing. The compartmentalization among beliefs that was just described differs from anti-selection for uptake in several ways. First, your belief that you

30 The Triesch experiment itself does not establish that the behavior it elicits is best explained by anti-selection for uptake, but this hypothesis is an open possibility.

Selection Effects on Experience

265

have an appointment with X at noon may be accessed by some strands of reasoning while not being accessed by other strands. For instance, the belief that you’ll see X at noon may ground your belief that you’ll soon see X and can give her back her umbrella, while being unaccessed by a strand of reasoning that would lead you to discover your conflicting appointments. In contrast, anti-selected experiences may be unaccessed by any strand of reasoning. Each of your appointment beliefs belongs to a separate compartment with its own internal rational complexity, whereas an anti-selected experience may belong to a compartment with no internal rational complexity to it. Second, in the Nassau Street case, the contents of the resident’s beliefs are inconsistent. The appointments case may involve contradictory contents as well, such as I will meet with X alone at noon and I will meet with Y alone at noon. Whether the belief states in these cases are inconsistent is a substantive question to which we’ll return shortly. In contrast, the content of an antiselected experience and its bypassing belief need not be inconsistent. Third, the subjects in the belief cases seem to fail to believe straightforward logical consequences of other things that they believe, and this failure seems due to the compartmentalization. In contrast, although the subject of experiences that are anti-selected for uptake will not believe the contents of that experience, or any of their consequences on the basis of that experience (since she does not believe anything on its basis), she may believe those contents on other grounds. Despite these differences, anti-selection of experience and compartmentalization of belief raise the same epistemological question: what rational bearing do the unintegrated states have, if any, on the rational status of beliefs that they cannot impact, so long as they remain compartmentalized? We can apply this general question to our cases of anti-selection of experiences for uptake by asking: When an anti-selected experience is pertinent to a bypassing belief, what impact does the anti-selection have, if any, on the rational status of the belief? There are two main options: no impact or some impact. Either the pertinent experience is rationally inert vis-à-vis the bypassing belief (the inertness option), or it isn’t (the impact option). The inertness option licenses combinations of experience and beliefs which, in a unified mind, would be irrational. In contrast, on the impact option, bypassed experiences retain their rational force with respect to all beliefs to which they are pertinent. Here, if a combination of experience and belief would be irrational if the experience was selected for uptake, it remains irrational, even when the experience is bypassed. On a global version of the impact option, all bypassed experiences retain whatever rational force that they would have if they were selected for uptake, and on a global version of the inertness option, anti-selection for uptake always removes any rational force that the bypassed experience would have, were it not bypassed. The impact and inertness options interact in a straightforward way with evidentialism about rational belief. In an evidentialist framework, the difference between the impact and inertness options directly translates into two

266

Susanna Siegel

positions on whether perceptual bypass is a means of ignoring evidence.31 The inertness option holds that anti-selecting a pertinent experience cannot make any corresponding bypassing beliefs irrational. In its evidentialist form, the inertness option entails that experiences provide evidence only if they are selected for uptake (not necessarily uptake into belief, but uptake into even the most minimal forms of inferential processes). In contrast, the impact option holds that anti-selection of a pertinent experience can make corresponding bypassing beliefs irrational. In its evidentialist form, the impact option entails that anti-selection of experience can be a means of ignoring evidence. I think a prima-facie case can be made for a version of the impact option, independently of the evidentialist framework. One could try to defend the global impact option, by appealing to a putative requirement of ideal rationality. According to the putative rational requirement, an ideally rational subject would always respond appropriately to her experiences, and perceptual impact thus leaves subjects one step closer to that ideal than anti-selection, where the subject does not have the opportunity to respond doxastically to her experience at all, either rationally or irrationally. According to this line of thought, anti-selection of experience that is pertinent to what the subject believes is a departure from ideal rationality. We can distinguish the line of thought just sketched from the idea that ideal rationality requires you to consider all your evidence in forming doxastic attitudes. What’s at issue between the global impact option and its opponents is whether bypassed experiences always have the status of evidence. The strategy sketched above would need to provide grounds for the idea that antiselected experiences always have this status. In contrast to the global impact and inertness options, in principle antiselected experiences might differ in whether they are rationally inert, depending on the etiology of the anti-selection. Here, certain routes of anti-selection for uptake make a case for at least a moderate impact thesis. In particular, when experiences are anti-selected for uptake in a way that leads to confirmation bias, it is plausible that the bypassed experience retains its rational force vis a vis the beliefs to which it is pertinent. Consider a subject S who comes to believe that the mustard jar in the fridge was full on the basis of Q’s testimony, and Q’s testimony gives S good grounds for this belief. But suppose that S overvalues Q’s opinion about nearly everything, and tends to give Q’s opinions more weight than they deserve, because in general, S wants Q to be right. S’s dispositions to overvalue Q’s opinion cause S (or causes S’s belief-formation system) to anti-select her experience when she opens the fridge and sees the nearly empty jar of mustard inside, with the result that after looking in the fridge, S keeps on believing what she believed before she looked in the fridge—that the mustard jar is full. Here anti-

31 At least, it can be so translated, given the assumption that experiences typically provide evidence for external-world beliefs.

Selection Effects on Experience

267

selection happens in the service of a type of confirmation bias, and a type that perceptual experience, in general, is well-suited to correct. In this type of case, S’s post-fridge mustard-belief seems epistemically less well grounded than the pre-fridge belief was, even though both beliefs were based on Q’s good testimony. A straightforward explanation of why it is less well-grounded is that S fails to heed the excellent reason, provided by her bypassed experience, to believe that the mustard jar is nearly empty. If her bypassed experience provides her with this type of reason for belief, then the impact option is correct. A second consideration targets the inertness option. In response to the cases of compartmentalized beliefs discussed earlier, one might hold that those beliefs are rationally inert with respect to one another, by virtue of the fact that they are never activated at the same time. For instance, Lewis’s reason for thinking that the New Jersey resident does not believe a contradiction is that the east-west belief and the north-south belief never “come into action” at the same time.32 And if it is part of believing a proposition that one is disposed to act in ways that would satisfy what one desires if the proposition is true, then it is hard to see what ways of acting those would be, if the proposition has the form P¬-P. (Here we return to the question set aside earlier about whether the belief states in the Nassau Street case are inconsistent). It seems compatible with Lewis’s own description of the Nassau Street case that even when the belief that p is dormant, it rationally impacts the activated belief that not-p. For instance, if the belief that p is well-grounded, whereas the belief that not-p is not, then the belief that p would make it less rational to believe not-p. That position would give beliefs rational impact on one another, even when they are compartmentalized. So Lewis himself seems uncommitted on the question whether compartmentalization per se neutralized rational impact that the belief in p would otherwise have with respect to a belief in not-p. But someone might hold that compartmentalized beliefs are always rationally inert with respect to one another, due to the fact that they never both “come in action” at the same time. What exactly is it for a belief to come into action? One option is that a belief comes into action, only while the subject is behaving as if the belief is true, relative to her desires (or to some of her desires). Another option is that a belief comes into action, only while the subject is (right then!) drawing inferences from it. We can set aside this important question, though, because on either of these ways of elaborating what it is for a belief to come into action, anti-selected experiences do not seem to come into action selectively, the way compartmentalized beliefs do. In our original mustard example, for example, the anti-selected mustard-experience fails to guide behavior at all. So if the fact that compartmentalized beliefs only selectively come into action is

32 “Different fragments came into action in different situations, and the whole system of beliefs never manifested itself all at once.” (Lewis, p. 436)).

268

Susanna Siegel

supposed to underwrite their rational inertness with respect to beliefs in other compartments, then this feature is not available to underwrite the inertness option for bypassed experiences. In a Triesch-style example of anti-selected experiences, the experience may feed into some beliefs but not into others. For instance, the experience may feed into beliefs about where the block is on its path from the moving belt it was plucked from to the pile, without feeding into beliefs about what color it is. This description of the case assumes that there is a single experience representing both color and movement, rather than ontologically separate experiences of motion and color, but it is convenient to use the labels “colorexperience” and “motion-experience” for two aspects of an overall experience that represents the block changing its color in mid-path. We could imagine a case in which, due to limited cognitive resources, after the color change, the motion-experience feeds into a belief about its trajectory, only if the colorexperience fails to feed into beliefs about its color, resulting in a color-belief that is not properly updated. In this type of case, the bypassed experience and the bypassing belief both guide behavior, just as both fragmented beliefs do. But unlike the fragmented beliefs, the bypassed experience guides behavior at the same time as the belief that bypasses it. So the feature of fragmented beliefs that might be taken to license rational inertness—the idea that they do not both “come into action” at the same time—does not apply to the case of anti-selected experiences.

CONCLUSION

Can selection effects on experience influence their rational role? I’ve suggested that some forms of selecting objects for experience can epistemically downgrade those experiences—for example when the selection effect is explained by its role in confirmation bias. For these cases, the proposed answer to the title question is Yes. I also suggested that when experiences are anti-selected for uptake, they can retain their rational force. For these cases, the proposed answer to the title question is No.

REFERENCES

Balcetis, E. and Dunning, D. 2006. “See What You Want to See: Motivational Influences on Visual Perception” Journal of Personality and Social Psychology 91(4), 612–25. Barrick, C. 2002. “Color Sensitivity and Mood Disorders: Biology or Metaphor?” Journal of Affective Disorders 68, 67–71. Byrne, A. 2009. “Experience and Content” Philosophical Quarterly 59(236), 429–451. Carey, S. 2009. The Origin of Concepts. New York: Oxford University Press. Chalmers, D. 2006. “Perception and the Fall from Eden”. Reprinted in The Components of Consciousness. New York: Oxford University Press, 2012.

Selection Effects on Experience

269

Churchland, P. 1988. “Perceptual Plasticity and Theoretical Neutrality” Philosphy of Science 55, 167–87. Conee, E. and Feldman, R. 2001. “Internalism Defended”. Reprinted in Feldman and Conee 2004, 53–82. 1998. “The Generality Problem for Reliabilism” Philosophical Studies 89(1), 1–29. Feldman, R. and Conee, E. 1985. “Evidentialism.” Reprinted in Feldman and Conee 2004, 83–107. 2004. Evidentialism: Essays in Epistemology. New York: Oxford University Press. Fodor, J. 1983. Modularity of Mind. Cambridge: MIT Press. Foley, R. and Fumerton, R. 1982. “Epistemic Indolence” Mind 91, 361, 38–56 Glüer, K. 2009. “In Defence of a Doxastic Account of Experience” Mind and Language 24(3), 297–327. Goldin, C. and Rouse, C. 2000. “Orchestrating Impartiality: The Impact of ‘Blind’ Auditions on Female Musicians” The American Economic Review 90, 715–41. Hansen, T. et al. 2008. “Color Scaling of Discs and Natural Objects at Different Luminance Levels”. Visual Neuroscience 23, 603–10. Huemer, M. 2013. “Epistemological Asymmetries between Belief and Experience” Philosophical Studies 162(3), 741–748. Kaplan, M. 1996. Decision Theory as Philosophy. Cambridge: Cambridge University Press. Kelly, T. 2002. “The Rationality of Belief and some other Propositional Attitudes” Philosophical Studies 110, 165–96. Levin, R. and Banaji, M. 2006. “Distortions in the Perceived Lightness of Faces: The Role of Race Categories” Journal of Experimental Psychology 135(4), 501–12. Lewis, D. 1982. “Logic for Equivocators” Noûs 16(3), 431–41. Lyons, J. 2011. “Circularity, Reliability, and the Cognitive Penetrability of Perception” Philosophical Issues 21(1), 289–311. McGrath, M. 2013a. “Cognitive Penetration and Bad Basis Counterexamples” in Tucker, ed. (2013). Seemings and Justification: New Essays on Dogmatism and Phenomenal Conservatism. McGrath, M. 2013b. “Siegel and the Impact for Epistemological Internalism” in Philosophical Studies, 162(3), 723–32. Martin, M. G. F. 2002. “The Transparency of Experience” Mind and Language 4(4), 376–425. Payne, B. K. 2001. “Prejudice and Perception: The Role of Automatic and Controlled Processes in Misperceiving a Weapon” Journal of Personality and Social Psychology 81(2), 181–192. Pollock, J. 1974. Knowledge and Justification. Ithaca: Cornell University Press. 1986. Contemporary Theories of Knowledge. Lanham, MD: Rowman and Littlefield. Popper, K. 1959. The Logic of Scientific Discovery. London: Hutchinson. Proffitt, D. 2006. “Embodied Perception and the Economy of Action” Perspectives on Psychological Science 1, 110–22. Raftopolous, A. 2009. Cognition and Perception: How do Psychology and Neural Science Inform Philosophy? Cambridge, MA: MIT Press.

270

Susanna Siegel

Siegel, S. 2002. “The Role of Perception in Demonstrative Reference” Philosophers’ Imprint, 2(1), 1–21. 2011. “Cognitive Penetrability and Perceptual Justification” Nous 46(2), 201–222. 2013a. “The Epistemic Impact of the Etiology on Experience” Philosophical Studies, 162(3), 697–722. and Silins, N. 2013. “The Epistemology of Perception.” In The Oxford Handbook of the Philosophy of Perception. Edited Mohan Matthen. Oxford: Oxford University Press. Stefanucci, J. K. and Proffitt, D. R. 2009. “The Roles of Altitude and Fear in the Perception of Height” Journal of Experimental Psychology, Human Perception and Performance 35(2), 424–38. Steinpreis, R., et al. 1999. “The Impact of Gender on the Review of Curriculum Vitae of Job Applicants and Tenure Candidates: A National Empirical Study” Sex Roles 41, 509–28. Sturgeon, S. 2000. Matters of Mind: Consciousness, Reason and Nature. London: Routledge. Triesch, J. et al. 2003. “What You See is What You Need” Journal of Vision 3, 86–94. Tucker, C., 2013. Seemings and Justification: New Essays in Dogmatism and Phenomenal Conservatism. Oxford: Oxford University Press. Watzl, S. 2010. The Significance of Attention. Columbia University PhD Thesis. 2011. Attention as Structuring of the Stream of Consciousness. In Attention: Philosophical and Psychological Essays. Edited by C. Mole, D. Smithies, and W. Wu. New York: Oxford University Press.

SYMPOSIUM

This page intentionally left blank

10. Knowledge as a Mental State Jennifer Nagel One striking feature of research into mental state ascription or ‘mindreading’ is the extent to which it has involved cooperation between psychology and philosophy. Contemporary empirical work on mindreading is often traced back to a 1978 target article in comparative psychology, an article which raised the question of whether chimpanzees attribute mental states to themselves and others (Premack and Woodruff, 1978). Three of the most influential responses published with that article were written by philosophers, each of whom drew attention to the importance of probing the capacity to represent states of ignorance and false belief (Bennett, 1978; Dennett, 1978; Harman, 1978). In the decades that followed, philosophers made significant contributions to the theoretical frameworks guiding empirical research into mental state ascription (e.g. Goldman, 2006; Nichols and Stich, 2003), while psychologists made significant contributions to philosophical debates about the nature of our access to our own mental states and those of others (e.g. Gopnik and Wellman, 1992; Perner, 1991). Major progress has also been made in work co-authored by philosophers and psychologists (e.g. Apperly and Butterfill, 2009; Gallese and Goldman, 1998). In general, disciplinary boundaries between philosophers and psychologists working on mindreading do not seem to put these two groups in opposition to each other. However, there is one question on which most philosophers and psychologists appear to be divided, and it is arguably a significant question in the theory of mental state ascription. This is the question of whether knowledge is a mental state. On the psychology side of the line, the common answer to that question is positive. Developmental, social, cognitive, and comparative psychologists explicitly and without hesitation classify knowledge as a mental state, alongside states of belief, desire, intention, perception and feeling (Apperly, 2011; Baron-Cohen et al., 1994; Call and Tomasello, 2008; De Villiers, 2007; Epley and Waytz, 2009; Heyes, 1998; Keysar, 2007; Lin, Keysar, and Epley, 2010; Premack and Woodruff, 1978; Saxe, 2005; Sodian, Thoermer, and Dietrich, 2006; Wellman and Liu, 2004). On the philosophy side of the line, the answer is almost unanimously negative. Knowledge is conspicuously absent from the lists of mental states presented by many prominent contemporary philosophers of mind (Davies and Stone, 2001; Goldman, 2006; Gordon, 1986; Stich and Ravenscroft, 1994). It is not simply by oversight that philosophers have failed to include knowledge in their lists: there are systematic reasons for the omission, to be discussed in what follows. In addition, philosophers have

274

Jennifer Nagel

presented various explicit arguments aimed at establishing that knowledge is not a mental state (Fricker, 2009; Magnus and Cohen, 2003; Molyneux, 2007). Although the current climate in philosophy is largely hostile to the idea that knowledge is a mental state, there is no historical incompatibility between this idea and a philosophical mindset: philosophers as different as Plato and Locke classified knowledge this way. In present day philosophy, there are some philosophers of mind who continue to list knowledge among the mental states (e.g. Carruthers, 2009), and there is one sustained defense of the recognition of knowledge as a mental state, in the work of Timothy Williamson (1995; 2000; 2009). This position has met considerable resistance from other philosophers. In responding to Williamson’s identification of knowledge as mental state, Anthony Brueckner is no doubt reporting the common sentiment among philosophers when he observes that ‘to many this is a highly counterintuitive claim’ (Brueckner, 2002, 197). When disciplines seem to disagree, it is natural to wonder whether the dispute is merely terminological. In this case one might suspect that the two sides currently mean something different by “mental state” or by “knowledge.” Certainly there are a number of words whose standard technical meanings are quite different in philosophy and psychology, sometimes for theoretical reasons and sometimes just because a different sense of the word is more commonly in play in a given discipline (e.g. “normative”; “modal”). Section 1 of this paper argues that, notwithstanding certain relatively superficial differences in usage, the relevant branches of the two disciplines generally have shared targets in their sights, both when they speak of mental states and when they speak of knowledge. If the debate is not merely verbal but substantive, at least one side is mistaken: this paper argues that the mainstream philosophers are the ones who have gone wrong. Section 2 discusses philosophical arguments on the matter and section 3 discusses empirical work. The difference between mainstream philosophers and psychologists is diagnosed as arising from a difference in the way the two sides see the relationship between the concepts of knowledge and belief. All parties to the dispute see these concepts as closely linked, and agree that belief is a mental state. Where they disagree is on the question of whether knowledge or belief is a more basic concept: mainstream contemporary philosophers take the concept of belief as their starting point, and see the concept of knowledge as something more complex, generated by adding various nonmental conditions to the uncontroversially mental state of belief. Williamson joins the psychologists in taking the concept of knowledge to be more basic: like most psychologists of mental state ascription, he sees the concept of belief as something derived from the concept of knowledge, where knowledge is from the start seen as a mental state in its own right. A central motivation for the mainstream philosophical approach appears to be the thought that our understanding of action is fundamentally dependent on belief attribution; in this view, the belief component of knowledge is what really matters to our natural understanding of what others do, and it

Knowledge as a Mental State

275

is uneconomical to admit knowledge-based explanations of action alongside belief-based ones. Advocates of the mainstream approach also appeal to a metaphysical intuition that mental states must be a function of strictly local conditions in the agent; observing that the truth of a known proposition is not typically something contained within the agent, they conclude that it would be metaphysically odd to characterize knowledge as a mental state. Against the first motivation for the mainstream view, it is not assumed in psychology that our understanding of other agents is ultimately belief-based; indeed, the capacity to attribute mere belief to others seems to be considerably harder to master and deploy than the capacity to attribute knowledge. This paper examines a range of work in developmental and comparative psychology supporting the conclusion that an ability to track what others would know seems to be the precondition, rather than the product, of an ability to track what they would believe. Mature social cognition is also relevant. By the time they reach maturity, human beings are able to recognize both knowledge and belief intuitively, but it seems the availability of belief ascription does not make knowledge ascription redundant in our understanding of the actions of others. On the metaphysical point, there is reason to resist the intuitively alluring thought that mental states must be localized within the agent: this sense of localization is arguably a natural illusion, like the parallel illusion in “naïve physics” which makes us feel that the motions of inanimate objects are determined by their own inner “impetus” and local interactions with what they contact, even though our intuitive calculations of those motions themselves make use of information about environmental factors such as gravitation. The intuitive sense that some human agent knows something is produced by intuitive mechanisms that make calculations about the target person and his relationship to the environment: for example, mechanisms that automatically track the target’s direction of gaze and note which objects lie in his visual field (Samson, Apperly, Braithwaite, Andrews, and Scott, 2010). Because these calculations are intuitive, we do not have explicit awareness of how they are made, and can have the intuitive sense that the target’s resulting mental state is “internal” to him, like the intuitively felt “internal impetus” of a falling object. Our actual use of the attributed mental state of knowledge in predicting and explaining the target’s subsequent actions does not, however, depend on that illusory impression: just as our intelligent interactions with falling objects incorporate our real capacity to take gravitation into account, so also our intelligent interactions with other persons incorporate our real capacity to take their relationships with the environment into account. The fact that the state of knowing incorporates a relationship with the environment does not disqualify it from counting as a state which is fundamental to our intuitive understanding of other intelligent beings. Evidence from social, developmental, and comparative psychology seems to support the view that knowledge is naturally seen as a mental state in its own right, and not as a composite of belief and non-mental factors. However,

276

Jennifer Nagel

one might grant that knowledge is naturally seen this way and still wonder whether knowledge really is a mental state. The final section of the paper discusses the relationship between our evidence about the way we are naturally inclined to see knowledge and our conclusions about the kind of state knowledge really is.

1.

W H AT T H E T W O D I S C I P L I N E S M E A N B Y “ M E N TA L S TAT E S ” AND “KNOWLEDGE”

One thing philosophy and psychology have in common is that neither discipline has produced a comprehensive and settled account of the nature of mental states or of knowledge. Ongoing controversy over these topics within each discipline makes it difficult to establish conclusively that philosophers and psychologists generally aim to speak of the same thing when they use the expression “mental state” (or the word “knowledge”). As each discipline produces and criticizes various successive “mental state” theories, for example, it is a live question whether there is a single phenomenon that researchers on both sides of the disciplinary divide are trying to explain, or whether we and our colleagues across campus are stalking different prey. Because active controversies could mask subtle but genuine differences between the disciplines, this section will not hope to demonstrate that philosophers and psychologists all have precisely the same thing in mind for each of our key expressions, but will instead try to show extensive similarities in the way “mental state” and “knowledge” are understood in the two disciplines, to shift the burden of proof onto those who claim that the general interdisciplinary disagreement over knowledge being a mental state is merely verbal. To give some rough indication of what they mean by “mental state,” both psychologists and philosophers often provide lists of examples. Setting aside the inclusion or omission of knowledge, these lists are strikingly similar across the disciplines. Psychologists and philosophers are widely agreed that the following are naturally classified as mental states: beliefs, desires, intentions, emotions, perceptions, and sensations such as pain and hunger. There is also wide agreement in both disciplines about the relationship between mental states and agency: we take1 a person’s actions (as opposed to mere reflexes

1 According to some philosophers, our natural inclinations to see human actions as caused by mental states are fundamentally mistaken: eliminitavists like the early Paul Churchland argue that the basic terms of folk psychology do not refer to anything with real causal powers (Churchland, 1981). As this is now very much a minority position within philosophy, and even Churchland himself has come to admit the reality of some mental states, eliminitavism will be set aside in what follows, and the reality of mental states such as belief and desire will be assumed as a starting point. A similar initial commitment to the reality of mental states is common in psychological work (see e.g. Apperly, 2011, p.2). One can of course admit the fallibility of folk psychology, and agree that the nature of those states cannot be read straight off our natural patterns of mental state ascription. Furthermore, even those who are still inclined towards an eliminitavist reading of much mental state discourse may still find some interest in

Knowledge as a Mental State

277

or automatic functions such as digestion) to be dependent on2 his or her epistemic and motivational mental states. There are well established systematic relationships between these mental states: for example, people generally intend to do what they believe will satisfy their desires. Given the universality of these relationships among mental states, a capacity to register the mental states of others—for example, what they believe, desire and intend— enables us to explain and predict what they will do. We do not simply expect old patterns of overt behavior to be repeated; because we can grasp the underlying causes of those patterns, we can anticipate how an agent will act and interact with us in novel circumstances. This social understanding may be imperfect in various ways, but it is strong enough to support successful communication, competition, and cooperation across a wide range of social situations. Of course there are a great many theories about the nature of mental states both in psychology and in philosophy, and some of these theories may have discipline-specific or idiosyncratic aims, but in broad brushstrokes recent philosophical and psychological theories about mental states generally seem to aim at capturing similar phenomena. The question of whether philosophers and psychologists mean the same thing by “knowledge” is more difficult. One of the characteristics of knowledge to which philosophers are particularly attentive is the truth condition: it is almost universally accepted in philosophy that knowing, unlike mere believing, is an attitude that one can have only to truths. One strand of evidence for this is linguistic: “know” is a factive verb, a verb that licenses an inference to its complement clause. From the truth of a proposition of the form “S knows that p” it would follow that p is the case. Nonfactive verbs such as “believes” do not license such inferences. If one follows philosophers such as Williamson in extending the terminology of linguistics over to epistemology, one can describe knowing itself as a factive propositional attitude, in contrast to nonfactive propositional attitudes such as thinking, hoping, and conjecturing (Williamson, 2000). Unlike the identification of knowledge as a mental state, the identification of knowledge as a factive state is not considered particularly controversial in philosophy: even the rare philosopher who doubts the strength of the linguistic evidence for the factivity of the verb “to know” does not see this as a reason to challenge the classification of knowledge itself as a factive state (Hazlett, 2010). Psychologists might at first seem not to take knowledge to be a factive state. For example, in the literature on calibration or the relationship between confidence and accuracy, some papers discuss the “accuracy of knowledge” the question of whether the (apparent) state of knowledge is a part of that domain, or whether knowledge ascription is a form of mental state ascription. 2 There are various views about the character of this dependence: many regard it as straightforwardly causal, but on some views (e.g. Dancy, 2000) mental states are characterized as necessary background conditions rather than causes of action. For ease of exposition I will describe the dependence as causal in what follows, but I believe that much of what I say could be reworded to fit views such as Dancy’s.

