Epistemic Contextualism: A Defense 0198754310, 9780198754312

Table of contents :
Acknowledgments
Contents
Introduction
Part I. Arguments
1. The Argument from Cases: Standard Contextualism and Standards Contextualism
2. The Argument from Reliability: The Role of Reference Classes
3. The Argument from Luck: The Role of Descriptions
Part II. Problems and Extensions
4. Skepticism, Lotteries, and Contextualist Solutions
5. Cross-Context Attributions and the Knowability Problem: Does Contextualism Lead to a Contradiction?
6. Beyond Knowledge: Action and Responsibility
Part III. Objections and Alternatives
7. Objections
8. Alternatives?
Bibliography
Index

Citation preview

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Epistemic Contextualism

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Epistemic Contextualism A Defense

Peter Baumann

1

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

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 © Peter Baumann 2016 The moral rights of the author have been asserted First Edition published in 2016 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 Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number: 2016937254 ISBN 978–0–19–875431–2 Printed in Great Britain by Clays Ltd, St Ives plc Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Acknowledgments My work on the topics of this book has a history reaching back to times when I had no clue that I would eventually write a book on contextualism. For comments on the manuscript as well as on the pieces out of which it developed I am grateful to the following (apologies to those who should be on this list but aren’t): Fred Adams, Joachim Aufderheide, Reinhard Baule, Friedhold Baumann, Kelly Becker, Sven Bernecker, Rüdiger Bittner, Martijn Blaauw, Jonny Blamey, Elke Brendel, Berit Brogaard, Audre Brokes, Tony Brueckner, Christopher Buford, Jon Cameron, Robin Cameron, Michael Clark, Stewart Cohen, Gisela Cramer, Martin Davies, Igor Douven, Fred Dretske, Christoph Fehige, Nick Fenn, Wolfgang Freitag, Martin Gierl, Alvin Goldman, John Greco, Thomas Grundmann, Steven Hales, Oswald Hanfling, Stephen Hetherington, Michaela Hohkamp, Gerry Hough, Christoph Jäger, Carrie Jenkins, Phillip Keller, Igal Kvart, Byeong D. Lee, Poong Shil Lee, Franck Lihoreau, Guido Melchior, Tony Milligan, Martin Montminy, Julius Moravcsik, Ryan Nichols, Günther Patzig, Daniel Pilchman, Bob Plant, Duncan Pritchard, Joachim Abdul Raffert, Manuel Rebuschi, Darrell Rowbottom, Bruce Russell, Joe Salerno, Jonathan Schaffer, Thomas Schmidt, Nick Shackel, Giacomo Sillari, Walter Sinnott-Armstrong, Barry Smith, Klaus Sommer, Mark Textor, Stefan Tolksdorf, Ted Warfield, Timothy Williamson, Lutz Wingert, Suck Young Won, René van Woudenberg, Crispin Wright, several referees, including two anonymous referees for Oxford University Press, and audiences at talks of mine in Aberdeen, Amsterdam, Bled, Bogotá, Dublin, Düsseldorf, Erfurt, Erlangen, Essen, Glasgow, Göttingen, Healdsburg, Irvine, Kirchberg, Mainz, Manchester, Nancy, Santiago de Compostela, Seoul, Stirling, Swarthmore, Tübingen, and Zürich. I am grateful to Swarthmore College for granting me a James A. Michener Faculty Fellowship. I have used material (sometimes more, sometimes less) from several prior publications: “Lotteries and Contexts” (Erkenntnis 61, 2004, 415–28), “Varieties of Contextualism: Standards and Descriptions” (Grazer Philosophische Studien 69, 2005, 229–45), “Contextualism and the Factivity Problem” (Philosophy and Phenomenological Research 76, 2008, 580–602), “Contrastivism Rather than Something Else? On the Limits of Epistemic Contrastivism” (Erkenntnis 69, 2008, 189–200), “Problems for Sinnott-­Armstrong’s Moral Contrastivism” (Philosophical Quarterly 58, 2008, 463–70), “Reliabilism: Modal, Probabilistic or Contextualist” (Grazer Philosophische Studien 79, 2009, 77–89), “Factivity and Contextualism” (Analysis 70, 2010, 82–9), “A Puzzle about Responsibility” (Erkenntnis 74, 2011, 207–24), “WAMs: Why Worry?” (Philosophical Papers 40, 2011, 155–77), “Knowing about Other Contexts” (Christoph Jaeger and Winfried Loeffler (eds), Epistemology: Context, Values, Disagreement, Frankfurt: Ontos, 2012, 63–79), “PS: Response to Schaffer’s Reply” (Stefan Tolksdorf (ed.), Conceptions of Knowledge, Berlin: de Gruyter, 2012, 425–31), “Gettier, Wissen, Zufall” (Gerhard

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

vi  Acknowledgments Ernst and Lisa Marani (eds), Das Gettierproblem. Eine Bilanz nach 50 Jahren, Paderborn: mentis, 2013, 9–27), “No Luck with Knowledge? On a Dogma of  Epistemology” (Philosophy and Phenomenological Research 89, 2014, 523–51), “A Contradiction for Contextualism” (Franck Lihoreau and Manuel Rebuschi (eds), Epistemology, Context, and Formalism, Heidelberg: Springer, 2014, 49–57). I am grateful to the publishers for permission to use material from these articles for this book. Finally, amongst those working for or partnering with Oxford University Press I would like to thank Matthias Butler, Eleanor Collins, Michael Janes, Peter Momtchiloff, Sarah Parker, Frank Pert (who produced the index), Dawn Preston, and Vaishnavi Venkatesan for all their help along the way.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Contents Introduction1

Part I.  Arguments 1. The Argument from Cases: Standard Contextualism and Standards Contextualism9 1.1. Cases for Contextualism 9 1.2. Parameters: Standards 14 1.2.1. Evidence, Reliability, Degrees of Belief 14 1.2.2. Epistemic Position 16 1.2.3. Ruling Out Alternatives 18 1.3. Determinants 20 1.3.1. Stakes 20 1.3.2. Purposes and Intentions 22 1.3.3. Conversational Contexts 23 1.3.4. Salience 25 1.3.5. Determinants for the Other Parameters. Norms and Conventions 27 1.4. Conclusion 31 2. The Argument from Reliability: The Role of Reference Classes

33 34 37 43 43 46 51 58 58 63

3. The Argument from Luck: The Role of Descriptions

65 66 68 74 79 83 86

2.1. Knowledge and Reliability 2.2. Reliability and Probability 2.3. Reliability and Context 2.3.1. Typing Topics 2.3.2. Typing Methods 2.3.3. Extension and Generalization: Reference Classes 2.3.4. Conclusion 2.4. Reliability and Modality 2.5. Conclusion 3.1. Luck, Its Varieties, and an Exclusion Claim 3.2. Lucky Knowledge: Variations on a Case by Russell 3.3. More Cases 3.4. Modal Luck 3.5. Probabilistic Luck 3.6. Conclusion

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

viii  Contents

Part II.  Problems and Extensions 4. Skepticism, Lotteries, and Contextualist Solutions 4.1. Traditional Skepticism 4.2. Lottery Skepticism 4.3. Knowing Lottery Propositions? 4.4. Lotteries and Closure 4.5. A Standard Contextualist Solution 4.6. Another Contextualist Solution 4.7. Conclusion

5. Cross-Context Attributions and the Knowability Problem: Does Contextualism Lead to a Contradiction? 5.1. Contradictions across Contexts? 5.2. Even Worse: Moore-Paradoxality and No Neutrality 5.3. A Problem for Contextualists Only? 5.4. Knowability Restrictions? 5.5. The Contextualist Way Out: Contextualist Closure 5.6. Conclusion

6. Beyond Knowledge: Action and Responsibility 6.1. A Puzzle 6.2. Responsibility 6.3. Reference Classes Again 6.4. No Straight Solution 6.5. Practical Contextualism

91 92 100 102 109 111 114 119 120 121 124 126 128 131 136 140 141 142 145 148 151

Part III.  Objections and Alternatives 7. Objections

159 160 162 166 172 173 177 182 190 190 193 195 197

8. Alternatives?

198 198 199 204 208

7.1. WAMs 7.1.1. Thought and Language 7.1.2. From WAMs to WBMs, and Back Again 7.1.3. Conclusion 7.2. Cappelen and Lepore’s Three Tests 7.3. More Linguistic Objections 7.4. Cases, Again 7.5. More Problems and Questions 7.5.1. Complexity 7.5.2. Normativity and Arbitrariness 7.5.3. Error, Blindness, and the Possibility of Communication 7.5.4. One Last Smaller Problem 8.1. Subject-Sensitive Invariantism 8.1.1. Some Questions about Stakes 8.1.2. SSI on Factors and Cases (Again) 8.1.3. Other Problems

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Contents  ix 8.2. Contrastivism 8.2.1. Knowledge without Contrasts 8.2.2. More Relata for the Third Slot 8.2.3. More Relativization 8.2.4. Conclusion 8.3. Relativism 8.3.1. Relative Truth, Monadic Truth, and Direct Expressibility 8.3.2. But Is It Truth? 8.3.3. Other Problems? 8.4. Conclusion

Bibliography Index

212 213 216 217 220 221 222 225 227 230 231 257

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Introduction Philosophical reflection about what we call “knowledge” has a natural starting point in the view that knowledge has certain fundamental characteristics and that it is one of the basic tasks of epistemology to identify these characteristics. No matter whether knowledge is taken to be a Platonic idea, a natural kind, something that can be captured by reductive definitions in terms of individually necessary and jointly sufficient conditions, or something else, and no matter whether knowledge is taken to involve certain characteristics rather than others (justification, reliability, etc.), the underlying idea of basic characteristics remains the same. Very often a second idea comes with it, namely the idea that we are always talking and thinking about the same thing when we use words like “knowledge.” In this sense, one could also say that knowledge is orderly. I will argue here that knowledge is not orderly but disorderly in a certain sense. More precisely, I will argue for a contextualist account of “knowledge” (but also see Lewis 1996, 566–7 for the legitimate dropping of quotation marks). According to such a view, the truth conditions or the meaning of knowledge sentences of the form “S knows that p” can vary with the context of the speaker; and they can vary because of some contextual variability in the meaning of the word “know” and of related words. Similar things can be said about knowledge attributions not expressed in a natural language but made in solitary thought. The content of a thinker’s thought about someone’s “knowledge” that p can vary with the thinker’s context, and it can vary because of contextual variation in the content of the concept of knowledge used by the thinker. If contextualism is correct, then the above, initial characterization of the task of epistemology is a bit misleading. According to the contextualist, talk about what we call “knowledge” concerns what we mean by that word in our particular context of using that word and thus lacks a certain generality of scope. Risking misunderstandings, one could put this in the following way: Contextualists talk about “knowledge” rather than knowledge. One might worry here (see, e.g., Sosa  2011, ch.5 and Blome-Tillmann 2014, 49–52) that contextualist epistemologists lose their subject matter and just talk about words for things and not the things themselves. However, this is not so and I hope that this worry will be allayed in the following chapters. Contextualism about “knowledge” and epistemic contextualism more generally has been discussed a lot over the last few decades (mainly in articles; there are surprisingly few monographs on the subject). Why then add another proposal and defense of

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

2  Introduction contextualism? The reason is simple and straightforward. The main and best arguments in favor of contextualism are, in my opinion, quite different from the ones typically put forward by defenders of contextualism. And the objections against contextualism should also be sought in other places than those usually occupied by critics. The virtues of contextualism—and perhaps also its vices—are sufficiently different from what they’re typically taken to be. Contextualism therefore deserves a new and fresh exposition and defense. Part I of the book deals with core arguments in favor of contextualism. One of the most popular such arguments starts with the consideration of examples and cases. Answering the question of what we would say about certain scenarios and how we would use words like “knowledge” in relation to such cases is supposed to give crucial support to contextualism. Even though I think that the importance of cases has been overestimated quite a bit, I start with the discussion of (mostly new) cases in Chapter 1. This is a somewhat natural starting point which also allows one to make some systematic points about our topic and show how contextualists can think systematically about cases. The most common form of contextualism (usually supported by arguments from cases) can be called “standard contextualism.” Standard contextualism is standards contextualism, that is, contextualism about epistemic standards. As the consideration of cases is supposed to show, the variation in the content of knowledge attributions is driven by the variation in the epistemic standards of the attributor. I propose to think more systematically about contextual parameters (aspects of epistemic contexts); as one will see in later chapters, epistemic standards constitute just one type of parameter. Here, I distinguish between standards concerning what needs to be ruled out by the subject, the subjects’ evidence, the reliability of their beliefs, the subject’s required degrees of belief and the required type of the subject’s epistemic positions (while I argue against the importance or relevance of other parameters often proposed in current discussions, like the conversational score or psychological salience). One general implication of this conception of parameters is that the common idea of a hierarchy of more or less demanding contexts is mistaken. It is also important to identify what I call the “determinants” of such parameters; the former fix the value of the latter. As far as the determinants of standards are concerned I propose to distinguish between attributors’ stakes, attributors’ purposes and intentions, and norms and conventions relevant in the attributor’s context. Even if the consideration of cases and of what we “would say when” is of some relevance, especially at the beginning, other, more theoretical arguments for contextualism are more important. The next two chapters deal with such theoretical support for contextualism. Chapter 2 starts with a defense of a very broad and general type of reliabilism according to which knowledge requires reliability. The core of the chapter is dedicated to an extension and strengthening of a well-known and obnoxious problem for reliabilism: the so-called generality problem. This problem is just an aspect of an underlying problem about the relevant reference class. I argue that the prospects for a “non-skeptical” solution of the problem are bleak. However, the apparent vice of

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

introduction  3 reliabilism can be turned into a virtue of contextualism. If one acknowledges that the reference class problem has no non-skeptical solution, then contextualism should be accepted as an interesting and promising “skeptical” solution. The contextualist way out of an important problem offers important support and ­motivation for contextualism. I also argue in favor of a probabilistic version of contextualism here and against a modal version of it. Thinking about what fixes the reference class leads to a  new set of contextual parameters (topics, methods, temporal frames, and spatial frames) and their determinants (attention, purposes and intentions, and norms and conventions). Chapter 3 argues against the very common view that knowledge excludes epistemic luck of a certain interesting and relevant type. I think that this view is just another dogma of epistemology and in need of criticism. This is interesting and important in itself but it also leads to another “theoretical” argument for contextualism. I argue that the notions of epistemic luck and of knowledge are context-sensitive in a parallel way. Again, I explain why I prefer a probabilistic account of knowledge to a modal account. The discussion of epistemic luck leads to the identification of a third type of parameters, “descriptions.” The relevant determinants are, again, attention and purposes and intentions (while it is left open here whether norms and conventions play a determining role). The arguments for contextualism offered in Part I make up the core of the book. However, one can often hear people say that contextualism also has to prove itself by offering promising solutions to important philosophical problems. I do think that the theoretical arguments in favor of a philosophical view are more important than its explanatory potential, but the latter aspect is still important enough to deserve extended discussion. Part II of the book deals with the way contextualism can respond to important philosophical problems (and with extensions to non-epistemic notions). Many defenders of contextualism hold that contextualism should (also) be accepted because it can respond to the problem of (Cartesian) epistemological skepticism. And many critics of contextualism reject it also because of its alleged failure to respond to the skeptic. I think that the fate of contextualism is largely independent of what it does and can say in response to Cartesian skepticism. There are other problems I find more pressing in this context. As a representative example, I discuss the lottery problem (the one brought up by Gilbert Harman) in Chapter 4. Lottery skepticism is in many ways more of a challenge to epistemology than Cartesian skepticism because it is much more mundane and harder to counter. I propose a contextualist response to the lottery problem which differs considerably from common contextualist takes on it. In contrast to most people I deny that one cannot truly be said to know a lottery proposition. The proposed solution of the problem involves a focus on the notion of an epistemic position as well as an independently plausible modified principle of epistemic closure (which in its unmodified, basic, and very rough form says that one can come to know something on the basis of inference from something else one knows).

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

4  Introduction Chapter 5 deals with a “homemade” problem for contextualism: the knowability problem. This problem concerns the evaluation of the truth value of knowledge ­attributions made in other contexts than the attributor’s one. It is astonishing how little has so far been said about this problem. I think that this is the hardest problem contextualism has to face. The problem consists in a threat of inconsistency. A contextualist could find herself in a context where she would have to deny that she “knows” a given proposition but at the same time, qua contextualist, would have to admit that some other subject does “know” that same proposition. A few argumentative steps lead to a contradiction. This chapter argues that the inconsistency can be avoided if one chooses a view according to which knowledge is a ternary relation between a subject, a proposition, and a contextual parameter (standards, reference classes, descriptions). This also suggests that the common tendency to see the semantics of “knowledge” as close to the one for indexicals and demonstratives is mistaken. Apart from a relational, ­ternary view of knowledge, the contextualist also needs to modify the closure principle further. If contextualism is the correct view of “knowledge,” then it would be surprising if there weren’t other philosophically important notions that also suggest a contextualist analysis. Why should “knowledge” be a unique case? And wouldn’t such uniqueness count against contextualism about “knowledge”? Chapter 6 deals with an extension and application of contextualism to non-epistemic notions and thus also attempts to reply to the uniqueness worry. As a representative I choose the notion of responsibility for one’s own actions. I start with a puzzle about responsibility and propose a solution that parallels the contextualist view of knowledge proposed before. Again, the distinction between parameters and determinants proves useful. One parameter consists in standards and thresholds of responsibility; its determinants consist in attributors’ stakes, character traits, and norms and conventions. Another parameter consists in reference classes again; they are determined by the factors of attention and of purposes and intentions. Even though the discussion in the first six chapters often goes into objections and alternative views, it is a good idea to dedicate a whole part (III) exclusively to these issues. Chapter 7 focuses on core objections to contextualist views. I discuss what is perhaps the most popular objection against contextualism, namely an objection which is based on the alleged possibility of so-called warranted assertibility maneuvers against contextualism. Furthermore, I discuss certain more linguistic objections against a contextualist semantics of “knowledge.” I also go back to the discussion of cases, especially cases that are often considered problematic for contextualists (this complements the discussion in Chapter 1). I end with some further problems for contextualism: mainly, a problem of complexity, an issue with normativity and arbitrariness, and finally problems concerning speaker’s error and semantic blindness. Contextualism remains quite unscathed in my view. Finally, Chapter 8 discusses three major alternatives to contextualism: first, subject-­ sensitive invariantism according to which the truth value of knowledge ascriptions can

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

introduction  5 vary with the context of the subject rather than the attributor (this section includes some detailed discussion of the notion of stakes—another neglected topic so far); second, epistemic contrastivism according to which knowledge is a ternary relation between a subject, a target proposition, and a contrast proposition; and finally, semantic relativism about “knowledge” which claims relativity of the truth value of knowledge ascriptions to the standards of the assessor of the claim. It turns out that these alternative accounts of knowledge do worse than contextualism and have to face much more severe problems. This completes the argument for contextualism.1 A positive view of knowledge, or better: of “knowledge,” is being developed here step by step. Instead of presenting the view in its entirety right at the beginning I find it preferable to develop it gradually while discussing arguments, problems, objections, and alternatives. I hope that contextualism appears attractive and in some new light here. I also hope that all this is interesting in a more general way, too, apart from the main topic of contextualism and with respect to other important topics in epistemology, like reliability, the role of epistemic luck or epistemic closure. Finally, many things I say here go against some currently widely shared views. I hope this will make the book more rather than less interesting to the reader.