278

Jennifer Nagel

in various domains, by which the authors mean the percentage of true out of total answers volunteered (e.g. Perfect, Watson, and Wagstaff, 1993). Epistemologists might observe that the accuracy of knowledge, strictly speaking, would always have to be 100 percent: any false answer volunteered would be mere belief at best, and fail to count as knowledge in the first place. But the charitable reader can readily make sense of the calibration literature by penciling in scare quotes where appropriate, for example by reading various calibration studies as probing something more like the accuracy of what the research subjects “know” (or perhaps take themselves to know). When the line between knowledge and what is taken for knowledge is particularly important and there is a real risk of confusion, psychologists do take more care in marking the distinction. In calibration research as elsewhere it is understood, and sometimes even underscored, that strictly speaking what is known must be true (e.g. Koriat, 1995). In the branches of psychology most directly engaged with mental state attribution, there is greater self-consciousness about the relationship between truth and knowledge, both because those who are examining the linguistic expression of mental state attribution take the factivity of the mental state verb “know” to be a core component of its meaning (e.g. Abbeduto and Rosenberg, 1985), and because much attention is paid to our competence for recognizing false beliefs, whether this competence is expressed linguistically or in other forms of behavior. Indeed, the threat of an interdisciplinary difference sometimes appears to loom larger in the treatment of falsity than factivity: in articles about the relative difficulty of attributing knowledge and false belief, the terms “belief” and “false belief” occasionally seem to be used interchangeably (e.g. in Kaminski, Call, and Tomasello, 2008). But this pattern of usage is also better explained by its expediency than by any interdisciplinary disagreement about the possibility of true belief. In experimental studies of the capacity to discriminate knowledge from mere belief, the easiest way to generate a situation involving a belief falling short of knowledge is usually to devise a situation in which the observed agent’s belief is false. Because some of the clearest tests of the capacity to represent belief (as contrasted to knowledge) are tests involving false belief, psychologists of mental state ascription sometimes gloss over the distinction between belief recognition and false belief recognition. This occasional elision in usage should not be taken to suggest any positive commitment to the thought that beliefs as such must be false, or to the thought that all true beliefs would have to count as knowledge. The relatively subtle capacity to distinguish knowledge from mere true belief (for example, belief generated by lucky guessing) emerges later than the capacity to distinguish knowledge from false belief: true belief that falls short of knowledge is surprisingly hard to ascribe. After they are capable of recognizing false belief, children pass through a stage in which they tend to describe a target as knowing if his action leads to successful performance—say selecting the box with the treat in it—even if the target had no evidential basis for his choice,

Knowledge as a Mental State

279

for example because he was absent when the treat was placed in the box (Miscione, Marvin, O’Brien, and Greenberg, 1978). The capacity to recognize the difference between knowledge and accidentally true belief is often ignored in studies of young children and non-human primates who fail to make even the more rudimentary distinction between knowledge and false belief.3 However, the more advanced capacity has also been studied heavily enough that there is no reason to think that psychologists of mental state ascription are generally oblivious to the distinction between knowledge and mere true belief (e.g. Johnson and Wellman, 1980; Moore, Bryant, and Furrow, 1989; Pillow, Hill, Boyce, and Stein, 2000; Sodian and Wimmer, 1987). A last reason to worry that psychologists may have something different in mind when they speak of “knowledge” is that they sometimes offer what seem to be the sort of analyses of knowledge that would meet considerable resistance in philosophy departments. For example, Bartsch and Wellman offer the following observation: ‘Adults use know to refer to a belief that is felt to be justified, assumed to be true, or that enjoys markedly higher conviction than one described by think’ (1995, 40). Meanwhile, according to Miller, Hardin, and Montgomery, “Young children may appreciate some of the criteria for ‘know’ (a true belief for which there is evidence) without fully understanding all of the criteria (a belief that is not only true but held with certainty)” (2003, 350). Some psychologists give even more explicit endorsement of theories of knowledge rejected in philosophy since Gettier (1963); for example, by claiming directly that “knowledge means justified true belief” (Reber and Unkelbach, 2010, 570). Others amend the classical analysis, but not in a manner that contemporary philosophers are inclined to accept as fully satisfactory, for example by simply adding a causal condition to the conditions of truth, justification, and belief (Dienes and Perner, 1999, 739; Perner, 1991, 304). Because the analysis-of-knowledge project commanded the attention of philosophers for decades after Gettier, counterexamples to all of these analyses will spring readily to mind for contemporary epistemologists. If there 3 One might complain that it is still controversial for psychologists to describe the more rudimentary capacity as distinguishing between knowledge and false belief, as opposed to true belief and false belief; perhaps the cases of accidentally true beliefs have special peculiarities that make it harder to handle them appropriately. But it is not in general true that false beliefs are harder to attribute than true ones; for example, children who are capable of recognizing false belief pass through a stage in which they explicitly represent others who in fact simply lack knowledge as holding false, rather than true, beliefs about propositions known to the child (Ruffman, 1996). Furthermore, the prototypical cases of knowing that the child recognizes (in particular, seeing that something is the case), are the same as those we recognize as adults. Although the adult has a deeper understanding of knowledge, insofar as the child’s early use of “know” is typically a response to what we also recognize as knowledge, she should charitably be seen as referring to knowledge, just as the ancient Romans are appropriately seen as having referred to gold with their word “aurum,” notwithstanding their failure to recognize it as the element with atomic number 79, and notwithstanding their relatively limited accuracy in sorting gold from its look-alikes. Thanks to Hagit Benbaji, Tamar Gendler, and Michael Thompson for pressing me on this point.

280

Jennifer Nagel

were evidence that psychologists wanted to insist on their analyses in the face of these counterexamples, one might worry about a real difference in what is meant by “knowledge” in the two disciplines. But an examination of the psychologists’ grounds for their general assertions about knowledge does not reveal any unorthodox responses to the currently standard epistemological counterexamples; rather, the relevant peculiar cases from the epistemology literature (cases involving deviant causal chains and the like) have not been the targets of theoretical attention in psychology, no doubt in part because such cases are ecologically rare. And, despite some perhaps overconfident phrasing—like the claim to have captured “all of the criteria” for knowledge— the general claims in the last paragraph are not typically treated even by the psychologists who advance them as definitive. In practice, psychologists do not, for example, insist that a mental state pass some test for “markedly higher conviction” or a special feeling of certainty before they are willing to classify it as knowledge. Everyday cases of perceptual observation, for example, are in standard practice taken to issue in knowledge without any prerequisite that they produce a particularly high level of confidence. So although the just-quoted general claims about knowledge may appear to philosophers as questionable analyses, they figure in their original contexts as rough empirical generalizations, typically aiming to deliver a quick outline of what it is that paradigmatic instances of knowing have in common. Reassuringly enough, lists of paradigm cases of knowledge are the same in the psychology of mental state ascription as they are in contemporary epistemology: perceptual observation in good conditions, sound inference, memory, and appropriate cases of testimony.4 If psychologists sometimes offer a questionable general recipe for what unites these cases, they are doing something that is often done in philosophy as well. When a philosopher advances what he describes as an analysis of knowledge, and this analysis is confronted with what we would ordinarily call an effective counterexample, it is more charitable to see the philosopher as having said something inaccurate about knowledge, than to see him as having attempted to introduce a new sense of the word “knowledge.” The same charity should be extended to psychologists: except in cases where for example it is clear that they mean to speak of “what is commonly taken to be knowledge” rather than knowledge itself, it is only fair to take their talk of knowledge at face value, as anchored in the same paradigmatic examples that drive philosophical work on knowledge, and as referring to a shared topic of pre-theoretical interest.

4 Philosophical skeptics have of course raised worries about whether knowledge is ever actually instantiated in any of these ways, but even they would not deny that these are the types of cases that people commonly take to be instances of knowing (Unger, 1971). Most contemporary epistemologists are not skeptics, and this paper will not address the challenge of skepticism. However, even skeptics who doubt the propriety of our ordinary ascriptions of knowledge may still have some interest in the question of whether what we ordinarily take to be knowledge is or is not naturally seen as a type of mental state.

Knowledge as a Mental State

281

If there are any underlying terminological differences between philosophy and psychology in their ordinary use of “knowledge” and “mental state,” these possible differences do not seem to be large enough to explain the starkly opposing positions generally taken in the two disciplines on the question of whether knowledge is a mental state. The next two sections examine the substance of what seems to be a substantive disagreement.

2.

PHILOSOPHICAL ARGUMENTS ABOUT WHETHER K N O W L E D G E I S A M E N TA L S TAT E

Most contemporary philosophers deny that knowledge is itself a mental state, but are happy to grant that knowledge incorporates a mental state, namely the state of belief.5 On the currently standard philosophical view, in order to know a proposition p, an agent must not only have the mental state of believing that p, but various further independent conditions must also be met: p must be true, the agent’s belief in p must be justified or well-founded, and so forth. At least some of these further conditions—most importantly, the truth of p, where p is some proposition about the external world6 —would clearly be nonmental conditions. States and events that are composites of the mental and non-mental are not naturally classified as mental: murder, for example, is not a mental event. Although every murder incorporates the mental condition of an intention to kill (indeed this mental condition, mens rea, is the part of the essence of murder that distinguishes it from involuntary manslaughter), an event cannot be a murder without also incorporating the non-mental condition of someone’s actual death. The inclusion of this non-mental conjunct disqualifies the event considered as a whole from counting as purely mental. Similarly, on the currently orthodox philosophical view, because it is thought to include the nonmental component of the truth of propositions about external reality alongside the mental component of belief, the state of knowledge is considered composite rather than purely mental. Enormous effort has been directed towards developing an account of the composition of this composite state. Epistemologists since Gettier have struggled hard to produce a closed list of factors that would need to be added 5 Some philosophers add further mental conditions, maintaining that the mental state or psychological component of knowledge is not simply belief, but particularly confident belief, or perhaps justified belief, where “justified” is understood in mental terms; these elaborations of the standard view will be discussed in due course. A few have argued that the state of knowledge positively excludes the state of belief; this type of position is now rarely held, and will not be discussed here (for criticism of such views, see Williamson, 2000, pp. 42–4). 6 Some of the propositions we might be said to know or believe are themselves about the domain of the mental; strictly for ease of exposition I am setting those cases aside in what follows. Of course the truth of a proposition about the mental is not itself a mental condition, but it is easier to see that the composite view of knowledge involves an element that is entirely independent of the mental realm if we focus on knowledge of propositions about the external world.

282

Jennifer Nagel

to true belief (or relationships in which truth and belief would need to figure) in order to constitute the state of knowledge, evidently without success (for surveys, see Plantinga, 1993; Shope, 1983; Steup, 2008). Increasingly complex analyses have been crafted to sidestep intuitive counterexamples to earlier analyses, but new counterexamples have steadily emerged to confront the new analyses (and perhaps not only steadily but inevitably, as argued in Zagzebski, 1994). Other, simpler efforts at the analysis of knowledge have collapsed into circularity, managing to analyze knowledge only by adding to true belief a condition which ultimately incorporates knowledge itself. Certain efforts to explain knowledge in terms of “well-foundedness,” in particular, look problematic in this regard, as do appeals to conditions such as epistemic adroitness or competence (e.g. Sosa, 2007; Turri, 2011), insofar as it is hard to see how to understand the relevant concepts of adroitness or competence without appeal to the notion of knowing how to do something (cf. Stanley, 2011; Stanley and Williamson, 2001). Although there is no simple formula for the ascent from belief to knowledge, the intuitive recognition of knowledge is supposed to start with the recognition of belief. According to the view that is currently standard in philosophy, in attributing knowledge of a proposition to some target agent, we attribute the mental state of belief, while simultaneously judging the agent’s situation to satisfy various non-mental conditions, in particular, the condition that the proposition believed be true, plus whatever further mental and nonmental features are necessary to distinguish knowledge from mere true belief. Attributions of knowledge, in this view, play no special or ineliminable role in our ability to explain and predict the actions of others.7 The purely mental factor of believing is what serves as the key input to our understanding of what the agent does, together with some representation of that agent’s desires. The guiding principle of the current orthodoxy is that our ordinary understanding of action is supplied by belief–desire reasoning (Davidson, 1963; Dennett, 1971; Fodor, 1987). There are certain sets of cases which make it seem perfectly natural to maintain this exclusive focus on belief as the pivotal mental state. When we are explicitly considering the relative advantages of explaining actions in terms of what is believed as opposed to what is known, it is easy to think of pairs of parallel scenarios in which an agent both knows and believes a proposition p in the first scenario, but merely believes p in the second scenario, and can be predicted to perform exactly the same salient action in both. For example, 7 The idea that attributions of knowledge play no special role in the success of our intuitive understanding of one another is not equivalent with the idea that knowledge plays no significant role in causal explanations. For example, Hilary Kornblith has argued at length for the causal and explanatory value of knowledge while expressing some skepticism about our natural intuitive capacity to register the presence or absence of knowledge, at least as this capacity is activated by thought experiments in philosophy (Kornblith, 2004). My own view is that our pre-theoretical intuitions about epistemological cases are the products of the same mindreading capacity that underpins ordinary mental state attribution (Nagel, 2012).

Knowledge as a Mental State

283

in (1) Janet knows (and a fortiori believes) that her mobile phone is in her handbag, remembering having seen it there a few moments before; in (2) Janet believes that her mobile phone is in her handbag, again remembering having seen it there a few moments before, but does not know that her mobile phone is in her handbag, because an extremely skilled pickpocket has just removed it without her noticing. If we add a certain motivational state to these two scenarios—say, a sudden desire to make a phone call—we will expect to see just the same immediate action performed in each (reaching into the bag). Given that her immediate action in both cases would be the same, it can seem not to matter to the explanation of this action whether Janet knows or merely believes. On the currently orthodox philosophical view, the relevant proximal cause of her action in both scenarios is Janet’s belief that her mobile phone is in her bag, which together with her desire to make a call and various background beliefs will determine what she does. On this view it would be inefficient to say that her action in the first scenario was produced by her knowledge of the key proposition while in the second it was produced by mere belief; for maximal generality we should posit the same set of mental causes in the two cases. At the first moment of reaching into the bag, the two scenarios may look very similar; however, when we extend the time frame and imagine the continuation of (1) and (2) we do expect differences to emerge. In (1) Janet will make her call; in (2) she will rummage in the handbag for a while and then come to the realization that something is amiss. But the advocate of the orthodox belief–desire view will not see this divergence as any reason to allow that knowledge has a special role in explaining action, or that the key mental states explaining action in (1) and (2) were perhaps different from the start. Rather, the difference in longer-range forecasts would be explained in terms of an interaction between the initial mental states (which on the orthodox beliefbased view are the same in both scenarios) and environmental conditions (which differ between the scenarios on the point of whether the mobile phone is actually present), where these evolving conditions will come to produce different mental states in the parallel agents, further shaping what each will do. A considerable part of the value of mental state attribution is its capacity to generate predictions across variations in environmental conditions. Positing just a single initial mental state—belief—across both of our cases looks like an economical thing to do. However, it will be a false economy if the state of knowledge is generally associated with patterns of behavior that are interestingly different from those produced by the state of mere belief. One could concede both that knowing entails believing, and even that in this particular pair of cases the same immediate action would be generated by knowledge or belief, without granting the general point that attributions of knowledge play no special role in our understanding of intentional action. Arguing against the orthodox philosophical view and in favor of the admission of knowledge as a mental state, Williamson has developed some examples of cases in which our expectations concerning

284

Jennifer Nagel

an agent’s extended pattern of conduct do seem to be sensitive to whether we see that agent as having knowledge or mere true belief on some point. A brief overview of his position, as presented in (Williamson, 2000), will facilitate discussion of these examples. Williamson stresses that his view is consistent with the orthodox view that knowledge in fact entails belief, so that any agent who knows that p could also be credited with a belief that p; what he denies is that knowledge is a composite of belief and other factors, or more generally, a hybrid state composed of mental and non-mental conditions. Knowledge, in his view, is a stronger mental state than belief, but it is nevertheless a purely mental state in its own right, so in attributing it to others we would not ordinarily have to pass through intermediate stages of attributing the weaker mental state of belief and also recognizing the believed proposition as true. To describe knowledge as a purely mental state is not to suggest that it is determined entirely by the internal physical or spiritual configuration of the agent, in a manner that is independent of what is going on around her. In Williamson’s view mental states like seeing or knowing that something is the case are not seen as, say, inner neurological states whose relationship to the environment is left entirely open, but rather as states in virtue of which an agent is related to an environment in certain ways (the thought that various neurological configurations can support our being in such states is undisputed). The relevant ways of being related to the environment mark another point of agreement with the orthodox view: Williamson agrees that knowledge is a factive state. Indeed, what distinguishes knowledge from other mental states is that knowledge is the most general factive mental state. Williamson allows that there are various different factive mental states—seeing that p, remembering that p, regretting that p, being acutely conscious that p and so on— but in his view all of these entail knowing. Factive mental states other than knowing are invariably just more specific ways of knowing. So, classifying knowledge as a state of mind does not mean giving up on the condition that what is known must be true as a matter of objective fact; the point is rather that states of mind can be constituted in terms of relationships between agents and the objective facts of their environment.8 The status of knowledge as our core factive mental state is what accounts for its special significance, according to Williamson: “factive mental states are important to us as states whose essence includes a matching between mind and world, and knowing is important to us as the most general factive stative attitude” (2000, 40). Knowledge on this view is a purely mental state rather than a composite of the mental state of belief plus non-mental factors. However, Williamson

8 It is important that these relations bind agents (or subjects—I use those terms interchangeably in this paper) directly to the environment. Orthodox analyses can also bind an agent to the environment, but only indirectly, through the mediation of the state of belief. For example, in Goldman’s causal theory the state of knowing that p was thought to consist in having a belief that p, where this belief was appropriately caused by the fact that p (Goldman, 1967).

Knowledge as a Mental State

285

does not oppose the orthodox view about the entailment between knowing and believing: any agent who knows that p also believes that p. While the orthodox approach since Gettier has explained this entailment in terms of a composite theory of knowledge that takes belief as one of its basic building blocks, Williamson instead allows the entailment by characterizing belief as a potentially diluted version of knowing. In his view, ‘believing p is, roughly, treating p as if one knew p’ (2000, 47). Matching how things are is not part of the essence of belief, although beliefs do sometimes happen to match how things are: we can believe false propositions, and we can believe propositions whose truth is not essential to our believing them. The attribution of knowledge is not taken to start from an attribution of belief, on Williamson’s view; rather, the capacity to recognize belief depends on some prior mastery of the concept of knowledge. Nevertheless, the weaker and derivative mental state of belief obtains wherever the stronger mental state of knowledge does.9 Williamson aims to support his view of knowledge as a distinctive and pure mental state by identifying some cases in which the simple attribution of knowledge yields a better explanation of action than the composite attribution of belief plus the truth of the proposition believed. One such case invites us to consider the difference between a burglar who knows that there is an especially valuable gemstone in a given house and a burglar who merely has a true belief to the same effect. Other things being equal, Williamson suggests, the first type of burglar would be expected to be more persistent in searching the house, if only because mere knowledge is in general more secure than mere true belief (2000, 62). Mere true belief has various vulnerabilities not shared by knowledge: for example, if a belief falls short of being knowledge because it is only accidentally true, it might be derived from a false premise. So, Williamson notes, the second sort of burglar could believe that the stone is in the house on the basis of a false report that it is hidden under the bed, and because of his reliance on false testimony fail to count as knowing that the stone is in the house. After a thorough search under the bed, this burglar would be likely to desist where his counterpart who knows the stone is in the house would continue. Because it is essential to states of knowledge that they match how things are, knowledge is more robust than true belief as such. Discovering that a certain burglar engaged in an exceptionally tenacious search of a house—having spent hours on the property in peril of being discovered— we would better explain this behavior by saying that he knew the stone was in the house, rather than that he had the true belief that it was. 9 This is not to say that an attribution of belief will always sound natural when an attribution of knowledge would be appropriate. Because we ordinarily aim to make our conversational contributions as relevantly informative as possible (Grice, 1975), by attributing the weaker condition of believing we would typically generate the conversational implicature that the stronger condition of knowing is not known to hold. So for example if I say, “Justin thinks that I want him to come to my party,” this carries the implication that I do not in fact want him there. But the implicature is cancellable and therefore merely pragmatic; I could say, “Justin thinks that I want him to come to my party; indeed, he knows that I do.”

286

Jennifer Nagel

This type of example has encountered resistance from defenders of the orthodox belief-oriented line. If resolute behavior can be explained by appeal to knowledge, they contend it can be explained even more successfully by appeal to certain kinds of belief, perhaps particularly confident or inflexibly held belief (Magnus and Cohen, 2003; Molyneux, 2007).10 As Williamson himself stresses, knowledge may be lost when one encounters misleading evidence; mere belief, on the other hand, may be dogmatic enough to shield its holder from reconsidering a given proposition in any circumstances. If we just want to explain exceptional persistence in a dangerous course of action, an attribution of obstinacy may serve us better than an attribution of knowledge. The advocate of knowledge-based explanations of action can point out that in some circumstances we may have good reasons not to attribute sheer obstinacy to a resolute actor, but it is open to the advocate of the belief-based approach to suggest that in those cases there could be further conditions added to belief that would all together explain the situation as well as knowledge. To avoid granting the point about the special explanatory power of knowledge attributions, these further conditions could not themselves be tantamount to knowledge; well-founded true belief will be off-limits, for example, if we are using the technical expression “well-founded” as some kind of abbreviation for “amounting to knowledge.” Magnus and Cohen suggest that a certain kind of rational-belief-that-will-tend-to-survive-scrutiny would be a good belief-based explanatory substitute for knowledge (2003, 49) but it is not clear we can make intuitive sense of such a concept except insofar as it is a rough characterization of knowledge itself. Magnus and Cohen do not elaborate on the nature of this tendency, but if they think that we intuitively explain action by appeal to such a concept, the burden of proof is arguably on them to establish that we have any intuitive notion of it. Furthermore, without an account of what grounds this special tendency, explanations that appeal to it would have limited explanatory value: it is not particularly informative to explain rationally tenacious action as being the kind of action that is based on beliefs that are, for some unexplained reason, rationally tenacious.11 10 An anonymous referee suggests that there may be a sense of “knowing that p” which means nothing other than “being subjectively certain (or highly confident) that p,” citing as evidence for this view comments such as “I just knew the butler was the murderer, until it turned out, in the last chapter, that he wasn’t.” On this view an attribution of “knowledge” would be a better explanation of the burglar’s persistence than an attribution of mere belief, but not in a way that would appeal to anyone committed to the essential factivity of “knows.” However, it is not clear that we do have reason to admit a strictly confidence-driven and nonfactive sense of “knows”: there may be a better way of making sense of utterances like the remark about the butler. To begin, the special emphasis on “knew” in that remark is plausibly a clue that “knew” is not being used in a straightforward fashion. Richard Holton argues that such utterances gain their dramatic flair from “protagonist projection,” in which one puts oneself in the position of another (or in this case, one’s past self), deliberately selecting words fitting that perspective rather than our current one (Holton, 1997). Note that one could say to similar effect, “Up until the last chapter, the butler was obviously the murderer, and then it turned out he wasn’t.” 11 This last point is owed to Tim Williamson, who notes “a salient similarity to dormitive virtue explanations” here (Williamson, p.c.).

Knowledge as a Mental State

287

If mere true belief can fall short of knowledge in an open-ended variety of ways, as the frustrations of the analysis-of-knowledge project would suggest, there is no quick formula to summarize the difference made by attributions of knowledge as opposed to true belief, and it is not immediately obvious which side has a better line on the explanation of action. Several points may still seem to favor belief-based over knowledge-based approaches to mental state attribution. There are costs associated with allowing the richer state of knowledge as an explainer: most notably, we are no longer in a position to give an exactly uniform explanation of what Janet initially does in our two mobile phone cases if her action is explained by knowledge in the first and mere belief in the second. The belief-based composite approach brings with it the main attraction of compositionality, its promise of greater fertility and generality. Molyneux suggests that knowledge-based explanations are too finely tailored to their circumstances to have useful explanatory strength. Noting the similarity of a pair of cases like our Janet cases, he presses the concern that “when an explanation type fails to preserve its force across extremely similar scenarios, it provides us with evidence that it does not bear up well through contingent variation of any sort” (Molyneux, 2007, 270). A somewhat related line of objection to Williamson would focus on the sorts of cases that he takes to illustrate the special value of knowledge-based explanation of action, and point out that these cases might be ecologically quite rare. If we take the usual non-skeptical stance that ordinary cases of perceptually, inferentially, and testimonially grounded true belief are knowledge, then true belief that falls short of knowledge is arguably going to be uncommon, and situations in which it figures are likely to be greatly outnumbered by situations involving ordinary false belief. If we need to evaluate the truth of the key proposition anyway in order to make an attribution of knowledge, it may seem largely redundant to admit knowledge as an independent explainer of action alongside mere belief.12 Situations in which we encounter beliefs that are true overlap so heavily with situations in which we encounter knowledge that it is hard to see how we would be ill-served by an approach that promised to make uniform sense of true belief and false belief situations, even if it did not make perfect sense of those rare cases in which an agent has accidentally true belief. Even if Williamson is right that knowers conduct themselves in subtly different ways from mere believers of the truth, the cases he is able to find to illustrate this point are exotic enough that we might hope to leave them on the sidelines as a range of curious exceptions to our general account of mental state attribution: perhaps the differences between knowledge and mere true belief could be handled by some codicil to a belief-based account of intuitive mental state ascription. 12 Or perhaps alongside highly confident mere belief, in the event that one’s composite concept of knowledge incorporates the requirement of passing a certain threshold of confidence. Another version of this objection would press a worry that knowledge explanations will be redundant with highly confident true belief explanations.

288

Jennifer Nagel

However, if natural social environments provide heavy overlap between observed true belief situations and observed knowledge situations, this is not necessarily bad news for a defender of the knowledge-based approach. If knowledge were a status only rarely attained by true belief, then the composite approach’s capacity to generate a smooth compositional story about true belief and false belief would be more attractive than an approach giving a special primary role to the ecologically rare state of knowledge. If the most common prototypical examples of judgment that we witness are in fact moments of knowing, on the other hand, then it becomes more plausible to maintain that recognition of knowledge is prior to recognition of mere belief (whether true or false). Furthermore, it is not obvious that we must separately evaluate the truth of the proposition believed in the course of attributing knowledge. On some views of the testimonial transmission of knowledge, we can reasonably see another person as knowing a proposition without conducting any independent check of that proposition.13 A defender of the composite approach could of course resist such views of testimony, or object that what the composite approach really needs is not that the mental state ascriber should consult independent evidence on any proposition another is taken to know, but just that the mental state ascriber needs to go through the work of representing that proposition as being true, in addition to being believed by the target. However, the advocate of the knowledge-based account might observe here that in some cases of attributing knowledge to another, we do not even have to specify the proposition believed, let alone represent it as true: for example, we can see others as knowing whether q, where the truth value of q is unknown to us. Not all knowledge attribution has the outward form “S knows that p.” More broadly, attributions of knowledge can take a range of interrogative complements: the mental state ascriber can see another as knowing when the party is supposed to start, where the drinks are, or who is invited, without any particular answer to these embedded questions being endorsed by the ascriber.14 Indeed, our ability to see others as knowledgeable on questions whose answer is unknown to us is arguably a core feature of our intuitive social understanding of others. The advocate of the composite belief-based approach could of course develop a composite treatment of this aspect of our social intelligence: perhaps what we really recognize in others is their reliability, or something of that nature, and perhaps attributions of knowing-wh are rather different in kind from ordinary attributions of knowing that p. But if we take them at face value and attempt to treat them in a uniform manner, attributions of knowing-wh do seem to count against the original suggestion that knowledge attributions always require the composite representation of a known proposition as true and as believed. 13

Thanks to Tim Williamson for this point. Thanks to Tim Williamson for this point, and to Jane Friedman for discussion of the issues here. 14

Knowledge as a Mental State

289

For many philosophers, the deepest problem with classifying knowledge as a mental state is quite different in character from the problems canvassed so far. It is a metaphysical problem about the localization of mental states. Because the essence of knowledge involves matching how things are in the world, a distal change in the environment can effect a change of mental state in a subject even if nothing in her internal consciousness or neurology changes at that moment. Early in April of 1865, an ordinary well-informed American citizen knew that Abraham Lincoln was President; at some point on the evening of the 15th her mental state switches from knowledge to mere belief.15 As soon as the President is assassinated and the proposition about his being President thereby becomes false, our citizen’s mental state will have undergone this change, even if she has not heard what has happened in the faraway theatre, and even if she would say and do just the same things immediately following this change as before it. In the same vein, we can think of Janet’s mental state before and after her mobile phone is stolen; when she first puts the phone in her bag she knows that it is there, and when it is stolen (or even just at very immanent risk of being stolen) she no longer has this knowledge of its location, but there need be no significant difference in either her internal neurological or conscious state between those two moments (assuming our pickpocket is stealthy enough to do nothing Janet consciously experiences). In short, the distal vulnerability of factive states may make them seem unfit to serve in causal explanations of behavior: critics of the knowledge-first approach have complained that positing knowledge as a mental state appears to violate strictures on action at a distance. Because nonfactive mental states like believing that Lincoln is President or believing the mobile phone is in the bag can persist through a change in the truth value of their contents, it can seem that they are better candidates for the causes of our actions. Pointing to intuitive theories according to which causal powers must always be in the same location as their effects, they object that ‘attributions of causal efficacy to knowledge work, if at all, by piggybacking on the causal efficacy of belief’ (Magnus and Cohen, 2003, 40). In response, Williamson claims that the best current accounts of the propositional attitude of belief do not localize it within the neurological or conscious states of the agent either; perhaps beliefs are also entangled with distal conditions in making reference to external particulars (2000, ch. 2).16 This paper will not attempt to evaluate the arguments of Williamson and others on the issue of whether we can find some way of making sense of belief as a local mental 15

This example is again Williamson’s (2000, 23). On a related note, knowledge is sometimes thought to differ from belief in having “normative” status: the person who is described as knowing is being positively evaluated, where the person who merely believes may be doing something wrong. Without disputing the claim that knowledge is normatively superior to mere belief, one might observe that belief itself has normative status: for example, as Plato observes in the Theaetetus, in order to have a belief about anything one must be appropriately acquainted with it, where it is at the very least far from obvious how one could cash out “appropriately” in a non-normative fashion (Plato, 1990). 16

290

Jennifer Nagel

state; serious treatment of these arguments would require engagement with a large and unwieldy body of work on the nature of mental content and mental states (for some discussion, see Fricker, 2009; Jackson, 2007; McDowell, 1986; Putnam, 1975; Williamson, 2009). What will instead be pursued in the next two sections is a complementary strategy of challenging the claim that good explanations must be anchored in strictly local conditions.17 If we consider a discipline that gives uncontroversially good explanations of its target domain—the science of physics as applied to the interactions of mid-sized inanimate observable objects—we find reasons to resist the condition that all real explanation is local. Our intuitive understanding of these interactions may however feel more local than it really is. The “naïve physics” which gives us an intuitive ability to predict the dynamic interactions among inanimate bodies is generated by a complex set of implicit computations involving the relationships among those bodies and so-called “environmental invariants” such as gravity and friction (McCloskey and Kohl, 1983). Interestingly enough, we have a natural tendency to “localize” the results of these computations, for example seeing the trajectory of an object as determined by its own private impetus or “representational momentum,” a strictly local power that is seen as vested in it as the result of its inner nature and contact with other things. However, this localization is a fiction, both because the principles of impetus physics are not entirely accurate, and because in computing it we draw essentially on the relationships between objects, relationships involving “action-at-a-distance” factors such as gravitation (Hubbard, 1995). We can recognize on reflection that an inanimate object’s motions do not depend strictly on local interactions between its inner condition and the objects it physically contacts, but this explicit recognition does not extinguish our natural intuitive tendency to see ordinary outer objects as driven in this way (Kozhevnikov and Hegarty, 2001). As a parallel,18 it is possible that we do in some sense intuitively represent the mental state of knowledge as localized within an agent, even if we actually intuitively compute the presence of knowledge on the basis of our grasp of relations between that agent and her environment, and even if we are able to recognize on reflection that this state must reflect how things are in the agent’s

17 Williamson remarks congenially on the possibility of attempting this strategy, but does not develop it himself (2000, 61). 18 It is plausible enough that there are parallels between naïve physics and mindreading, given the overall structural similarities in the function of these capacities. A psychological capacity for naïve physics underpins our ordinary experience of the motions of inanimate objects, enabling us to anticipate (and sometimes manipulate) their motions; mindreading underpins our ordinary experience of the actions of animate agents, enabling us to anticipate (and sometimes manipulate) what they do. Both capacities require very rapid assimilation of a broad range of complex information about our environment, much of it below the threshold of conscious or explicit reasoning. Psychologists do sometimes discuss the natural strengths and limitations of one domain in terms of an analogy with the other (e.g. Saxe, 2005).

Knowledge as a Mental State

291

environment. In intuitive judgment, we are aware of the final product or outcome of a process of computation, and not of the processing that gives rise to it (Sloman, 1996); while reflective and deliberate attributions of knowledge may make us conscious of the role played by distal factors in knowledge attributions, it is entirely possible for intuitive attributions of knowledge to incorporate distal factors without alerting us to this fact. If the epistemic state underpinning action is something that subjectively seems local, our intuitive representation of knowledge may well provide this subjective impression: we could at some level feel that an agent’s knowledge is localized within her, just as we feel that an object’s motion is sustained by something within it.19 As far as objective reality is concerned, the example from intuitive physics should give us second thoughts about the wisdom of a general principle ruling out appeal to non-local factors in our explanations of what is really going on, and even in our explanations of what is really going on in the generation of our intuitive assessments. There is something jarring about the reflective observation that a subject can switch from one mental state to another (e.g. from knowing to not knowing) through something that happens remotely. However, the attribution of quite different mental states to the agent before and after the change may be natural enough if there is a genuine change in how we intuitively represent them as agents, or if suitably distinct procedures naturally apply to representing knowledge and false belief. If we take belief to be a state that is derivative from the state of knowledge, then the more complex attribution of a false belief that p could nevertheless in some circumstances enable a prediction of the same immediate action that would be produced by knowledge that p. This observation could be applied to the explanation of the similar immediate behavior of our two Janets, one of whom knows that her phone is in her bag while the other falsely believes it. Williamson’s view does not sever the link between knowledge and belief altogether: if believing that p is understood as a state that involves “roughly, treating p as if one knew p,” then the predicted outcome of immediate action given the false belief that p may be systematically similar to the predicted outcome of immediate action given the knowledge that p, although the former type of prediction may be expected to have some additional level of complexity. At this point it may be helpful to examine some independent evidence on the nature and relative complexity of our intuitive attributions of knowledge 19 Tamar Gendler observes (p.c.) that there may be a significant relationship between this tendency and the Fundamental Attribution Error, a bias which inclines us to assign unduly heavy weight to inner personality characteristics and insufficient weight to situational factors in our explanations of the actions of others (Jones and Harris, 1967; Ross, 1977). This is an intriguing suggestion, detailed exploration of which cannot be attempted here, if only because it would lead us into some very complex issues involving self/other asymmetries in the explanation of action, and into problems concerning the natural extent and character of misrepresentation in ordinary action explanations.

292

Jennifer Nagel

and belief, and on the question of whether the concept of knowledge is in some sense prior to the concept of belief, or whether it is composed from that concept and further conditions. In particular, it may be useful to examine developmental work on the emergence of the concepts of knowledge and belief: arguably, if intuitive representation of knowledge really is a composite involving intuitive representation of belief, the capacity to represent knowledge should not be available unless and until the capacity to represent belief is in place. The next section of this paper examines some recent empirical work which addresses this issue, and the closely related issue of the relative complexity of ordinary intuitive ascriptions of knowledge and belief.

3.

E M P I R I C A L W O R K O N T H E R E L AT I O N S H I P B E T W E E N THE CONCEPTS OF KNOWLEDGE AND BELIEF

The standard view in psychology is that knowledge is a mental state; indeed, knowledge typically features prominently in lists of paradigmatic mental states (e.g. Apperly, 2011; Heyes, 1998; Premack and Woodruff, 1978; Sodian and Kristen, 2010; Wellman and Liu, 2004). Reading the empirical literature on mental state attribution, one sometimes finds an article which omits mention of knowledge as a mental state (e.g. Russell, 2005), but I do not know of any article in this body of literature in which it is explicitly argued or even claimed that knowledge is not a mental state. Marking a further point of difference from the mainstream philosophical view, psychologists of mental state attribution do not seem to assume that the concept of knowledge is an elaboration of the concept of belief; in fact, it is generally agreed that belief is the more sophisticated concept, and is harder to attribute than knowledge.20 In comparative psychology, there is nothing alarming about entitling an article, “Chimpanzees know what others know, but not what they believe” (Kaminski et al., 2008). In developmental psychology it is widely held that children acquire the concept of knowledge before the concept of belief.21 After a very brief review of psychological work on the relationship between the concepts of knowledge and belief, this section will discuss the reasoning behind the psychological classification of knowledge as a mental state. There is a wide range of evidence relevant to the question of whether the concept of knowledge is prior to the concept of belief. One might begin by looking at work on the acquisition of the mental state lexicon. Children use “know” both earlier and more heavily than they use “think” (Bartsch and

20 This observation is of course compatible with the notion that knowledge entails belief. A creature must be a mammal in order to be the conspecific of a primate—being a conspecific of a primate entails being a mammal. Nevertheless, many primates have the conceptual capacity to recognize their conspecifics while lacking the conceptual capacity to recognize mammals as such. 21 Williamson mentions the potential relevance of this fact to his “knowledge first” approach, but does not elaborate (2000, 33).

Knowledge as a Mental State

293

Wellman, 1995; Shatz, Wellman, and Silber, 1983). Both the earlier age of initial use and the increased frequency of a word for knowing over a word for thinking appear cross-culturally, for example in languages such as Mandarin and Cantonese (Tardif and Wellman, 2000). Drawing on a database of over 200,000 spontaneous utterances by English-speaking children up to the age of six, Bartsch and Wellman found that the verb “know” figured as the main verb in 70 percent of children’s epistemic claims, with “think” following at 26 percent (other epistemic mental state verbs, such as “believe,” “wonder” and “expect,” were rarely used) (Bartsch and Wellman, 1995).22 Before the age of about four, children have trouble with the word “think,” showing some tendency to take it as factive (Abbeduto and Rosenberg, 1985), although children early into their third year do use the word and are sometimes even able to produce the contrastive utterances taken to be the best evidence of an understanding of nonfactive verbs—for example, “I thought there wasn’t any socks, but when I looked, I saw them” (Bartsch and Wellman, 1995, ch.3). Explaining the initial acquisition of epistemic verbs is no easy matter. In the ordinary course of development, there is a cross-culturally robust and very significant lag between the emergence of action and activity verbs like “go,” “eat” and “push,” and mental state verbs like “think” and “know” (Choi and Gopnik, 1995). Summarizing a few of the relevant difficulties, Papafragou and colleagues note that verbs like “think,” “remember,” and “know” do not refer to perceptually transparent properties of the reference world; they are quite insalient as interpretations of the gist of scenes; they appear frequently in maternal speech to babies and yet occur in the child’s own speech comparatively late; the concepts that they encode are evidently quite complex or abstract; and they are hard to identify from context even by adults who understand their meanings. (Papafragou, Cassidy, and Gleitman, 2007, 126)

They go on to make the following observation about their approach to the problem: The idea here is that words that refer to mental states and events lack obvious and stable observational correlates: as a general rule, it is easier to observe that

22 The dominance of “know” over “think” continues in adult usage but is less pronounced. “Know” and “think” are respectively the 8th and 12th most common verbs in English, according to the Oxford English Corpus. In the 425-million word Corpus of Contemporary American English these words occur at a ratio of 53:47. In the context of explicit action explanations there is some evidence that knowledge dominates. In COCA, for example, the ratio between instances of the strings “because s/he knew” and “because s/he thought” is 68:32. Including the next most common non-factive (“believe”) alongside “think” drops the ratio to 58:42. This is only superficial evidence of the importance of knowledge ascription in action explanation; a detailed study of this issue would have to examine corpus data for a wider range of factive and nonfactive verbs and constructions, and address questions about the relationship between our ordinary understanding of action and our verbal reports of this understanding. If we tend to remark on people’s reasons for acting when their actions are out of line with our expectations, as Gricean norms might predict, then the method of counting occurrences of “know” and “think” would presumably under-report reliance on attributions of knowledge.

294

Jennifer Nagel

jumpers are jumping than that thinkers are thinking. Assuming that word learning— especially in its initial stages—relies heavily on establishing relations between what is said and what is (observably) happening in the extralinguistic environment, it follows that there will be barriers to acquiring mentalistic terms even where the concepts themselves fall within the learner’s repertoire. (Papafragou et al., 2007, 128)

To adults, mental state talk feels so natural that it takes some imaginative work to appreciate the mental state verb acquisition problem from the perspective of the child. Although children hear verbs like “think” and “know” heavily enough, it is not easy for them to forge the link between what is heard and what it refers to. Some aspects of the difficulty here can be reproduced even in adults through an experimental protocol called the Human Simulation Paradigm (HSP). In one version of the HSP, adult subjects watch a series of brief silent videos of mother–infant interaction and try to guess the mystery word spoken by the mother at the moment of an electronic beep; for each mystery word subjects were told in advance whether it was a noun or a verb, and shown a series of six taped interactions in which the word was used. At the end of the series the subjects could reflect on all that they had seen and offer a seventh and final conjecture as to the identity of the target mystery word. None of the chosen words were particularly uncommon: Gillette and colleagues tracked the twenty-four nouns and the twenty-four verbs that occurred most frequently in their videotaped samples. Rates of correct identification were strikingly different for different classes of words: on average 44.9 percent of the nouns were correctly identified by the final try, versus only 15.3 percent of the verbs. Concrete nouns like “piggy” (89.3% right), “ball” (78.6%) and “plane” (100%) did relatively well, as did some verbs: “come” (75.0%), “throw” (85.7%) and “push” (42.9%). Mental state verbs did very badly: both “think” and “know” were identified by zero percent of subjects (Gillette, Gleitman, Gleitman, and Lederer, 1999). Given that adults already have mature mental state concepts, it should be easier for them than it is for infants to pick out adult uses of mental state words; the very poor performance of adults in recognizing mental state words in the HSP experiments suggests that the problem of finding “observational correlates” of mental state verbs is a serious one. One reason why the task was difficult was that common mental state verbs are often used for reasons other than referring directly to a salient mental state; for example, in fixed expressions (such as the attention-getting “know what?”) and to perform social functions such as softening a request (“I think it’s time to go”). These “conversational” uses of mental state verbs are relatively frequent (Bartsch and Wellman, 1995) and are agreed to offer very little information to the language learner (Shatz et al., 1983). In their initial HSP experiments, Gillette and colleagues had made no special effort to find scenarios in which mental states would be particularly notable. Papafragou and colleagues experimented with various types of scenario to see whether some circumstances would improve performance on mental state verb recognition.

Knowledge as a Mental State

295

They contrasted scenarios in which someone was fooled or mistaken (FB scenarios) with closely matched counterparts in which nothing went wrong (TB scenarios). In a typical FB scenario, a woman drinking tea while reading a newspaper did not observe her cup being moved by someone wiping the table and a teapot being set in its place; she reaches for the teapot and brings it close to her mouth before noticing her error. In the control TB scenario there is the same sequence of object displacements but the woman is not misled and picks up the cup. Children and adults then had to speculate about what would be said by a mother viewing the scene with her child. As in the earlier HSP experiments, most responses focused on the concrete and observable aspects of the scenarios. Overall, looking at the TB and FB experiments together, both child and adult responses were skewed towards action verbs (43.9% and 31.6% respectively); among mental state verbs the verbs of desire were dominant (23.5% and 30.9% of child and adult responses referred to motivational mental states). Epistemic mental state verbs were used in 11.5 percent and 23.5 percent of the child and adult responses. But strikingly enough, the FB scenarios produced a much higher incidence of epistemic verb use: in comparison to the TB scenarios, the FB scenarios elicited more than triple the rate of epistemic verb use for children, and almost double for adults (Papafragou et al., 2007). These experiments do not show that we typically fail to register mental states in witnessing the successful actions of others. The discovery that epistemic mental states become particularly salient when actions do not unfold as expected might rather suggest that our background expectation is that agents will act knowledgeably, and we find it remarkable when they do not. The default expectation of knowledgeable behavior on the part of others is thought to be manifest in a well-known sequence of nonlinguistic mental state ascription tasks. The first of these, passed in some versions even by some nonhuman primates, measures the capacity to distinguish knowledge from ignorance; the second task measures the capacity to attribute false belief, and is passed only by humans, and only later than the knowledge-ignorance task. Some description of the tasks may help to clarify why psychologists of mental state ascription generally take the concept of knowledge to be prior to the concept of belief. In one version of these tasks, pairs of children were given a familiar container with familiar contents to examine (a domino box with a picture of dominos in it). One child from each pair was then sent out of the room, and in his absence the other witnessed the familiar contents being replaced with a novel item. The second child was then asked (in random order) two questions: (1) Does [name of absent child] know what is in the box now, or does he not know? (2) If we ask [name of absent child] what is in the box, what will he say? Among three-year-olds, 39 percent answered question (1) correctly, and only 6 percent answered question (2) correctly; four-year olds improved to 81 percent and 44 percent, and five-year olds were right 88 percent and 76 percent

296

Jennifer Nagel

of the time on the two questions (Hogrefe, Wimmer, and Perner, 1986).23 If we generally made judgments about the presence or absence of knowledge by attributing belief and then evaluating the truth or falsity of this belief, we would not expect to see such a lag between the capacity to recognize the absence of knowledge and the capacity to attribute a false belief.24 Young children are not random or indiscriminate in the way they fail the knowledge-ignorance task: they have a general tendency to over-attribute knowledge to others. This tendency has been found across a great variety of different tasks; in general, the trend as children mature is towards the recognition of more and more of the restrictions on knowing that are imposed by the various limits of sensory modalities, perspective, perceptual access, and experience (e.g. Apperly and Robinson, 2003; Mossler, Marvin, and Greenberg, 1976; Wimmer and Perner, 1983; Woolley and Bruell, 1996). But studies of immature and evolving mental concept use naturally raise questions about whether young children are really referring to knowledge as such when they come to succeed at tasks of the sort just given. In this context, it is useful to examine work in comparative psychology, where there has been a very active debate on the question of just what should count as genuine mental state recognition. Non-human primates can be very responsive to each others’ behavior. For example, they track where their conspecifics are looking, following the direction of their gaze (Tomasello, Call, and Hare, 1998) in a way that is sensitive to the presence or absence of opaque barriers blocking the gazer’s line of sight (Okamoto-Barth, Call, and Tomasello, 2007). The question of whether they are capable of mental state recognition is, however, difficult to assess. Early research on mental state reasoning in non-human primates seemed to show that they lacked the capacity to distinguish knowledge from ignorance. Chimpanzees will make visible begging gestures equally to human trainers who can see them and to those who have their vision obstructed by a blindfold or bucket, although they do discriminate between trainers who are and are not facing them (Povinelli and Eddy, 2000). In a food-hiding game, chimpanzees also fail to discriminate between the helpful pointing gestures of a trainer they have seen hiding food (behind a barrier, in a location out of the chimpanzee’s view) and a trainer who was absent at the time of hiding (Povinelli, Rulf, and Bierschwale, 1994). Negative results such as these seemed to support the view

23 Difficulty with the second (false belief) type of question is still encountered if the child simply has to select between two pictures representing the two possible mental contents of the other person (Custer, 1996; Woolfe, Want, and Siegal, 2002). 24 This interpretation of the contrast is somewhat controversial, not least because there are various ways of understanding the difference between the knowledge-ignorance task and the false belief attribution task. Some psychologists have raised the concern that the knowledgeignorance task is inherently easier for various irrelevant reasons, such as having a higher baseline of being answered correctly simply by chance (e.g. Perner, 1995). For a response to such concerns in the context of a large-scale meta-analysis of studies showing knowledge-ignorance to be easier than false belief attribution, see (Wellman and Liu, 2004).