1   A few side remarks about what I will not do in this book might be useful here. I will say a bit but not too much about the semantics of knowledge attributions. I agree with Cohen 1999, 61 that there isn’t that close a connection between epistemological and semantic theories of “knowledge” (but see for an overview Stojanovic 2008 and for a recent discussion Moghaddam 2015). I am skeptical of contextualism’s potential to solve the “Gettier-problem,” that is, to offer a reductive definition of “knowledge” (but cf., e.g., Cohen 1998a). I don’t think such a definition can be had at all (see Baumann 2013). Finally, I am having doubts about recent proposals, contextualist or not, of knowledge rules of assertion (“Assert only what you know”; see Unger 1975, ch.6 and Williamson 2000, ch.11) or of practical reasoning and action (“One knows p if/ only if/if and only if p is an appropriate premise for one’s practical reasoning and an appropriate basis for one’s actions”; see Fantl and McGrath 2009, Hawthorne and Stanley 2008, and Williamson 2005c, 231). I think that there is some connection between knowledge and practical reasoning and also assertion but it is not as close as suggested by many authors these days (see Baumann 2012c and also 2014c). I won’t say much at all here about such knowledge rules.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

PA RT I

Arguments

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

1 The Argument from Cases Standard Contextualism and Standards Contextualism 1.1  Cases for Contextualism One way to support a view on the nature or characteristics of knowledge (and related epistemic states) begins with the consideration of cases and examples. Data about “what we would say when,” that is, data about usage of “knowledge” (and related terms) in such cases by ordinary speakers of the language are then used as support for or objections against the epistemic views under discussion. This kind of “argument from cases” has been extremely popular amongst contextualists as well as opponents of contextualism (see for many DeRose 2005 or Stanley 2005). The aim of this book is rather different: to offer more theoretical arguments for contextualism (see Chapters 2 and 3). I don’t want to put much weight at all on arguments from cases. However, since it is in general a good idea to start with the phenomena—and also because the argument from cases has been so important to discussions of contextualism—this first chapter is dedicated to the discussion of cases. Central to it is a distinction between parameters—or aspects of epistemic contexts—and determinants—or factors which fix the parameters. I distinguish between several different parameters as well as determinants. All this is meant to show how contextualists can think systematically about cases. The direction of the discussion is not so much bottom-up (cases supporting a view) but rather topdown: showing how an abstract view (supported by other arguments) can be applied to cases. I will revisit cases later, including alternative interpretations and explanations of cases (especially in Chapters 7 and 8). The type of contextualism favored here will be developed and argued for in the following chapters—in a more theoretical way and step by step. But first to cases. Here is a story: Weather Forecast Mary works as a weather forecaster for the local weather station. While on her lunch break with her friend Charlotte and her colleague Frank, she is discussing with them the possibility of having dinner outside together later in the day. At one point, Mary looks at the sky and notices some dark clouds approaching quickly. She says “Oops, I think we can forget about dinner in

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

10  The Argument from Cases the garden today: it is going to rain within the hour. I can see dark clouds approaching rapidly.” Mary is pretty good at making predictions like that. And, indeed, within the hour rain is falling where they are. Just half a minute after Mary spoke, both Charlotte’s and Frank’s cell phones ring. Charlotte receives a call from her husband who is asking about the dinner plans. Charlotte tells him that they have to call the whole thing off because of incoming rain. “Really?” asks the disappointed husband, “Does Mary know it’s going to rain?” “Yes,” answers Charlotte, “she knows that.” Frank’s call is by the station manager who wants to know whether Mary or he already know whether it is going to rain later in the day. Frank replies “No, we don’t know yet. We haven’t seen the most recent data yet. But we’ll get to it first thing after lunch.”1

There seems nothing remarkable or even slightly interesting about either Charlotte’s or Frank’s phone conversation taken in isolation. However, taken together they constitute a puzzle. Why? When Charlotte says “Mary knows that it’s going to rain,” she seems to be saying something true while Frank also seems to be speaking truly when he says “Mary doesn’t know yet whether it’s going to rain.”2 We can easily assume that Charlotte and Frank are speaking at the same time, referring to the same time and using the same evidence concerning Mary’s epistemic situation. Aren’t they contradicting each other? But how can that be the case if they both speak the truth?3 We can bring out the puzzle even more by assuming that Charlotte is also a colleague: Weather Forecast—Even Worse Mary works as a weather forecaster for the local weather station. While on her lunch break with her friend and colleague Charlotte, she is discussing with her the possibility of having dinner outside together later in the day. At one point, Mary looks at the sky and notices some dark clouds approaching quickly. She says “Oops, I think we can forget about dinner in the garden today: it is going to rain within the hour. I can see dark clouds approaching rapidly.” Mary is pretty good at making predictions like that. And, indeed, within the hour rain is falling where they are. Just half a minute after Mary spoke Charlotte receives a call on her cell phone and a Skype call on her laptop. She is great at multitasking and takes both calls at the same time. On the phone, it is her husband who is asking about the dinner plans. Charlotte tells him that they have to call the whole thing off because of incoming rain. “Really?” asks the disappointed husband. “Does Mary know it’s going to rain?” At the same time, the station manager is asking on Skype whether Mary or Charlotte already know whether it is going to rain later in the day. To her husband, Charlotte says “Yes, she knows that it’s going to rain.” To their station manager, 1   Skeptics concerning the possibility of knowledge of future contingents can easily modify the example without changing anything relevant here. 2   We can assume here that the negation of a truth is false. We can also assume, as part of the description of the case, that there is no further circumstance that clearly makes one of the speakers wrong while keeping the other one right. 3   Further questions arise when we take into account that Mary is making an assertion (“It is going to rain within the hour”) and makes a decision based on her prediction. If appropriate assertion requires knowledge of what is asserted (Unger 1975, ch.6; Williamson 2000, ch.11) and if practical reasoning is acceptable only if based on knowledge (Williamson 2005c, 231; Hawthorne and Stanley  2008; Fantl and  McGrath  2009) then there is a puzzle about Mary’s assertion and practical reasoning, too. See Baumann 2012c.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  11 Charlotte replies two seconds later, “No, we don’t know yet. We haven’t seen the most recent data yet. But we’ll get to it first thing after lunch.”4

How can two apparently contradictory statements both be true? Is the concept of knowledge incoherent? Or should we give up classical logic? Both reactions seem premature and only justifiable if other, less drastic, attempts to explain what is going on in cases like Weather Forecast fail. And there are alternative and less revisionary accounts. One of them is epistemic contextualism, more precisely: contextualism about “knowledge” and closely related terms.5 According to contextualism, the truth conditions and the meaning of sentences involving terms like “knowledge,” “know,” etc.—sentences of the form “S knows that p”—can vary with the context of the speaker who is uttering the sentence (and this because of the context-sensitivity of “know”).6 We can apply contextualism easily to the cases above. In Weather Forecast, for instance, the words “Mary knows that it’s going to rain” when uttered by Charlotte mean something different than when uttered by Frank. According to one at least prima facie plausible way to explain this further, the standards for knowledge—more precisely: the standards a subject has to meet if they can be correctly said “to know” the relevant proposition—vary between Charlotte’s and Frank’s context. In Charlotte’s context, the standards are those of meteorological lay persons and not very demanding, whereas the standards operative in Frank’s context are much more demanding. Because this element of the context, the standards for knowledge, varies and because these epistemic standards partly determine the meaning of “know,” the meaning of Charlotte’s and Frank’s sentences differ. This also explains how they can both be right. Mary meets the laxer standards operative in Charlotte’s context but not the more demanding standards operative in Frank’s context; hence, Charlotte is right to ­attribute “knowledge” to Mary while “Frank” is right to deny “knowledge” to Mary. Standard contextualism is standards contextualism (see, e.g., DeRose 1996a, and also Greco 2010, 103). There are many questions already at this point. For instance: What exactly are standards? What fixes the standard in some situation? What is a context after all? Are there 4   Weather Forecast—Even Worse is similar to DeRose’s multitasking example (see DeRose 2009, 269–77). There are several examples and cases like Weather Forecast in the literature; see DeRose’s bank cases (DeRose 1992, 913), his Thelma and Louise case (DeRose 2009, 3–6), or Cohen’s airport cases (Cohen 1999, 58). 5   Often I will refer to this view just by “contextualism.” 6   Since “knowledge” is a vague term, we might also expect—if one holds certain views about vagueness—that not just the truth value of a knowledge-sentence can vary with context but also whether it does have a truth value: in one context the sentence might be true, in another context false, and in still another context lack a truth value. However, we can disregard this complication here (especially if we adhere to the view that vagueness does not entail possible truth-value gaps). Nothing hinges on the question here whether it is sentences or utterances of sentences that are the primary or exclusive bearers of meaning. For the sake of simplicity, I will speak of sentences having meaning. It is also not intended here to identify truth conditions with meaning; what is very plausible though is that truth conditions cannot vary without a variation of meaning. For classic statements of contextualism see Cohen 1986, 1987, 1988; DeRose 1992, 1995, 1999, 2009; Lewis 1979a, 1996; Unger 1984, 1986. See also Sosa 1986, 1988; Fantl and McGrath 2009, ch.2.; and the debate between DeRose 1992, Brueckner 1994a, and DeRose 2006, as well as the debate between Conee 2005a, Cohen 2005a, Conee 2005b, and Cohen 2005b.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

12  The Argument from Cases other contextual parameters, apart from standards, that determine the meaning of “knowledge,” “know,” and related words? Can standards and other contextual parameters always be characterized as more or less demanding? In other words, is there a hierarchy of contexts? Apart from that, one would also want to know, for instance, what exactly the semantic features of the context-sensitivity of words like “know” are; some contextualists see “know” as indexicals like “I,” “here,” or “now” (see, e.g., Cohen 1987, 15, 1988, 97, 1990, 166). We will get to all these and some other questions soon. It would be misleading to say that contextualism is a view about knowledge. The last word of the last sentence can mean different things according to contextualism. The contextualist thesis entails that “Contextualism is a view about knowledge” can mean different things in different contexts. Does the contextualist then just want to talk about what his word “knowledge” in his context refers to? Hardly. Are there as many different contextualist theses as there are relevantly different contexts in which one can utter the words “Contextualism is a view about knowledge”? Hardly. The contextualist typically rather wants to make one general claim about all the different uses of “knowledge,” not just the one within the contextualist’s context. If this is what the contextualist has in mind, then to say that “contextualism is a view about knowledge” is incompatible with contextualism itself. Contextualism is a view about “knowledge” (and related English words as well as translating counterparts in other languages), not really about knowledge. However, for the sake of simplicity of expression and when there is no danger of confusion the quotation marks may be dropped every now and then (see Lewis 1996, 566–7).7 Finally, to say that contextualism is a view about “knowledge” and not about knowledge can also be a bit misleading insofar as it could be taken to suggest that contextualism is about words only and not about what those words are about. But that would be odd and as we will see contextualism should not be understood as being solely about words. The remarks above concern words, sentences, and utterances. They are about knowledge attributions made by using a (natural) language. However, knowledge attributions need not be expressed in language. Whatever one’s view about the relation between thought and language, it should be relatively uncontroversial that one can think the content of a sentence without uttering it and that one can make a knowledge attribution in thought without saying anything using a language. For that reason, we need both an explanation of contextualism for non-linguistic attributions of knowledge and a more general explanation of contextualism covering both the linguistic and the non-linguistic case (see Section 7.1). One can say this here: contextualism says that the content of knowledge attributions (whether expressed in a language or not) can vary with the context of the attributor (speaker, thinker). However, when possible I  will often, for the sake of simplicity, use pars pro toto and talk about linguistic knowledge attributions also when I have knowledge attributions in general in mind. 7   I will also use ordinary quotation marks instead of corner quotes when the latter would be appropriate; I trust that no confusion will arise from using less cumbersome notation.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  13 Finally, one should expect that contextualism about “knowledge” can be extended to other epistemic (e.g., “justification”) and non-epistemic (e.g., some moral terms) expressions. I have presented a case, Weather Forecast, which one can take as suggesting a certain epistemic view, contextualism. I have also given a first and rough outline of this view. It is a good idea to look at more cases. Here are Keith DeRose’s (1992, 913) very well-known and much discussed bank cases: Bank Case A. My wife and I are driving home on a Friday afternoon. We plan to stop at the bank on the way home to deposit our paychecks. But as we drive past the bank, we notice that the lines inside are very long, as they often are on Friday afternoons. Although we generally like to deposit our paychecks as soon as possible, it is not especially important in this case that they be deposited right away, so I suggest that we drive straight home and deposit our paychecks on Saturday morning. My wife says, “Maybe the bank won’t be open tomorrow. Lots of banks are closed on Saturdays.” I reply, “No, I know it’ll be open. I was just there two weeks ago on Saturday. It’s open until noon.” Bank Case B. My wife and I drive past the bank on a Friday afternoon, as in Case A, and notice the long lines. I again suggest that we deposit our paychecks on Saturday morning, explaining that I was at the bank on Saturday morning only two weeks ago, and discovered that it was open until noon. But in this case, we have just written a very large and very important check. If our paychecks are not deposited into our checking account before Monday morning, the important check we wrote will bounce, leaving us in a very bad situation. And, of course, the bank is not open on Sunday. My wife reminds me of these facts. She then says, “Banks do change their hours. Do you know the bank will be open tomorrow?” Remaining as confident as I was before that the bank will be open then, still, I reply, “Well, no. I’d better go in and make sure.”

In these cases the subject whose epistemic state is under discussion is identical with the attributor. In order to keep subject and attributor more clearly distinguished we can look at Stewart Cohen’s airport case (Cohen 1999, 58; see also DeRose 2009, 3–6 and Fantl and McGrath 2002, 67–8 with their well-known train cases): Mary and John are at the L.A. airport contemplating taking a certain flight to New York. They want to know whether the flight has a layover in Chicago. They overhear someone ask a passenger Smith if he knows whether the flight stops in Chicago. Smith looks at the flight itinerary he got from the travel agent and responds, “Yes I know—it does stop in Chicago.” It turns out that Mary and John have a very important business contact they have to make at the Chicago airport. Mary says, “How reliable is that itinerary? It could contain a misprint. They could have changed the schedule at the last minute.” Mary and John agree that Smith doesn’t really know that the plane will stop in Chicago. They decide to check with the airline agent.

If one wants to take a contextualist perspective on cases like the above, then it is ­useful  to distinguish between what one can call “parameters” and “determinants.” Contextually variable parameters determine the meaning and truth conditions of the relevant knowledge-sentences. These parameters are aspects of context and are, in turn, fixed by contextually variable determinants. Let us take the parameters on first (1.2) and then move on to determinants (1.3).

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

14  The Argument from Cases

1.2  Parameters: Standards I propose to distinguish between five epistemic parameters of a context: standards for  evidence, reliability, degrees of belief, epistemic position, and for ruling out alternatives.

1.2.1  Evidence, Reliability, Degrees of Belief Each of the examples above involve different speakers in different contexts with different operative conditions or standards for “knowledge.” In Weather Forecast Charlotte speaks in a rather personal context where Mary’s lunchtime evidence (a look at the sky, given her experience) is sufficient to meet a necessary condition, an evidence condition, for the truth of “Mary knows that it’s going to rain.” However, for Frank, who finds himself in a much more professional context (being on the phone with the station manager), the situation is different: Mary’s lunchtime evidence is not sufficient to meet the evidence condition because lab evidence is needed (but not in yet). Lab evidence would be good enough for Charlotte’s context whereas lunchtime evidence would not be good enough for Frank’s context. As the case is construed, the evidential standards for knowledge vary between Charlotte’s and Frank’s context. They are higher in the latter context. Sometimes a more demanding context requires just more of the same kind of evidence that is already present in the less demanding context. Physicists need to repeat experiments to be able to claim to “know” certain things. Sometimes, e.g., in Weather Forecast, the evidence in the more demanding context is of a different kind than the evidence in the less demanding one. One should not be misled by all this talk about more or less demanding contexts (which is very prevalent in contextualist discussions) into thinking that contexts can always or even often be ranked as more or less demanding. The contextually variable standards might just be different without one being more demanding than the other. Consider Courts A traffic accident has happened: a blue car hit a red car. Police Officer P witnessed the ­accident with his own eyes and is convinced that the blue car was speeding. Missing speed cameras at the location, one could still draw certain inferences from the damage done to the red car but P has not checked the red car. It is not clear in which state the case will go to court. Traffic courts in state A put a lot of weight on eye witnesses’ testimonies and very little weight on the kinds of inferences just mentioned. Traffic courts in state B go the other way and put a lot of weight on exactly these kinds of inference but not much on eye witnesses’ reports. A judge in state A would come to the conclusion that “P knows that the blue car was speeding” while a judge in state B would rather hold that “P does not know that the blue car was speeding.”

It might not be quite as plausible as in the other cases above that both speakers could be speaking truly but I think it still plausible enough to back the claim that evidential standards for knowledge can just be different without one being more demanding than

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  15 the other. It is important not to assume that whenever there are different evidential standards or standards in general there is also a difference in demandingness.8 One important question here is this: Aren’t there context-invariant truths about the relative quality of different kinds of evidence? Isn’t, for instance, eye-witness testimony like the one in Courts either better or worse or equally good as evidence from the kind of inference mentioned (and no matter what the context)? Is the contextualist then committed to one of two claims: either to the claim that all relevant differences of evidential standards are differences of demandingness or to the claim that kinds of evidence are incomparable in the sense that for at least some pair of types of evidence neither is better than the other nor are they equally good? Or could the contextualist also argue that claims about the relations of quality between different kinds of evidence are context-sensitive themselves? I lean towards incomparabilism here but we need not go further into this. So much about one kind of standard of knowledge: standards concerning the kind, quality, and quantity of evidence necessary for a true attribution of knowledge to a subject (for this kind of standard see, e.g., Cohen 1987; Neta 2002 and 2003a proposes contextualism about “evidence”; see also Prijic-Samarzija  2007). There are others kinds of standards, as the other cases above illustrate. Mary in Cohen’s airport scenario raises the question “How reliable is that itinerary?” which suggests that standards of reliability—of sources of information or epistemic processes more generally—can also vary between contexts. Similar things hold here as for evidential standards (I am avoiding the repetitions). Even standards concerning degrees of belief might vary contextually in the relevant way. Consider this case (modeled after Radford 1966): Exam Linda is taking an oral exam in geography. At one point the examiner asks her what the capital of the Netherlands is. Linda has always been a bit confused about the status of The Hague and she starts to hesitate. Finally she gives the correct answer “Amsterdam.” This is good enough for the examiner who judges that Linda knows this. In the audience (it is a public exam) is her friend Anna who judges that “Linda doesn’t know this” and is happy that she got the correct answer out after all. The examiner does not care much about strength of belief or degree of hesitation and is content with a relatively low degree; he cares much more about the correctness of the final answer. Anna, in contrast, takes the strength of conviction much more seriously and expects much more conviction. 8   In current discussions of contextualism (especially in relation to skepticism) it is often assumed, for the sake of some example, that the standards are higher in the philosophy seminar room than in the pub. So much depends on the details here that this claim must look a bit suspicious. First of all, a lot will depend on the topic: the standards for claims concerning the special taste of some Belgian beer might well be much tougher in the pub than in the seminar room. But even when we restrict the topic to philosophy, it is far from clear whether the above claim is true: What if what matters most in the seminar room is whether people can impress each other and improve their relative standing in the informal departmental ranking of students while in the pub it is nothing but the truth that matters? All this suggests that one has to be quite careful and specific with claims concerning the contextual variation of the demandingness of standards for knowledge.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

16  The Argument from Cases Linda and the examiner can both be right, given the details of the case: they can be because they have different standards for the degree of belief required for knowledge. It is true in all contexts that “Knowledge requires belief ”9 but how much strength of belief is required can vary between contexts. If one believes that knowledge requires outright belief, then one needs a threshold for degrees of belief strong enough to count as outright belief. If one does not believe that there is such a thing as outright belief, one would still need a threshold for degrees of belief strong enough for knowledge. Since standards for the strength of belief can vary between contexts and since the knowledge-belief connection holds in every context, the truth conditions of knowledge sentences can also vary with contextually determined standards for belief.