Knowledge as a Mental State

297

that non-human primates track surface behavioral regularities (like “facing me—likelier to give food”) but do not understand the relationship between seeing and knowing, and fail to attribute mental states such as knowledge or ignorance (Heyes, 1998; Povinelli and Vonk, 2003; Tomasello and Call, 1997). On this view, these primates’ navigation of social situations is guided strictly by learned or innate “behavioral rules” linking certain observable situations or patterns of behavior and their typical behavioral consequences. This superficial grasp of behavior patterns would enable non-human primates to predict behavior in a fixed set of familiar situations; unlike mindreading abilities that discern the deeper mental causes of action, the “behavior-rule” ability would not generate predictions flexibly, across novel circumstances. More recent work has reopened the question of whether non-human primates might have some grasp of the unobservable mental states underlying behavior. Artificial cooperative games with humans might not be the best setting to test a chimpanzee’s mindreading abilities, and a number of primatologists wondered whether chimpanzees would perform better in a competitive setting involving their conspecifics. Exploiting the fact that subordinate chimpanzees will not challenge dominant chimpanzees for food, Hare and colleagues allowed subordinates to watch food being hidden outside their cages while a caged dominant either was or was not also watching within sight of the subordinate. Both chimpanzees were then released from their cages, with the subordinate getting a brief head start so he could not simply react to the dominant’s behavior. Subordinates were able to distinguish between knowledgeable and ignorant competitors, preferring to go for the food when the dominant competitor was ignorant of its location (Hare, Call, and Tomasello, 2001). A series of control conditions ruled out various potential non-mental explanations of this behavior: for example, chimpanzees did not simply regard the food that had been seen as tainted or dangerous, but kept track of the relationship between the food and the specific individual who had seen it. If a different dominant animal was brought in rather than the one who had seen the food, the subordinate no longer avoided it. Surveying these and an array of other recent experiments, leading comparative psychologists who had earlier concluded that non-human primates do not grasp mental states (Tomasello and Call, 1997) came to revise their opinion, arguing that the total current body of evidence supports the conclusion that “chimpanzees, like humans, understand that others see, hear and know things” (Call and Tomasello, 2008, 190). It is possible to maintain a strict behavior rules’ interpretation of the recent experimental results, for example, by positing that what the subordinate animal really knows is not that the dominant knows where the food is, but that any dominant will go for food that has recently been in its line of sight (Povinelli and Vonk, 2003). However, this strategy may put us on a slippery slope: if we are allowed to formulate behavior rules ad hoc, then a behavior rules interpretation could be produced for any pattern of apparent mental state ascription, even one involving advanced social interactions among adult

298

Jennifer Nagel

humans (Fletcher and Carruthers, 2013). The fact that a certain interpretative strategy is in principle available—as behaviorism continues to be available, in principle, for explaining even human social interaction—does not mean that it is advisable, or that it is the strategy that would yield the best explanation of our data. Considerations of parsimony and fruitfulness also matter. It is not trivial to show that the mindreading interpretation is more parsimonious than its behavioral rules counterpart as an explanation of the data involving nonhuman primates; in particular, the inherent complexity of the “mindreading module” which links knowledge, goals, and behavior must itself be taken into account, and cannot be treated as a “black box” which the theorist gets for free (Perner, 2010). However, considerations of parsimony weigh more and more heavily in favor of a mindreading interpretation as creatures exhibit their competence in predicting the behavior of others across a broader range of situations. On the mindreading interpretation, the chimpanzee can systematically combine attributions of knowing where something is with attributions of wanting that thing to generate predictions about behavior in novel circumstances, where there could be various different outward signals of knowing or wanting; meanwhile, the behavior rules’ interpretation has to posit a separate rule to connect each type of outward signal with each type of behavioral consequence (Fletcher and Carruthers, 2013). Interestingly, support for a mindreading interpretation of non-human primate abilities arises also from evidence about their limitations in social reasoning. There are some mental state attribution tests that non-human primates are apparently unable to pass, although as Fletcher and Carruthers point out in their (2013), it is possible to concoct relatively simple behavior rules that would dictate a successful pattern of reasoning for these tasks. The crucial feature that distinguishes these harder tasks from the ones that the chimpanzees succeed at is that they involve belief rather than knowledge. Non-human primates consistently fail false belief tests (Call and Tomasello, 1999; Hare et al., 2001), even in competitive situations and using apparatus that enables them to pass very closely matched knowledge-ignorance tests (Kaminski et al., 2008; Krachun, Carpenter, Call, and Tomasello, 2009). For example, when two chimpanzees or bonobo apes in view of each other have seen a reward placed in one container, and then the target animal is away while the reward is moved to another container, the subject animal does not seem to be able to anticipate that the returning target animal will seek the reward where it was originally placed, even when there would be a competitive advantage to this recognition. If apes were able to pass this test, we could easily write a behavior rule to underwrite their passing it: Fletcher and Carruthers suggest the rule, ‘A potential competitor will approach the previous location of food if the competitor was absent when the food was moved to its present location’. This rule is not inherently any more complex than the behavior rule explaining successful performance on a matched knowledgeignorance task, so the advocate of the behavior rules’ approach faces an extra burden in explaining why such a rule does not become part of the ape’s

Knowledge as a Mental State

299

repertoire. It is hard to see why only one of these patterns of action can be anticipated by apes without helping ourselves to the idea that the contrast in mental states makes a difference: knowledge attribution is easier than belief attribution.25 One way of appreciating the greater simplicity of knowledge attribution is to follow the earlier line according to which knowledge is a state that essentially involves matching how things are, where belief is essentially a state that may or may not match reality. The additional degree of freedom in belief attribution poses an additional computational burden, which matters because a very significant challenge in explaining mature human mindreading is explaining how it is computationally possible, given the open-ended character of the information that might be task-relevant. In contrast to our attributions of a state like jumping, our attributions of mental states are not narrowly constrained by what is immediately observable. Ian Apperly has emphasized this difficulty in connection with the classic test of false belief. Sally sees an object in the round box, and leaves the scene. In her absence, Andrew moves it to the square box, and we are asked where Sally will expect to find it on her return. The “right” answer—the answer that counts as a “pass” on false belief tests—is: “in the round box where she left it.” But Apperly points out that many other mental state attributions are logically consistent with the scenario as given: perhaps Sally will guess correctly that Andrew has moved the object; perhaps she will know that he has moved it, on the basis of what he has told her in the past; perhaps she believes that it has spontaneously teleported to the square box, or to some quite different location (2011, 118). Given the right supplementary evidence about this scenario, any one of those other attributions might be correct. The default understanding of the scenario makes sense, however, and will ordinarily serve us well: according to Apperly, we ordinarily construct simple and stereotyped models of situations involving other agents, drawing on our background knowledge of interactions like the ones we are observing, in a 25 By observing that chimpanzees have some capacity to recognize the state of knowledge, one need not thereby credit chimpanzees with any very sophisticated understanding of the nature of knowledge, nor even with the capacity to recognize knowledge across a great range of situations. For example, they may be unable to pass tests requiring the attribution of knowledge gained through hearing rather than vision (Brauer, Call, and Tomasello, 2008). Ian Apperly has expressed concern that a modality-restricted conception of knowledge would be seriously impoverished: commenting on chimpanzees’ apparent failure to pass auditory tasks, he notes “This would be significant because a core feature of the concept of ‘knowledge’ is that it provides some unification over the results of a variety of perceptual and inferential processes. If chimpanzees’ understanding of ‘knowledge’ is modality-specific then it falls short of providing this conceptual unification” (2011, 53). Nonetheless, Apperly himself stresses the importance of rudimentary mental state concepts (or proto-concepts) in paving the way for our more sophisticated human capacities for explicit mindreading; he also grants the general priority of knowledge recognition over belief recognition. His overall assessment of the comparative literature is that “there is good evidence that some non-human species ‘understand something about’ seeing or knowing, but do not understand these mental states completely, and have no understanding of other mental states, such as believing” (2011, 109).

300

Jennifer Nagel

process he compares to the situation modeling approach to discourse comprehension (e.g. Zwaan and Radvansky, 1998). The default state of knowledge plays an important role in this process of situation modeling. What makes it reasonable to grade “Sally thinks the object is in the round box” as the right answer to the problem (or to attribute this mental state to Sally in a real-world version of the problem) is that this agent would know that the object is in the round box if things had gone for her as they normally do in a person’s momentary absence. If recently having seen something somewhere were not generally a reliable source of knowledge of its location, we would have no reason to attribute that particular belief to Sally.26 So a prior sense of how knowledge ordinarily arises helps to make belief attribution a tractable problem. The relative simplicity of factive attitudes also plays a role here. Apperly observes that “making a mindreading inference consists of constructing a model of ‘what is in the head’ of the target other person, but this model is both related to and part of our larger model of the situation in which that person figures” (2011, 131). Of the various different relationships that might hold between another person’s mental state and the world, the factive ones are arguably the simplest: where the other person is seen as knowing, her mental state essentially reflects how things are around her. When we have reason to do so, we can posit restrictions on what others are taken to know, and see their mental states as deviating, or even just potentially deviating, from reality, but there are computational costs associated with broadening our horizons to include these more complex relationships.27 Given these costs, there is a genuine difference in how we intuitively evaluate an action performed by someone who knows some relevant proposition and an action performed by an agent in a parallel scenario who falsely believes a parallel proposition: predicting the second kind of action is a more complex task, requiring the construction of a situation model to represent the agent’s belief, where this embedded model will be at odds with the larger model representing the agent’s situation. It will even increase the complexity of our prediction to see someone as truly believing without knowing; in order to represent a belief that is only accidentally right, we need to model situations in which someone could have had a similar false belief. When we think about the relative advantages of explaining action strictly in terms of belief (whether true or false) as it interacts with other mental and

26 In a similar vein, Fletcher and Carruthers observe that “the false-belief data make no sense in the absence of a capacity to attribute goals, perceptions, and knowledge to other agents, since the experiments are all designed around just such an assumption, and since no one has any idea how a set of behavior-rules could fit together with belief understanding to issue in the patterns of behavior we observe” (2013, 11). 27 It may be more accurate to say that there are computational costs involved in admitting false belief when we reach the level of explicit mental state ascription: in Apperly’s interpretation, recent work on implicit false belief recognition in infants (e.g. Onishi and Baillargeon, 2005) suggests that early implicit recognition of false belief may be fully automatic (cf. Apperly, Riggs, Simpson, Chiavarino, and Samson, 2006).

Knowledge as a Mental State

301

environmental conditions, it can seem that such explanations have a costfree advantage over knowledge-based explanations. Once we are imagining pairs of scenarios in which a key proposition is known in the first, but merely believed in the second, we have already done the work of constructing the more elaborate set of models for the second scenario: for example, we have already imagined Janet as the unwitting victim of a theft, thinking her phone is in her bag when it is not. In this frame of mind, it may seem natural to explain Janet’s action in terms of her belief across the pair of cases, but the contemplation of ways in which an agent might be mistaken is not a natural part of the ordinary ascription of knowledge, so our impression of the simpler scenario may already be distorted by our focus on the second. Apperly’s talk of “what is in the head” may seem uncongenial to a treatment of knowledge as a factive mental state, but there is reason to think his scare quotes should be taken seriously here. He unpacks the metaphor elsewhere by saying that mental states themselves are not immediately visible but are known to us through their outward signs, claiming for example that “we do not have direct access to what other people know, want, intend or believe, but must infer these mental states on the basis of what they do and say” (2011, 1). His more detailed discussion of the manner in which mental states are attributed makes it clear that these attributions draw not only on the target agent’s local actions and utterances, but also on environmental conditions around the agent. In attributing knowledge of the external world to an agent, we see that agent as related in a certain way to that world.

4.

CONCLUSION

This paper might be faulted for attempting to answer the question of whether knowledge actually is a mental state by arguing that it is naturally perceived as one. Even if our natural mindreading systems parse agency with the help of the epistemic mental state concepts of knowledge and belief rather than belief alone, one might worry that mental reality could fail to correspond to our natural way of tracking it. Someone might be satisfied that we intuitively attribute knowledge to others as a state which explains their actions, without being satisfied that it really has the role it is intuitively seen to have. If we take mental states to be those states which actually do explain the actions of intelligent agents, we might wonder whether the intuitively recognized state of knowledge merely seems to explain what people do, while the real underlying explanation is substantially different, and perhaps more accurately expressed in terms of the weaker mental state concept of belief, plus various further conditions. Indeed, once we are wondering about the relative merits of knowledge- and belief-based explanations of action, we may start to wonder whether there is actually any such thing as being related directly to one’s environment in a way that essentially involves getting it right. It may come to seem that when

302

Jennifer Nagel

we are being our usual lazy selves, we assume that all is well with ourselves and other agents; the application of a little cognitive effort is all that it takes to recognize that what seems very much like knowledge could always fail to hit the mark. If I find myself intuitively attributing knowledge to someone— Janet just put her phone in her bag, so she knows where it is—it is open to me to rescind that generous ascription, on reflection, in favor of something more limited. Dwelling on counterfactual possibilities in which Janet might have been robbed without her knowing it, or more extreme counterfactual possibilities in which the phone spontaneously dissolves, I might start to feel inclined to attribute to Janet a state of mind that falls short of essentially matching how things are: perhaps she is just lucky that the world she inhabits is not one of the worlds I have just imagined, and in that case should be credited only with true belief. Insofar as the application of greater cognitive effort often accompanies an increase in the precision of our judgment, we may feel that we are being more accurate in taking this skeptical stance. However, a shift towards skepticism does not necessarily make us better at predicting and explaining how others will act: by adding extra possibilities to our models of their relationships with their environment, we may be weakening rather than strengthening our understanding of what they are doing. The skeptical stance generates an awkwardly large space of possibilities to be contemplated, and our capacity to reason accurately about all those possibilities is limited. If simplicity is generally a virtue in explanations, the greater simplicity of knowledge-based explanations of action should be counted as a point in their favor. Of course, thinking about possibilities of error sometimes really can increase the accuracy of our mental state ascriptions: the original philosophical critics of Premack and Woodruff were right to observe that the capacity to attribute false belief is a powerful weapon in the mindreader’s arsenal. Other things being equal, creatures who are unable to imagine a conflict between belief and reality will not be as accurate in their predictions or explanations of action as creatures who are able to attribute false belief. But this is not to say that predictions and explanations of other agents generally become better the more we think about them, even when we have no positive grounds to suspect an agent is mistaken; serial contemplation of various possibilities of error threatens to leave us with suspension of judgment about the minds of others, where trust in our natural instincts about what they see, want, and know would have been more helpful, and would have enabled us to make better predictions. The fact that we are capable of applying extra caution and retreating from an ascription of knowledge down to an ascription of mere belief does not entail that ascribing mere belief (plus some environmental conditions about the truth of the proposition believed) is always the more accurate thing to do.28

28

For further discussion of this point, see (Nagel, 2011).

Knowledge as a Mental State

303

Furthermore, the defender of knowledge-based mental state ascriptions can point out that under a more consistently applied skepticism, ascriptions of belief would themselves become problematic. The same forces that naturally generate doubts about whether Janet knows that her phone is in her bag can equally drive us to wonder whether she even believes as much: if we think about the range of possible attitudes she might have on this question, and especially if we do not allow ourselves to assume that her beliefs are a function of what a person in Janet’s circumstances would ordinarily know, then the question of what she should be seen as believing becomes very hard to answer. There are many circumstances in which careful reflective re-thinking of our initial intuitive attitudes may well be warranted, but it is probably not a good idea, in our efforts to understand other agents, to aim for pure reflection without reliance on any intuitive input. This paper has argued that the identification of knowledge as a mental state is one of the central principles of our intuitive mindreading system. Taking a non-skeptical attitude towards that system means agreeing that when we are guided in our understanding of others by an intuitive sense of what they know, we gain significant insight into what they do. If the test of a theory is its capacity to generate good predictions and explanations, then anyone with a generally non-skeptical attitude towards intuitive mindreading should see the thesis that knowledge is a mental state as well confirmed. This is not to say that we have to accept at face value all of our intuitive impressions about the minds of others: like our perceptual systems, our mindreading capacities have certain natural limitations. These limitations—the bias towards egocentrism, for example—are subject to reflective correction, but such corrections can be better executed in light of a clear understanding of the core principles of the natural system that gives us access to other minds.29

REFERENCES

Abbeduto, L., and Rosenberg, S. (1985). Children’s knowledge of the presuppositions of ”know” and other cognitive verbs. Journal of Child Language, 12(03), 621–41. Apperly, I. (2011). Mindreaders: The Cognitive Basis of “Theory of Mind”. Hove and New York: Psychology Press. and Butterfill, S. (2009). Do humans have two systems to track beliefs and belieflike states? Psychological Review, 116(4), 953–70. Apperly, I., Riggs, K., Simpson, A., Chiavarino, C., and Samson, D. (2006). Is belief reasoning automatic? Psychological Science, 17(10), 841.

29 Thanks to Kent Bach, Richard Holton, Patrick Rysiew, Jonathan Weisberg, and members of the Van Leer Institute Epistemology group for discussion of the issues in this paper. Special thanks to Jane Friedman, Tamar Gendler, Sergio Tenenbaum, Tim Williamson, and an anonymous referee for this journal for helpful comments on an earlier draft. I am grateful to the Social Sciences and Humanities Research Council of Canada for funding my research.

304

Jennifer Nagel

Apperly, I. and Robinson, E. (2003). When can children handle referential opacity? Evidence for systematic variation in 5-and 6-year-old children’s reasoning about beliefs and belief reports. Journal of Experimental Child Psychology, 85(4), 297–311. Baron-Cohen, S., Ring, H., Moriarty, J., Schmitz, B., Costa, D., and Ell, P. (1994). Recognition of mental state terms: Clinical findings in children with autism and a functional neuroimaging study of normal adults. The British Journal of Psychiatry, 165(5), 640–649. Bartsch, K. and Wellman, H. (1995). Children Talk About the Mind. New York: Oxford University Press. Bennett, J. (1978). Some remarks about concepts. Behavioral and Brain Sciences, 1(04), 557–60. Brauer, J., Call, J., and Tomasello, M. (2008). Chimpanzees do not take into account what others can hear in a competitive situation. Animal Cognition, 11(1), 175–8. Brueckner, A. (2002). Williamson on the primeness of knowing. Analysis, 62(275), 197–202. Call, J., and Tomasello, M. (1999). A nonverbal false belief task: The performance of children and great apes. Child Development, 70(2), 381–95. (2008). Does the chimpanzee have a theory of mind? 30 years later. Trends in Cognitive Sciences, 12(5), 187–92. Carruthers, P. (2009). How we know our own minds: The relationship between mindreading and metacognition. Behavioral and Brain Sciences, 32(02), 121–38. Choi, S., and Gopnik, A. (1995). Early acquisition of verbs in Korean: A crosslinguistic study. Journal of Child Language, 22(03), 497–529. Churchland, P. M. (1981). Eliminative materialism and the propositional attitudes. The Journal of Philosophy, 78(2), 67–90. Custer, W. L. (1996). A comparison of young children’s understanding of contradictory representations in pretense, memory, and belief. Child Development, 67(2), 678–88. Dancy, J. (2000). Practical Reality. New York: Oxford University Press. Davidson, D. (1963). Actions, reasons, and causes. The Journal of Philosophy, 60(23), 685–700. Davies, M., and Stone, T. (2001). Mental simulation, tacit theory, and the threat of collapse. Philosophical Topics, 29(1&2), 127–73. De Villiers, J. (2007). The interface of language and Theory of Mind. Lingua, 117(11), 1858–78. Dennett, D. C. (1971). Intentional systems. The Journal of Philosophy, 68(4), 87–106. (1978). Beliefs about beliefs. Behavioral and Brain Sciences, 1, 568–70. Dienes, Z., and Perner, J. (1999). A theory of implicit and explicit knowledge. Behavioral and Brain Sciences, 22(05), 735–808. Epley, N., and Waytz, A. (2009). Mind perception. In S. T. Fiske and D. T. Gilbert and G. Lindsay (eds.), The Handbook of Social Psychology, 5th Edition (pp. 498–541). New York: Wiley. Fletcher, L., and Carruthers, P. (2013). Behavior-Reading versus Mentalizing in Animals. In J. Metcalfe and H. Terrace (eds.), Agency and Joint Attention. New York: Oxford University Press.

Knowledge as a Mental State

305

Fodor, J. A. (1987). Psychosemantics: The Problem of Meaning in the Philosophy of Mind: Oxford, England: British Psychological Society; Cambridge, MA: The MIT Press. Fricker, E. (2009). Is Knowing a State of Mind? The Case Against. In P. Greenough and D. Pritchard (eds.), Williamson on Knowledge (pp. 31–60). New York: Oxford University Press. Gallese, V., and Goldman, A. (1998). Mirror neurons and the simulation theory of mind-reading. Trends in Cognitive Sciences, 2(12), 493–501. Gettier, E. L. (1963). Is Justified True Belief Knowledge? Analysis, 23, 121–3. Gillette, J., Gleitman, H., Gleitman, L., and Lederer, A. (1999). Human simulations of vocabulary learning. Cognition, 73, 135–76. Goldman, A. (1967). A Causal Theory of Knowing. The Journal of Philosophy, 64(12), 357–72. (2006). Simulating Minds: The Philosophy, Psychology, and Neuroscience of Mindreading. New York: Oxford University Press. Gopnik, A., and Wellman, H. (1992). Why the child’s theory of mind really is a theory. Mind & Language, 7(1 2), 145–71. Gordon, R. (1986). Folk psychology as simulation. Mind & Language, 1(2), 158–71. Grice, H. P. (1975). Logic and conversation. Syntax and Semantics, 3, 41–58. Hare, B., Call, J., and Tomasello, M. (2001). Do chimpanzees know what conspecifics know? Animal Behaviour, 61(1), 139–51. Harman, G. (1978). Studying the chimpanzee’s theory of mind. Behavioral and Brain Sciences, 1(04), 576–7. Hazlett, A. (2010). The myth of factive verbs. Philosophy and Phenomenological Research, 80(3), 497–522. Heyes, C. M. (1998). Theory of mind in nonhuman primates. Behavioral and Brain Sciences, 21(01), 101–14. Hogrefe, G. J., Wimmer, H., and Perner, J. (1986). Ignorance versus false belief: A developmental lag in attribution of epistemic states. Child Development, 57, 567–82. Holton, R. (1997). Some telling examples: A reply to Tsohatzidis. Journal of pragmatics, 28(5), 625–8. Hubbard, T. L. (1995). Environmental invariants in the representation of motion: Implied dynamics and representational momentum, gravity, friction, and centripetal force. Psychonomic Bulletin and Review, 2(3), 322–38. Jackson, F. (2007). Is belief an internal state? Philosophical Studies, 132(3), 571–80. Johnson, C. N., and Wellman, H. M. (1980). Children’s developing understanding of mental verbs: Remember, know, and guess. Child Development, 51(4), 1095–102. Jones, E. E., and Harris, V. A. (1967). The attribution of attitudes. Journal of Experimental Social Psychology, 3(1), 1–24. Kaminski, J., Call, J., and Tomasello, M. (2008). Chimpanzees know what others know, but not what they believe. Cognition, 109(2), 224–34. Keysar, B. (2007). Communication and miscommunication: The role of egocentric processes. Intercultural Pragmatics, 4(1), 71–84. Koriat, A. (1995). Dissociating knowing and the feeling of knowing: Further evidence for the accessibility model. Journal of Experimental Psychology: General, 124(3), 311–33. Kornblith, H. (2004). Knowledge and its Place in Nature. Oxford: Oxford University Press.

306

Jennifer Nagel

Kozhevnikov, M., and Hegarty, M. (2001). Impetus beliefs as default heuristics: Dissociation between explicit and implicit knowledge about motion. Psychonomic Bulletin and Review, 8(3), 439–53.. Krachun, C., Carpenter, M., Call, J., and Tomasello, M. (2009). A competitive nonverbal false belief task for children and apes. Developmental Science, 12(4), 521–35. Lin, S., Keysar, B., and Epley, N. (2010). Reflexively mindblind: Using theory of mind to interpret behavior requires effortful attention. Journal of Experimental Social Psychology, 46(3), 551–6. McCloskey, M., and Kohl, D. (1983). Naive physics: The curvilinear impetus principle and its role in interactions with moving objects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9(1), 146–56. McDowell, J. (1986). Singular thought and the extent of inner space. In J. McDowell and P. Pettit (eds.), Subject, Thought and Context (pp. 137–68). Oxford: Clarendon Press. Magnus, P., and Cohen, J. (2003). Williamson on knowledge and psychological explanation. Philosophical Studies, 116(1), 37–52. Miller, S., Hardin, C., and Montgomery, D. (2003). Young children’s understanding of the conditions for knowledge acquisition. Journal of Cognition and Development, 4(3), 325–56. Miscione, J. L., Marvin, R. S., O’Brien, R. G., and Greenberg, M. T. (1978). A Developmental Study of Preschool Children’s Understanding of the Words “Know” and “Guess”. Child Development, 49(4), 1107–13. Molyneux, B. (2007). Primeness, Internalism and Explanatory Generality. Philosophical Studies, 135(2), 255–77. Moore, C., Bryant, D., and Furrow, D. (1989). Mental terms and the development of certainty. Child Development, 60(1), 167–71. Mossler, D. G., Marvin, R. S., and Greenberg, M. T. (1976). Conceptual perspective taking in 2-to 6-year-old children. Developmental Psychology, 12(1), 85. Nagel, J. (2012). The psychological basis of the Harman-Vogel paradox. Philosophers’ Imprint, 11(5), 1–28. Nagel, J. (2011). Intuitions and Experiments: A defense of the case method in epistemology. Philosophy and Phenomenological Research, 85(3), 495–527. Nichols, S., and Stich, S. (2003). Mindreading: An Integrated Account of Pretence, Selfawareness, and Understanding other Minds. New York: Oxford University Press. Okamoto-Barth, S., Call, J., and Tomasello, M. (2007). Great apes’ understanding of other individuals’ line of sight. Psychological Science, 18(5), 462–68. Onishi, K. H. and Baillargeon, R. (2005). Do 15-month-old infants understand false beliefs? Science, 308(5719), 255–58. Papafragou, A., Cassidy, K., and Gleitman, L. (2007). When we think about thinking: The acquisition of belief verbs. Cognition, 105(1), 125–65. Perfect, T. J., Watson, E. L., and Wagstaff, G. F. (1993). Accuracy of confidence ratings associated with general knowledge and eyewitness memory. Journal of Applied Psychology, 78(1), 144. Perner, J. (1991). Understanding the Representational Mind. MIT Press Cambridge, MA. (1995). The many faces of belief: Reflections on Fodor’s and the child’s theory of mind. Cognition, 57(3), 241–69.

Knowledge as a Mental State

307

(2010). Who took the cog out of cognitive science? Mentalism in an era of anticognitivism. In P. French and R. Schwarzer (eds.), Cognition and Neuropsychology: International Perspectives on Psychological Science (Vol. 1, pp. 241–61). New York: Psychology Press. Pillow, B. H., Hill, V., Boyce, A., and Stein, C. (2000). Understanding inference as a source of knowledge: Children’s ability to evaluate the certainty of deduction, perception, and guessing. Developmental Psychology, 36(2), 169. Plantinga, A. (1993). Warrant: The Current Debate. New York: Oxford University Press. Plato. (1990). The Theaetetus of Plato (M. J. Levett, Trans.). Indianapolis: Hackett. Povinelli, D. J., and Eddy, T. J. (2000). What Young Chimpanzees Know about Seeing. New York: Wiley-Blackwell. Povinelli, D. J., Rulf, A. B., and Bierschwale, D. T. (1994). Absence of knowledge attribution and self-recognition in young chimpanzees. Journal of Comparative Psychology, 108(1), 74. Povinelli, D. J. and Vonk, J. (2003). Chimpanzee minds: Suspiciously human? Trends in Cognitive Sciences, 7(4), 157–60. Premack, D., and Woodruff, G. (1978). Does the chimpanzee have a theory of mind? Behavioral and Brain Sciences, 1(04), 515–26. Putnam, H. (1975). Mind, Language and Reality. Cambridge: Cambridge University Press. Reber, R., and Unkelbach, C. (2010). The Epistemic Status of Processing Fluency as Source for Judgments of Truth. Review of Philosophy and Psychology, 1, 563–581. Ross, L. (1977). The intuitive psychologist and his shortcomings: Distortions in the attribution process. Advances in Experimental Social Psychology, 10, 173–220. Ruffman, T. (1996). Do children understand the mind by means of simulation or a theory? Evidence from their understanding of inference. Mind & Language, 11(4), 388–414. Russell, J. (2005). Justifying all the fuss about false belief. Trends in Cognitive Sciences, 9(7), 307–8. Samson, D., Apperly, I., Braithwaite, J., Andrews, B., and Scott, S. (2010). Seeing it their way. Journal of Experimental Psychology-Human Perception and Performance, 36(5), 1255–66. Saxe, R. (2005). Against simulation: The argument from error. Trends in Cognitive Sciences, 9(4), 174–9. Shatz, M., Wellman, H. M., and Silber, S. (1983). The acquisition of mental verbs: A systematic investigation of the first reference to mental state. Cognition, 14(3), 301–21. Shope, R. K. (1983). The Analysis of Knowing: A Decade of Research. Princeton, NJ: Princeton University Press. Sloman, S. (1996). The empirical case for two systems of reasoning. Psychological Bulletin, 119(1), 3–22. Sodian, B., and Kristen, S. (2010). Theory of mind. In B. M. Glatzeder and V. Goel and A. von Muller (eds.), Towards a Theory of Thinking (pp. 189–201). Berlin: SpringerVerlag. Sodian, B., Thoermer, C., and Dietrich, N. (2006). Two-to four-year-old children’s differentiation of knowing and guessing in a non-verbal task. European Journal of Developmental Psychology, 3(3), 222–37.

308

Jennifer Nagel

Sodian, B. and Wimmer, H. (1987). Children’s understanding of inference as a source of knowledge. Child Development, 58(2), 424–33. Sosa, E. (2007). A Virtue Epistemology: Apt Belief and Reflective Knowledge (Vol. 1). New York: University Press. Stanley, J. (2011). Know How. New York: Oxford University Press. and Williamson, T. (2001). Knowing how. The Journal of Philosophy, 98(8), 411–44. Steup, M. (2008). The analysis of knowledge. Stanford Encyclopedia of Philosophy, . Accessed 28 November 2012. Stich, S. and Ravenscroft, I. (1994). What is folk psychology? Cognition, 50(1–3), 447–68. Tardif, T. and Wellman, H. M. (2000). Acquisition of mental state language in Mandarin-and Cantonese-speaking children. Developmental Psychology, 36(1), 25–42. Tomasello, M. and Call, J. (1997). Primate cognition. New York: Oxford University Press. and Hare, B. (1998). Five primate species follow the visual gaze of conspecifics. Animal Behaviour, 55(4), 1063–9. Turri, J. (2011). Manifest failure: The Gettier problem solved. Philosophers’ Imprint, 11(8), 1–11. Unger, P. (1971). A defense of skepticism. The Philosophical Review, 80(2), 198–219. Wellman, H. M. and Liu, D. (2004). Scaling of Theory of Mind Tasks. Child Development, 75(2), 523–41. Williamson, T. (1995). Is knowing a state of mind? Mind, 104(415), 533–565. (2000). Knowledge and its Limits. New York: Oxford University Press. (2009). Replies to Critics. In P. Greenough and D. Pritchard (eds.), Williamson on Knowledge (pp. 279–84). New York: Oxford University Press. Wimmer, H. and Perner, J. (1983). Beliefs about beliefs: Representation and constraining function of wrong beliefs in young children’s understanding of deception. Cognition, 13(1), 103–28. Woolfe, T., Want, S. C., and Siegal, M. (2002). Signposts to development: Theory of mind in deaf children. Child Development, 73(3), 768–78. Woolley, J. and Bruell, M. (1996). Young children’s awareness of the origins of their mental representations. Developmental Psychology, 32(2), 335–46. Zagzebski, L. (1994). The inescapability of Gettier problems. The Philosophical Quarterly, 44(174), 65–73. Zwaan, R. A. and Radvansky, G. A. (1998). Situation models in language comprehension and memory. Psychological Bulletin, 123(2), 162–185.

11. What Does Knowledge Explain? Commentary on Jennifer Nagel, ‘Knowledge as a Mental State’ Stephen A. Butterfill

1.

INTRODUCTION

Nagel contrasts two views. On the view she opposes, adult humans’ ‘understanding of action is fundamentally dependent on belief attribution’ in such a way that attributing knowledge somehow depends on being able to attribute belief (chapter 10, section 1, para.1). On the view Nagel defends, ‘the capacity to recognize belief depends on some prior mastery of the concept of knowledge’ (section 2, para. 10).1 Put roughly, the contrast concerns whether being able to recognize knowledge depends on being able to recognize belief or whether the converse dependence holds. In part of what follows I shall argue, contra Nagel, that currently available evidence fails on balance to support either view. The evidence points to a less straightforward but more interesting picture of mindreading. Seeing this will require considering a broader range of evidence than Nagel discusses. Nagel’s primary aim, of course, is to defend the claim that knowledge is a mental state in its own right (rather than being reducible to belief, truth, and other ingredients). As she sees things, defending a view about mindreading is useful, and perhaps necessary, for defending the claim about what knowledge is. I shall object to this way of seeing things. And, in the final section, I shall sketch an alternative view, one which involves holding on to the claim that knowledge is a mental state while remaining neutral on Nagel’s views about dependence and priority. On this alternative, some reasons for regarding knowledge as a mental state are closely related to reasons for regarding intention as a mental state. Knowledge and intention play complementary and interlocking roles in planning and practical reasoning. It is these roles (rather than claims about dependence or priority) which block attempts to identify either knowledge or intention with special kinds of belief and desire. 1 Nagel also puts her view by saying that ‘an ability to track what others would know seems to be the precondition, rather than the product, of an ability to track what they would believe’ (chapter 10, para. 6). It may be important to distinguish mastery of the concept of knowledge from abilities to track what others know.

310

2.

Stephen A. Butterfill

M I N D R E A D I N G I S N ’ T C O N C E P T UA L A N A LY S I S

In defending the claim that knowledge is a mental state, Nagel aims to show that ‘the identification of knowledge as a mental state is one of the central principles of our intuitive mindreading systems’ (last paragraph). Is it true that intuitive mindreading systems identify knowledge as a mental state? Nagel argues for a positive answer, partly on the grounds that knowledge features in intuitive explanations of action. This by itself is not sufficient grounds. Even if knowledge features in intuitive explanations of action, it doesn’t follow that knowledge is identified as a mental state. For things other than mental states, such as facts, can feature in intuitive explanations of action. And there is no reason to suppose that all such things are intuitively identified, incorrectly, as mental states. So it would be a mistake to suppose that knowledge is identified as a mental state just because it features in intuitive explanations of action. But do facts really feature in intuitive explanations of action? The grounds for holding that they do are closely related to those for holding that knowledge so features, and the case for facts is stronger than the case for knowledge. Consider, for example, this explanation: Ayesha went inside because it was getting dark. Note that this explanation could hold even if Ayesha neither knew nor believed that it was getting dark (perhaps, for instance, the change in lighting, although unnoticed, affected her mood or made her feel tired). As this indicates, appealing to facts allows us to explain actions which we could not explain by appeal to mental states only, and such explanations have greater generality in one dimension than comparable explanations involving mental states. This is one reason, not decisive but significant, for holding that facts feature along with mental states in intuitive explanations of action. Not everything which features in intuitive explanations of action should be identified as a mental state. Are there any reasons to doubt Nagel’s claim that intuitive mindreading systems identify knowledge as a mental state? One reason is that there seems to be no need for such systems to make an identification either way. Among their functions are prediction and explanation of thought and action. Performing functions such as these surely involves identifying factors which predict or cause actions. But it doesn’t seem to require taking a view on whether these factors are mental, nor even on whether they are states.2 If I were a mindreading system, I would want to remain neutral on which things are mental states. So far I have only been skimming the surface of Nagel’s argument. A core aim of hers is to oppose the claim that knowledge is a composite of belief, 2 Hyman [27, p. 451] argues that propositional knowledge is an ability ‘to act, to refrain from acting, to believe, desire or doubt for reasons that are facts.’ I am not persuaded that he is right, but I don’t think intuitive mindreading systems need to risk the possibility that he is by identifying knowledge as a state.

What Does Knowledge Explain?

311

truth, and other factors. Nagel sees this claim as a key reason for denying that knowledge is a mental state. And she suggests that truths about the role of knowledge in intuitive mindreading provide reasons to reject the claim. In outline, one strand of her argument is this: 1. The capacity to represent belief is not in place until after the capacity to represent knowledge is. This entails that: 2. Intuitive representation of knowledge cannot be ‘a composite involving intuitive representation of belief’. This in turn supports the view that: 3. ‘knowledge is naturally seen [by ordinary mindreaders] as a mental state, and not as a composite of belief and non-mental factors’. Which in turn is evidence that: 4. Knowledge is not ‘a composite of belief and non-mental factors’. It is not my intention to argue that knowledge is composite or that it is not a mental state. I am not committed to either claim. But I do think there are several problems with the above line of argument. Below I shall suggest that, on balance, the currently available evidence does not support (1). But first, does (2) really support (3)? To see that it does not we need to be careful about the distinction between a representation of something as non-composite and a representation which is non-composite. Nagel’s (2) is about representations which are non-composite. Tracking knowledge by means of non-composite representations does not necessarily make it natural to see knowledge as non-composite. To see why not, consider a parallel. Imagine individuals who can represent coffee but not caffeine. These individuals’ intuitive representation of coffee cannot be a composite involving an intuitive representation of caffeine. But, you are to imagine, coffee features in their explanations of action. For instance, they explain variations in their own and others’ performance by appeal to coffee consumption. And in many cases appeal to coffee consumption allows them to give better (relative to their own ends, at least) explanations than they could give if they were to appeal to caffeine or other coffee components. Clearly none of this is evidence that coffee is not a composite involving caffeine. Nor does it suggest that they naturally see coffee as non-composite, since they have no reason to do so. As this example indicates, non-composite representations of things which are in fact composite are not necessarily misrepresentations and not necessarily defective relative to the ends they serve.3 3 Fricker [18, p. 51] makes a different but related point: ‘There is absolutely no tension between knowing’s being a good explanatory state, and each instance of knowing being a conjunctive, hybrid phenomenon.’

312

Stephen A. Butterfill

This is why having a non-composite representation of knowledge would not make it natural to see knowledge as non-composite. So even assuming ‘a nonskeptical attitude’ (chapter 10, last para.) to intuitive mindreading, (2) does not support (3).4 Perhaps I have misinterpreted Nagel’s argument.5 It may be that she did not intend to offer (2) plus premises I have missed as evidence for (3). Instead her claim may be that philosophers’ motivation for denying (3) or (4) involves an assumption that (2) is false. Perhaps, then, Nagel’s argument for (2) is designed only to remove some of the motivation for rejecting (3) or (4). One difficulty with this interpretation of Nagel is that the opponents she mentions do not seem to be so motivated. For instance, Fricker (2009, p. 35) allows (possibly only for the sake of argument) that knowledge is ordinarily taken to, and does, explain action before arguing against the claim that knowledge is a mental state. And Magnus and Cohen (2003, pp. 39–40) claim that knowledge is explanatory of action only insofar as it has a narrow component, and that this narrow component would be the only causally efficacious ingredient in knowledge. This does not seem to commit them to any view about intuitive mindreading, and their claim is motivated by metaphysics not psychology. We should also be cautious about the inference from (1) to (2) in the above argument. Recall the coffee-representing individuals but now imagine a further stage of their development in which they acquire a capacity to represent caffeine. Now we can no longer be sure that their representation of coffee is not a composite involving representations of caffeine. After all, this further stage might involve a change in their representation of coffee. Similarly, the fact (if it is a fact—see below) that humans acquire a capacity to represent knowledge before they can represent belief does not entail that their representation of knowledge is not eventually a composite involving representation of belief. So far I have argued that Nagel demands too much of intuitive mindreading systems. As far as we know, their functions are bound up with the particular, with, say, what Miss Kelly knows about Tony and how this will shape her actions towards Eilis. Fulfilling these roles doesn’t seem to require identifying knowledge as a mental state (or otherwise), nor does it require representing in ways which reveal whether or not it is a composite involving belief.

4 Nagel also appears to suggest that anyone who, like Williamson [53], holds that belief should be explained in terms of knowledge has reason to hold that ‘the capacity to recognize belief depends on some prior mastery of the concept of knowledge’ (section 2, para. 10). This is not straightforward if, as Nagel allows elsewhere (in footnote 20), it is possible to recognize something without knowing everything about it and, in particular, without knowing every conceptual truth about it. 5 I offer the above interpretation first in part because Nagel writes that ‘Evidence from social, developmental and comparative psychology seems to support the view that knowledge is naturally seen . . . not as a composite of belief and non-mental factors’ (last paragraph of the introduction to the paper).

What Does Knowledge Explain?

3.

313

NO EVIDENCE FOR PRIORITY

The previous section examined some ways Nagel connects controversy over whether knowledge is a mental state with claims about mindreading. In this section I want to set aside that controversy in order to focus just on mindreading. Right at the core of her paper, Nagel argues that ‘the concept of knowledge is in some sense prior to the concept of belief’ (last para. of section 2). Why think that the concept of knowledge is in any sense prior to the concept of belief? Nagel is motivated in part by developmental evidence.6 As she notes, earlier research shows that children can reliably answer questions about knowledge and ignorance months or years before they can answer questions involving false belief (Hogrefe et al. 1986). So a 3-year-old might be able reliably to report whether someone knows something and, relatedly, which of two or more people know it while systematically failing to correctly answer questions about what someone with a false belief will think, do, or say (Wellman et al. 2001). Further, children’s sensitivity to knowledge and ignorance appropriately guides a range of decisons, such as whether to rely on what someone says (Robinson et al. 1999; Robinson and Whitcombe 2003) and whether to provide information about the location of an object (Dunham et al. 2000; Liszkowski et al. 2008). Nagel seems to interpret these findings as evidence that ‘the concept of knowledge [is] prior to the concept of belief’ (section 3).7 What Nagel doesn’t mention, however, is that the picture becomes more complicated when we take a wider view encompassing both earlier and later points in development. Infants are sensitive to others’ false beliefs from around seven months of age or earlier (Kovács et al. 2010). From soon after their first birthday or earlier infants manifest sensitivity to belief in a variety of ways. It is not just that infants look longer when an agent who apparently has a false belief acts as if she knew, which is evidence that such actions violate infants’ expectations (Onishi and Baillargeon 2005; Surian et al. 2007). It is also that infants’ eye movements anticipate actions based on false beliefs (Southgate et al. 2007), and that facts about others’ false beliefs shape some communicative actions (Knudsen and Liszkowski 2012) and to some extent guide word-learning (Carpenter et al. 2002) as well as modulating attempts to help others (Buttelmann et al. 2009). These findings complicate the interpretation of older children’s failure to pass standard false belief tasks.8 6 Nagel also considers linguistic development and offers an a priori conjecture about the relative computational costs of attributing knowledge and belief. While Nagel’s discussion of cognitive development is quite brief, it may be the strongest part of her case for a priority claim. 7 Nagel doesn’t explicitly say that there is evidence for this claim. What she says is just that the view is ‘widely held’ and that the experimental work she cites ‘may help to clarify why psychologists . . . generally take the concept of knowledge to be prior to the concept of belief’ (section 3, para. 8). The key issue, though, is surely what the evidence shows. 8 While this research has recently attracted renewed interest, some of the credit should also go to much earlier work which established sensitivity to false belief significantly before children

314

Stephen A. Butterfill

Of course none of the evidence from infants establishes (at least not in any straightforward way) that typically developing humans do not deploy a concept of knowledge before they deploy a concept of belief. After all, it is an open question whether infants’ abilities are best explained by their having a concept of belief. And if we do accept that these abilities are evidence that infants have a concept of belief, it remains possible that capacities to represent knowledge appear before capacities to represent belief. My point is just that currently available evidence does not straightforwardly support the view that children deploy the concept of knowledge before they deploy the concept of belief. But to fully appreciate this point, we also need to look at later developments in children’s understanding of knowledge. Children’s competence in dealing with knowledge appears to develop over several years. Around the fourth, fifth, and sixth years of life there are marked improvements in children’s understanding of expertise (Lutz and Keil 2002; Sobel and Corriveau 2010), of the links between what people know and what they say or might be able to tell them (Robinson 1994; Robinson et al. 2010), and of sources of knowledge (O’Neill et al. 1992; O’Neill and Chong 2001; Robinson et al. 2008). As this might suggest, there is debate about whether 2- and 3-year-olds’ talk about, and sensitivity to, knowledge is best explained by supposing that it involves deploying a concept of knowledge. The leading alternative is not the conjecture that these children (or other animals showing related knowledge-tracking abilities) are merely deploying ‘behavioural rules’. It is the conjecture that these individuals have a fragmentary and limited understanding of epistemic phenomena, somewhat as earlier scientists were able to identify several electrical phenomena including charge and current without yet having understood how these are connected (or even realizing that they are connected). According to this conjecture, children and possibly other animals are sensitive to whether others are engaged or disengaged in an event and, when helpful, seek to provide updates about events accordingly (O’Neill 2005, pp. 88–9; Virányi et al. 2005).9 Children are also sensitive to whether others have a history of reliability and they can use reliability in accepting or rejecting information offered by others (Koenig and Harris 2005; Birch et al. 2008). But these two patterns of sensitivity, to engagement and to reliability, may be only weakly integrated at first (Nurmsoo and Robinson 2009a, b). A conjecture along these lines can explain how children and other animals are able, within limits, to track others’ knowledge and ignorance and requires no commitment either way on the issue of whether they have a concept of knowledge.10 pass standard false belief tasks. See Clements, Garnham and Perner, Garnham and Ruffman, Ruffman et al. [12, 20, 21, 47]. 9 For notions related to O’Neill’s engagement, see Doherty, Gomez [14, 23] on intentional relations to objects, Call and Tomasello [10] on tacking targets of visual access, and Butterfill and Apperly [9] on encountering and registration. 10 If it is possible to use words to refer to things while knowing little or nothing about what one is referring to, such a conjecture may also be consistent with Nagel’s suggestion

What Does Knowledge Explain?

315

None of this shows, of course, that children at some particular point in their development lack the concept of knowledge. But there does seem to be a dilemma for the interpretation Nagel seems to rely on. If we take 2- and 3-year-old children’s abilities to discriminate knowledge and ignorance as evidence that they are deploying a concept of knowledge, then it will be hard to justify denying that early sensitivity to belief does not involve deploying a concept of belief. If, on the other horn of the dilemma, we insist that these discriminatory abilities are not sufficient for concept possession, then we can no longer infer from 2- and 3-year-old children’s failure on standard false belief tasks that children deploy a concept of knowledge before they deploy a concept of belief. Either way, current evidence does not (at least not in any straightforward way) support the claim that ‘children acquire the concept of knowledge before the concept of belief’ (section 3, first para.).11 It is perhaps tempting to conclude that developmental evidence is just too messy to be relevant to philosophy. But there is at least as much justification for seeing things the other way around. Some of the puzzles which arise in studying development are due to inadequacies in the philosophical groundwork. Few philosophical theories of things like action, concepts, knowledge, mindreading, and mental states are sufficiently developed to be useful in constructing and testing theories aimed at explaining how minds function, develop, or evolve. In some cases, for example, we are challenged to divide development into two phases, pre- and post-concept-of knowledge, or challenged to divide animals into those with and those without this concept. The complex pattern of findings in developmental and comparative research (not just in the case of mindreading, but also in research on physical reasoning,12 number cognition13 and awareness of speech14 ) indicates that such divisions are sometimes unilluminating. We need better conceptual tools.

4.

A N A LT E R N AT I V E

I want to finish by comparing and contrasting Nagel on knowledge with Bratman on intention. The comparison is inexact but points to an alternative to Nagel’s strategy for defending the claim that knowledge is a mental state, one that avoids commitment either way on claims about dependence and priority.

that ‘the child’s early use of ‘know’ . . . should charitably been seen as referring to knowledge’ (footnote 3). 11 It is easy to miss this point by emphasizing a distinction between verbal and non-verbal measures. But there appear to be insufficient grounds for conjecturing that, in the experiments under discussion, this distinction maps onto a distinction between tasks which do involve some mastery of the concept of knowledge and tasks which don’t. 12 e.g. Berthier et al., Baillargeon, Hood et al. [3, 2, 26]. 13 e.g. Xu, Feigenson and Carey, Gallistel and Gelman, Gelman and Butterworth [54, 17, 19, 22]. 14 e.g. Eimas et al., Jusczyk, Anthony and Lonigan, Liberman and Liberman [16, 28, 1, 32].

316

Stephen A. Butterfill

Bratman (1999; 1987) aims to show that intention is a state distinct from belief and desire. For instance, he opposes the view that intentions are special kinds of beliefs about the future (1999, pp. 257ff). Relatedly, Nagel argues that knowledge is a mental state and opposes attempts to analyse knowledge in terms of belief, truth, and other ingredients. Bratman identifies functional and normative roles for intention which cannot be played by belief or desire or any combination of them. In particular, he suggests that intentions play a distinctive role in deliberative planning linked to coordination of present and future action (1999, p. 223). One consequence of this is that if someone were to refrain from ascribing intention and confine herself to belief and desire only, her abilities to explain thought and action would be compromised. Relatedly, Nagel argues that ascribing knowledge sometimes yields better explanations than ascribing belief would. The similarities are not striking, I admit. But the differences are interesting. As we saw earlier, Nagel aims to show that knowledge not a special kind of belief by arguing that ‘the concept of knowledge is in some sense prior to the concept of belief’ (section 2, last para.). She associates this claim about priority with the view that belief can be analysed in terms of knowledge and with the view that ‘the capacity to recognize belief depends on some prior mastery of the concept of knowledge’ (section 2, para. 10). Bratman’s position, by contrast, involves no such claims about priority. It does not require supposing that belief or desire can be analysed in terms of intention. And it does not require holding that capacities to recognize belief or desire depend on having the concept of intention. It is consistent with Bratman’s view (but not required) to hold that some mindreaders can ascribe beliefs and desires but not intentions, and that an ability to ascribe intentions would be a further, more sophisticated achievement. I mention this because earlier I offered some objections to Nagel’s arguments for claims about priority. Now I want to consider whether, taking Bratman’s approach as a model, it is possible to hold that knowledge is a mental state without commitment to any sort of priority. If this is indeed possible, objections to arguments for priority are not necessarily objections to the thesis that knowledge is a mental state. There is a conceptual distinction between, first, an agent’s having beliefs and desires which rationally (in a decision-theoretic sense of ‘rational’) guide her actions and, second, an agent’s deliberatively planning her actions. We need this distinction in order to recognize the characteristic roles of intention. But, as we shall see, the distinction is also linked to characteristic roles of knowledge, roles that distinguish it from belief. Hawthorne (2004, pp. 29–31) defends the view that, with some exceptions, we should take as premises in our practical reasoning only propositions that we know. To illustrate, take Rose who is deciding whether to accept a job offer. She believes with justification and conviction that her grant application will be rejected but does not actually know this. Hawthorne’s view implies that, in deciding whether to take the job, Rose should not rely on her grant application

What Does Knowledge Explain?

317

being rejected. Note that this is a normative claim. The claim is not that Rose won’t in fact rely on being rejected in her practical reasoning. It is that (with exceptions) she should not do so. Hawthorne’s normative claim is linked to a claim about a role for knowledge: it connects an agent’s actions to facts about her environment in such a way that they can be reasons justifying her actions from her own point of view. Intention exists in part so that agents can coordinate their actions over time; and knowledge exists in part so that agents can base their plans on facts which are recognizably reasons. On this view, then, knowledge and intention play complementary roles in practical reasoning. It is because they play these roles that ascribing knowledge and intention makes it possible to explain some events more fully than could be achieved by ascribing belief and desire only. And it is these roles (rather than any priority claim) which complicate attempts to identify either knowledge or intention with special kinds of belief and desire.15 Here, then, in barest outline, is an alternative approach to defending the claim that knowledge is not a special kind of belief. No doubt this alternative faces many objections. But one attraction is that it avoids commitment to the sorts of priority claim Nagel endorses. It is possible to agree with Nagel that knowledge is a mental state while remaining neutral on whether the concept of knowledge is in her sense prior to the concept of belief.