1.2.2  Epistemic Position One can also argue that there are contextually variable standards concerning the kind of epistemic position sufficient for knowledge. Consider Cheating Al has a strong hunch that his boss has been cheating on his wife for some time. And he is right. However, Al can’t prove anything and does not possess any real evidence. Carl is a colleague of Al and a very close ally of the boss; he introduced the boss to his current mistress. Carl has found out that Al has this hunch and that Al is seriously thinking about informing the boss’s wife. The boss has recently noticed a behavior change in Al. Over lunch with Carl he asks him “Does Al know anything about me and, you know?” Carl bluntly replies, “Yes, Al knows that you’re cheating on your wife. But he can’t prove anything.” The next day the boss discusses the case with lawyer friend Lou. After explaining everything to Lou, she replies “Don’t you worry. Al might have a hunch but he doesn’t know anything; he cannot prove anything. You’re safe!”

In Carl’s context little (if anything) more than true belief is sufficient for a true ascription of knowledge (see also Sartwell 1991, 1992) while in Lou’s context more is required. The standards concerning the type of epistemic state sufficient for knowledge vary between contexts. For that reason, the truth conditions of “Al knows that his boss is cheating on his wife” differ between Carl’s and Lou’s context in such a way that Carl can say something true with this sentence while Lou doesn’t. The contrast between true belief and true belief plus something else is just one of many possible contrasts between epistemic states considered sufficient (or necessary for that matter) for knowledge. One could object to this analysis of Cheating and claim, for instance, that it is just the standards of evidence that vary between Carl and Lou, not the type of epistemic position considered relevant. Carl’s standards are just much lower than Lou’s and that’s all there is to it. However, in reply to this objection one can simply make it explicit in the description of the case that in Carl’s context Al’s evidence (if any) is negligible.   As well as that knowledge requires truth.

9

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  17 There are other variations of this type of context-sensitivity of standards. Consider Fermat Kurt just learned from Wikipedia that Andrew Wiles has proven Fermat’s last conjecture. Kurt has not had any advanced mathematical education and would not be able to understand the proof. Later, Kurt is having the following conversation with his friend Bud: Bud: Kurt, you know nothing. I wonder how you get through life. Kurt: Wait a minute! I know lots of things. For instance, that Fermat’s last conjecture is true. Ha! Bud: How do you know that? Kurt: I just read it in . . . Bud: . . . Wikipedia, right? A bit later Bud tells his mathematician friend Matt about Kurt: Bud: Can you believe it? Math is starting to be part of pop science culture. Even someone like Kurt is aware of the fact that there is such a thing as Fermat’s conjecture. He even read on Wikipedia that it’s been proven. Matt: Wait a minute! What does Kurt know? C’mon. He might have heard about Wiles and Fermat and so on but he certainly does not know that Fermat’s conjecture is true!

Both Kurt and Matt seem right. They can be because the truth conditions of the sentence “Kurt knows that Fermat was right” are determined by contextually variable standards concerning acceptable sources or types of epistemic position. In Kurt’s context, testimony can be sufficient for knowledge while in Matt’s it isn’t (as is typical of mathematical contexts). Hence, they can both be right. There are many different distinctions between epistemic positions sufficient for knowledge: I have mentioned mere true belief versus more than true belief as well as true belief based on testimony versus true belief based on expertise. One could add Ernie Sosa’s difference between animal and reflective knowledge (see Sosa 2007, 2009) or the difference between knowing and understanding some proposition and knowledge without understanding of the proposition (I can know that what my friend, the biologist, says about cell structure is true without understanding it; in a sense I then know the relevant proposition). I do not want to get into further details here but rather mention one worry about the alleged context-sensitivity of standards for types of epistemic position: Isn’t all this just a case of ambiguity of “knowledge”? For instance, sometimes we talk about lay knowledge and sometimes about expert knowledge (e.g., in Fermat) but the fact that we apply the same word “knowledge” without further qualification to the relevant cases does not mean that “knowledge” is context-sensitive; it is simply ambiguous. One might suspect that there is a quite general anti-contextualist strategy at work here, an “ambiguity maneuver” (AM), which tells us to analyze every alleged context-sensitivity as a case of lexical ambiguity. In reply to this AM against the idea of context-sensitive standards for types of epistemic position one can say several things. First, we would not want to run an AM against the claim that “here” is context-sensitive; different uses  of “here” have too much semantics in common (e.g., a Kaplanian character;

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

18  The Argument from Cases see Kaplan 1989; see also DeRose 1992, 921) for “here” to be plausibly analyzed as ambiguous. Similarly, I think, for the above uses of “knowledge.” Second, in the case of lexically ambiguous terms there is a very good chance not to have the ambiguity reproduced in all other natural languages; however, the variation we see with respect to “knowledge” seems to hold across all kinds of languages (at least I know of no contrary evidence).

1.2.3  Ruling Out Alternatives There is another, quite popular way of describing and explaining the variation between different standards for knowledge. In Cohen’s airport case, Mary comes up with an idea: “They could have changed the schedule at the last minute.” Mary and John feel that neither they themselves nor Smith can rule out this possibility. They also judge that therefore neither they themselves nor Smith know that there will be a stop in Chicago. We, the readers, are invited to agree. Interestingly, we are also invited to judge that Smith, the passenger with the itinerary, does speak the truth when he claims to know that there will be a layover in Chicago. Smith, however, also (so it appears) cannot rule out Mary’s possibility. The crucial difference between Smith’s and Mary’s and John’s context is that in the former the subject can be truly characterized as a knower even if he cannot rule out Mary’s possibility, while in the latter context a true knowledge ascription to Smith requires that Smith can rule out that possibility. Contextually varying standards of what needs to be ruled out by the subject explain different truth conditions of the knowledge sentences.10 Another way to put this is to say that Mary’s possibility (change of schedule and no stop in Chicago) constitutes a relevant alternative for Mary and John but not for Smith; hence, the inability to rule it out stands in the way to a true knowledge ascription in Mary’s and John’s context but not in Smith’s context.11 Something similar is present in DeRose’s bank cases. An inverse way of putting all this talk about ruling things out is to talk about what can be taken for granted (and thus need not be ruled out). Apparently, there is a lot of contextual variability here (see, e.g., Blome-Tillmann 2009a, 2012). To use one of Michael Williams’ examples (see Williams  2004, 470–1): A historian can correctly claim to “know” certain historical facts even if he cannot rule out Russell’s doubt that the world came into existence five minutes ago; the historian can take it for granted, even implicitly, that the world is much older. However, a philosopher dealing with epistemological skepticism and the reality of the past cannot just take this for granted. Hence, both the historian who claims that “I know that Napoleon was born more than 200 years ago” and the philosopher who claims that “No, you don’t know that” can be 10   More can be said about what “ruling out” involves but this is not necessary here; an intuitive understanding is sufficient for our purposes. Suffice it to briefly add that I want to use “ruling out” both in an active sense of a subject doing the ruling out and in a passive sense of a subject’s evidence being incompatible with certain possibilities (for the latter see, e.g., Lewis 1996, 553). 11   For more on the notion of a relevant alternative see Stine 1976 and Vogel 1999.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  19 right. The truth conditions of their sentences vary with standards for what can be taken for granted.12 So much for now about the standards parameter in contextualist views. There is an orthodox view amongst contextualists according to which the main or only parameter consists in standards: standard contextualism is “Standards Contextualism.” Below I will argue that there are other parameters to be added to the picture. I have distinguished between five kinds of standards for knowledge: evidential standards (kind, quality, and quantity), standards of reliability, standards concerning strength of belief, standards for types of epistemic positions relevant to knowledge, and standards for ruling out (taking for granted). It is easy to see how these standards can combine with each other. For instance, one standard tells us what needs to be ruled out and another standard tells us what kind of evidence we need to successfully rule out what we need to rule out (see, e.g., DeRose’s bank cases and Cohen’s airport case). Is our list of five standards complete or is there more to be added? Are the different standards mutually independent (in that they can vary independently)?13 Are these standards reducible to fewer? I wouldn’t be surprised if more kinds of standards can be identified or if it turns out that the list can be simplified. I think one can and should wait and see what happens in further discussions of the topic. 12   Let us assume (nothing hinges on this here) that what is or isn’t being ruled out are possible worlds. Very often people make two related assumptions:

1. Possible worlds can be ordered according to closeness or remoteness to the actual world (see Lewis 1973, 1986); 2. It is easier to rule out closer rather than more remote worlds. Even if we leave the controversial first assumption aside here (for problems see Chapter 2), there are serious problems with the other assumption. Let us assume (and the adherent of the first assumption should not resist this kind of assumption) that a world where I am a monk on some mountain top now is more remote than a world where I have one hair less now. However, it seems much more difficult to rule out the latter than to rule out the former possibility (the difference between modal distance and difficulty to rule out seems to play some role in Murphy 2005). Finally, a remark on a third and also quite popular assumption: 3. A (main) difference between standards of ruling out consists in how many possibilities need to be ruled out and larger sets of relevant alternatives contain all the smaller sets of relevant alternatives. This assumption, however, also does not seem warranted; the contextually variant sets of possibilities to be ruled out might just be different. Consider the case of an athletic competition, say long jump. Everyone agrees that Jackie is the winner. However, amongst those who do the measurements and determine how far Jackie leaped it is crucial to rule out certain kinds of mismeasurement; for those who judge whether Jackie misstepped at the take-off point, other alternatives are relevant. What we have is a difference without one set of alternatives being a proper subset of the other (and no matter how hard each of these jobs is in comparison). 13   Standards of evidence (or reliability) and standards of ruling out are closely related (though not the same thing); very often what needs to be ruled out determines to some degree what evidence is needed, and what evidence is needed determines to some degree what needs to be ruled out. However, the two standards still can vary independently. Fix what needs to be ruled out and the question of what kind of evidence is needed to do that can still be open (see, as an analogy, Richard 2004, 229–30, 237–8 on the word “tall” and how whether someone can truly be called “tall” depends both on the comparison class—basketball players, jockeys, etc.—and the cut-off point; see also Richard 2008, 90, fn.1, 101–2, 175). Fix what evidence is needed and the question of what needs to be ruled out can still be open.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

20  The Argument from Cases Are all these standards comparable or commensurable with each other in the sense that one can weigh the relative importance of meeting one standard to a certain degree against meeting another standard to a certain degree? Should one think that there could be, in principle but perhaps not in practice, an overall measure of degrees of meeting the set of standards? I cannot show that this isn’t possible but I am skeptical and much more inclined to assume that there is a multidimensional “space” of standards which are independent and irreducible and which cannot be “commensurated” in the sense above. This also suggests that an idea quite popular amongst contextualists is mistaken: namely that there is a clear hierarchy of standards according to demandingness (see Ludlow 2005, 25–6; DeRose 1995; Greco 2003, end of sec.4; on this see also the remarks above on evidential standards in particular).

1.3 Determinants What determines the parameters? What fixes the standards for knowledge? One may think of several different factors as determinants: stakes, purposes and intentions, certain aspects of the conversational context, salience, and norms and conventions.

1.3.1 Stakes Let us start with standards for ruling out possibilities. Both Cohen’s airport case and DeRose’s bank cases deal with that. In both kinds of scenarios what is practically at stake for the attributor differs in the two contrasting contexts (see DeRose 1992, 914–15, 2009, 53–6; Cohen 1999, 59; Fantl and McGrath 2009; see also for versions of contextualism with a special focus on stakes: McKenna  2011,  2013, and  2014; Hannon 2013; see in favor of salience rather than stakes: Schaffer 2006; see also the critique in Schiffer 2007 and in Nagel 2008; for a different kind of connection between knowledge and stakes see Schroeder 2012). Acting on the assumption that there will be a stop at Chicago when there won’t be one would be very bad for Mary and John but not for Smith (if it makes a difference at all to him). Acting on the assumption that the bank will be open the next day when it is not would be very bad for Keith in bank case B but only mildly annoying in bank case A. What follows from this? The worse the consequences of being wrong about something the smaller the acceptable risk of being wrong about it. So, Mary and John should check further and try to rule out that the schedule has been changed at the last minute; Keith and his wife (in case B) should also check further or, if that is not possible, not act on the assumption that the bank will be open the next day and rather get in line now. They have very good practical reasons to behave like that; it would be foolish to just rely on the assumption in one’s behavior that there will be a stop in Chicago or that the bank will be open on Saturday. However, what is demanded from Mary and John (or Keith in case B) by practical reason is not demanded from Smith (or Keith in case A): his stakes are much lower. The contextualist can accept that and add to,

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  21 e.g., the airport case a remark by John like “Lucky guy—he apparently does not need to check further.” The decisive step in the contextualist account comes now—and it is a subtle one. The attributor (Mary) sees herself as addressee of a certain demand of practical reason: to try to rule out certain possibilities, to check further, etc. She can at the same time admit that this demand of practical reason does, due to different practical circumstances, not apply to the subject (Smith). However, what applies to Smith is a corresponding epistemic standard: if the subject can, in the attributor’s context, truly be called a “knower” of a given proposition, then the subject must meet the same standard that the attributor must meet for the sake of the quite different reasons of practical reason. The attributor turns her own practical reasons into epistemic standards for the subject. One need not assume that all attributors explicitly reason like this but this is the rationale or the explanation of their practice of attributing knowledge, according to the contextualist. If the contextualist is right about the data—as I think they are—then the practice of knowledge attributors conforms to the following principle (whether attributors are guided by it or not): Egocentrism of Standards: My practical reasons are/determine your epistemic standards.14 But, one might ask, is it reasonable to turn one’s own practical reasons into other people’s epistemic standards? Isn’t a practice that conforms to a principle like Egocentrism of Standards just crazy? This, however, is not the relevant question to ask here. The contextualist claims (or  many contextualists do so) that as a matter of fact knowledge attributions are ­context-sensitive because epistemic standards are and that the latter are driven by the attributors’ stakes. All that matters here is whether the data—like ordinary folks’ judgment about what we would say in cases like DeRose’s bank cases or Cohen’s airport case15—confirm the contextualist analysis. If they do, then we should not complain that all this sounds crazy but rather admire the surprising complexity and subtlety of our concept of knowledge (as well as the contextualists’ discovery). Many more questions need to be asked. In DeRose’s bank cases and in Cohen’s airport case, the attributors are in high-stakes cases while the subjects are in low-stakes cases. What if it is the other way around and the subjects are in high-stakes cases while the attributors are in low-stakes cases? What if attributors (or subjects) are ignorant of or wrong about their stakes? There has been quite a bit of discussion of all these ­different kinds of cases recently, including empirical studies (see, e.g., Sripada and 14   As soon as one realizes that third parties have diverging practical interests and reasons one is also close to realizing that there is a problem about the evaluation of knowledge attributions made in other contexts. On this see Chapter 5. 15  DeRose aptly speaks of an “ordinary language basis” for contextualism (see DeRose  2005 and 2009, ch.2).

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

22  The Argument from Cases Stanley 2012). I will go into this and the notion of stakes more generally below (see Sections 7.4 and 8.1 in particular).

1.3.2  Purposes and Intentions At the beginning of an article Stewart Cohen remarks that “the truth value of a sentence containing the knowledge predicate can vary depending on things like the purposes, intentions, expectations, presuppositions etc. of the speakers who utter these sentences” (Cohen 2000, 94). He does not elaborate much at all on this remark but we may use it as a hint at another kind of determinant of standards, especially the standards of ruling out. Let us consider the idea that the purposes and intentions of an attributor (see also Montminy 2013) can determine the standards for what needs to be ruled out by the subject in order to qualify, in the context of the attributor, as “knowing” the relevant proposition (“Cohen’s principle”). Consider the following case: Two Allergies Mary and John are on a plane. Mary has a nut allergy and John has an egg allergy. Mary asked the flight attendant twice and got confirmation that the food contains no nuts; John asked another flight attendant about eggs and got confirmation that there is no problem for him either. Both also then look at the flight information leaflet and take it from there, correctly, that the food contains nothing problematic for people with allergies. They both say to themselves “Now I know that the food won’t cause any problems for people with allergies.”

Let us assume that Mary knows that the food does not contain nuts and John knows that it does not contain eggs. But do they both know that the food does not contain anything problematic for people with allergies? It seems plausible to say so. For both the stakes are high (so that merely consulting the leaflet would not suffice for “knowing” that the food is OK) but in different ways. They both have to rule out the possibility that the food might contain something problematic but given their different allergies they have to rule out different specific possibilities: Mary has to rule out that it contains nuts and may ignore the question whether it contains eggs; John needs to rule out that the food contains eggs and may ignore whether it contains nuts. Since they both asked the flight attendants, we can assume that Mary and John can be said to know that the food does not contain anything problematic for people with allergies. Cohen’s principle tells us what exactly, given the stakes, a subject has to rule out in order to be correctly characterized, in the context of attribution, as a “knower” of the relevant proposition. It is thus a neat complement of the stakes principle. By the way, one can spell out the case in such a way that Mary could at the same time correctly say about John “He doesn’t know that the food does not contain anything problematic” because he does not meet the standards in Mary’s context which demand that one rule out the specific possibility that the food contains nuts (and consulting the leaflet won’t be sufficient for that). Similarly, John could then correctly say about Mary “She doesn’t know that the food does not contain anything problematic” because she does not meet the standards in his context which demand that one rule out the specific

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  23 possibility that the food contains eggs (and consulting the leaflet won’t be sufficient for that). This is what the context-sensitivity of knowledge attributions can come to. One might object that in the above example it is rather basic needs and interests, or stakes, than purposes and intentions that determine specific standards. However, one can easily accommodate this and, e.g., change the case in such a way that neither Mary nor John have any allergies but that for reasons having to do with lost bets Mary intends not to have any nuts (which she loves) for a week while John plans to stay away from eggs (which he loves) for a week.

1.3.3  Conversational Contexts How about other determinants of standards like the standards for ruling things out? David Lewis claimed (Lewis 1996, 559) that there is a “rule of attention” according to which any possibility that comes up in a conversational context is thereby such that it cannot be properly ignored and must be ruled out by the subject if they are to count as “knowers” in that context. There is a narrower and a wider interpretation of this rule. According to the first, only what has been explicitly mentioned in a conversation is relevant to the participants of the conversation. According to the wider interpretation one also has to include possibilities that were not explicitly mentioned but that participants are considering because of something that was said in the conversation; what not all participants or just one participant happens to be thinking of would, however, not be part of the conversational context. It is quite difficult to draw a line between what is being said explicitly and what is being triggered by what is being said explicitly without being said explicitly; and even if we could draw such a line, we would have to ask why the difference should matter here. Hence, it is plausible to go with the wider interpretation of the rule of attention (see also DeRose 1992, 915).16 One issue with this proposal has to do with a lack of specificity in the notion of a conversational context.17 What are the identity criteria for conversational contexts? What makes an episode of interaction one conversation rather than a part of an overarching conversation? Suppose in a variation of DeRose’s bank case A Keith and his wife talk over lunch about whether to deposit their paychecks later that day or on the following day. After Keith’s wife mentions the possibility that the bank might be closed on Saturday, Keith replies that he was there two weeks ago on a Saturday and that it is open until noon. He then adds “Wait, did I not hear somewhere the other day that some banks—I cannot recall which—were intending to stop Saturday hours completely?” Even when the stakes are low, we can imagine the case to be such that Keith introduces an additional possibility that, according to Lewis, cannot be ignored but also cannot be ruled out by the participants; in the lunch context “Keith knows that the bank will be open” would thus be false. Not having reached a decision, Keith and his wife go back 16   It is not clear at all whether Lewis’ idea can deal with knowledge attributions made outside of a conversation, say, in solitary thought. I leave this problem aside here. 17   This is a serious problem for Lewis and those contextualists who give the notion of a conversational context a crucial role in the theory (which I, for instance, don’t do).

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

24  The Argument from Cases to work after lunch. Hours later, in the car on the way back home, they return to the topic. It so happens that they have both forgotten about Keith’s additional possibility. She says “You know, why don’t we go to the bank tomorrow? If you were there on a Saturday two weeks ago, we’re fine.” With Keith’s additional possibility “missing” in the car, we can imagine that Keith and his wife can rule out all the possibilities raised in the car; hence, in the car context “Keith knows that the bank will be open” would be true. The interesting question is whether the lunch context and the car context are just parts of one overarching context or rather two different though closely related contexts. The trouble is that this seems completely indeterminate (we could make it determinate but then we would add something artificial to our ordinary notion of a conversational context). Hence, it would be indeterminate whether “Keith knows that the bank will be open” is true or false; one might even wonder whether there is a truth value gap here.18 Perhaps sentences concerning the identity conditions of conversational contexts are themselves context-sensitive (and thus refer to contexts of “the second order”)? Similar issues arise concerning the social boundaries of a conversational context. Suppose Hank, Frank, Lou, and Sue are having a chat. Hank is a five-year-old, Frank is very tired and close to falling asleep, Lou checks her phone every now and then, and Sue is pretty much committed to the conversation. Apparently, there are different degrees of participation; apart from that, one can also participate partly. How then should we count participants of a conversation? If that is indeterminate (and it looks like it) then what belongs to the set of relevant possibilities in this conversational context is also indeterminate (because what is on the mind of only some participants would presumably not be part of the conversational context). We would get the same kind of problem as the one above again. Finally, people can talk about more than one topic at the same meeting. Is the identity of a conversational context tied to the identity of a topic such that one cannot have different topics in the same conversational context? If not, then all kinds of things could be relevant in one conversational context and it is not easy to see how to make sense of that. If, however, contexts are built around topics, then we would need to know what is part and what isn’t part of a given topic. This is often controversial and the reason there is a controversy might be that there is no fact of the matter that determines what exactly the topic is. Again, we get into the above kinds of problems. But even if one can precisify the notion of a conversational context in a non-­ arbitrary way a more serious problem remains for the rule of attention and the idea that whatever comes up in a conversation is thereby relevant and needs to be ruled out by subjects if they can truly be called “knowers” of the relevant propositions in that context. There are certainly cases (e.g., Cohen’s airport case or DeRose’s bank cases) where the possibilities that need to be ruled out are mentioned or considered by the attributors. But it seems false that whenever a possibility is mentioned or considered, 18   The idea of score keeping (see Lewis 1979a) is not of much help here: one would need to know what the conversation is in the first place for which one is keeping the score.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  25 it becomes relevant and needs to be ruled out. Here is a case from Richard Holton (2003, 291): Hasn’t this gone too far? The murderous axeman knocks at your door; you have hidden your friend upstairs. “Do you know where he is?” demands the axeman. “Interesting that you ask,” you reply, “since I was just thinking that this might all be a bad dream. No, I don’t know where he is.” This might indeed confuse the axeman, but it scarcely gets one off the charge of lying. It seems that the parties to the conversation have to somehow accept the new standards as relevant; here the mere mention of sceptical possibilities does not bring such acceptance. Indeed, even if the axeman had briefly considered the possibility you mentioned, he could surely have returned to the original standards easily enough (“Enough of this tomfoolery: do you really not know where your friend is?”). Such worries do not discredit Lewis’ pragmatic framework for the account of knowledge; but they do raise questions about the particular rules he gives.

Cases like this one suggest very strongly that just mentioning (or just considering) a possibility is not sufficient for having standards in place according to which one has to rule out that possibility in order to count as a “knower.” There is ample room here for  “C’mon” maneuvers (see Cohen  1999, 85, fn.27; Cohen 2005c; DeRose 2004a; Hawthorne 2004a, 84, 161; see also Daukas 2002, 66–7). No defense lawyer would build his defense on the idea that it could have all been a dream and that nobody can rule that out. Mentioning (or considering) a possibility also does not seem necessary for securing standards which demand that that possibility be ruled out. If stakes also determine standards, then the possibility of ignorant high stakes suggests that a possibility might be in need of being ruled out even if the attributor (or the subject) are not even thinking about that possibility. Consider, for instance, a variation of Cohen’s airport case where John and Mary are not aware at all that they have a strong reason to stop at Chicago: Wouldn’t they be mistaken if they attribute knowledge to Smith? Whatever the other determinants of standards are, we can easily imagine cases where the operative standards require the ruling out of possibilities that have not been mentioned or considered. Even if there should be special kinds of cases where mentioning or considering a possibility is sufficient or necessary for certain standards being operative, it seems that in general, at least, this is not the case. In general, it seems that ­mentioning or considering a possibility is just irrelevant to the determination of contextually operative epistemic standards.

1.3.4 Salience Sometimes people argue (see, e.g., Cohen 2005a) that it is the salience, in some context, of certain chances of error that determines which possibilities need to be ruled out by the subject in order to be correctly characterizable, in that context, as “knowing” the relevant proposition. In both of DeRose’s bank cases Keith’s wife makes a certain possibility (the bank is not open on Saturdays; the bank has changed its hours) and the corresponding chance of error (assuming that the bank is like many other banks open on Saturdays when it’s not, or assuming it has not changed its hours when it has) salient.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

26  The Argument from Cases Similarly, and even more explicitly in Cohen’s airport case, where Mary is quoted as saying, “How reliable is that itinerary? It could contain a misprint. They could have changed the schedule at the last minute.” This is similar to the above ideas about considering and mentioning possibilities. Analogous problems arise. Even though often possibilities of error that need to be dealt with are salient in a context (the bank cases, the airport case, etc.), it is often not the case that such salience has any influence on the standards. Consider this fictional dialog: andrew:  Now I know that Fermat was right and that his famous conjecture is true. questioner:  Are you sure? andrew:  Yes! questioner:  Did you check? There might be some mistake in the proof. andrew:  I myself checked it many times and so did the leading experts. questioner:  But you and all these experts could, for instance, all have made the same mistake without ever noticing it. andrew:  Sure, but so what? questioner:  Well, if you can’t prove that that hasn’t happened, then you don’t know that the proof is right. andrew:  C’mon! (I have a reply I can’t put on paper.) Perhaps what is needed for the salience of a chance of error is the salience of a specific chance of error. Whatever the criteria for specificity are, here is a case that seems straightforward on that front: bert:  Hey Ernie, hey Ernie, look: I got a new haircut! ernie:  Oh, are you sure that’s your hair? Because if it’s not, then it isn’t your haircut (giggles). bert:  What? Sure! How could it not be my hair? What do you mean? ernie:  Your hairdresser could have played a prank on you. They could have put a drug into the shampoo that made you hallucinate that you were getting an ordinary haircut. But then (giggles) they could have ripped out all your natural hair and replaced it with an incredibly well done chewing gumbased imitation of natural hair. Who knows—perhaps they then just snipped their fingers, the hallucination ended, you paid, convinced that you got a real haircut. And still have your real hair (giggles). bert:  (shakes his head). Whatever, Ernie! I’m going to have one of your bananas! Again, it seems very plausible to say that the salience of a (particular) chance of error is not sufficient for having standards in place which demand the ruling out of the corresponding possibility. It also seems hard to see how this could be necessary. Like the mentioning or considering of possibilities, the salience of a chance of error

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  27 seems irrelevant to the determination of contextually operative standards (e.g., for ruling out).

1.3.5  Determinants for the Other Parameters. Norms and Conventions So much about the factors which determine the standards for ruling out possibilities. What about the other four kinds of standards? Are they governed by the same determinants? Let us look at evidential standards, that is, standards concerning the kind, quality, and quantity of evidence. Plausibly, how high or low the stakes are matters also in this respect. More, better, or more diverse kinds of evidence are needed when a lot depends on the truth of the relevant proposition. Mary’s critical remark in Cohen’s airport case—“How reliable is that itinerary? It could contain a misprint. They could have changed the schedule at the last minute”—can also be interpreted as a request for more and better evidence (“Ask that guy, he is just looking this up on the internet!”) or for different kinds of evidence (“Asking people like Smith won’t help much. Let’s ask at the gate”). Without the relevant change of the evidential situation, Mary would not attribute knowledge to John that there will be a stop in Chicago and it is plausible to say that an attribution of knowledge by her would not be true. Similarly, Cohen’s principle (see above) would have it that the purposes and intentions of the attributor can determine how much evidence of what kind and quality is needed to qualify as a knower of the relevant proposition. Consider Two Allergies again. Mary and John need different kinds of evidence (Mary about nuts, John about eggs) before they can correctly say about themselves that they “know” that the food does not contain anything problematic; both also need more and better evidence than the traveler who does not and need not care at all about allergies. Like in the case of stakes the points about purposes and intentions of the attributor are parallel to the ones made above about standards for ruling out and need not be spelled out in more detail here. The critical points about the alleged determining role of the conversational context and the role of salience for standards for ruling out also apply, mutatis mutandis, to evidential standards. However, there is reason to think that there is one determinant of standards concerning the quantity, quality, and kind of evidence that we don’t see at work in the case of standards for ruling out. Consider Weather Forecast again: Weather Forecast Mary works as a weather forecaster for the local weather station. While on her lunch break with her friend Charlotte and her colleague Frank, she is discussing with them the possibility of having dinner outside together later in the day. At one point, Mary looks at the sky and notices some dark clouds approaching quickly. She says “Oops, I think we can forget about dinner in the garden today: it is going to rain within the hour. I can see dark clouds approaching rapidly.” Mary is pretty good at making predictions like that. And, indeed, within the hour rain is falling where they are. Just half a minute after Mary spoke, both Charlotte’s and Frank’s cell phones ring. Charlotte receives a call from her husband who is asking about the dinner plans. Charlotte tells him that they have to call the whole thing off because of incoming rain.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

28  The Argument from Cases “Really?” asks the disappointed husband, “Does Mary know it’s going to rain?” “Yes,” answers Charlotte, “she knows that.” Frank’s call is by the station manager who wants to know whether Mary or he already know whether it is going to rain later in the day. Frank replies “No, we don’t know yet. We haven’t seen the most recent data yet. But we’ll get to it first thing after lunch.”

Why would an attributor in a lab context require more, better, and different evidence than the attributor at lunch? Sure, stakes as well as purposes and intentions might play a role but can these factors explain everything? Apart from that, Mary might just be doing her job and her stakes or her purposes and intentions might not drive the standards up very much (if at all). To be sure, perhaps it is what is at stake for the general public or the purposes and intentions of many people that matter here and drive up the standards in the lab. What is at stake for Mary must not be restricted to what is at stake for herself personally. Just think of someone who is trying to defuse a bomb in the center of some big city. But still, isn’t there more at work in cases like Weather Forecast—something like institutional conventions and norms? We can easily and plausibly imagine the station manager remind Frank that “At Weather for You we do things more carefully than at other places!” Conventions and norms in general appear to be an independent factor determining evidential standards. In Weather Forecast the norms and conventions of the lab require different kinds of evidence of a certain quality and quantity whereas nothing like this is the case in the lunch context. This also explains the different truth conditions of the different knowledge sentences. Perhaps the practical aspects in Weather Forecast occlude the norms and conventions at work there, too. One could, as an alternative, look at Courts again but let us rather look at still another case where it is harder to think of any practical dimension of the operative standards and where the conventional or norm-related aspect might be clearer: Amateur Genealogists Lineage and Genealogy (LaG) is a pretty ambitious association of amateur genealogists. At their meetings they discuss topics like “Who was Jack the Ripper’s maternal grandfather?” U. N. B. Daft is not a member but visiting today because he might be interested in becoming a member. He volunteers his true opinion that so-and-so was Jack the Ripper’s maternal grandfather and mentions that he’s read that in a popular biography. The members of LaG frown upon this, politely shake their heads and whisper amongst themselves that “This guy doesn’t know a thing about the Ripper family.” They then proceed to explain to Daft that they themselves make trips to different archives and check on different hypotheses using different kinds of sources before they even participate in a general discussion of cases at LaG: “We can’t help it but we just take research very seriously!”

In an LaG context there are norms or conventions that raise the evidential standards so much that it is not easy to be correctly characterized, in that context, as a “knower” of LaG specific propositions. In a different context, say at Daft’s bowling club, the norms and conventions for evidential standards concerning genealogical claims might be much more relaxed; in that context it might be true to say that Daft “knows” that

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  29 so-and-so is Jack the Ripper’s maternal grandfather (“Wow, did you say you read a whole book about this?”). What I said about evidential standards here also holds, mutatis mutandis for standards of reliability and we don’t need to explore the close parallels in any detail here. This leaves us with the remaining two types of standards: standards concerning strength of belief and standards for types of epistemic positions. Let us look at the former first and go back to Exam. Exam Linda is taking an oral exam in geography. At one point the examiner asks her what the capital of the Netherlands is. Linda has always been a bit confused about the status of The Hague and she starts to hesitate. Finally she gives the correct answer “Amsterdam.” This is good enough for the examiner who judges that Linda knows this. In the audience (it is a public exam) is her friend Anna who judges that “Linda doesn’t know this” and is happy that she got the correct answer out after all. The examiner does not care much about strength of belief or degree of hesitation and is content with a relatively low degree; he cares much more about the correctness of the final answer. Anna, in contrast, takes the strength of conviction much more seriously and expects much more conviction.

Leaving aside salience and the conversational context as not even clearly applicable in such a case, one might wonder whether norms and conventions could regulate how much conviction the subject would need to be correctly called a “knower” in a given context. Do Anna and the examiner follow different norms and comply with different conventions? Are there such norms and conventions for strength of belief in the first place? I think one should be skeptical here and be careful with attributing any role here to norms and conventions. What about stakes? We are able to imagine that more is at stake for Anna than for the examiner but why should that have an influence on the  threshold for the degree of belief required for the correct application of the ­knowledge-predicate in a given context? Again, we have reason to remain skeptical. All this leaves us with one factor: the purposes and intentions of the attributor. Anna is Linda’s friend and expects to deal with her on many other occasions in the future. Friendship requires some stability of interaction over time which, in turn, requires that it is possible to expect some stability of belief over time from each other. Hence, Anna will expect a higher degree of conviction (necessary for more stability of conviction) from Linda than the examiner (assuming he is not also a friend of Linda). Add the assumption to the case that Anna and Linda are thinking about traveling to Europe together: Wouldn’t Anna mind if Linda showed no interest at all for European geography and no somewhat stronger views on it? The examiner in contrast just wants to take a snapshot and is not interested in Linda’s future. He cannot, we may assume, be bothered with paying much attention to degrees of conviction of this or that particular candidate; he rather just wants to see whether they get a sufficient proportion of answers right. He would pause if the candidate appeared to blindly pick answers but that is not the case with Linda. So, Exam is a case where we have different contexts with

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

30  The Argument from Cases different intentions and purposes of the attributor which determine different standards for degree of conviction (required for knowledge) and thus also different truth conditions for “Linda knows that Amsterdam is the capital of the Netherlands.” Exam can also be read (perhaps slightly modified) as illustrating the presence of standards for types of epistemic positions required for knowledge. In Anna’s context the standards for knowledge require that the subject have more than a mere true belief while in the examiner’s context true belief seems sufficient. Or consider Cheating, again: Cheating Al has a strong hunch that his boss has been cheating on his wife for some time. And he is right. However, Al can’t prove anything and does not possess any real evidence. Carl is a colleague of Al and a very close ally of the boss; he introduced the boss to his current mistress. Carl has found out that Al has this hunch and that Al is seriously thinking about informing the boss’s wife. The boss has recently noticed a behavior change in Al. Over lunch with Carl he asks him, “Does Al know anything about me and, you know?” Carl bluntly replies, “Yes, Al knows that you’re cheating on your wife. But he can’t prove anything.” The next day the boss discusses the case with lawyer friend Lou. After explaining everything to Lou, she replies “Don’t you worry. Al might have a hunch but he doesn’t know anything; he cannot prove anything. You’re safe!”

In Carl’s context, the standards of epistemic positions require less than true belief while in Lou’s context much more is required. Hence, the difference in truth conditions between these two contexts. What determines these standards? Again, we can leave out the conversational context and salience here; it is hard to see how they could apply to this kind of standard. What about norms and conventions? Consider Exam again. It might well be due to the norms regulating geography exams that not much more than true belief is required for truly characterizing, in such a context, a subject as “knowing” the relevant proposition. Fermat and Cheating suggest the same general point except that in the latter two cases the norms and conventions of the professional context are more demanding than the norms and conventions of the non-professional context. But this need not be so: Why should professional contexts always push the standards up (see also note 8 above)? Do stakes make a difference to the standards concerning the epistemic position of the subject? In Cheating the stakes are much higher in lawyer Lou’s context than in Carl’s context. Lou cannot afford to just rely on hearsay and hunches. She needs legal proof. In her context, the standards for knowledge require that there is more than just true belief in order to make “Al knows that the boss is cheating” true. Similar things can be said about the purposes and intentions of the attributor. In Exam, for instance, Anna has certain plans involving Linda which the examiner lacks. In her context, but not in his, much more than true belief is necessary to make the relevant knowledge attribution to Linda true. Similarly in Cheating: given the role of the lawyer, in Lou’s context the subject needs nothing short of legal proof in order to count as “knowing.” One can also look at Fermat in this light: the practice of mathematics differs in many

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

The Argument from Cases  31 ways from “ordinary” practices and one difference lies in the purposes of the practitioners. Kurt’s and Matt’s contexts are very different in that respect which explains why both Kurt’s self-attribution of knowledge and Matt’s denial that Kurt “knows” are true.