REFERENCES

Anthony, J. L. and Lonigan, C. J. (2004). The nature of phonological awareness: Converging evidence from four studies of preschool and early grade school children. Journal of Educational Psychology, (1), 43–55. Baillargeon, R. (2002). The acquisition of physical knowledge in infancy: A summary in eight lessons. In U. Goswami (ed.), Blackwell Handbook of Childhood Cognitive Development (pp. 47–83). Oxford: Blackwell. Berthier, N. E., De Blois, S., Poirier, C. R., Novak, M. A., and Clifton, R. K. (2000). Where’s the ball? Two- and three-year-olds reason about unseen events. Developmental Psychology, (3), 394–401. Birch, S. A., Vauthier, S. A., and Bloom, P. (2008). Three- and four-year-olds spontaneously use others’ past performance to guide their learning. Cognition, (3), 1018–34. Bratman, M. E. (1999). Faces of Intention. Cambridge: Cambridge University Press. (1987). Intentions, Plans, and Practical Reasoning. Cambridge MA: Harvard University Press. (2009). Modest sociality and the distinctiveness of intention. Philosophical Studies, (1), 149–65. 15 The view sketched is compatible with there being other roles for knowledge, as there certainly are for intention. For instance, knowledge and intention may be characterized in part by their roles in social interaction Craig, Bratman [13, 6].

318

Stephen A. Butterfill

Buttelmann, D., Carpenter, M., and Tomasello, M. (2009). Eighteen-month-old infants show false belief understanding in an active helping paradigm. Cognition, (2), 337–42. Butterfill, S. and Apperly, I. A. (submitted 2011). How to construct a minimal theory of mind. . Call, J. and Tomasello, M. (2005). What chimpanzees know about seeing revisited: An explanation of the third kind. In N. Eilan, C. Hoerl, T. McCormack, and J. Roessler (eds.), Joint Attention:Communication and other Minds (pp. 45–64). Oxford: Oxford University Press. Carpenter, M., Call, J., and Tomasello, M. (2002). A new false belief test for 36-montholds. British Journal of Developmental Psychology, (20), 393–420. Clements, W. and Perner, J. (1994). Implicit understanding of belief. Cognitive Development, (9), 377–95. Craig, E. (1990). Knowledge and the State of Nature. Oxford: Clarendon Press. Doherty, M. J. (2006). The development of mentalistic gaze understanding. Infant and Child Development, (15), 179–86. Dunham, P. J., Dunham, F., and O’Keefe, C. (2000). Two-year-old’s sensitivity to a parent’s knowledge state: Mind reading or contextual cues? British Journal of Developmental Psychology, (18), 519–32. Eimas, P. D., Siqueland, E. R., Jusczyk, P., and Vigorito, J. (1971). Speech perception in infants. Science, (3968), 303–6. Feigenson, L. and Carey, S. (2005). On the limits of infants’ quantification of small object arrays. Cognition, (3), 295–313. Fricker, E. (2009). Is knowing a state of mind? The case against. In P. Greenough & D. Pritchard (eds.), Williamson on Knowledge. Oxford: Oxford University Press. Gallistel, C. and Gelman, R. (2000). Non-verbal numerical cognition: From reals to integers. Trends in Cognitive Sciences, (2), 59–65. Garnham, W. and Perner, J. (2001). Actions really do speak louder than words—but only implicitly: Young children’s understanding of false belief in action. British Journal of Developmental Psychology, (19), 413–32. Garnham, W. and Ruffman, T. (2001). Doesn’t see, doesn’t know: Is anticipatory looking really related to understanding or belief? Developmental Science, (1), 94–100. Gelman, R. and Butterworth, B. (2005). Number and language: How are they related? Trends in Cognitive Sciences, (1), 6–10. Gomez, J.-C. (2007). Pointing behaviors in apes and human infants: A balanced interpretation. Child Development, (3), 729–34. Hawthorne, J. O. (2004). Knowledge and Lotteries. Oxford: Oxford University Press. Hogrefe, G., Wimmer, H., and Perner, J. (1986). Ignorance versus false belief: A developmental lag in attribution of epistemic states. Child Development, (3), 567–82. Hood, B., Cole-Davies, V., and Dias, M. (2003). Looking and search measures of object knowledge in preschool children. Developmental Science, (1), 61–70. Hyman, J. (1999). How knowledge works. Philosophical Quarterly, (197), 433–51. Jusczyk, P. (1995). Language acquisition: Speech sounds and the beginning of phonology. In L. Miller, Joanne and P. D. Eimas (eds.), Speech, Language and Communication. San Diego: Academic Press. Knudsen, B. and Liszkowski, U. (forthcoming 2012). 18-month-olds predict specific action mistakes through attribution of false belief, not ignorance, and intervene accordingly. Infancy.

What Does Knowledge Explain?

319

Koenig, M. and Harris, P. (2005). Preschoolers mistrust ignorant and inaccurate speakers. Child Development, (6), 1261–77. Kovács, Á. M., Téglás, E., and Endress, A. D. (2010). The social sense: Susceptibility to others’ beliefs in human infants and adults. Science, (6012), 1830–4. Liberman, I. Y. and Liberman, A. M. (1990). Whole language vs. code emphasis: Underlying assumptions and their implications for reading instruction. Annal of Dyslexia, (40), 51–76. Liszkowski, U., Carpenter, M., and Tomasello, M. (2008). Twelve-month-olds communicate helpfully and appropriately for knowledgeable and ignorant partners. Cognition, (3), 732–9. Lutz, D. J. and Keil, F. C. (2002). Early understanding of the division of cognitive labor. Child Development, (4), 1073–84. Magnus, P. and Cohen, J. (2003). Williamson on knowledge and psychological explanation. Philosophical Studies, (116), 37–52. Nurmsoo, E. and Robinson, E. J. (2009a). Children’s trust in previously inaccurate informants who were well or poorly informed: When past errors can be excused. Child Development, (1), 23–7. (2009b). Identifying unreliable informants: Do children excuse past inaccuracy? Developmental Science, (1), 41–7. O’Neill, D. K. (2005). Talking about ‘new’ information: The given/new distinction and children’s developing theory of mind. In J. Astington and J. A. Baird (eds.), Why language matters for theory of mind (pp. 84–105). Oxford: Oxford University Press. Astington, J., and Flavell, J. (1992). Young children’s understanding of the role that sensory experiences play in knowledge acquisition. Child Development, (2), 474–90. O’Neill, D. K. and Chong, S. (2001). Preschool children’s difficulty understanding the types of information obtained through the five senses. Child Development, (3), 803–15. Onishi, K. H. and Baillargeon, R. (2005). Do 15-month-old infants understand false beliefs? Science, (8), 255–8. Robinson, E. (1994). What people say, what they think, and what is really the case: Children’s understanding of utterances as sources of knowledge. In C. Lewis and P. Mitchell (eds.), Children’s Early Understanding of Mind: Origins and development (pp. 355–381). Hove: Erlbaum. Butterfill, S., and Nurmsoo, E. (2010). Gaining knowledge via other minds: Children’s flexible trust in others as sources of information. British Journal of Developmental Psychology (29), 961–980. Robinson, E., Champion, H., and Mitchell, P. (1999). Children’s ability to infer utterance veracity from speaker informedness. Developmental Psychology, (2), 535–46. Robinson, E., Haigh, S. N., and Pendle, J. E. C. (2008). Children’s working understanding of the knowledge gained from seeing and feeling. Developmental Science, (2), 299–205. Robinson, E. and Whitcombe, E. (2003). Children’s suggestibility in relation to their understanding about sources of knowledge. Child Development, (1), 48–62. Ruffman, T., Garnham, W., Import, A., and Connolly, D. (2001). Does eye gaze indicate implicit knowledge of false belief? Charting transitions in knowledge. Journal of Experimental Child Psychology, (80), 201–24.

320

Stephen A. Butterfill

Sobel, D. M. and Corriveau, K. H. (2010). Children monitor individuals’ expertise for word learning. Child Development, (2), 669–79. Southgate, V., Senju, A., and Csibra, G. (2007). Action anticipation through attribution of false belief by two-year-olds. Psychological Science, (7), 587–92. Surian, L., Caldi, S., and Sperber, D. (2007). Attribution of beliefs by 13-month-old infants. Psychological Science, (7), 580–6. Virányi, Z., Topál, J., Miklósi, á., and Csányi, V. (2005). A nonverbal test of knowledge attribution: A comparative study on dogs and children. Animal Cognition, (9), 13–26. Wellman, H., Cross, D., and Watson, J. (2001). Meta-analysis of theory of mind development: The truth about false-belief. Child Development, (3), 655–84. Williamson, T. (2000). Knowledge and Its Limits. Oxford: Oxford University Press. Xu, F. (2003). Numerosity discrimination in infants: Evidence for two systems of representations. Cognition, (89), B15–25.

12. Knowledge, Causal Explanation, and Teleology Johannes Roessler Can knowledge make a difference? Is Janet’s reaching into her bag causally explained, partly, by the fact that she knows that her mobile phone is in the bag? In ‘Knowledge as a Mental State’ Jennifer Nagel spearheads a debate across three perspectives on this question. The standard view in philosophy is that knowledge is not causally relevant: it’s belief that ‘does the causal work’. Connectedly, on this view, knowing that p is a mental state only in what Timothy Williamson labels the ‘dull’ sense, i.e. courtesy of the mental state it entails or involves, viz., believing that p. Knowing that p is not a mental state in the contentious and interesting sense—the one intended, and endorsed, by Williamson—that there is a kind of mental state that is both necessary and sufficient for knowing that p. (Williamson 1995: 533) It’s here that Nagel finds a head-on collision between current orthodoxy in philosophy and psychology. ‘The standard view in psychology’, she maintains, ‘is that knowledge is a mental state; indeed, knowledge typically features prominently in lists of paradigmatic mental states’ (chapter 10, the beginning of section 3). Towards the end of her article, the two disciplines are joined by what might loosely be described as a third discipline: the view of commonsense psychology or, in the Humean phrase, the ‘vulgar’ view. Nagel characterizes this view as follows: ‘our natural mindreading systems parse agency with the help of the epistemic mental state concepts of knowledge and belief rather than belief alone’ (the beginning of section 4). Let’s call the view that knowledge is a mental state (in the contentious sense) mentalism. Nagel’s project may be described, in abstract terms, as aimed at showing that the ‘vulgar’ view is mentalist, and that developmental psychology can help to deepen our understanding of its mentalist commitments. I enthusiastically agree with both points. But I would like to question some of Nagel’s more specific claims. A chief concern of her paper is with the idea of conceptual priority. As she presents the debate, a view shared by mainstream psychologists and Williamson is that the concept of belief is ‘something derived from the concept of knowledge.’ (paragraph 5 of the paper, my emphasis) I find both attributions, and the view itself, doubtful, and in any case I think mentalism is not committed to it. I develop and defend these points in the first two sections. In section 3, I focus on the ‘vulgar’ view’. Why do we find it so natural to explain Janet’s action in terms of her knowledge?

322

Johannes Roessler

In Nagel’s discussion, commonsense mentalism features as a naïve theory of the causes of behaviour. I want to suggest that this account conceals the rationale of commonsense mentalism, which lies in the role of knowledge in providing us with reasons. We can deepen our understanding of that rationale by looking at the pre-history of commonsense psychology: specifically, at young children’s teleological schema of action explanation (on the hypothesis proposed by Perner and Roessler 2010). This genealogy helps to bring out the connection between mentalism and the pre-theoretical conception we have of ourselves and others as rational agents.

1.

C O N C E P T UA L P R I O R I T Y

According to a strong tradition in philosophy, the concept of belief is in an important sense more basic that the concept of knowledge. This is evidently a view Williamson rejects. He frequently inveighs against the idea that ‘belief is conceptually prior to knowledge’ (2000: 8). Nagel’s Williamson not only rejects the traditional priority thesis; he also endorses its converse. He claims that the concept of knowledge is ‘more basic’ than the concept of belief, and that the latter is ‘derived’ from the former. There is a tendency, in Nagel’s discussion, to conflate mentalism with this converse priority claim. I want to suggest that the two views need to be sharply distinguished, and not just in the interest of proper bookkeeping. Mentalists are not compelled to trade one priority claim for another. They may opt for a ‘no priority’ view. What does it mean to say that one concept is prior to another? It’s natural to think that conceptual priority has something to do with the order of analysis. For example, on the traditional view, the concept of belief is prior in the sense that it denotes one of the necessary and jointly sufficient conditions for knowledge, where these conditions are seen as providing for a non-circular, reductive analysis of the concept of knowledge. Let’s call this analytic priority. A related idea might be that mastery of the concept of knowledge requires mastery of the concept of belief but not vice versa. Let’s call this psychological priority. Note that psychological priority has a certain modal force: it’s impossible to grasp knows without grasping believes. This is one respect in which the claim differs from what might be called developmental priority, the idea that normally developing humans grasp the concept of belief before they grasp the concept of knowledge. According to the converse priority claim Nagel is interested in, and appears to endorse, knows is prior to believes in all three senses. In the next section, I look at the evidence regarding developmental priority. In the rest of this section, I consider these questions: (1) Does mentalism commit one to Nagel’s priority thesis? (2) Does Williamson endorse the thesis? (3) How plausible is the thesis? Mentalism is a claim about the state of knowing, not about the concept of knowledge. You might wonder, therefore, whether as a metaphysical thesis,

Knowledge, Causal Explanation, and Teleology

323

it may not simply be neutral on conceptual issues. This would be a mistake, though. Mentalism arguably does commit one to the rejection of the traditional priority view. Crudely put, the idea is this. Suppose believes is analytically prior to knows. This would imply that there is a successful reductive analysis of knows in terms of the more basic concepts believes, is true, and so on. These more basic concepts would, in turn, enable us to identify the circumstances whose obtaining would be sufficient for the state of knowing. So knowledge would turn out to be a ‘composite’ state, involving, among other things, the mental element of believing that p and the (typically) non-mental element of p being true. If this were so, believing that p would be a kind of mental state that’s necessary for knowing that p, but there would be no kind of mental state that is necessary and sufficient for knowing that p. So knowing would not be a mental state. There is more to be said about this argument, but in the current context the important question is: does mentalism also have positive implications as regards conceptual priority? Does it commit one to accepting the priority of knows over believes? One reason for denying this is that mentalism would appear to be neutral on the question whether believes is open to any interesting conceptual analysis at all (let alone one in terms of knows). So it’s hard to see how mentalism could entail the analytic priority of knows. Again mentalism surely carries no commitment that one can grasp knows independently of mastery of the concept of belief: a mentalist might coherently think of the two conceptual abilities as interdependent. Finally, mentalists can afford to keep an open mind on the developmental trajectory. Certainly if there were convincing evidence that knows is understood before believes this would put pressure on the traditional priority view, and would perhaps count in favour of mentalism. But if it turned out that the two concepts were acquired in the reverse order, mentalists should lose no sleep over this. They never claimed that either concept was more basic than the other. Nagel’s priority claim is inspired, partly, by a striking remark of Williamson’s, that ‘believing p is, roughly, treating p as if one knew p’ (2000: 47). The remark may seem to suggest that we can explain what it means to believe something in terms of a prior, independent understanding of knowledge. So we should not just reject, but reverse the traditional priority claim. But it’s not clear that the text licenses such an interpretation. Notice, first, that Williamson’s slogan is not intended as an attempt at conceptual analysis. The slogan is preceded by this disclaimer: ‘Although a full-blown exact conceptual analysis of believes in terms of knows is too much to expect, we can still postulate a looser connection along these lines’ (2000: 47). A claim to analytic priority does not seem to be what Williamson has in mind at all. Then how should the ‘looser connection’ be understood? Williamson apparently intends the slogan to express what he regards as an important normative thesis, stated in the sentence succeeding the slogan: ‘Knowledge sets the standard of appropriateness of belief’ (2000: 47). The idea here is that belief does not just ‘aim at truth’ but at knowledge: ‘Mere believing is a kind of botched knowing.’

324

Johannes Roessler

(2000: 47). This raises a number of questions, but the point that matters in the present context is that Williamson’s normative thesis is neutral both on the analytic and psychological priority of knows. It would presumably entail that understanding belief requires understanding knowledge. But it’s consistent with allowing that understanding knowledge in turn requires understanding belief. It’s consistent with a ‘no priority’ view. One reason Williamson shies away from a ‘full-blown conceptual analysis’ is that he thinks there are counterexamples: ‘one might know p while in a sense treating p as if one did not know p’ (2000: 47). But there is also the question whether the slogan can be elaborated without invoking belief. What does it mean to treat p as if one knew p? Williamson says it is ‘to treat p in ways similar to the ways in which subjects treat propositions which they know’ (2000: 46–7). One thing he has in mind is reliance on a proposition in practical reasoning. Evidently more would need to be said about what counts as relevantly similar here. The way we treat p when we pretend that p is not dissimilar to the way we treat propositions we know. If we mark the difference by saying that the relevant ways of treating p should involve acceptance of p, not much constructive progress towards a full-blown analysis would have been made. These points do not amount to decisive objections to the priority of knows. What they bring out is that it would be precipitate to expect that an analysis of believes in terms of knows will not set in motion the cycle of counterexamples, modifications, and further counterexamples, familiar from discussions of the traditional priority claim. Nagel appears to think her converse priority claim is needed to accommodate the fact that knows entails believes. She writes that Williamson ‘allows the entailment by characterizing belief as a potentially diluted version of knowing’ (chapter 10, section 2, para. 10). The assumption here seems to be that to ‘allow’ the entailment we have to make it intelligible in terms of an analysis (or characterization) of one of the concepts involved. It’s not clear exactly how the idea that believing p is treating p as if one knew p is supposed to make the entailment intelligible. In any case, the assumption is far from trivial, and it is one that Williamson explicitly rejects.1

2.

KNOWLEDGE FIRST?

Nagel offers two reasons for her claim that the ‘standard view in psychology is that knowledge is a mental state’. One is the observation that knowledge frequently features in psychologists’ lists of mental states: it’s considered to be one of the states children need to learn about in acquiring a ‘theory of mind’. The second reason turns on Nagel’s reading of the current state of play in psychology regarding the relative priority of knows and believes, specifically on her claim that it is ‘generally agreed [in psychology] that belief is the more

1 ‘More generally, the existence of conceptual connections is a bad reason to postulate an analysis of a concept to explain them.’ (2000: 33)

Knowledge, Causal Explanation, and Teleology

325

sophisticated concept, and is harder to attribute than knowledge’ (section 3, para. 1). The inclusion of knowledge in psychologists’ lists is certainly a striking datum, but I’m not convinced it can bear the weight Nagel assigns to it. The question whether knowledge is a mental state, not just in what Williamson calls the ‘dull’ sense but in the contentious sense, is a distinctly philosophical question. It turns on issues such as whether knowledge is a conjunctive state, whether the concept of knowledge is analyzable, or whether the apparent causal relevance of knowledge can be reductively explained in terms of other factors. That psychologists tend to include knowledge in their lists provides no evidence that they are inclined to take a stand on these philosophical problems. In my experience psychologists are disinclined to do so even when pressed. A plausible interpretation of the inclusion of ‘knowledge’ in the lists is that a ‘theory of mind’ is assumed to involve the kinds of concepts we use in making sense of our own and others’ actions. That knowledge plays a prominent role in this enterprise is something not even proponents of the standard ‘philosophical’ view of knowledge would deny. (I’ll come back to this in the final section.) Recent findings on children’s understanding of knowledge can be sorted into three groups: (i) Findings that appear to show that attributions of knowledge are easier than attributions of belief. For instance, Sodian et al. 2006 found clear evidence that 3-year-olds are able to discriminate between good and bad informants: when asked to choose between two possible helpers in a hide-and-seek task, they tend to select the knowledgeable helper— the one who saw the object being hidden. This contrasts with 3-yearolds’ notoriously poor performance on classical false-belief tasks. (ii) Findings that appear to show that the two kinds of attributions are equally hard. For instance. Gopnik and Graf 1988 found that 4-year-olds were much better than 3-year-olds at answering questions about how they knew what was inside a box (whether they saw it, were told, or received a clue). (iii) Findings that appear to show that attributions of knowledge are harder than attributions of belief. Several kinds of studies suggest that up until the age of about 5 children dramatically overestimate the range of facts made manifest to us by perceiving an object. For instance, 4-year-olds confidently predict that someone will be able to recognize a depicted animal on the basis of seeing a small, unidentifiable fragment of the drawing (Chandler and Helm 1984). This complex picture of the developmental progression raises several theoretical possibilities. One is that the easier tasks fail to probe children’s understanding of propositional knowledge. For example, it might be said that 3-year-olds’ adeptness at simple ignorance tasks reflects their understanding

326

Johannes Roessler

of the role of seeing or looking as an enabling condition of successful action, with the word ‘knowledge’ being used simply to express a disposition for acting successfully (Perner 1991: 304). A second possibility is that children’s mastery of the concept of knowledge gradually matures between the ages of 3 and 6. Thus Elizabeth Robinson summarizes her recent review of the evidence by saying that there is, over this time span, ‘no single point at which we would deem a child to have achieved understanding [of the connection between perceptual access and knowledge state]’ (Robinson 2010: 338; see Doherty 2009: 67 for a similar view). Finally, it may be held that 3-year-olds have an unimpaired grasp of the concept of knowledge and that their difficulties with the harder tasks reflect extraneous factors, such as higher demands on working memory. Only the third theoretical possibility is consistent with Nagel’s developmental claim that belief is ‘harder to attribute than knowledge’. The first possibility would suggest that belief and knowledge are either equally hard or that knowledge is harder. From the second possibility we could conclude, at best, that a less-than-fully-comprehending use of ‘knows’ is easier than a comprehending use of ‘believes’. Nagel does recognize the experimental challenge facing her priority claim. She raises the question whether children are ‘really referring to knowledge as such’ when tackling simple ignorance tasks, and she refers to ‘studies of immature and evolving mental concept use’ (section 3, para. 10) Having raised the question, Nagel unexpectedly turns to comparative psychology, with a view to ascertaining whether non-human primates may find knowledge easier to attribute than belief. I won’t go into this part of her discussion, but it seems to me that even if, as she argues, there is strong support for the view that chimpanzees are able to recognize knowledge (but not belief), it’s hard to see how this helps to resolve the debate over the interpretation of human 3-year-olds’ performance on ignorance tasks. As indicated, the evidence is far from clear cut. But it seems to me closer scrutiny suggests that the third theoretical possibility is the least plausible one. Advocates of the third possibility think that success on simple ignorance tasks brings to light a real understanding of the connection between seeing and knowing: 3-year-olds realize that the reason only one of two protagonists knows which kind of object is inside a box is that only she saw the object. It’s not just that children understand something about object perception, for example that you can’t see an object when it’s placed inside a non-transparent box (unless the lid if lifted). They also grasp that seeing an object makes it possible to see where the object is and what it is like, for example to see (and hence know) that it is a pen. With this basic explanatory schema in place, children can be said to grasp the difference between knowledge and ignorance. On this analysis, even 3-year-olds know what knowledge is, notwithstanding their well-documented tendency to attribute perceptual knowledge a bit promiscuously. It’s not clear, though, that this basic explanatory scheme is sophisticated enough to warrant the attribution of the concept of knowledge. The problem is that someone’s seeing an object O is not, on its own, an adequate explanation

Knowledge, Causal Explanation, and Teleology

327

of her knowledge that O is F. There is a second dimension to our conception of the conditions that make epistemic seeing possible. The subject also has to be able to recognize the object as having the relevant property or falling under the relevant type. Whether in seeing O the subject can see (without inference) that O is F depends on familiar conditions such as which kind of property is in question, the subject’s recognitional abilities, distance, viewpoint, lighting, and so forth. Suppose you don’t believe in the importance of such conditions. You think that seeing an object O that is in fact F is sufficient for seeing that O is F. I think this would raise legitimate doubts as to whether you know what epistemic seeing is. It might be said that you simply subscribe to an eccentric theory, on which seeing an object, through the operation of some magical mechanism, makes the perceiver omniscient about the object. But that would be to assume that you do appreciate the need for some further explanation of how object perception yields knowledge (it’s just that your explanation is inadequate). If, as we are supposing, you don’t see that need at all, it’s hard to understand just what you have in mind. On one interpretation, findings of type (ii) and (iii) suggest that your puzzling conception of epistemic seeing is the one that prevails among 3-year-olds. They show that young children are oblivious to the second dimension of our ordinary conception of the enabling conditions of epistemic seeing. For example, Martin Doherty hypothesizes that ‘children simply think that any direct perceptual access to an object is sufficient to know all aspects of the object’ (2009: 65). If this interpretation is correct, we should be reluctant to credit 3-year-olds with a proper understanding of epistemic seeing, despite their success on simple ignorance tests.

3.