1.4 Conclusion We can wrap all this up and pull it together in Table 1.19 Is the list of factors which determine epistemic standards complete? I don’t claim this and would welcome further additions, though I myself cannot think of any. However, given the difficulties of completeness arguments in such areas, I doubt there is any way of showing whether such a list is complete or not. Are the items on the list mutually independent and not reducible to each other? I think so but won’t go beyond an appeal to plausibility here.20 What is the relative strength of the different factors? I don’t think that there can be a general answer to this question: Why should this not depend on particularities of the case at hand? One might also raise the question whether having a particular standard operative in a given context rather than another standard can be justified (either in an epistemic way or in a pragmatic way). Or is it all brute determination by the factors mentioned above? If the latter: Doesn’t this make our practice of attributing knowledge look arbitrary in a worrisome way? I would reply that one should not expect too much justification here. There can be explanations of why we have the concept of knowledge and why it has the features it has but it seems misguided to expect a justification for all that. Sometimes requests for justification are out of place (see, e.g., Wittgenstein 1969, §§ 110, 192, 204; see also Section 7.5.2). Table 1.  Standards and factors Determinants: →

Stakes

Purposes and intentions

Norms and conventions

√ √ √ − √

√ √ √ √ √

− √ √ − √

Parameters: standards for: ↓ Ruling out Evidence Reliability Degree of belief Epistemic position

19  “√” indicates that the determinant fixes the parameter. “−” indicates that there is no such fixation or that there is at least some doubt about it. My reason for assuming that standards for ruling out are not determined by norms and conventions is mainly that these norms and conventions would have to be specifically about particular alternatives, and that it is hard to imagine how this could be the case. However, this is not to rule out categorically that someone could make a case for the existence of such a connection. 20   Stakes and purposes or intentions can vary independently from each other: in Two Allergies, for instance, the stakes are the same but the purposes and intentions of the attributors differ. And in the bank cases the stakes differ but the relevant purposes and intentions of the attributor seem the same.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

32  The Argument from Cases In this chapter I have mainly done two things. First, I have made at least a prima facie case for contextualism based on the ordinary usage of words like “know”; there are, it seems, many cases which suggest that “know” and related words are context-sensitive. However, one should not overestimate the weight of such arguments. Second, in the process I have given an outline of what contextualism could amount to—with a special focus on parameters and determinants (there has been surprisingly little systematic inquiry into these). To be sure, there is more than one version of contextualism and the version proposed here differs from standard contextualism; it will be developed in more detail over the next chapters.21 The “ordinary language-based” argument (see DeRose 2005 and 2009, ch.1) for contextualism raises some questions and has come under attack from two directions. First, there are alternative explanations of what is going on in cases like DeRose’s bank cases or Cohen’s airport case; I will go into this below (especially in Chapters 7 and 8). Second, there is a growing body of research in “experimental philosophy,” some of which seems to support the contextualist and some of which seems to provide evidence against it. It seems to me that this research has been somewhat inconclusive. We don’t need to go into this here—also because the case from cases is of limited importance here. There is a more general question as to what, in principle, one can expect from ordinary language arguments like the ones presented in this chapter. Austin famously remarked (see Austin 1970, 185) that ordinary language does not have the last word in philosophy but the first word. I find this to be a very good attitude in general. In the case for contextualism, though, I do not want to put too much weight on the argument from cases (against a majority of contextualists) and rather make a more “theoretical” argument for it (see also Fantl and McGrath 2012a). This will be the subject of the following chapters and the core of the approach proposed here. Apart from that, contextualism can also explain certain things. Many contextualists entertain the hope that it can give us a promising response to the skeptic. I am skeptical about that but there are certainly lots of things contextualism can explain better than the alternatives (see Chapter 8). But again, the main weight here shall be on the more theoretical reasons to adopt contextualism. Finally, like every other view contextualism creates its own problems. We will go into the main ones as we go along spelling my preferred version of contextualism out in what follows.

21   Many contextualists assume, implicitly or explicitly, that standards can be ordered hierarchically. I have already indicated some skepticism about this idea: at least in many cases this seems not to be the case.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

2 The Argument from Reliability The Role of Reference Classes Many make the case for contextualism from cases, as laid out in Chapter 1. However, there are also more “theoretical” considerations concerning the concept of knowledge that suggest that contextualism is true. I take these considerations to have even more weight than the case from cases. They will also show that contextualism has more dimensions than just the dimension of standards. The main argument in this chapter very roughly goes like this: knowledge requires reliability; reliability is context-sensitive; therefore knowledge is context-sensitive. Or, less informally: The Reliability Argument for Contextualism (RA) (1) If “S knows that p” is true in C, then “S’s belief that p is reliable” is true in C;1 (2) The truth conditions of sentences of the form “S’s belief that p is reliable” can vary with the context of the speaker; (C) Hence, the truth conditions of sentences of the form “S knows that p” can vary with the context of the speaker. That there is talk of contexts in (1) does not beg the question in favor of contextualism; it just keeps the doors open for contextualism; (1) does not imply contextualism but is  fully compatible with the view that the two attributions mentioned in it have ­context-invariant truth-conditions (in which case it would be merely odd but not false to mention context here). For the sake of simplicity and when there is no danger of misunderstanding, I will in the following often use the word “knowledge” when I should, strictly speaking, mention it. The above sketch of an argument is rough also because the inference from (1) and (2) to (C) is, obviously, not valid. That a necessary condition of F has a certain property P does not entail that F has P. For instance, that C is a necessary but not sufficient condition of F does not entail that F is a necessary but not sufficient condition of itself. With respect to our case: suppose that “reliability” is context-sensitive and that knowledge requires reliability. This does not rule out that knowledge requires not just reliability but, say, maximal reliability and that “maximal reliability” is not context-sensitive. In   “S” stands for subjects, “C” for contexts, and “p” for propositions. I assume that (1) holds necessarily.

1

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

34  THE ARGUMENT FROM RELIABILITY that case “knowledge” could still be context-invariant; “maximal reliability” would “screen off ” the context-sensitivity of “reliability.”2 However, the argument can still be a good, non-deductive, defeasible argument for the context-sensitivity of “knowledge.” This will be my line here (rather than the attempt to add further premises which would give us a deductively valid argument). The context-sensitivity of “reliability” is a major explanandum of the context-sensitivity of “knowledge” and cannot be reduced to some more basic factor. This does not mean that there aren’t other factors, too, that explain why “knowledge” is context-sensitive; however, ceteris paribus the context-sensitivity of “reliability” correlates with and explains the context-sensitivity of “knowledge.” Let us start with the relation between knowledge and reliability (2.1) and reliability and probability (2.2), before we move on to the core of the argument (2.3) (followed by some remarks on modal views of reliability).

2.1  Knowledge and Reliability (1) says that knowledge requires reliability: necessarily, if S knows that p, then S’s belief that p is reliable. To say that a belief is reliable I take to mean that it has been acquired in a reliable way. It does not matter here whether the subject consciously and conscientiously applies a certain method of settling whether p or whether the belief comes about in ways the subject is not aware of. Also, there are different competing explanations of “reliability.” According to one sort of modal views, a reliable process of belief acquisition could not have easily (not in close possible worlds) led to a false belief; according to another sort of modal view, the subject’s reliable belief could not have easily (not in close possible worlds) been false. According to probabilistic views, a reliable process has a high (enough) probability of leading to a true belief. Below I will say much more in detail about these different views of reliability and argue for a probabilistic view. Before I go into that, I need to address an immediate, basic worry concerning the overall argument (RA). In order to do that I only need a general, intuitive understanding of what “reliability” of a belief and its acquisition amounts to. The worry is simply that reliabilism is just one amongst many different theories of knowledge; so, in order to make (RA) convincing as a general argument for contextualism, one first needs to show that reliabilism is true. I haven’t done that and won’t do it here. But doesn’t (RA) then collapse immediately? Alternatively, wouldn’t one have to restrict (RA) massively so that it only works for reliabilists and does not need to concern anyone else? It is important to distinguish between reliabilism as a view about knowledge and the reliability requirement for knowledge. Reliabilism as a view about knowledge says, roughly, that knowledge is true and reliable belief (as classic statements of the view see Ramsey 1990a, 91–4, 1990b, 110; Goldman 1992a, 1992b; see also Armstrong 1973; 2   See, e.g., for the case of a necessary condition of knowledge not being closed under known implication: Warfield 2004.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  35 Dretske 1981a; and Nozick 1981). Whether this is meant as a reductive definition listing individually necessary and jointly sufficient conditions for knowledge or not can be left open here. The reliability requirement amounts to much less: it just states a necessary condition of knowledge. It is only the latter that one needs for (RA). But still, doesn’t the worry persist? Isn’t it mainly reliabilists who would subscribe to (1) or the reliability requirement for knowledge? Wouldn’t one still have to make the case for reliabilism or, alternatively, restrict (RA) considerably to reliabilist views? The answer is “no”: We only need a pretty basic notion of reliability and in that sense (almost) everyone would agree with the reliability requirement (but cf. Sartwell 1991 and 1992 according to whom true belief is sufficient for knowledge). In other words, in the very basic sense that is relevant here, (almost) everyone is a reliabilist. One slight modification of this claim and of (1) is necessary because (as was argued in Chapter 1) in some contexts a subject’s true belief is sufficient to make the corresponding knowledge attribution true. Since this is only a minor modification which does not affect the substance of (1) or (RA), we can, for the sake of simplicity, just leave it aside here. The main question is: Why should one think that (almost) everyone agrees with the reliability requirement for knowledge? Consider the view (see as an example Lehrer and Paxson 1969) that knowledge is  true belief based on sufficient justification of a certain kind (so as to survive Gettier-type examples; see Gettier 1963). One can make the view as “internalist” as one wants and, e.g., hold that whatever does or carries the justification is a mental state of the subject that is unproblematically accessible to the subject and gives the subject a good epistemic (truth-related) reason to hold the relevant belief (for internalism and externalism see, e.g., Fumerton 1988; Alston 1989; Kim 1993). Such a view is very far removed from typical externalist or reliabilist views of knowledge. But still, there is a reliability condition built into even such an extremely internalist account—whether it is explicit or implicit. On any plausible such account it cannot be sufficient for knowledge that there is some such justification for the relevant belief; the subject must also “base” the relevant belief on that justification (see, e.g., Feldman and Conee  1985, 15, 24 on justified and well-founded beliefs; see also Comesaña 2006, 45–6, fn.14; on the basing relation see, e.g., Korcz 2015). Whatever the details of the basing relation, there must be some process leading from the justifier to the justified belief. And that process better be reliable if the relevant belief is supposed to constitute knowledge. Consider an example. Assume that Mary believes truly that the cat is either in the kitchen or in the bathroom, and that she also believes truly that the cat is not in the kitchen; assume that she then makes the relevant inference and comes to truly believe that the cat is in the bathroom. If we want to think of her belief that the cat is in the bathroom as a piece of inferential knowledge—and if she is not making any other relevant inferences at the moment—then we have to assume that her belief that the cat is in the bathroom is in some sense based on her two other beliefs and that there was some mental, psychological process leading from the two to the one. And if that process is

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

36  THE ARGUMENT FROM RELIABILITY not reliable, then it is hard to see how Mary could come to know that the cat is in the bathroom. But how, one might ask, could that process of deductive inference not be reliable? The logical relation between the premises and the conclusion of her inference is impeccable; the conclusion cannot be false, given the truth of the premises and the validity of the argument. Mary’s process of inference must therefore be perfectly reliable. However, this objection confuses the logical and the psychological aspects of inference.3 It is one thing to say that the premises logically entail the conclusion—that there is a certain logical relation between the propositional contents of Mary’s beliefs. It is another thing to say that Mary has to go through some process (presumably some psychological process) to get from the premises to the conclusion. The psychological process of making a certain inference is not already given by the fact that there are certain logical relations. And the former is needed for knowledge. Now, Mary might make her inference by constructing mental models for the ­disjunctive syllogism and her work with whatever mental model she uses might be more or less reliable. Suppose she imagines a crossed-out cat in the kitchen to represent that the cat is not in the kitchen; it does not take much to imagine how this way of thinking about things can easily lead to confusion and, e.g., make Mary forget about the cross and belief that the cat is in the kitchen. Things like that happen all the time (see Johnson-Laird 1983, Johnson-Laird and Byrne 1991, Kahneman et al. 1982). This illustrates the point that even internalist views of knowledge contain, if they are plausible at all, a reliability requirement for knowledge (whether implicitly or explicitly): the relevant processes the subject is going through have to be reliable in order to allow for knowledge (see, e.g., paradigm internalists Feldman and Conee 1985, 25, 27 on the “equivalence” of their evidentialism with a sort of reliabilism of the abstract sort laid out here). A “justificationist” view of knowledge like the one discussed above is often considered to be very far removed from reliabilist views of knowledge. Can we expect that there are any other views of knowledge that are not committed to (1)? Whatever one’s view of knowledge—if more is considered necessary for knowledge than true belief—it is hard to see how any further necessary condition could be compatible with the lack of reliability in the relevant processes. In that sense, (almost) everyone is committed to (1) in (RA), the reliability requirement for knowledge. It is not controversial at all, even if it might seem so. There is thus no need to restrict (RA) in a substantial way by cutting down on (1). But even if this were not so (see, e.g., Turri 2015a), it would be possible to restrict (1) and (RA) in a not so dramatic way: even if knowledge in general should not require reliability, a lot of knowledge does. And that would still be sufficient to be interesting. 3   See here Harman’s distinction between argument and reasoning or between implication and inference (Harman 1986, 3–4). In the background is Frege’s distinction between psychology and logic (see Frege 1884, xxii). As far as the notion of justification is concerned, we are dealing here also with the difference between doxastic and propositional justification.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  37

2.2  Reliability and Probability The above argument for the reliability requirement for knowledge—for (1) in (RA)—did not require any specific view on reliability; on the contrary, in order to keep the argument as “ecumenical” as possible I used a very general notion of reliability. In order to proceed to making the case for (2)—the claim of the context-sensitivity of reliability—I need to be more specific and explain the notion of reliability in some detail. There are two main views: a modal and a probabilistic one.4 I favor the latter. I will introduce it first here and describe the source of context-sensitivity afterwards. Later (see Section 2.4) I will discuss modal views of reliability and explain what I think the major problems with such accounts are; nevertheless, it will also turn out that there is context-sensitivity for reliability according to the modal views of reliability, too. According to the probabilistic view of reliability, a belief is reliable just in case it has been acquired in a reliable way or, in other words, just in case it results from a reliable process; furthermore, the reliability of the process of acquisition consists in a certain probability that that process would lead to a true belief. As pointed out above, “acquisition” should be interpreted in a broad way so as to cover the conscious application of some cognitive procedure (e.g., thinking hard about some mathematical problem, using some algorithm) as well as processes the subject is not even aware of (e.g., visual depth perception in normal subjects). In the following, I will adhere to common usage and use the expressions “method” and “use of a method” to cover both of these cases (both of what Goldman 1986, 93 calls “processes” and “methods” in a more narrow sense). Consider an uncontroversial case. There is a cup on the table in front of me. By looking at it, I form the belief that there is a cup on the table. Circumstances are not abnormal so that we can also assume that my belief is reliable: the process of visual perception is reliable insofar as looking at close midsize objects like cups (in a situation like the one considered here) typically leads to a true belief about the closeness of the location of the object. We can generalize this case. A belief-token5 (e.g., my belief now that there is a cup on the table) is reliable just in case it has been brought about by a process-token6 (e.g., what I just did: looking at the table) that belongs to a type of belief-producing process (e.g., visual perception of close midsize objects) which has a high probability of bringing about a true belief about the relevant type of topic or of correctly answering the relevant type of question (e.g., “What is the location of the object under discussion?”). 4   The modal account can do without probabilistic notions and the probabilistic account can do without modal notions, at least as far as the basic account is concerned. Whether there is a secondary way in which probabilistic notions can play some role in a modal account or modal notions in a probabilistic account can be left open here. 5   “Knowledge” applies to belief-tokens; hence, we can focus on tokens rather than types of beliefs. 6   We can also plausibly assume that the process-token only brings about one belief about the subject matter (“Is there a cup in front of me?”). Looking around normally does not make me believe that there is a cup and also make me believe that there is no cup.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

38  THE ARGUMENT FROM RELIABILITY Let us call the belief the reliability of which we want to explain the “target belief.” Let “P(y/x)” refer to the conditional probability that y, given that x (for some x and y).7 Let “t” abbreviate “The subject acquires a true belief about the relevant type of topic (the target belief ’s type of topic)” and “m” abbreviate “The type of belief-producing process that brought about the target belief happens.” Finally, “n” is some constant value. Then we can take a first shot and say: (REL) S’s belief that p is reliable iff P(t/m) ³ n. A few explanations are in order.8 First, one should not put the wrong weight on the biconditional form of (REL). I intend (REL) to be what Carnap called an “explication” (see Carnap 1958, sec.2): it has elements both of a stipulative definition and a descriptive definition capturing the actual usage of a term. So, I only attempt to get close enough to the actual usage of “reliable” (especially in epistemic contexts); a reductive definition of “reliable” as it is used in the language is probably not possible and it is also not necessary here. Apart from that, some element of stipulation is useful here and probably inevitable. (REL) explains the reliability of a belief-token with reference (on the right-hand side) to two types: types of processes and types of topics. This will be very important in what follows. It is also inevitable. It is hard to defend the idea of the probability that a subject acquires a true belief about some topic, given that a particular token-process takes place. The token-process is the process that brought about the target belief-token and that belief-token is either true or it isn’t true. How could the reliability of a belief-token then consist in the probability that it is true? Either it is not clear what this could mean in the first place or it is but then it seems to go against a core of the notion of reliability (which allows for degrees). The typing of processes seems inevitable and also brings with it a typing of beliefs acquired. This can also be seen by looking at the example of an indexical belief like my belief that I have a cup in front of me now. Insofar as I can consider this belief to be reliable, I have to refer to beliefs of a certain type and say something like “Vision is great for figuring out what’s around oneself.” Otherwise 7   It does not matter here whether we take “x” and “y” to range over events, states of affairs, or something else. It also does not matter here which interpretation of the probability calculus we are choosing (see Kyburg 1970, chs.3–7 and Gillies 2000); nothing hinges on that here.—As readers will notice, I do not have much use for the notion of an unconditional probability here. 8   I assume that (REL) holds necessarily. I am not committed to “ ≥” rather than “ >”. One obvious implication of (REL) is that both true and false beliefs can be reliable. One could restrict (REL) to true beliefs because knowledge entails truth; then reliabilists could simply state their view by saying that knowledge is reliable belief. However, I will go with the unrestricted (REL) here; there is no reason not to do so. (REL) is compatible with the rejection of bivalence. What matters for reliability is whether the subject acquires a true belief, given some process or method; what happens if it doesn’t—whether it acquires false beliefs or beliefs with a different truth-value than truth or falsehood—does not matter here. What is assumed here is that the process produces some belief. According to Goldman 1986, 103, for reliable processes and methods the ratio of true resulting beliefs to all resulting beliefs is greater than ½. This way of putting things seems to suggest a frequentist interpretation of reliability (but cf. Goldman 1986, 50 where he suggests a propensity view of probability). For another interesting probabilistic conception of knowledge see Kvart 2006.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  39 I would have to explain the reliability of my belief via the probability that I acquired this true belief-token, given the use of that process. If then by “process” we mean a token, then the probability would be 1 whereas if we mean a type, then the probability might be indefinitely low because there are indefinitely many alternative process-­ tokens (leading to different indexical beliefs). Both options are unacceptable. Or, making the point from a different direction, consider the weird case of someone who typically acquires true beliefs about the location of close midsize objects by looking around (under normal circumstances) but in addition to that also acquires complicated false mathematical beliefs whenever he looks around. We still want to say that looking around is a reliable method for something and therefore have to restrict the types of beliefs considered to be relevant to beliefs about a certain kind of topic (the location of close midsize objects) and exclude certain other kinds of topics (certain mathematical topics which are better not dealt with simply by looking around). Reliability admits of degrees but there is also a threshold that separates the reliable from the unreliable: this is the value of “n” in (REL). It is a bit of an idealization to assume, as in (REL), that there is a precise value for “n”. Like many other notions, the notion of reliability is vague and admits of borderline cases. It is only for the sake of simplicity that I am making this idealization here; nothing substantial hinges on it. One should expect the value of “n” to lie below 1; it is hard to think of any case of perfect reliability (at least in the case of humans and similar mortals). The value of “n” should also lie above .5 (for yes-no questions)—otherwise the process could hardly count as a reliable one. Where between .5 and 1 it lies can vary from context to context; this constitutes one kind of context-sensitivity of “reliability” which is easily inherited by “knowledge” (see Chapter 1 on standards of reliability).9 However, I will not focus on this kind of context-sensitivity here. Could reliability also be determined by qualitative aspects like, e.g., the importance of a belief? Suppose A and B are organisms that perceive their environment and scan it for interesting (potentially beneficial or dangerous) objects. A gets most of his object identifications wrong but he gets almost all of the few important ones right; A is just very good at identifying beneficial or harmful objects. B gets most of his object identifications right but gets almost all of the few important ones wrong; B is not good at all at detecting threats and potentials. Should we say that A is still more reliable in some sense than B? I don’t think so. B is much more reliable than A in identifying objects in general while A is much more reliable than B in identifying interesting objects. What if there is a better alternative method m* such that P(t/m*) > P(t/m)? Should we require for a reliable method that there is no such alternative? No. Using a telescope can be a good way of figuring certain things out about other planets—even if 9   According to a frequentist interpretation of “probability,” “n” will be the value of a limit. There are some special issues connected with this but we do not need to go into this here.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

40  THE ARGUMENT FROM RELIABILITY much better telescopes become available later (or are already available at the time). A fortiori, the existence of equally reliable or less reliable alternative methods does not affect the reliability of m. What if there is a method m’ which is a component of m (such that m cannot be used without also using m’ while m’ can be used without using m) such that P(t/m’) > P(t/m)? Does that mean that m is not reliable? No, it just means that m is not as reliable as m’ and that the most efficient subject would use m’ without m. But it does not mean that m is not reliable. Looking around for my car keys might be a good way of finding them even if I do at the same time also think about the groceries and even if looking around without multitasking would be more reliable. A fortiori, the existence of equally reliable or less reliable component methods does not affect the reliability of m. What if the use of the method does not raise (or even lowers) the probability of acquiring a true belief? In the case in which it raises this probability (Raise-a) P(t/m) > P(t)10 or, alternatively (Raise-b) P(t/m) > P(t/not m).11 In the case in which the use of the method lowers the probability of acquiring a true belief (Lower-a) P(t/m) < P(t) or, alternatively (Lower-b) P(t/m)< P(t/not m). Finally, in the case in which the use of the method neither raises nor lowers the probability of acquiring a true belief (Equal-a) P(t /m) = P(t) or, alternatively (Equal-b) P(t /m) = P(t/not m). Suppose the use of m does not raise but also not lower the probability of t. Would that mean m is not reliable? I don’t think so. In this case, the subject could as well not have used m but as long as P(t/m) ≥ n there is no reason to deny that m is reliable. Even if the use of m lowers the probability—why deny that it is reliable if it is still true that   − with “P(x)” referring to the unconditional probability of x.   We do not need to go into the question here which of the two captures better the idea of raising the probability (the same holds for the related two pairs of principles below). 10 11

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  41 P(t/m) ≥ n? Suppose Jackie believes that she is alive. Sometimes she philosophizes about this and then usually but not always comes to the conclusion that she is alive. When she does not philosophize about this, some ordinary process almost always (and more often than philosophy) convinces her that she is alive. When she philosophizes, this ordinary process is not running and she uses some explicit reasoning instead. Even if the ordinary process is more reliable than philosophizing and even if philosophizing lowers the probability of acquiring a true belief (P(t/philosophizing) < P(t/ordinary processing)), philosophizing is still a reliable method for Jackie for figuring out whether she is alive.12 There is some reason to think that the right-hand side of (REL) is too strong: some processes are reliable but P(t/m) < n. How can that be? Consider deductive inferences (similar things hold for non-deductive inferences). Not every valid inference is sound; if the premises are false, then the conclusion might well be false, too. However, this does not make inference an unreliable method of belief acquisition, even if it should happen a lot that false premises lead to false conclusions and even if this happens so often that P(t/m) < n. The culprit here is not the method of inference but the “input” of beliefs on which the method or process depends. In general, methods or process where the truth value of the resulting belief also depends on the truth values of some input belief are called “belief-dependent” (see, e.g., Goldman 1992b, 117). In order to make sure that belief-dependent methods can qualify as reliable, too, no matter how often false input beliefs lead to false “output” beliefs, we should modify (REL) in the following way:13 (REL*) S’s belief that p is reliable iff (P(t/m) ≥ n and if m is belief-dependent, then P(t/m) ≥ n for true input beliefs).14 Do we have to require the input beliefs to be reliable, too? One might worry about the danger of a regress here. Apart from that and more importantly, it seems that our ordinary notion of reliability does not offer a clear answer here; we could stipulate it one way or the other. Fortunately, nothing substantial hinges on this question here and we can leave it open here. For the sake of simplicity, we can even go back to (REL); whatever will be said below also holds, mutatis mutandis, if we consider (REL*).15 12   According to probabilistic accounts of causality, a necessary (and perhaps also sufficient) condition of m causing t is that (Raise-a) or (Raise-b) holds (see, e.g., Hitchcock 2011). Whether I should claim a noncausal relation holding between m and t (especially if this probabilistic account of causality is true) or whether I am committed to a different account of causality can be left open here. 13   I assume that (REL*) holds necessarily. 14   One could wonder whether this modification is necessary since good inference from false beliefs also often lead to true beliefs so that the unmodified (REL) might also unproblematically apply to belief-­ dependent methods. I will leave this question open here. 15   One worry concerns the possibility of distinguishing in principle between belief-dependent and belief-independent methods. If one is a holist about belief, then all beliefs depend in some sense on a whole system of beliefs. If holism is understood as a semantic claim about the content of beliefs, then it does not show much here because it does not follow from it that all beliefs are also epistemically dependent on other beliefs; one would have to argue for this claim separately. But even if one manages to do so, there still seems

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

42  THE ARGUMENT FROM RELIABILITY I mentioned above that the notion of probability should not be understood as a modal one. The probabilistic notion of reliability in (REL) contains no modal notions either. However, some authors have proposed that reliability also has a counterfactual, modal aspect (see Goldman 1992a, 85, 98, 1986, 45–8, 106–7, 1988, 61–3; Becker 2007, 11–12, 32, 89–92 but cf. also McGinn 1984, 537–9; and Williams 1996, 324). According to these authors it is not sufficient for the reliability (in this, actual, world) of a belief or a method that the relevant method produces a good enough share of true beliefs in this, the actual, world but it is also necessary that it would do so in close possible (and non-actual) worlds. This thought would motivate something like the following modification of (REL):16 (REL-modal) S’s belief that p is reliable in the actual world iff P(t/m) ≥ n in the actual world as well as in (some or many)17 close possible worlds. There are serious problems with the idea of closeness and remoteness of possible worlds and I will go into them below (Section 2.4). Now, I rather want to raise the question why one should accept something like (REL-modal). Here is a thought. My perceptual belief that there is a cup in front of me now is based on a process, namely visual perception, that is reliable enough in this world. We can also assume that in a close possible world which differs from the actual world in not much more than a slightly different color of my hair the process of visual perception would be unaffected and give me the same share of true beliefs as in the actual world. What, however, if that were not so and if in such a close possible world visual perception would produce much fewer true beliefs? It is not easy to imagine how that could be but let that go. One could also suspect that a world where only my hair color and the tendency of visual perception to produce true beliefs differ would not really be a close world; put this aside, too.18 Should one then say that if such a possible world is really close to the actual world, then my actual belief is not reliable? I do not see why. Even if the actual world were the only world in which visual perception tends to produce a good amount of true beliefs (which is hard to imagine) and even if there were lots of close but non-actual possible worlds where visual perception exists but does not deliver a good amount of true beliefs—even then I would not see why we should deny that my actual belief is reliable in the actual world. Nothing to be a difference in principle between the kind of epistemic dependence which one has in an inference (between beliefs in premises and beliefs in conclusions) on the one hand and, on the other hand, the kind of epistemic dependence of, say, a perceptual belief on background beliefs. That is all we need for our distinction between belief-dependent and belief-independent processes and methods.   I assume that it would hold necessarily.   Whether one world would be sufficient or many if not all would be necessary can be left open here. 18   If one holds that all non-actual close possible worlds are such that P(t/m) is the same (or at least roughly the same) in both the actual and the relevant non-actual possible worlds—that is, if one lets closeness of worlds be (partly) determined by “sameness of reliability”—then it would be trivially true that one cannot have a reliable belief in the actual world without the process behind it also being good at producing true beliefs in close worlds. 16 17

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  43 changes if we make the even wilder assumption that visual perception only exists in the actual world. Perhaps we should expect that in close possible worlds our methods work more or less in the same way in which they work in the actual world. But that does not mean that if that weren’t so, then our actual methods wouldn’t be reliable. Sure, this last claim is not plausible if one starts with a modal notion of reliability but I haven’t done that; one would still have to show why the probabilistic notion of reliability in (REL) needs some modal additions. Not that such an addition—like in (REL-modal)—would change things drastically or make any drastic difference to the main point of (2) in (RA) and to what follows below; one could add this modal element to (REL) but I don’t see why that should be necessary or helpful (for further issues with such accounts, though, see Section 2.4).

2.3  Reliability and Context So much for a sketch of a probabilistic notion of reliability. I will now argue that given such a notion, contextualism about “reliability” and “knowledge” becomes very plausible. Let us start with the two typings in (REL): its right-hand side is about types of methods or processes and about types of topics. Let us consider the latter first.

2.3.1  Typing Topics Consider the visual perception example again: I am looking around and come to see that my mug is on the table in front of me now. Let us assume for the sake of the problem that my method consists in looking around.19 Is it—is my belief that my mug is on the table in front of me now—reliable? According to (REL), this depends on whether it tends to lead to true beliefs (more precisely: whether the probability of acquiring true beliefs, given the use of the method, is high enough). So, does looking around tend to produce true beliefs? This question is underspecified: true beliefs about what? One needs to specify the kind of topic. Suppose looking around is good20 for figuring out whether a red coffee mug is in front of me now, very good for figuring out whether some kind of object is in front of me now, not great for figuring out whether my favorite red coffee mug with the piece missing in the handle is in front of me now, and pretty useless for figuring out whether there is a clay mug in front of me now. To which type does my belief that my favorite red coffee mug with the piece missing in the handle is in front of me now belong? Does it belong to (i) the singleton class of my belief at a certain time t that I could express at t (and at t only) by uttering the sentence “My favorite red coffee mug with the piece missing in the handle is in front of me now,” or to   We can also assume, for the sake of the problem, that circumstances are normal.   For me though perhaps not for others.

19

20

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

44  THE ARGUMENT FROM RELIABILITY (ii) the class of all my beliefs at different times that I could express at the time by uttering the sentence “My favorite red coffee mug with the piece missing in the handle is in front of me now,” or to (iii) the class of all my beliefs about the presence of a red mug at the time of the belief(-tokening), or to (iv) the class of all beliefs by anyone about the presence of some midsize dry object in front of them, or to (v) some other class? Depending on the answer we get different results concerning the reliability of the method used. The trouble is that there are too many answers and nothing like the single right one. One might perhaps think that the right way of typing in the last case is the one in (i) but that would go against the principle that we need to determine types of topics with more than just one instance. But, one might reply, isn’t the relevant type of the topic determined by the content of the target belief, along the lines of (ii) above? But if we are happy to determine the type of topic via the class of all beliefs that share the same Kaplanian character (see Kaplan 1989; see also Section 7.1) with the target belief as in (ii), then we’ve already opened the group up to beliefs with other contents than the target belief. Why then not broaden the type further and let, say, all beliefs about mugs in?21 There is no reason not to do that. But does, one might object, the subject not have ideas about what the topic is? Not necessarily. But even if the subject has ideas about the topic—why should it be any less problematic for some subject to determine what the relevant type of topic is? Does the subject have some special “authority” here? If yes, one would want to know what that authority is based on. It is hard to see how it can be some kind of first person authority but if isn’t that then we’re back with our problem. One might also point out that the subject typically has a question on her mind and uses a method to answer that question. Isn’t the relevant topic determined by the subject’s question? Not all cases of belief acquisition are cases where the subject is trying to settle a question. But even if that were the case, this would not help us here. What is the subject’s question exactly? If it is the yes-no question that is answered by the content of the target belief (e.g., “Is my favorite red coffee mug in front of me?”), then we’re not typing questions and thus not typing topics. As soon, however, as we’re looking for the relevant type of question to which the subject’s question belongs, we are back with our original problem.

21   Or all kinds of beliefs whatsoever, no matter whether they are perceptual or not. A mathematician might work better on his problems if he looks around while he is thinking about his problems. Looking around would be part of the method the mathematician uses. One could perhaps block this  last extension by insisting that the type of belief has to be the typical outcome of the method used. We can leave things open here because the indeterminacy of topics presented is already significant enough.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  45 The upshot of all this is that there is a genuine and irreducible indeterminacy of the type of topic. There is what one could call a “generality problem” for the individuation of topics, parallel to the well-known and much discussed generality problem for the individuation of methods (to which we will turn below). In contrast to the latter, the generality problem for topics has not been discussed much, if at all.22 What is important about it here is that (REL) types topics. Given that there is a plurality of equally legitimate ways of typing the topic, and given that the value of P(t/m) for a given m varies with types of topic in such a way that the same method can, depending on the type of topic, get a “reliability value” above or below n, it follows that it is also indeterminate whether a given method is reliable and to what degree. It is not hard to see how this can make “knowledge” indeterminate. If the same target belief can be counted both as “reliable” and as “not reliable” (depending on the topic typed), and if nothing apart from the lack of reliability could stand in the way to this belief constituting knowledge, then whether it constitutes knowledge is also indeterminate and varies with the topic typed. All this might sound very alarming. However, one can and should make a virtue of necessity. Biting the bullet is not always that hard and sometimes there isn’t really that hard a bullet to bite. Contextualism is the way out of the problem of the indeterminacies of topic, reliability, and knowledge.23 It is not hard at all for people to come up with ideas about the relevant topic, reliability, and knowledge when thinking about the epistemic state of some subject. The contextualist take on the problem that I want to propose here is that a speaker who is in the process of attributing epistemic states to a subject typically24 has a type of topic on his mind which then also guides his judgment about whether the subject has a reliable belief or knows the target proposition. In other words, the type of topic used by the attributor is a contextual parameter; I will talk about determinants of this and other parameters like it below (for the distinction between parameters and determinants, see Chapter  1). The contextually variable ­topics parameter drives the attributor’s knowledge attributions; it helps explain the context-sensitivity of “knowledge” and related terms. What is the evidence for this claim? Why should one believe that there are contextual parameters like this one, different from the standards parameters? Here the evidence does not lie in cases and “what we would say when”; or at least, the connection with ordinary usage of the relevant terms is not as direct or close as in the case of the standards parameters. The evidence is more theoretical and more based on conceptual considerations. Given certain principles about knowledge, reliability, and their relation to each other (and including the assumption that there is no further condition of knowledge that would “screen off ” the context-sensitivity of reliability), contextualism 22   But see the related brief remark on relativization to types of belief in Feldman 1985a, 164 and also: McEvoy 2005, 21–5; Kappel 2006; Comesaña 2006, sec.4; Lepock 2009, 276. 23   There might be other ways out but I, at least, cannot think of convincing or plausible alternatives. 24   Often but not always; if not, then the speaker or thinker fails to make an attribution.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

46  THE ARGUMENT FROM RELIABILITY looks like a very good theoretical principle and explanation which helps to create a satisfying reflective equilibrium.25 Lacking better alternatives, we should embrace it. But isn’t this contextualist solution simply too hard to believe? Even putting antecedent anti-contextualists views aside, one might still get the impression that this kind of contextualism gives us a massively overintellectualized picture of what ordinary subjects are doing. Can we really believe that knowledge attributors have topics, types thereof, probabilities, and more on their mind when they make ordinary knowledge attributions (see, e.g., Conee and Feldman 1998, 22–4, and Kappel 2006, 544–5)? In order to respond to this worry one needs to remind oneself that there are different levels of cognitive processing; only some of them are accessible to the subject and the subject is aware only of some of them. A reader of, say, David Marr’s classic “Vision” (1982) might wonder how it could possibly be true that he himself is doing “all those complicated things” when, for instance, looking at a banana. The puzzle disappears as soon as one accepts that a lot of processing is implicit, not explicit. Knowledge attributions are not different here. To reject the idea that knowledge attributions could be that complex is a bit like rejecting the idea that perceiving a banana as a colored, spatial object involves “so much computation.” It also reminds one of the predicament of Molière’s Mr. Jourdain who was surprised to learn that he had been speaking prose all his life.

2.3.2  Typing Methods So much about one kind of typing involved in (REL). The other kind concerns the typing of methods. What is the issue? Consider Anna who is looking at the sky, spots an airplane and comes to believe truly that it is an Air France plane. Suppose that she has her glasses on and has also taken the extraordinary FarSight eyedrops; visibility and other circumstances are normal. Given all this, Anna had a very good chance at arriving at a true belief about the airline to which the plane belongs. Is her belief reliable? This depends on the method she used. So, what is the method used by Anna? If it consists in looking at the sky under normal conditions with glasses on and after having taken FarSight eyedrops, then the method used is reliable, and so is Anna’s belief. However, if the method consists in looking at the sky, then (as we can assume in the example) the chances of arriving at a true belief about the airline by using the method are pretty low, even if we add that circumstances are normal; given this method, Anna’s belief is not reliable because the method isn’t. Between these two methods might lie another method: looking at the sky with one’s glasses on. If knowledge requires reliability and if reliability depends on the method used by the subject, then it seems that we should be able to identify the unique method used by the subject in the case at hand. Several authors have argued—and I agree—that this constitutes a problem and that there simply is no unique method to be determined. If   For a discussion of alternative theoretical explanations see Chapter 8.