REASONS

Anti-mentalists deny that Janet reaches into her bag because she knows her mobile is there. The causally relevant mental state, they think, is belief. This doesn’t mean that anti-mentalists are unaware that we frequently make sense of what people do in terms of what they know. In a passage often cited as a canonical statement of an ‘internalist’ approach to causal relevance, Jaegwon Kim writes: ‘I know that if I turn this knob counterclockwise the burner will go on. Since I want the fire to go on, I turn the knob. My knowledge that turning the knob will cause the burner to go on plays a causal role in the explanation of my action of turning the knob. This is a simple and familiar sort of action explanation’ (1993: 188). Note that Kim doesn’t simply dismiss such talk, as you might dismiss appeal to the putative efficacy of telepathy. Our mistake in crediting knowledge with causal powers, on his view, is merely that we conflate what are in fact two distinct explanations. First, there is the explanation of his turning the knob, in terms of his belief that turning the knob causes the burner to go on. Kim thinks that only this part of the overall explanation is correctly labelled psychological. It’s no business of psychology, he insists, to explain why the burner went on. Second, his turning the knob,

328

Johannes Roessler

together with the non-psychological fact that doing so causes the burner to go on, explains his turning on the burner. Actually, if the ostensible explanatory role of knowledge is to be successfully dismantled, we need to distinguish a third explanatory project (though Kim does not mention this). Suppose we are interested in whether turning on the burner was something he did intentionally. That the action is explained by his wanting to turn on the burner and his believing that he can do so by turning the knob provides no adequate answer to this question. If his belief is true by sheer luck—if, for example, he has conclusive evidence that the burner is not operational—the action won’t be intentional under the description ‘turning on the burner’.2 On the face of it, the fact that he acts on the basis of knowledge is highly pertinent to our understanding of the action as intentional. On Kim’s reductive account, this will have to be analyzed in terms of non-factive notions, such as his acting on the basis of a justified belief (no doubt plus further conditions). Nagel’s distinctive response to this reductive project is to highlight the naturalness of the natural view. In her concluding section she writes that knowledge is ‘naturally’ perceived as a mental state, that ‘our natural mindreading systems parse agency with the help of epistemic mental state concepts’ (section 4, para. 1), and that we should place ‘trust in our natural instincts’ about people’s mental states (section 4, para. 4). Notice, though, that the reductionist can agree with a lot of this. Kim characterizes explanations that give a causal role to knowledge as ‘simple and familiar’. He should be equally happy to acknowledge that they are natural and intuitive. His question is whether they are correct. The most explicit statement I could find of Nagel’s position on this question is the following: ‘If the test of a theory is its capacity to generate good predictions and explanations, then anyone with a non-skeptical attitude towards intuitive mindreading should see the thesis that knowledge is a mental state as well confirmed’ (last para. of the article). It does seem plausible that if intuitive mindreading yields good—i.e. correct— explanations, that would be a compelling reason to adopt a non-skeptical attitude towards it. But reductionists think that the intuitive explanatory role of knowledge fails to make salient the real causal-explanatory structure of the situation. Strictly speaking, such explanations are not correct. Appeal to the mindreading system’s alleged capacity to generate good explanations will cut no ice with a reductionist. Of course reductionists can and should grant that explanations in terms of knowledge may be ‘good’ in other ways. They are simple, useful, and very natural to us. That point, though, is not enough to establish that mentalism is a well-confirmed thesis. The real issue surely is not whether reductionism is revisionist—it’s agreed on all hands that it is—but whether the reductive project is mandatory, and whether it can succeed. This is of course a large and difficult question. A number of debates in the philosophy of mind are relevant to it. I won’t try to

2

Gibbons 2001 makes a persuasive case for this.

Knowledge, Causal Explanation, and Teleology

329

answer the question here, but I want to end with a suggestion as to what makes the ‘vulgar’ view so natural. This is not the same as the question whether the view is correct, but to assess the credentials of the vulgar view we need a proper understanding of its rationale. We can usefully approach this issue by reflection on the pre-history of the commonsense ‘theory of mind’. 3 Three-year-olds find it hard to understand how someone’s beliefs, or for that matter knowledge, affect their intentional actions. Does that mean 3-yearolds lack the very notion of intentional action? If to act intentionally is to act for a reason, and practical reasons are pairs of beliefs and desires, it’s hard to see how the idea of someone doing something intentionally could be available to 3-year-olds. It would be no help, for example, to suggest that at least they have some understanding of the explanatory role of desires. A desire, on its own, does not amount to a practical reason. The key to the solution to this problem lies in young children’s performance on standard false belief tasks. The striking finding here is that when asked where the protagonist will go (in order to retrieve a desirable object, surreptitiously displaced by someone else) children are not guessing. Rather, they consistently and confidently give the wrong answer. They are adamant that Maxi is heading towards the cupboard (where the chocolate is) rather than the drawer (where he put it). Children assume, it seems, that Maxi will do what it makes sense for him to do, i.e. what he has reason to do. The point can be obscured by the widespread tendency amongst philosophers of mind to equate reasons with the explanantia of ‘rationalizing explanations’, i.e. mental states such as beliefs and desires. Ordinarily, though, we don’t think of reasons in that way. Suppose you advise Maxi on where he should go. The obvious point to make would be that he should go to the cupboard, given that this is where the chocolate is. It’s the fact that the chocolate is in the cupboard that is a reason for him to go there. It would be perverse for you to tell Maxi that he should go to the drawer, on the grounds that he believes that that’s where the chocolate is. True, there is a sense in which Maxi should go to the drawer: it’s rational for him to do so, given his belief. But there is also an obvious sense in which it’s the cupboard he should go to. What we need here is a distinction between two kinds of practical ‘ought’: in Nico Kolodny’s terms, the ‘ought of rationality’ and the ‘ought of reasons’ (Kolodny 2005). You may expect that someone will do what he has reason to do because you think he is well-informed and sensible. But in the case of young children a more plausible hypothesis is this: they expect that people act intentionally, and they take intentional actions to be explained by reasons, rather than by ‘rationalizing’ mental states. In other words, prior to the acquisition of a ‘theory of mind’ children use a ‘teleological’ schema of action explanation, invoking evaluative and instrumental facts. One attraction of this hypothesis

3 For a more detailed exposition and defence of the developmental hypothesis I’m about to sketch, see Perner and Roessler 2010; Roessler and Perner (in press).

330

Johannes Roessler

is that it points to a solution to the problem of how it is possible for children to have even a rudimentary idea of what it means to act intentionally. Another attraction is that the hypothesis may explain various shortcomings of children’s explicit thinking about actions.4 There is an obvious point of contact between children’s primitive teleology and the ‘vulgar’ view of knowledge: both assign causal relevance to reason-giving facts. The teleological schema does this simply by ignoring the explanatory role of agents’ perspective on what they have reason to do. Under children’s austere version of the principle of charity, it’s assumed that agents will do what they ought to do (in the ‘ought of reasons’ sense). Mature commonsense psychology is mindful of the importance of the agent’s perspective, but conceives of perspectives as involving factive mental states. If Janet reaches into her bag because she needs to make a phone call and knows that her mobile phone is in the bag, the attitude that makes her action rational is one that entails that there is a reason for what she is doing. 5 Mentalism thus protects an element of the primitive teleology via which we are inducted into the commonsense conception of people as intentional agents. Perhaps we find mentalism natural partly for that reason. More significantly, this commonality between teleology and mentalism brings out one sense in which the reductionist dismantling of the causal role of knowledge would seem to involve ‘explanatory loss’. If you explain Janet’s action by invoking her belief rather than her knowledge, the explanation no longer provides any reassurance as to whether she reached into her bag for a reason. Your explanation is after all consistent with Janet’s mobile phone having been stolen, in which case she would have been mistaken in believing that there was a good reason for her to reach into her bag. In extracting the mobile, the pickpocket would have simultaneously deprived Janet’s action of its point. It’s here, I think, that we may be able to see the beginnings of an explanation of what makes the ‘vulgar’ view natural to us. We like to think of our actions as not merely rational—something that is consistent with their lacking any real point—but done for reasons. This is certainly a natural view to take for a deliberating agent. The point of deliberation is to determine what one will do through reflection on one’s reasons, and it would be nonsensical to deliberate if one didn’t take deliberation to be (in general) a way of settling what one will in fact do. As Richard Moran remarks, ‘there is no point in calling it “deliberation” any more, if [the deliberator] takes it to be an open question whether this activity will determine what he actually does or believes’ (2001: 127). Moran rightly characterizes the perspective of deliberation as one ‘where reasons that justify are at issue’. Less plausibly, he contrasts this with the ‘stance of causal explanation’. If a deliberator is committed to thinking of deliberation as 4 See Perner and Roessler 2010, for a review of relevant experimental findings, and Priewasser, Roessler, and Perner (in press) for new evidence. 5 For elaboration and defence of the view that acting for the justifying reason that p requires knowing that p, see Hornsby 2008. For dissent, see Dancy 2011.

Knowledge, Causal Explanation, and Teleology

331

(generally) effective, it’s hard to see how ‘reasons that justify’ can be insulated from concerns with causal explanation. A deliberator will naturally think of such reasons as the very reasons for which she is acting. She will be aware, of course, at least by the age of 4, that to make a difference, reasons must be apparent to the agent. But this realization does not compel a separation of the ‘stances’ of justification and explanation. A reason that’s apparent to the agent need not be a merely apparent reason. It may be a consideration the agent knows to be true.6 If this analysis is on the right lines, mentalism is natural to us, not because it is a theory we (or ‘our natural mindreading systems’) take to generate valuable explanations and predictions of people’s conduct. We find mentalism intuitive because we naturally think of our own and others’ intentional actions from the perspective of deliberators, rather than from the perspective of theorists. Whether mentalism can be sustained is of course a further question, but the teleological analysis of what makes it natural to us certainly does not make things easier for the reductionist. She may either seek to analyze the content of our natural belief in the causal relevance of reasons in terms she regards as unproblematic. For example, she might insist that all we should be taken to have in mind is that actions are caused by beliefs and desires. Alternatively, she might debunk the natural belief in the efficacy of reasons as a remnant of a teleological, ‘pre-disenchanted’ (and possibly infantile) view of causal explanation. The first option is arguably implausible as an account of reasons for actions, and distorts the way we ordinarily think about such reasons. The second option, if Moran is right, would jettison a commitment integral to our ordinary sense of self. Whether this amounts to an objection is a question that would require detailed investigation. It may turn out that reductionism would take with it rather more than one might at first realize.7

REFERENCES

Chandler, M. J. and Helm, D. (1984). ‘Developmental changes in the contribution of hsared experience to social role-taking competence’, International Journal of Behavioural Development 7, 145–54. Dancy, J. (2011). ‘Acting in ignorance’, Frontiers of Philosophy in China 6, 345–57. Doherty, M. (2009). Theory of Mind. Hove and New York: Psychology Press. Gibbons, J. (2001). ‘Knowledge in action’, Philosophy and Phenomenological Research LXII, 579–600. Gopnik, A. and Graf, P. (1988). ‘Knowing how you know—young children’s ability to identify and remember the sources of their beliefs’, Child Development 59, 1366–71. Hornsby. J. (2008). ‘A disjunctive conception of acting for reasons’, in A. Haddock and F. Macpherson (eds.), Disjunctivism. Oxford: Oxford University Press. 6 Of course the truth of the consideration is merely a necessary, not a sufficient condition for its amounting to a reason for some particular course of action. 7 Many thanks to Josef Perner for extensive knowledgeable help.

332

Johannes Roessler

Kim. J. (1993). ‘Psychophysical supervenience’, in his Supervenience and Mind. Cambridge: Cambridge University Press. Kolodny, N. (2005). ‘Why be Rational?’, Mind 114, 509–63. Moran, R. (2001). Authority and Estrangement. Princeton: Princeton University Press. Nagel, J. (this volume). ‘Knowledge as a mental state’. Perner, J. (1991). Understanding the Representational Mind. Cambridge, MA: MIT Press. and Roessler, J. (2010). ‘Teleology and causal understanding in children’s theory of mind’, in J. Aguilar and A. Buckareff (eds.), Causing Human Actions. Cambridge, MA: MIT Press. Priewasser, B., Roessler, J. and Perner, J. (in press). ‘Competition as rational action: Why young children cannot appreciate competitive games’, Journal of Experimental Child Psychology (available online November 2012). Robinson, E. (2010). ‘Development of understanding of the causal connection between perceptual access and knowledge state’, in J. Roessler, H. Lerman, and N. Eilan (eds.), Perception, Causation, and Objectivity. Oxford: Oxford University Press. Roessler, J. and Perner, J. (in press). ‘Teleology: Belief as Perspective’, in S. BaronCohen et al., Understanding Other Minds (Third Edition). Oxford: Oxford University Press. Sodian, B., Thoermer, C, and Dietrich, N. (2006). ‘Two- to four-year-old children’s differentiation of knowing and guessing in a non-verbal task’, European Journal of Developmental Psychology 3, 222–37. Williamson, T. (1995). ‘Is knowing a state of mind’, Mind 104, 533–63. (2000). Knowledge and its Limits. Oxford: Oxford University Press.

13. Is Knowledge a Non-Composite Mental State? Patrick Rysiew

1.

INTRODUCTION

In “The Need to Know,” Fred Dretske writes: Getting things right is not just a useful skill. It is a biological imperative. Behavior has to be coordinated with the external conditions on which its success depends. An animal doesn’t want to be running all the time; only when there is something chasing it. Courtship and mating activities are nice, but only with a partner. (Dretske 1989: 89)

For social creatures such as ourselves, the need to reliably track certain states of our fellows is no less real or pressing than the need to track certain features of the physical landscape: if “needing to know” about “the external conditions” is “a biological imperative,”1 so is needing to know such things as what others believe, and whether they possess knowledge. Coordinating with them requires it, as does being in a position to effectively exploit them as useful sources of information. According to epistemological orthodoxy, there is a certain priority amongst the states of others we need to track: knowing, the orthodox thinking goes, involves having a belief that meets certain further conditions—most centrally, the belief must be true (as the factivity of “know(s)” suggests); and the belief must be justified, warranted, or what have you.2 So knowledge is a composite state. It is, moreover, not merely composite but hybrid (Williamson 2000)— that is, it is a composite state that incorporates both clearly mental factors (belief, of course; and, plausibly, at least some factors relevant to whether it is warranted) and clearly non-mental ones (truth, of course; and, on certain views, at least some factors relevant to whether it is warranted). Adam Leite sums up the common thinking nicely: As the traditional conception of knowledge has it . . . one’s knowing a particular proposition about the world is often a complex state or condition comprising both

1 Granted, in the quoted passage, it’s ‘getting thing right,’ which may or may not require knowing, that Dretske mentions. As his discussion (not to mention the paper’s title) reveals, however, it is knowledge he’s referring to. 2 There has been much discussion about which among the latter such factors and terms is the most apt and/or promising. (See note 4 below.) Fortunately, such disputes don’t have any direct bearing upon the present discussion.

334

Patrick Rysiew

purely mental factors (such as belief) and non-mental, environmental factors. The term “mental state” can, of course, function as a term of art and may reasonably be extended in certain ways for theoretical purposes. But we have a pretheoretical conception of the mental which prompts ready agreement with the traditional view. The claim that knowledge is nothing but a mental state comes as a surprising philosophical thesis. (2005, 166).

In “Knowledge as a Mental State,” Jennifer Nagel looks to turn this traditional thinking on its head: not only is our “pretheoretical conception of the mental” actually such that we naturally treat knowledge as a mental state alongside belief, desire, and so on, but prominent arguments against regarding knowledge as a mental state can be reasonably challenged; further, there is empirical evidence which strongly suggests that the concept of knowledge is prior to that of belief, and not the other way around. More broadly, and as befits the (as she sees it) thoroughly mental status of knowledge, Nagel seeks to portray our ordinary knowledge-ascriptions as the product of our general mind-reading capacity, and our ability to track others’ epistemic states as a reliable and computationally feasible solution to the specifically psychological ‘need to know’ described above. Nagel’s paper gives us much to admire, and to talk about. It’s a model of how to usefully integrate empirical considerations and findings with philosophical arguments and theories in a way that’s likely to benefit researchers in both disciplines. The result is a compelling case, if not for abandoning epistemological orthodoxy, then at least for thinking hard about it in a way we seldom do. In the end, however, it’s not clear whether Nagel’s discussion, and in particular the psychological work she discusses, really is a threat to epistemological orthodoxy. This is not because of any suspicions about that work itself, but rather because (Section 3) it is not clear whether it really is at odds with the epistemological orthodoxy Nagel seeks to dislodge. Before turning to that, however, we should consider just what’s at issue in debates around whether knowledge is a mental state, and wherein exactly the idea that it is conflicts with the orthodox view.

2.

K N O W L E D G E A S A M E N TA L S TAT E ?

While the ostensible topic of Nagel’s paper is whether knowledge is a mental state, it’s important to note that it’s not around precisely that issue that most of the discussion and argument revolves. Rather, the central issue is whether knowledge is—as, according to the orthodox view, it’s naturally and best conceived of as being—composite, in the above-indicated sense. Whether knowledge is composite, and whether it’s mental, are issues that, in fact, we might do well to keep distinct.3

3

I should stress that I don’t mean to be implying any kind of confusion on Nagel’s part here.

Is Knowledge a Non-Composite Mental State?

335

But isn’t the thought that knowledge is composite part-and-parcel of the view that it isn’t a mental state? Certainly, people’s views on the two issues tend to align, but it’s not clear whether that’s because of any essential connection between them. Once again, on the orthodox view, being in the state of knowledge is a composite state, comprised of having a belief that is true, justified, and so forth. On anyone’s view—at least, on the view of anyone party to the present debate—it’s uncontroversial that belief is a mental state, and that truth is not. So knowledge is, again, a hybrid state. But isn’t that enough to give us the result that it’s non-mental? It might seem so. As Nagel observes: States and events that are composites of the mental and non-mental are not naturally classified as mental: homicide, for example, is not a mental event. Although every homicide incorporates the mental condition of an intention to kill (indeed this mental condition, mens rea, is the part of the essence of homicide that distinguishes it from involuntary manslaughter), an event cannot be a homicide without also incorporating the non-mental condition of someone’s actual death. The inclusion of this nonmental conjunct disqualifies the event considered as a whole from counting as purely mental. (Section 2, para. 2)

However, while the addition or involvement of an external factor sometimes renders a hybrid state non-mental, it seems that it doesn’t always obviously do so. For instance, the fact that externalist semantic theses, as applied to belief, are a going theory shows that the mere involvement of “external” factors doesn’t always make a state non-mental in the obvious way in which homicide is. On a familiar “two-factor” view of belief, for example, “we can think of belief as built up from a narrow factor, which supervenes on what is in the head, together with a broad factor, which fails to supervene on what is in the head” (Magnus and Cohen 2003, p. 39). This type of view may or may not be correct, and it’s open to question which of the two “factors” (broad or narrow) does the work in psychological explanations and/or in virtue of what exactly propositional attitudes are “causally efficacious” (Magnus and Cohen 2003, p. 40). The point, however, is that it’s not usually said, in connection with such views, that the involvement of “a broad factor” in our understanding of the relevant state has the immediate, not to mention surprising, consequence that beliefs are not mental after all. In the same vein, there are a variety of states naturally classed as mental, at least for everyday purposes, that are clearly world-involving. These are factive states like remembering, regretting, resenting, and so on (cf. Williamson 2000, p. 22). Perhaps these states are not mental in some strict sense; perhaps they are mental only in the same way that many orthodox theorists hold the composite state of knowledge to be—maybe they too are hybrid and at most only “impurely” mental. Once again, though, the salient point is that they are not obviously non-mental in the way that homicide is. So while it’s not being suggested here that these states are all best viewed as mental, there’s

336

Patrick Rysiew

an important sense in which they could be. For just this reason, it’s open to a proponent of the view that knowledge is a composite of belief, truth, and various other things, to hold that that composite is, in the best theory, properly regarded as a mental state. Putting the point the other way around: from just the fact that a theorist (philosopher or not) lists knowledge as among the mental states, we can’t infer their position as to the compositeness (or not) of that state. All of this, I take it, is harmless and uncontroversial enough. But, just as obviously, it doesn’t yet engage with the views of those who—like Williamson and Nagel—seek to oppose the orthodox view. For their point isn’t merely that knowledge is best classed as a mental state. Rather, they aim to establish the stronger thesis that knowledge is purely mental—that is, that it is not metaphysically hybrid. As Williamson puts it, the claim that knowing is a state of mind is to be understood as the claim that there is a mental state being in which is necessary and sufficient for knowing p. In short, knowing is merely a state of mind. (2000, p. 21)

And of course it’s clear from her discussion that, for Nagel too, the core issue is whether knowledge is “a mental state in its own right, and not . . . a composite of belief and non-mental factors” (last para. of the introduction; emphasis added). So it’s not the mentality (or not) of knowledge per se that’s at issue, but rather whether it’s hybrid—and perhaps, therefore, at best impurely mental—or non-composite and mental through-and-through. Of course, there are familiar objections to thinking of knowledge as a mental state. Nagel discusses two of the more prominent lines of argument—first, that knowledge-based explanations of action are unnecessary and uneconomical; and second, that genuinely mental states are local, and not the function of such “external” factors as whether the state matches reality (introduction). Nagel’s conclusion, on both of these counts, is that the argument is inclusive at best, and that on some of these matters—e.g., on the question of what is and isn’t essential to our understanding of intentional action—things in fact probably tip in favor of the non-orthodox view. A good part of the case for the latter assessment rests, again, upon a consideration of empirical findings that seem to point towards the relative priority of knowledge over belief, and the relatively uneconomical character of belief-based explanation. Before turning to those findings, however, it’s worth reflecting very briefly on the two abovementioned arguments themselves, if only because doing so serves to show, once again, just what is/isn’t at the heart of the dispute between Nagel and the orthodox view. While it may be true that many proponents of the orthodox view would hold that “attributions of knowledge play no special role in our understanding of intentional acting” (section 2, para. 7), it’s important to stress that this—like the claim that knowledge, because it is composite, is therefore not mental—is an optional add-on to the orthodox view, and not something entailed by it. Thus, there may be certain types of behavior—and, of course, certain patterns

Is Knowledge a Non-Composite Mental State?

337

of robustly successful behavior—that are best explained by citing the fact that the agent knows. However, as Elizabeth Fricker says, “There is absolutely no tension between knowing’s being a good explanatory state, and each instance of knowing being a conjunctive, hybrid phenomenon” (2009, p. 51). In other words, that knowledge has distinctive value in explaining certain types of action doesn’t show that it’s a pure mental state (section 2, para. 11). In order to establish the latter, it would have to be shown that knowledge as the orthodox theorist conceives of it is of inferior explanatory value to knowledge thought of as a pure mental state. But whether and how this might be shown is far from clear.4 In permitting that in certain cases knowledge has special and distinctive value as explanans, of course, the orthodox theorist would sacrifice a certain form of economy: not all intentional behaviors would be best explained in terms simply of belief, or of belief divorced from any features essentially tied to matters epistemic. Even so, because knowledge on the orthodox view has belief as an essential component, the fact that knowledge has sometimes-special explanatory power is compatible with a more fundamental, underlying theoretical parsimony. For, in explaining action sometimes by belief, sometimes by knowledge (and sometimes plausibly by something in between), the orthodox theorist is not alternately appealing to two different, sui generis types of states—two types of states-“in-their-own-right”. Rather, in all such cases beliefs are an essential part of what accounts for what we want to explain. What different situations call for is, in effect, the invocation of different forms or features of beliefs as explanatorily salient. Potentially, various cases will call for explanation in terms of (bare) belief, or belief that’s justified, or (merely) true, or justified but false, or false and unsupported, or satisfying all the conditions on knowing. Of course, if one is arguing that belief simpliciter is always the best or only explainer of action—that belief is “the pivotal mental state” (section 2, para. 5)—invoking some of these explanans will, as Nagel points out (section 2, para. 13), be problematic. But, again, there’s no reason why a proponent of the orthodox must take that line.5 Similar thoughts apply to the issue of the supposed “locality” of genuinely mental states: one can entertain suspicions about whether this is really 4 Part of the difficulty, of course, is as Nagel notes that there’s no agreed-upon method for a complete spelling out of the conditions that must be added to true belief in order to yield knowledge. In fact, it’s not clear even that such a thing is to be had—that knowledge admits of any neat analysis. At certain points, Williamson seems to regard this as indirect evidence against the orthodox view. But it’s not clear why that should be: there are plenty of reasons why, while composite, knowledge might not be fully analyzable in terms of necessary and sufficient conditions. (See Leite 2005, section 1; Dougherty and Rysiew, forthcoming.) 5 Nagel argues that everyday attributions of knowledge-wh, of which they are plenty, pose an additional problem for the orthodox view, since they appear to provide counter-examples to the idea that “knowledge attributions always require the composite representation of a known proposition as true and believed” (section 2, para. 17). (A judgment such as, Jane knows where to go, often occurs in ignorance of the answer to the implied question.) The latter idea may indeed be problematic. As I argue below, however, the orthodox theorist need not commit to such a claim.

338

Patrick Rysiew

so regardless of where one stands on the issue of whether knowledge is composite. As Nagel says, “[t]here is something jarring about the reflective observation that a subject can switch from one mental state to another (e.g. from knowing to not knowing[6] ) through something that happens remotely” (penultimate para. of section 2). But perhaps that’s because we are (most of us, anyway) steeped in a long tradition of internalistic theorizing—and not just about the mental, but about specifically epistemic matters too. Further, it may be that there’s some good psychological explanation to be had, both of our tending to see mentality (including states of knowledge) as localized within agents, and of why the products of this tendency should not be given too much weight: [I]t is possible that we do in some sense intuitively represent the mental state of knowledge as localized within an agent, even if we actually intuitively compute the presence of knowledge on the basis of our grasp of relations between that agent and her environment, and even if we are able to recognize on reflection that this state must reflect how things are in the agent’s environment. In intuitive judgment, we are aware of the final product or outcome of a process of computation, and not of the processing that gives rise to it (Sloman, 1996); while reflective and deliberate attributions of knowledge may make us conscious of the role played by distal factors in knowledge attributions, it is entirely possible for intuitive attributions of knowledge to incorporate distal factors without alerting us to this fact. If the epistemic state underpinning action is something that subjectively seems local, our intuitive representation of knowledge may well provide this subjective impression: we could at some level feel that an agent’s knowledge is localized within her, just as we feel that an object’s motion is sustained by something within it” (antepenultimate para. of section 2; cf. the introduction).