25

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  47 that is so, then it is easy to show that an indeterminacy of reliability follows from such an indeterminacy of method; the consequences for the concept of knowledge are not hard to grasp either (see the parallel point about topics above). This problem is known as the “generality problem.”26 Some think that it is a problem for reliabilism in particular but if the remarks above are correct then it is a problem for everyone (see on this Bishop 2010 and Conee 2013 as well as Alston 2005, 114–15, 132–5): the problem appears already with the minimal reliability requirement that everyone should accept.27 It seems relatively uncontroversial that one should not individuate methods extremely narrowly: in our example, one should not pick the one unique process-token that happened when Anna formed her belief. We have already seen above that we need to determine types of beliefs. Another problem with this proposal consists in the fact that a method only used once (and necessarily so) would either lead to a true or to a false belief (leaving aside third options here). This means that the method would have a value of either 1 or 0 for P(t/m); it would be either completely reliable or completely unreliable. But this seems clearly unacceptable because lots of methods (if not all) have a degree of reliability > 0 and < 1—something that would be excluded by the extremely narrow individuation of methods. What remains are broader individuations of methods. But how broad should they be? Often people remark that the individuation should neither be extremely narrow nor extremely broad but lie somewhere between the extremes. That it shouldn’t be extremely narrow is very plausible but I don’t see what speaks against even the broadest individuation as a “method of belief acquisition.” There seems to be no non-arbitrary way of picking one description of the method over all others as long as they are characterized with some degree of generality. The problem can also be seen if one puts it in terms of the distinction between method and circumstances of the application of the method. Given a characterization of a method, everything that is not part of that description can be subsumed under “circumstances” of the application of the method. For instance, if we say that Anna’s method consists in looking at the sky, then normal visibility as well as having one’s glasses on and having taken FarSight eyedrops count as circumstances of the application 26   See for instance: Goldman  1992b, 115–16, 1986, 49–51; Pollock  1984, 108–10; Foley  1985, 196–7; Feldman 1985a, esp. 160–2; Alston 1995 (and 2005, ch.6, esp. 125–43); McEvoy 2005 (who is using ideas of Alston); Conee and Feldman  1998; Wallis  1994; Wunderlich  2003; and Comesaña  2006. See also Brokes 2001 who argues that externalists need not and should not see this as a problem. For another argument to the effect that there isn’t a serious problem in the first place, see Adler and Levin 2002 (responding to Conee and Feldman on Alston; see also Feldman and Conee 2002 for a response back). Lepock 2009 emphasizes the positive aspect of having a plurality (narrower or broader ones) of ways of evaluation at hand. Kappel 2006, sec.6–7 defends a no-determination view and offers this as a solution to the generality problem but one does wonder whether it rather reinforces the problem. Schmitt 1992, ch.6 offers “pragmatic constraints” on the selection of methods but does not claim to have a more “principled” solution to the problem (see also Schmitt 1984, 3–5; and Feldman 1985b in reply). 27   Comesaña 2006, 45–6, fn.14 argues that every adequate epistemological theory needs to appeal to the basing relation—which is, according to him, crucial to any solution of the generality problem.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

48  THE ARGUMENT FROM RELIABILITY of the method. If, in contrast, we take Anna’s method to consist in looking at the sky under normal conditions, with her glasses on and after having taken FarSight eye drops, then some of the factors have changed status from parts of the circumstances to parts of the method. Where should one draw the unique line between method and circumstances? It seems there is no such unique line. The arbitrariness of any specific way of drawing the line expresses the same underlying fact as the indeterminacy of the method. A similar indeterminacy also characterizes the notion of “normal” circumstances of the use of a method. How should one distinguish between a method, normal circumstances of its application, and non-normal circumstances? A typical new car will drive well on streets, not so well in the woods, and not at all under water. Should one say that being underwater with one’s car is a non-normal circumstance of the use of the car while being on an ordinary street with it is a normal circumstance of its use? Or should we rather say that being on an ordinary street is part of what it means to use the car? The relation between trains and tracks makes this point even more vivid: Is being on tracks part of the normal circumstances of the use of the train or rather part of the use of the train? Similar for epistemic methods: Is a certain degree of visibility part of the normal circumstances of the use of one’s perceptual faculties or rather part of the use of those faculties? It seems there is no unique non-arbitrary way to draw a line between the use of a method and normal circumstances of its application. The notion of non-normal circumstances also is more suspect than it might seem prima facie. Are we still using the Renault as a car when we’re underwater with it? Am I still using my visual senses when it’s pitch dark and I cannot see anything? Why should one see these “special conditions” as circumstances, even non-normal ones, of the use of the relevant method? On the other hand, is there a compelling reason not to do so? It seems that whether and if yes where we draw lines between the method, normal, and non-normal circumstances of application is not determined by any relevant fact of the matter.28 There is one proposal of a solution of the generality problem that has more support in the literature than any other proposal (see, e.g., Goldman 1986, 49–50, and more recently Becker 2007, 104–9, 2008). The basic idea is that the relevant process is the narrowest type of process which is causally operative and causally sufficient for the production of the target belief. The narrowest type is the one that is most specific and has the fewest instances. This proposal reflects the widespread tendency to think that the relevant process or method has to be individuated narrowly rather than broadly. It is clear that this alone won’t suffice: one cannot go for the narrowest individuation and also not for the second narrowest: that would seem ad hoc and apart from that there might not even be such a thing as the second narrowest individuation of the method (the scale of more or less narrow individuations might be such that for any two individuations of different narrowness there is one with an intermediate degree   For this reason the notion of normal circumstances plays no role in (REL).

28

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  49 of narrowness). Hence the addition of the condition of causal sufficiency and causal operativeness. There is reason to doubt that this proposal will solve the problem. Let us take a closer look at it. Consider Anna who has her glasses on and is looking at Fido, the dog, while whistling an aria. She acquires the belief that Fido is in front of her. What is the relevant method? Is it (i) looking at Fido, (ii) looking at Fido with her glasses on, or (iii) looking at Fido with her glasses on while whistling an aria? Let us assume that all of these are causally sufficient for the acquisition of the target belief (and we can assume that none of them is causally necessary: Anna might also just use her ears to figure out that Fido is in front of her). Let us also assume that whistling an aria is irrelevant here insofar as Anna would acquire the target belief no matter whether she whistles an aria or not. Let us further assume that Anna would acquire the target belief no matter whether she wears her glasses or not. However, while whistling an aria has no effect at all on the process of belief acquisition, wearing glasses makes it easier for Anna to acquire her target belief. Should we say that all of (i)–(iii) are also causally operative? We may use a probabilistic notion of causal operativeness here29 according to which (CO) A factor C30 is causally operative for some phenomenon E just in case (a) P(E/C) > P(E) (or, alternatively: P(E/C) > P(E/not C), and (b) there is no factor C* less specific than C31 such that P(E/C*) = P(E/C) or P(E/C*) > P(E/C). There is also relative causal unproductiveness (where (a) but not (b) of (CO) holds— the case (iii) above), causal indifference (where P(E/C) = P(E) or P(E/C) = P(E/not C)) and causal disturbance (where P(E/C) < P(E) or P(E/C) < P(E/not C)). Given all that, we should then say that though all three of (i)–(iii) are causally sufficient, only (i) and (ii) are also causally operative. According to the proposal under discussion we should then finally apply the narrowness criterion and decide (assuming that we’re not missing any potentially relevant further description of Anna’s method) that since (ii) is narrower than (i), looking at Fido with glasses on is the relevant method that produced Anna’s belief that Fido is in front of her. Looking at Fido isn’t it. The basic idea here is that while the condition of causal operativeness pulls us away from being too specific in the individuation of the relevant method, the condition that it should be the narrowest such individuation 29   Only in order to clarify and illustrate a point, not in order to commit ourselves to some view on causality. 30   One that is instantiated. 31   In the sense that C’s presence entails C*’s presence but not vice versa. I am leaving aside the mere possibility that two factors might be equally specific; nothing depends on this here.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

50  THE ARGUMENT FROM RELIABILITY pulls us away from being too unspecific in the individuation of the relevant method; the result of this is, hopefully, just the right, unique equilibrium. However, it is not clear at all that there is such a difference between (i) and (ii). One could argue for each of them. Why not argue that all Anna is doing is just to look at Fido, with or without glasses? Whether she is wearing the glasses does not make any crucial difference as to whether she acquires the target belief. On the other hand, one could argue that when the astronomer looks at Venus with his sophisticated instruments, then looking at Venus with these sophisticated instruments is the relevant method, not just looking at Venus (which is also accessible to the non-astronomer). In other words and more generally: both principles of individuation of the relevant method are equally plausible. One can go with the narrowest causally sufficient and operative method as well as with the broadest causally sufficient and operative method. Hence, the indeterminacy of methods remains. No argument for preferring narrower over broader methods is in sight. Furthermore, does the criterion of the narrowest (or broadest) causally sufficient and operative method help distinguish between the use of a method and the circumstances of its application? According to the “broadness” version, wearing glasses is part of the circumstances of using the method; according to the “narrowness version,” it is part of the use of the method itself. If there are reasonable doubts that one of these two views has an advantage over the other, then there are also reasonable doubts about any of these views constituting the right way of individuating methods and distinguishing them from circumstances of application. Finally, we could have explained “causal operativeness” (being a partly technical term) in a slightly different, less restrictive way and allow for the possibility that the less specific method is causally more efficient than the more specific method: (CO*) A factor C is causally operative for some phenomenon E just in case (a) P(E/C) > P(E) (or, alternatively: P(E/C) > P(E/not C)), and (b) there is no factor C* less specific than C such that P(E/C*) = P(E/C). What should we then say about a case where the broader method has a higher probability of producing the target belief than the narrower method? It is not clear at all whether this should speak in favor of the broader method being the relevant one. Again, it is very hard to see how this question could be decided in principle. This adds another problem to the proposal discussed here.32 Sometimes people propose to determine the relevant method by identifying it with the one the subject identifies as relevant (see Hudson 2004) or with the one the subject intends to use (Leplin 2007a, 35–6; see also Christensen 2007, 45–7 and Leplin 2007b). One problem with this proposal is that often the subject lacks relevant views or intentions concerning the method used. But perhaps this problem can be overcome using the idea of subconscious attitudes. More importantly, one should 32   I am putting to one side the case where there is no factor C* less specific than C such that P(E/C*) > P(E/C).

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  51 allow room for the difference between using a method and thinking that one is using it. Also, intending to use a certain method does not entail that one will use it. And why should we go with the subject’s identification in the first place? Even if different people tend to identify methods in the same way (see Jönsson 2013) why should be believe they are right? Like in the case of the indeterminacy of types of topics we are facing an indeterminacy here that looks very bad, perhaps even worse. More problems will be added in the next subsection. If there is an irreducible and genuine indeterminacy of methods and if different equally legitimate pickings of a method lead to very different values of P(t/m), with some above n and some below n, then it will in many cases also be indeterminate whether a given belief is reliable. Again, indeterminacy of reliability will often lead to indeterminacy of knowledge. Like in the case of typing topics, however, there is a very attractive theoretical way out of the alleged theoretical mess: contextualism (see Heller 1995 who pioneered this way out; also see Conee and Feldman 1998, 20–4 on Heller; see further Kappel 2006, especially 544–5; Greco 2010, 76–80; and already the brief remark in Cohen 1988, 115). I cannot think of anything better (and it is, I think, good enough) than to assume that attributors typically type methods one way rather than another and that this guides their judgments about whether a given subject has a reliable belief in some proposition and knows the proposition. Like the type of topic, the type of method used by the attributor is a contextually variable parameter which determines the truth conditions of knowledge attributions. This is yet another aspect of the context-sensitivity of “knowledge.” Concerning the evidence for this claim and its believability I just refer back to the parallel considerations concerning the typing of topics (see also Conee and Feldman 1998, 22–4 for a worry here).

2.3.3  Extension and Generalization: Reference Classes There are thus two independent indeterminacies and two context-sensitive parameters for reliability so far: one concerning the typing of topics, another one concerning the typing of methods. Even with a fixed method, the typing of topics can still vary contextually. And vice versa: even with a fixed topic, the typing of methods can still vary contextually. The two parameters can combine in many different ways across different contexts. The generality problem for methods is already well known; it has been extended here to a generality problem for topics. All this can now be further extended and, in a sense, generalized. The above two problems can be seen as special cases of the so-called reference class problem (see, e.g., Reichenbach 1949, 372–8; Salmon 1966, 90–2; Salmon  1970, 40–51; Hempel  1965, 53–79, 397–403; Kyburg 1977; Levi 1977; Fetzer 1977; Hájek 2003 and 2007; Gillies 2000, 119–25; for very early remarks on the problem see Venn 1876, 194–6 (part II, ch.VIII, sec.11–14)). What’s that? Consider Alastair who has just read a trustworthy report about the life expectancy of Scottish bus drivers. Alastair is a bus driver in Glasgow. The report says that Scottish

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

52  THE ARGUMENT FROM RELIABILITY bus drivers of Alastair’s age group have a life expectancy of 78 years. Should Alastair expect to live 78 years (and, for instance, adjust his retirement plans on this basis)? Alastair also smokes two packets of cigarettes per day and the health bureau tells us that a smoker like Alastair has a life expectancy of 72 years. What should Alastair make of this? Should he just go with the life expectancy for bus drivers and neglect the other one, or vice versa? Or should he combine both and expect to live 75 years (computing the arithmetic mean)? Or should he compute his life expectancy in a different way (e.g., via the geometric mean)? What if life expectancy in Glasgow is lower than in other parts of Scotland? Let us also assume that Alastair hasn’t always lived in Glasgow. Let us add that he also plays golf quite regularly and Golfers World Wide have found out that golfers in Alastair’s age group can expect to live for 83 years. We should also mention that Alastair is slightly overweight but his blood pressure is low and his heart is quite healthy. How should one then compute Alastair’s life expectancy?33 Alastair belongs to all these different reference classes and the life expectancies for the different classes differ significantly. Which one is the relevant reference class for Alastair when it comes to his life expectancy? Is it any one particular class? Or the intersection of all these classes? More generally: What is the probability that an individual a has property P, given that a belongs to different groups of Fs, Gs, Hs, and so on, and that the conditional probabilities for P, given membership in these groups, differ? The problem of finding an answer to this question is a case of the reference class problem. One might want to toy with the idea that the relevant reference class for Alastair is the class with himself as the only member. But what exactly could that mean? And does this tell us anything about Alastair’s life expectancy? It seems it rather presupposes that Alastair’s life expectancy is already given. It seems to boil down to an instance of the general tautology that the probability that a has property P is determined by the probability that a has P. One would seem to do better to say that Alastair’s relevant reference class is the class of all people who have exactly the properties Alastair has. But this class will also have only one member, namely Alastair. However, now we have a clearer idea at hand: the relevant reference class for a given individual a is the class of all individuals that have all and only the properties of a. However, it seems that some properties are “irrelevant” to Alastair’s life expectancy: for instance, the property that the sum of the numbers on his current car’s license plate is prime. So, we should modify the idea under discussion one more time: the relevant reference class for Alastair’s life expectancy is the class of all those who have exactly those properties of Alastair that are probabilistically relevant to life expectancy. The notion of probabilistic relevance can be explained in a way reminding us of its special case (CO) above: 33   Insurance companies have a pragmatic way of dealing with this problem, and they focus on samples of cases instead of individual cases; they don’t need to worry too much about the philosophical problem. See also Brokes 2001, 149, 152 on this point.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  53 (SR) A condition C34 is statistically relevant to some phenomenon E just in case (a) P(E/C) ≠ P(E)35 and (b) there is no condition C* less specific than C such that P(E/C*) = P(E/C). If E is Alastair’s property that he will get to age 80 (with “C1 and C2 . . . and Cn” referring to the conjunction of all the properties statistically relevant to E), then the probability of Alastair getting to the age of 80 is given by P(E/C1 and C2 . . . and Cn). Call an “admissible reference class” a reference class that is determined exclusively by statistically relevant properties. According to all the above, the relevant reference class is the narrowest admissible reference class (assuming there is one), that is, the reference class determined by the narrowest36 bundle of statistically relevant properties (excluding complex properties that add statistically irrelevant properties to statistically relevant properties: see (SR-b)). This is the most favored proposal to solve the reference class problem.37 Does it? First of all, one might wonder whether there really are, strictly speaking, any statistically irrelevant properties. Don’t probabilities like P(E/C), P(E), and P(E/C*) always differ, even if only very little? Could such probabilities ever be identical and not just roughly identical (that is, lie within the same very narrow interval)? It seems quite plausible to say that the notion of relevance is graded and that all that could matter here is the degree of relevance or irrelevance but not absolute relevance or irrelevance.38 But then our notion of statistical relevance never applies and we are back with the extreme case of reference classes with exactly one member. An alternative would be to introduce a threshold of relevance in addition (in whatever way the degree of relevance will be measured). A problem with that would be that this threshold would have to be quite arbitrary and ad hoc—except if one wants to turn this into another contextual parameter.39   One that is instantiated.   An alternative would say that P(E/C) ≠ P(E/not C). Nothing depends on such details and we can safely ignore them here. 36   The narrowest bundle is the “referentially” narrowest one: it bundles the most properties. 37   See Reichenbach 1949, 372–8, esp. 374 and Hempel 1965, 53–79, 397–403. Following Salmon 1966, 90–2 (see also Salmon 1970, 40–7), many people call this the “broadest statistically homogeneous reference class.” See also with reference to the generality problem: Beebe 2004, 181 as well as the critical reply to Beebe by Dutant and Olsson 2013. The latter show that the application of Reichenbach’s and Salmon’s ideas to the generality problem leads to trivialization: every neither perfectly reliable nor perfectly unreliable process has statistically relevant subtypes, namely those processes that only lead to true (false) beliefs; hence, reliability collapses into perfect reliability—Goldman’s take on the generality problem (see above) is very close to the above idea. 38   The degree of relevance could be measured by the difference between the values of P(E/C), P(E/C*), and P(E). One can imagine there to be different alternative measures but these details need not concern us here. 39   See on this also Dutant and Olsson 2013, sec.5. There certainly are pragmatic thresholds determined by the pragmatic interests of those who do the inquiry but this won’t help here. Another possibility would be to give up the idea of precise probabilities and work with indeterminate probabilities (see, e.g., Hájek and Smithson 2012). This raises a lot of further questions, for instance, questions about the identity conditions of indeterminate probabilities (is it indeterminate whether P(X) and P(Y) are indeterminate in exactly the same way?). 34 35