Of course, in explaining something away there’s always the risk that something one wants in fact to preserve will be carried along and explained away as well. In the present case, insofar as we naturally tend to regard mental phenomena as “internal” and “covert” (Miscione et al. 1978, p. 1108), it could be that, to the extent that we naturally regard knowledge as an uncontroversially mental state, that is precisely because of our natural (but perhaps ultimately mistaken) tendency to “localize” such states as knowledge. Perhaps if we had no such tendency, it would be even clearer (if it is) that knowledge isn’t a mental state, or certainly not a purely mental state. Perhaps. But that’s just a possibility. And no doubt, the distinction between intuitive judgments and reflective and deliberate ones is real and important. Further, the evidence on behalf of regarding knowledge as a non-composite state goes beyond whatever intuitiveness it enjoys: there is empirical evidence as to “the nature and relative complexity of our intuitive attributions of knowledge, and on the question of whether the concept of knowledge is in some sense prior to that concept of belief, or whether it is composed from that concept and further 6 Remember: it was suggested above that the orthodox theorist can consistently hold that knowledge is a mental state.

Is Knowledge a Non-Composite Mental State?

339

conditions” (section 2, last para.). It’s to a consideration of that evidence that we now turn.

3.

K N O W L E D G E A S A N O N - C O M P O S I T E S TAT E ?

If the orthodox view is correct and knowledge is a composite state, then, whether or not that composite is best classed as mental, it can seem as though the concept of belief can’t be more complex and difficult for one to get a handle on than that of knowledge. After all, getting a handle on knowledge requires getting a handle on belief plus some other things (truth, and so on). So too, it looks like acquisition of the concept of belief would have to precede any ability to reliably think about and track others’ knowing states. As Nagel forcefully argues, however, evidence concerning such things as the acquisition of mental7 state verbs and nonlinguistic mental state ascription tasks suggests that the reality is exactly the reverse of what the orthodox view thus predicts, for “know(s)” is among our most common mental verbs, well ahead of ‘believe’ or ‘think’, and it’s acquired relatively early on. And children appear to be able to successfully distinguish between knowledge and ignorance (non-knowledge) well before they’re able to perform respectably in attributing false belief. As Nagel says: If we generally made judgments about the presence or absence of knowledge by attributing belief and then evaluating the truth of falsity of this belief, we would not expect to see such a lag between the capacity to recognize the absence of knowledge and the capacity to attribute false belief. (section 3, para. 9; cf. section 2, para. 4)

Whereas, of course, the apparent lag is exactly what we might expect if we’re taking knowledge to be a mental state in its own right, rather than a composite of belief plus other factors. After all, we have independent reason to think that certain features of the concept of knowledge—not least, its factivity and/or its essentially involving matching how things are (section 3, third and fifth paras from the end)—are going to make it more tractable than that of belief—which, after all, permits all sorts of such failures of fit. So, what’s an orthodox theorist to say? It won’t do, I think, for the orthodox theorist (or anyone else) to attempt to question whether children who pass the knowledge-ignorance task really are engaging in genuine metal state recognition: even if (as seems unlikely) we could explain what they’re doing without crediting them with any real grasp of “knowledge as such,” or indeed of anything specifically mental, that might only be setting ourselves up, on pain of inconsistency, for having to explain away our mind-reading skills, our 7 In using this label, I’m just following Nagel and the standard way of referring to such items in the relevant literature. For the remainder of this discussion, I set aside the question of whether the relevant states, including knowledge, are all (properly, purely, etc.) mental; I want instead to focus on the question of the relation between knowledge and belief, and between the corresponding concepts.

340

Patrick Rysiew

ability to track others’ mental states, in similar terms (section 3, seventh para. from the end). So too, there is reason to not restrict the mental states being tracked to something short of knowledge—to not insist that ‘the knowledgeignorance task’ is a misnomer, that children aren’t tracking knowledge but merely true belief (footnote 3). Of course, they may yet lack anything like a complete or full understanding of knowledge. There is an evident lag, for example, between success on the knowledge-ignorance task and the ability to distinguish between knowledge and mere true belief (section 1, para. 6; see Miscione et al. 1978). (Another finding that’s entirely unsurprising given a non-composite view of knowledge.) Yet, as Nagel says, “the prototypical cases of knowing that the child recognizes (in particular, seeing that something is the case), are the same as those we recognize as adults”; and it would be bad general policy to think that successful reference to some state, kind, or thing, requires a full understanding of its underlying nature (footnote 3).8 All of this is good sense. It is also, in my view, very much of a piece with a more promising line of response on behalf of the orthodox theorist to the empirical data Nagel cites. As Nagel presents it, what those data show is that the concepts of belief and knowledge have a priority that’s the reverse of what the orthodox view predicts. But that is not clear; for the type of priority the orthodox view asserts between the concepts of belief and knowledge is not the same as that to which the empirical data speaks. The orthodox theorist claims that belief is a constituent of knowledge, and therefore that belief is prior to knowledge “in the order of analysis”—or, better perhaps (see footnote 4, this chapter), in the order of theoretical understanding—such that one needs a good grasp of belief in order to have a full or proper theoretical understanding of knowledge. Of course, orthodox mainstays like the “JTB” conception of knowledge are often said to be “intuitive.” But, charitably understood, what this means is that the orthodox view is a way of thinking about the relevant states, and of the relations between them, that those who’ve got mastery of the corresponding concepts9 are able—using reflection on examples, some gentle argumentation, and other familiar devices—to arrive at without a great deal of difficulty. On its own, however, the orthodox view is silent as to the relative ease or order in which those concepts are acquired. So too, the orthodox view as such is silent on10 the psychology of knowledge attributions—such questions as whether

8 “Although the adult has a deeper understanding of knowledge, insofar as the child’s early use of ‘know’ is typically a response to what we also recognize as knowledge, she should charitably be seen as referring to knowledge, just as the ancient Romans are appropriately seen as having referred to gold with their word ‘aurum’, notwithstanding their failure to recognize it as the element with atomic number 79, and notwithstanding their relatively limited accuracy in sorting gold from its look-alikes” (footnote 3). 9 Here, I mean practical mastery of the sort that enables one to apply the concepts in a manner that’s on the whole robustly reliable. 10 Which is not to say that orthodox theorists are uniformly good about observing this, or the previous, point.

Is Knowledge a Non-Composite Mental State?

341

“the intuitive recognition of knowledge [starts] with the recognition of belief” (section 2, para. 4), proceeds to determining whether the belief is true, and then, if so, whether further conditions on knowing are plausibly met. As Nagel says in responding to the claim that all mentality must be “local”: we must be careful to distinguish intuitive from reflective and deliberate judgment: in the former, we’re not aware of the processing that gives rise to it, and facts about the former can’t be read off of either the judgments themselves or features of their production in the latter type of case. This seems right, so far as it goes. But of course it leaves unaddressed the really hard question: how is it possible, if knowledge requires belief, for children to be reasonably reliable trackers of knowledge before they have any comparable handle on the concept of belief? Merely distinguishing between reflective and intuitive judgment, and so on, doesn’t help with that. True enough. But the very features, noted above, that lead us to expect knowledge to be a simpler and easier concept to grasp than belief also suggest an answer to the question of how one might grasp and track knowledge without being able to grasp and track one of its essential components. In particular, an understanding of the factivity of knowing, together with modest information about the relation of the subject to a given state of affairs—the sort of information about the relation to the environment that Nagel suggests might be unconsciously and automatically incorporated into the production of our intuitive knowledge ascriptions—would enable a kind of handy shortcut to determining whether they know. As Hogrefe et al. put it: Ignorance should be understood earlier [than false belief] because children have to judge only whether the other had or did not have epistemic access to the target proposition. (1986, p. 579)11

This point, note, holds whether or not one takes belief to be a component of knowledge: either way, given the factivity of knowledge, ignorance is going to be easier to get an intuitive handle on than false belief. More to the point: an intuitive linking of seeing with knowing and not seeing with not knowing— what Ted Ruffman dubs “a ‘seeing = knowing’ rule” (1996, 390)—has the result that belief drops out as irrelevant, for purposes of assessing knowledge. Yet, when coupled with “mechanisms that automatically track the target’s direction of gaze and note which objects lie in his visual field” (para. 7 of the article; see Samson et al. 2010), it yields a means of assessing knowledge that delivers exactly the right result in the knowledge–ignorance task.12 11 Broadly similar remarks are made by Abbeduto and Rosenberg 1985, p. 621, and Miller et al. 2003, p. 350. 12 Employment of this rule, of course, doesn’t explain why, after they are capable of recognizing false belief, children pass through a phase in which they tend to ignore the question of whether the subject has an evidential connection to the proposition in question, ascribing knowledge so long as the subject performs some task successfully (section 1, para. 6). There is no reason, though, to think that children don’t employ a variety of “rules” or “heuristics,” successively or in concert, in their epistemic evaluations. Thus, for instance, Perner (1991) has

342

Patrick Rysiew

It is an open question, of course, whether the account just quickly sketched is correct. But it provides an illustration of the sort of thing a proponent of the orthodox view can say by way of making sense of the data Nagel describes. Of course, it’s part of the story that the children in question have only an imperfect grasp of knowledge (it’s false that seeing = knowing); and, whether it’s thought of as being achieved via “representational reinterpretation” (Karmiloff-Smith 1995) and/or ‘System-2’ processing (Apperly 2011, Applerly and Butterfill 2009; Sloman 1996), there is plenty of room for improvement in their understanding of it through reflective thinking about the subject matter—not to mention, through further experience with the relevant phenomena. But that’s agreed upon by all (cf. footnote 12 of this chapter); and we have no antecedent reason to think that the acquisition of the concept of knowledge, or of any concept for that matter, is going to be an all-or-nothing matter (Apperly and Butterfill, “Conclusion”; Miscione et al., p. 1107). Recall the idea with which we began: we have a prior and pressing need to be able to reliably track others’ states, including what they know. In this respect, knowledge resembles health, another state of our fellows that’s of long-standing and immediate practical concern to us. Now, whatever health is, it’s almost certainly going to be a complex state; so, anyone with a complete theoretical understanding of it will need to have the conceptual sophistication required to understand all the various things it involves. But of course this doesn’t mean that a registering of the presence of all of those various components will be reflected in ordinary ascriptions of health, or that they must be if those ascriptions are to be reliable. Much more likely that we’ll employ such tricks as taking an individual’s complexion or posture, say, as readily available signs of their general state of health—an imperfect strategy, of course; but one that’s eminently affordable, computationally speaking, and perhaps on the whole pretty reliable.13 It’s in an analogous manner, I’m suggesting, that we intuitively track others’ knowledge, and in so doing we may elide any specific consideration of belief and so avoid the comparative difficulties that that concept involves. One final thought. As noted early on, part of Nagel’s concern in her paper is to make it plausible that ascriptions of knowledge are the product of an already-acknowledged intuitive mind-reading system. To a large extent, however, that idea too can be preserved from within the orthodox view. On Nagel’s view, recall, the counter-intuitiveness of this idea can be mitigated claimed that younger children take knowing to essentially amount to “getting it right” (see too Miscione et al. 1978), which of course squares with the fact just noted. Whatever its correct explanation, however, that fact doesn’t pose any special problem for the present line of thinking: anyone who thinks that children’s getting the knowledge–ignorance task right is evidence of their ability to track knowledge, and of their having a grasp of the relevant concept, is going to have further explaining to do here. 13 I don’t know what indicators of health we do actually intuitively rely upon, but I assume that there are some such, and that they’re not horribly unreliable—or, at least, not obviously less reliable than our intuitive knowledge-tracking strategies.

Is Knowledge a Non-Composite Mental State?

343

by noting that it is possible—perhaps even plausible, given the example of intuitive physics—that our intuitive attributions of knowledge incorporate our grasp of relations between that agent and her environment, and perhaps other distal factors, without alerting us to this fact. But this can occur even if the relevant “purely mental” base is restricted to belief. So the orthodox theorist can agree that “[t]he fact that the state of knowing incorporates a relationship with the environment does not disqualify it from counting as a state which is fundamental to our intuitive understanding of other intelligent beings” (para. 7 of the introduction). Granted, if the purely mental base of knowledge is restricted to belief, the obtaining of the requisite relationship with the environment will be regarded as at most an extra-credal component of knowledge; and we can perhaps disagree about whether or in what sense that composite state is best regarded as mental. It is not obvious, however, that anything of substance in one’s account of how we come by knowledge of what others know turns upon the outcome of that question.

4.

CONCLUSION

I have suggested that the orthodox view of knowledge as a composite state is compatible, not only with regarding knowledge as a mental state, but also with the empirical findings Nagel discusses, and therefore that those findings don’t in fact support the view that knowledge is a non-composite state, a mental state in its own right. The latter claims, in particular, might seem like points of serious disagreement with Nagel. It should be emphasized, however, that in many ways the present discussion has been complementary to Nagel’s: that knowledge might properly be regarded as a mental state; that philosophical arguments to the contrary are open to question; that we are in need of careful empirical investigation into the form and mechanisms of our intuitive ascriptions of knowledge (something that the orthodox view itself doesn’t give us); and that those intuitive ascriptions might be importantly based in our natural mind-reading abilities. These are all points on which, as far as the present discussion goes, Nagel and the orthodox theorist can happily agree.14

REFERENCES

Abbeduto, L. and Rosenberg, S. (1985). Children’s knowledge of the presuppositions of know and other cognitive verbs. Journal of Child Language 12(3), 621–41. Apperly, I. (2011). Mindreaders: The Cognitive Basis of “Theory of Mind.” Hove and New York: Psychology Press. and Butterfill, S. (2009). Do humans have two systems to track beliefs and belieflike states? Psychological Review 116(4), 953–70. 14 For helpful comments and discussion, my thanks to Mike Raven, Jennifer Nagel, and Cindy Holder.

344

Patrick Rysiew

Dougherty, T. and Rysiew, P. (forthcoming). Experience first (and replies), Contemporary Debates in Epistemology, 2nd edition, edited by Ernest Sosa, Matthias Steup, and John Turri. Blackwell. Dretske, F. (1989). The need to know. In M. Clay and K. Lehrer (eds.), Knowledge and Scepticism. Boulder: Westview Press, pp. 31–60. Fricker, E. (2009). Is knowing a state of mind? The case against. In P. Greenough and D. Pritchard (eds.), Williamson on Knowledge. New York: Oxford University Press, pp. 31–60. Hogrefe, G. J., Wimmer, H., and Perner, J. (1986). Ignorance versus false belief: A developmental lag in attribution of epistemic states. Child Development 57, 567–82. Karmiloff-Smith, A. (1995). Beyond modularity. Cambridge, MA: The MIT Press. Leite, A. (2005). On Williamson’s arguments that knowledge is a mental state. Ratio (new series) 18(2), 165–75. Magnus, P. and Cohen, J. (2003). Williamson on knowledge and psychological explanation. Philosophical Studies 116(1), 37–52. Miller, S., Hardin, C., and Montgomery, D. (2003). Young children’s understanding of the conditions for knowledge acquisition. Journal of Cognition and Development 4(3), 325–56. Miscione, J. L., Marvin, R. S., O’Brien, R. G., and Greenberg, M. T. (1978). A developmental study of preschool children’s understanding of the words ‘know’ and ‘guess’. Child Development 49(4), 1107–13. Perner, J. (1991). Understanding the Representational Mind. Cambridge, MA: MIT Press. Ruffman, T. (1996). Do children understand the mind by means of simulation or theory? Mind & Language 11(4), 388–414. Samson, D., Apperly, I. A., Braithwaite, J. J., Andrews, B. J., and Bodley Scott, S. E. (2010). Seeing it their way: Evidence for rapid and involuntary computation of what other people see. Journal of Experimental Psychology: Human Perception and Performance 36(5), 1255–66. Sloman, S. A. (1996). The empirical case for two systems of reasoning. Psychological Bulletin 119(1), 3–22. Williamson, Timothy. (2000). Knowledge and Its Limits. Oxford: Oxford University Press.

INDEX

absence of evidence norm, 60 Adams, E.W., 22 Adams’ Thesis, 22–6, 29 Qualitative, 26–7 Aikin, S., 35 agnosticism, see suspension of judgment Allendorf, F., 132–3 Apperly, I., 301–3 a priori, 68, 70–2, 171, 176, 179–81, 184 deeply contingent, 60, 71, 176, 184 Bayesianism, 4, 20, 53–4, 67–8, 75, 82, 86 subjective, 82 belief, degrees of, see credences compartmentalized, 244 implicit, 115–16, 120 sets of possible worlds as modeling, 93 perceptual, 240–2 processes of forming, 256–64 relation of action to, 274–5, 282–9, 300–3 Benacerraf-Field problem, 214 biological sex, 134–8 Bogardus, T., 49 Bratman, M., 315–16 Burge, T., 71, 179 bypassing beliefs, 262, 265–6, 268 Cassidy, K., 293–4 choiceworthiness, 83–9, 94–101 Chisholm, R., 201 Christensen, D., 44, 48–51, 100, 195, 223 Churchland, P., 276 circularity, epistemic, 179, 181–4 premise, 174–5 cognitive bias, 108–9, 158 cognitive penetration, 240 Cohen, J., 286, 289, 312, 335 Cohen, S., 94 common sense, 185–212 conceptual priority, analytic, 322–4 psychological, 322–4 developmental, 322 Conceptual Role Semantics, 227–8 Concessive Absence of Support Thesis, 27–9

conditionalization, 15, 67–8, 84–6, 100 conditionals, acceptability and assertability of, 23–9 causal, 12 concessive, 27 inferential, 12, 24–5 possessing the concept of, 226–7 probabilities of, 6–22 psychological studies on, 11–14, 17–18, 24–9 semantics of: possible worlds account, 5, 8–10 material conditional account, 4–5, 7, 11 non-propositional account, 4, 10, 22–3 simple, 13–14 confirmation bias, 242–5, 262, 266–8 conformism, 48–51 conjunctions, 62–6, 224–5 conjunction fallacy, 7 conjunctive content, 252–3 conservatism, 187, 198–9, 206–11 conservativeness, 224 contextualism, 94–7 core cognition, 259 credences, 57–80, 88, 259–60, 263 imprecise, 75 standard, 58–9, 64–6, 72–3, 78, 79 prior, 67–73 ur-, 68–70 credence function, 58 credence gaps, 66 decision theory, 84–96, 101–3 Dawkins, R., 34 DeRose, K., 96 Devitt, M., 133–41 disagreement, 18, 43, 47–51, 92, 194–5 disjunctions, 62–6, 225 dogmatism, 34–55 anti-Quinean, 40–1, 53 backward-looking, 37–8 forward-looking, 37–8 stymied, 39, 41–52, 53 super-stymied, 40, 41 dogmatism paradox, 45–6 Dretske, F., 333

346

Index

dual process theories, 234 Dutch books, 7, 15, 97–102 downgrading, epistemic, 243, 251, 256–61 Elga, A., 48–50, 195 epistemic dependence, 198, 208–11 essence, 109–58 quint-, 110–36, 155–8 young children’s beliefs about, 117–20, 130 essentialism, biological, 110, 124, 131–42 chemical, 131, 134, 142–58 psychological, see essence, quintethical relativism, 196 Evans, J.St.B.T., 11 evidentialism, 249–51, 265 Evidential Support Thesis, 27–9 expected utility maximization, 82–9, 92, 101–4 experiences, compound, 251–8, 261 differences between beliefs and, 259 individual, 252, 255 explanation, evolutionary, 215–17, 220–3 causal, see knowledge, causal explanations and localization of, 290–1 Fantl, J., 98 Feldman, R., 47–8 Fine, K., 185–6 flat dismissal, 34–41 Fodor, J., 241 fragmentation, 243, 264 Fricker, M., 311–12, 337 Gambler’s fallacy, 204, 206 Garber, D., 90 Gelman, S., 116–20, 130 gender, 113, 115–18, 122, 135 Gendler, T., 291 generalization, 245–52, 255–8 over-, 257–8 genetics, 121–3, 130–40 Gibbs’ phase rule, 148, 155 Gleitman, L., 293–4 Goldman, A., 284 Goodman, N., 198, 203 Grice, H.P., 285 Gupta, A., 188, 190

Hájek, A., 19–22, 23, 102 Harman, G., 45, 187, 198–9, 206–11 Hawthorne, J., 84, 88, 316–17 Hegel, G.W.F., 186 high-stakes bets, 86–9 Hogrefe, J.G., 341 Holmes, S., 119; see also evidentialism Hume, D., 34–5, 41, 50 idealization, 102–3 Implicit Association Test, 115–16 intention, 315–17, 328–31, 336–7 internalism (about justification), 247, 249, 256 intuitions, 108–110, 125–126, 129, 151, 158, 188, 282 irrationality, 257–61, 263–6 Jeffrey, R., 85 Johnson-Laird, P.N., 12–13 Joyce, J., 75 Kant, I., 186 Kaplan, M., 248–9 Kelly, T., 41, 185, 187–8, 190, 198, 200–7, 211 Kitcher, P., 131, 141 Kolodny, N., 329 Kornblith, H., 282 Kripke, S., 70, 109, 124–7, 132, 143, 147–8, 152 knowledge, analysis of, 279–82, 287 attributions of, as typically true, 96 concessive, 83 causal explanations and, 321, 327–31 epistemic probability of, 82–105 problem of easy, 166–84 relation of belief to, 274–303, 310–15, 321–31, 333–43 relation of action to, 97, 274–5, 282–8, 301–3, 310, 312, 316–17, 321–2, 324, 327–31, 336–7 knowledge as a mental state, 273–343 Kyburg, H., 4 Lackey, J., 48 LaPorte, J., 139–40, 155 Leary, R., 132–3 Leite, A., 333–4 Lewis, D., 8–9, 14–16, 18, 26, 92–3, 185–6, 198, 206–7, 264, 267

Index logic, definition of, 217 objectivity of, 214–15 reliability about, 214–37 logical concepts, possession of, 223–7 lottery paradox, 59, 169 Locke, J., 274 Lockeanism (about belief), 73–6 mathematical Platonism, 214 Lucian, 34 Lycan, W., 185, 187–8, 190, 199 Magnus, P., 286, 289, 312, 335 McGrath, M., 98 McGrath, S., 35 mentalism, 321–3, 328–31 mental states, factive, 284 localization of, 275, 289–90, 338 methodism, 201–6 Mill, J.S., 227, 231 mindreading, 273, 282, 291, 297–303, 309–15, 321, 328, 331, 334, 339, 342–3 Molyneux, B., 286–7 Monty Hall Problem, 200 Moore, G.E., 35, 37, 39, 186–8, 198–201 Moore’s plausibility principle, 199–201, 211 Moran, R., 330 natural kinds, 109–11, 113, 115, 117–18, 124–9, 132, 158 natural selection, see evolutionary explanations Needham, P., 144–51 Neo-Lorentzianism, 189–90, 193–4 No Feedback principle, 169–72 no-lose investigations, 169, 172–4 Nolan, D., 9–10 Nozick, R., 231 Ockham’s razor, 190 Okasha, S., 139–40 ontological nihilism, 196 Over, D.E., 11 ‘ought’ of rationality vs ‘ought’ of reasons, 329 Papafragou, A., 293–5 particularism, 201–6 Payne, K., 240–1, 245, 246 perception, 215, 224–5, 241–2, 326–7 Perner, J., 341

347

phenomenalism, 127–30 planning, 235, 309 Plato, 274, 289 preface paradox, 169 Principle of Modesty, 41, 47 probability, conditional, 8, 12, 26, 85 epistemic, 83–94 relation to contingency, 91–2 objective, 82 probability function, 82–6 prodigality problem, 84–8 propositions, logical, 214, 217 logically equivalent, 217, 229 ordinary contingent, 60 Pryor, J., 37–8, 206 Putnam, H., 109, 124–8, 131, 142–3, 147 questions, treating as open or closed, 254 race, 113, 115–17, 122, 129–31 rationality vs reasonability, 83 rational support, 242–5, 255, 259–61 reasoning, 241, 260–1, 265 abductive, 21 belief-desire, 284 deductive, 170, 215–20, 228–9, 233–7 explicit, 292 inductive, 116, 169, 170 physical, 317 practical, 89, 309, 316–17, 324 probabilistic, 14, 21 reflective equilibrium, 187, 198, 206–11 reliabilism, 247 salience structure, 254, 258 same substance relation, 142–4, 146–52 selection effects (on experience), 240–68 anti-, 243–4, 261–8 semantic externalism, 335 skepticism, 35, 38–9, 45, 84, 87, 186, 191–2, 196–207, 212, 280 Salmon, N., 127 Skyrms, B., 225 special relativity, 186, 189–98 spinelessness, problem of, 48 Stalnaker, R., 5, 8, 93 Stalnaker’s Hypothesis, 8, 11–14, 20–2, 25, 29 Generalized, 15–19 Stanley, J., 84, 88, 98 Stich, S., 232–4

348

Index

Straightforward Reduction, 57–8, 64–73 subject-sensitive invariantism, 97–102 suspension of judgment, 57–80, 258 Titelbaum, M., 166, 169, 172–4, 176–7 Tragic Mulatto, 129–31 transcendental arguments, 21 transmission failure, 169, 174–9, 181, 183–4 Triesch, J., 264 triviality arguments, 9, 11, 14–22, 23 Tucker, C., 166, 169, 174–7, 182 Twin Earth, 125–6, 130–1, 150–1

updating rules, 15, 100 utility, see expected utility maximization van Fraassen, B., 16, 18–19 Verbrugge, S., 17, 19, 24–9 Weatherson, B., 98 Weisberg, J., 166, 169–71, 176–77 Williamson, T., 67, 82–9, 274, 283–92, 321–5, 333, 335–7 Wright, C., 166, 174, 176, 180 Zeno’s paradox, 34, 36, 39, 42, 46, 52, 54