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

54  THE ARGUMENT FROM RELIABILITY But even if one can somehow solve the problem of the distinction between relevance and irrelevance there would be other questions. What, for instance, if there are infinitely many individuals in the reference class or if there are infinitely many statistically relevant properties determining the reference class? Assuming that this is a mere technicality there still remains the problem that narrowest admissible reference classes will at least often have exactly one member. Very often we simply do not have enough statistical information concerning the single case. Apart from that, the notion of single case probability is not philosophically unproblematic (see Gillies 2000 and Baumann 2005a). And what is the point of looking for a reference class for an individual if one only gets back to exclusive consideration of that individual? However, the most serious problem is another one: Why should the narrowest amongst all admissible reference classes be the relevant one? Why is “narrow better than broad”? It is hard to find an argument here—despite or perhaps because of the prima facie plausibility of the tendency towards narrowness. But the question remains: Why not, for instance, a class of intermediate generality? Does the preference for the narrowest class express an implicit commitment to the idea that ideally one would deal with a single case? To be sure, the narrower a reference class the more specific it is to the individual under consideration; however, it is also true that the broader a reference class is, the more robust on the whole will the statistical data be that we get; apart from that the need for a reference class is the need to get away from the individual case. Consider Alastair again. He wants to know what his personal life expectancy is and in order to figure that out he needs to consider some broader class; however, the broader the class is the less it seems to have to do with him. The quest for the relevant reference class seems to be plagued by a combination of two incompatible pulls going into opposite directions: a pull towards broadness and a pull towards narrowness. Some kind of compromise is called for but it is quite unclear what a non-arbitrary one could look like. It is therefore no surprise that there is widespread pessimism concerning a “non-skeptical solution” (see Hume 1975, sec.V for this notion in a different context) of the reference class problem, that is, a solution that is based on non-arbitrary criteria of relevance. The reference class problem also arises for epistemic cases, especially for (REL). Even if we fix the type of topic and the type of methods, the reference class problem remains (see also Wallis 1994 for the distinction between what he calls the “characterization problem” (the generality problem) and the “relevance class problem” (the reference class problem); also see Bedke 2010, sec.1.1). Consider the following case. Kurt is in court, as a witness. Is asking Kurt about the time the defendant left the bar a reliable method of finding out the truth about that? Suppose Kurt is being asked between 4:00 p.m. and 4:05 p.m. on a particular day of the trial. He gives the correct answer and everyone in court believes him. However, as it happens Kurt missed lunch that day because he had to wait to be called into the witness stand. Therefore, he was low on blood sugar when he was questioned. Under such circumstances, Kurt’s tendency to answer questions like the one asked in court correctly is not very high. However,

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  55 during the three weeks of the trial Kurt was also questioned on two other occasions; apart from these five minutes he was always in excellent shape during these three weeks and much more trustworthy than almost anybody else. The trial takes place in the summer and Kurt is in general a very sober guy; however, during Winter and Fall he drinks and during those darker seasons his tendency to give correct answers to questions like the one asked in court approaches randomness. If “m” stands for “asking Kurt” and “t” refers to topics like the one he is being asked about, then P(t/m) < n during 4:00 p.m. and 4:05 p.m. on the particular day, P(t/m) > n during the whole three weeks of the trial, and P(t/m) < n during the whole year. This means that the court’s method of asking Kurt about the time the defendant left the bar is not reliable during 4:00 p.m. and 4:05 p.m. on the particular day, reliable during the whole three weeks of the trial, and not reliable during the whole year. If we want to know whether asking Kurt about the question is a reliable method, then we need to know what the relevant time is—what the relevant temporal reference class is. So, what is it?40 One could choose a very narrow time interval (reference class): let’s say the few seconds during which Kurt was asked the question, thought about it, and gave his answer. But why? Sure, that is the time when Kurt was questioned and gave his answer. But that does not imply that that is the relevant time for judgments about reliability. Sure, the court might specify that they are interested in Kurt’s reliability during the time of the question, say, between 4:00 p.m. and 4:05 p.m. A responsible judge might even ask Kurt “Are you ready? We know that you had no lunch and we can ask you later if you don’t feel fine.” However, this is no objection but rather a confirmation of the point that we need to specify a time in order to be able to make a judgment about the reliability of questioning Kurt. If we leave out the time frame, then the reliability of the method is indeterminate. You ask me whether my laptop is reliable. I answer “yes” even though it just froze. I did not forget that it just froze but rather took your question as one asked about reliability in the longer run; that does not mean that I think it’s reliable while it freezes. If one takes the question about some method’s reliability in its generality and without antecedent temporal specification, then one has to accept a genuine and irreducible indeterminacy of temporal reference classes. Sure, there is a strong temptation to go, in the abstract, for the narrowest temporal reference class.41 But like in the general case 40   For a similar problem concerning the generality problem see Weatherson 2012a. See also Baumann 2000, 2001. 41   The narrowest “meaningful” or “admissible” temporal reference class: a split second is too short in the Kurt-Court case. We don’t need to go into what determines the constraints on “meaningful” or “admissible” temporal reference classes here. The problem discussed here does not depend on this technicality.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

56  THE ARGUMENT FROM RELIABILITY discussed above, there seems to be an argument missing and there is a problem with opting for single cases. Like in the non-epistemic case of the laptop, in epistemic cases we often do choose a broader temporal reference class when judging reliability—and there often seems nothing wrong with that. There thus seems to be no way around accepting the indeterminacy of temporal reference classes. There are spatial cases, too. Consider one version of the well-known barn cases (see Goldman 1992a, 86 and Brandom 1998, sec.V as well as Brandom 1994, 210–12). Mary is standing in front of a barn, looks at it, and comes to believe truly that there is a barn in front of her. However, unbeknownst to her, the situation is the following: all the other barn-like structures in this part of the village are fake barns; all the other barns in the village are genuine barns; all the other villages in the county are full of fake barns whereas all the other counties in the state only have real barns. Suppose Mary’s method (m) is looking at objects around oneself and the question or topic (t) to be settled is whether there are barns close to oneself. Is Mary’s method reliable? The trouble is that P(t/m) > n in front of Mary’s barn, P(t/m) < n in this “special” part of the village, P(t/m) > n in the (rest of the) village, P(t/m) < n in the (rest of the) county, and P(t/m) > n in the (rest of the) state. Hence, Mary’s method is reliable in front of Mary’s barn, not reliable in this “special” part of the village, reliable in the (rest of the) village, not reliable in the (rest of the) county, and reliable in the (rest of the) state.

Like in the temporal case, we need to know what the relevant reference class is, here a spatial reference class, in order to figure out whether Mary’s method is reliable. Going with the narrowest spatial reference class42—a close area around Mary and the barn—won’t help us, for reasons parallel to the ones in the temporal case above. Sure, Mary was in this exact location, not in some other part of the village, county, state, or universe; she was in front of that particular real barn and not in front of one of the many fake barns in the vicinity. But that does not mean that the narrowest spatial reference class is the relevant one when it comes to judgments about the reliability of Mary’s method. Sure, if one asks questions about the reliability of Mary’s methods “in that location,” then that is the relevant spatial reference class—but only because this spatial specification has been added to the description of the method. Without antecedent 42   The narrowest “meaningful” or “admissible” spatial reference class: a tenth of a square inch is too short in the barn case. We don’t need to go into what determines the constraints on “meaningful” or “admissible” spatial reference classes here. The problem discussed here does not depend on this technicality.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  57 spatial specifications, there is a genuine and irreducible indeterminacy of the spatial reference class.43 Is there also a similar problem about “social” reference classes? Assume that Kurt is very good at answering questions even when his blood sugar level is low, much better than most other people. Does this matter? Is the method asking Kurt or asking an adult? Assume that there are gender differences (for whatever reason) in the ability of detecting fakes and that women are very good at it, in contrast to men. Assume further that Mary is exceptionally bad at detecting fakes. What then is the method—Mary looking around, a woman looking around, or a human being looking around? Perhaps there is an indeterminacy of social reference classes in addition to the indeterminacy of spatial and temporal reference classes? I want to leave this question open here. The main aim here was to show that there is a problem of the relevant reference class in the case of the reliability of epistemic methods and processes of belief acquisition. I argued that there is a genuine and irreducible indeterminacy of spatial and temporal reference classes. This does not mean that there is universal indeterminacy in all cases. What is essential to the cases considered above is a certain heterogeneity in the distribution of a certain probability: as one extends the time interval considered in the court case, the value of P(t/m) goes up and down considerably; as one extends the spatial area considered in the barn case, the value of P(t/m) goes up and down considerably. I am not claiming that this kind of heterogen­ eity is pervasive and universal but it is enough to say that it happens quite often—as the cases above and their realistic character suggest.44 What should we make of the indeterminacy of epistemic reference classes? If there is a plurality of equally appropriate pickings of reference classes and if the value of P(t/m) varies with reference classes, then we also get varying results for the reliability or unreliability of the method used. Since knowledge requires reliability, we can also expect that in many cases the truth conditions of knowledge attributions will depend on and vary with the reference classes picked. Again, this might look alarming but, again, there is a good contextualist way out of this problem.45 Attributors typically pick reference classes which guide them in their judgments about reliability and knowledge. One need not take this in an overly “intellectualist” way; as pointed out above in the case of the typing of topics and of methods a lot of cognitive processing is not conscious. So, reference classes are contextual parameters which also explain the context-sensitivity of “knowledge.”46

  All this should also raise doubts about the use of fake barn cases as Gettier-cases.   One might want to replace the example of fake barns by examples about more realistic kinds of fakes, though. 45   Again, it won’t help to appeal to the beliefs or intentions of the subject or the attributor. The reason why not is, mutatis mutandis, like the reason why appeal to beliefs and intentions does not help with the indeterminacy of topics or methods. 46   What kind of evidence is there for this reference class-contextualism? It is the same kind of evidence that also supports contextualism about topics and methods. 43 44

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

58  THE ARGUMENT FROM RELIABILITY

2.3.4 Conclusion In this section I argued that not only is there a generality problem for the individuation of methods but that there is also one for the individuation of topics. Furthermore, I argued that the generality problem can be extended and generalized into a problem about relevant reference classes (of which the generality problem is just a particular case). At the most abstract level, the basic point here is that there is a widespread and irreducible indeterminacy of reference classes when it comes to the reliability of methods and processes of belief acquisition. Given the close connection between reliability and knowledge, we must expect a similar “indeterminacy of knowledge” in many cases. Awful as this might sound at first, there is a good solution to this “problem”: contextualism. This solution might not be the kind of solution one would expect and hope for at the beginning; perhaps it deserves the name of a “skeptical solution.” But that is good enough. And sometimes that is all we can reasonably expect. One can take all this as an occasion to update (REL) (or (REL*) for that matter) and make the relativization to topics, methods, and reference classes (R) more generally explicit: (REL-contextual) S’s belief that p is reliable, given R iff P(t/m) ≥ n, given R.47 A further task would consist in explaining how the different types of reference class parameters combine into judgments about reliability. I won’t go into that here. The overall claim here is not restricted to reliabilists; given the undemanding and plausibly universal reliability requirement for knowledge, the result concerns internalists as well. But perhaps it is due to a special interpretation of “reliability” as probabilistic? In the following I will discuss the main, modal, alternative to the probabilistic view of reliability and see what implications it might have for the overall argument (RA) for contextualism.

2.4  Reliability and Modality There are two main varieties of modal accounts of reliability: a sensitivity account, modeled after sensitivity accounts of knowledge (see Nozick  1981, 172–8; Dretske 1971; Goldman 1992a; and Carrier 1971), and a safety account, modeled after safety accounts of knowledge (see Sosa 1999 and Williamson 2000). There are two varieties of each, depending on whether they are applied to methods and processes of belief acquisition or directly to beliefs (with “⇒” for the subjunctive conditional):48 (Sensitive Method) S’s belief that p and the method m that produced it are reliable iff [(a) p ⇒ m produces a belief in S that p, and 47   I assume that (REL-contextual) holds necessarily. For the sake of ease of exposition I have performed semantic descent here. In metalinguistic form the principle would say that: “S’s belief that p is reliable” is true, given appeal to R in C iff “P(t/m) ≥ n” is true, given appeal to R in C. 48   What follows need not be understood as reductive descriptive definitions. I assume that each of the four holds necessarily.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  59 (b) not p ⇒ m does not produce a belief in S that p]; (Sensitive Belief) S’s belief that p is reliable iff [(a) p ⇒ S believes that p, and (b) not p ⇒ S does not believe that p]; (Safe Method) S’s belief that p and the method m that produced it are reliable iff (m produces a belief in S that p) ⇒ p; (Safe Belief) S’s belief that p is reliable iff (S believes that p) ⇒ p. Since subjunctive conditionals do not contrapose, the right-hand sides of (Safe Method) and (Safe Belief) are not equivalent to (Sensitive Method-b) or (Sensitive Belief-b). It is only for simplicity’s sake that any reference to methods has been dropped in (Sensitive Belief) and (Safe Belief).49 The above principles have problems with the case of necessary propositions but we can let that go here, together with some other issues that need not be raised here (see, e.g., Roland and Cogburn 2011).50 There are different ideas about the truth conditions of subjunctive conditionals. Quite influential in epistemology is the idea that a subjunctive conditional is true just in case in possible worlds in which the antecedent is true and which are close to the actual world, the consequent is also true (see Nozick  1981, 173; see also 174–5).51 Whether the consequent would have to be true in all, most, or some close antecedent-­ worlds (or in the unique closest antecedent-world, if there is such a thing) can be left open here. What is crucial here, is the idea that possible worlds can and must be ordered according to closeness to or remoteness from the actual world. Closeness is usually explained in terms of similarity: one world is close to another world just in case it is similar to it; world B is closer to world A than world C is to A just in case B is more similar to A than C is to A (see Lewis 1973, 48–52, 66–7, 75–6; Lewis 1979b, 472, 1986, 20–7; see also Fine 1975; Jackson 1977, 4–8; Slote 1978, 20–5; Bowie 1979; and Krasner and Heller 1994). Everything is similar to everything else in some respect and also ­dissimilar to everything else in some respect. For the explanation of the notion of closeness one needs some idea about how to construct a notion of overall similarity. It is not easy to see how that can be done (for some general issues concerning the notion of similarity see Goodman 1972; Tversky 1977; and Morreau 2010). But even if it can be done, it is very hard to see how there could not be a plurality of different, mutually incompatible but equally legitimate constructions of overall similarity. The truth 49   In all four cases, the generality problem and related problems come up again, on top of the ones ­discussed in this section. 50   One might want to object that the two belief-principles aren’t principles of reliability. Given, however, that the explanation of “reliability” will have to have some stipulative aspects, I would not take this objection too seriously. 51   If we assume that the actual world is close to itself, then the four principles above are met only by true beliefs. All reliable beliefs would have to be true beliefs. One might hold that this in itself speaks against the above principles but one could easily use “close” in a sense in which it does not apply to the actual world.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

60  THE ARGUMENT FROM RELIABILITY c­ onditions of overall similarity judgments depend on which respects the thinker or speaker selects as the relevant ones and how he weighs the different respects against each other. There is no unique correct way to do this. So, attributions of similarity are context-sensitive (the contextual parameter being the attributor’s ideas about relevant respects). Insofar as closeness of worlds just consists in their similarity, “closeness” will be as context-sensitive as “similar.” David Lewis—who has, perhaps, championed this kind of account more than anyone else—himself accepts the context dependency of “similarity” (see Lewis 1973, 91–5; see also Lowe 2002, 189) and thus of the “closeness” of possible worlds (see Lewis 1973, 50–2, 66–7, 1979b, 466; see also Heller 1995, 505–7, and Heller 1999a, 116; for the general problem with ideas of a closeness-ranking see also Lowe 2002, 81, 146, 189 and Zalabardo 2012, 111–14). One can make the case for the context-sensitivity of “closeness” also in a more direct way, without using ideas about similarity and just relying on some intuitive notion of closeness. Take the actual world (ACTUAL) with its laws of nature and other features; assume for instance that we are not brains in vats and that we are also not in any of the other notorious skeptical scenarios. Consider two non-actual possible worlds: one world (VAT) where the laws of nature are the same as in the actual world but we are all envatted (whoever belongs to “us”), and another world (WEIRD) where nobody is envatted but the laws of nature are very different from what they are in the actual world. How close are VAT and WEIRD to ACTUAL and to each other? Is WEIRD closer to ACTUAL than VAT is to ACTUAL? Or are they equally close to ACTUAL? If one asks an epistemologist one typically gets the answer that WEIRD is closer to ACTUAL than VAT is; however, if one asks, for instance, a physicist, one rather gets the answer that VAT is closer to ACTUAL than WEIRD is. There is disagreement but a faultless one: nobody is wrong. The two closeness attributors are simply choosing different criteria of relevance and are thus focusing on different aspects of the relations between the different worlds. One could say that the epistemologist is using epistemological criteria while the physicist is using metaphysical criteria of relevance (in a broad sense of “metaphysical”). Since examples like the one just given can be constructed easily, the case easily generalizes: attributions of closeness are context-sensitive and the contextual parameter are the attributors’ ideas about relevance.52 Some contextualists about “knowledge,” like Keith DeRose (but interestingly not David Lewis: see Lewis 1996), make heavy use of the idea of a closeness ordering of 52   For some hints at skepticism about the very common idea amongst epistemologists that vat-worlds are very remote in an absolute sense see: Grobler 2001, 293; Neta 2003a, 16, fn.51; Freitag 2013, 138; see also Baumann 2005b, sec.1, and 2009a, sec.2, and for a very detailed treatment Wilburn 2010. It does not help to insist on epistemological criteria since we are discussing epistemological topics here like reliability and knowledge. Why should the latter speak in favor of the former? And: Can one explain what the epistemological criteria for closeness are—which we need to explain reliability and knowledge—without using these very notions, the ones we are trying to explain? An account that does not lead into circularity or into a bad infinite regress is needed and, as far as I can see, (still) lacking. For this problem see Wilburn 2010, 260 and passim; see also Baumann 2005b, sec.1, and 2009a, sec.2.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  61 possible worlds (see, e.g., DeRose 1995, esp. 36–7). One core idea is that the standards for what possible worlds one needs to be able to rule out in order to count as “knowing” the relevant proposition vary with context: in very demanding contexts the subject needs to be able to rule out “remote” possibilities like the one that one is a handless brain in a vat in order to count as “knowing” that one has hands while in relaxed contexts the subject “only” needs to be able to rule out more mundane and “close” possibilities in order to count as “knowing” that one has hands. If “closeness” itself is context-sensitive, then we get a two-tiered contextualism. First, what counts as “close” or “remote” itself varies with context; second, given one particular closeness ordering, how far out into modal space one’s powers of ruling possibilities out have to reach in order for one to count as “knowing” some proposition also varies with context. There need not be anything in principle wrong with such a two-tiered version of contextualism except perhaps that it makes contextualism more complex than expected or assumed. It is not clear to me how many contextualists would not mind such a further complication of their views (see, interestingly, the remarks in DeRose 1995, 21). Apart from that, certain cherished contextualist arguments look much more difficult to make, like for instance some anti-skeptical arguments by contextualists. For instance, insofar as the contextualist wants to distinguish between ordinary, less demanding, and skeptical, more demanding contexts, it will be very difficult to spell this out in terms of close and remote possible worlds if the typical skeptical scenarios can also be seen as close, and the typical ordinary scenarios as remote (for more on contextualism and skepticism see Chapter 4). The context-sensitivity of “closeness” has implications for the four modal explanations of reliability above. Consider (Sensitive Method) and two non-actual possible worlds in their relation to the actual world. Assume that I am going on a week-long trip and have just reassured myself that I have turned off the oven. My target belief is that I have turned the oven off. In the non-actual possible world HAPPY I have indeed forgotten to turn the oven off but am so happy about going on a trip and careless about possible problems that I come to believe that I have indeed turned the oven off. In the non-actual possible world WORRY I have also forgotten to turn the oven off but I am quite nervous and obsessed with the possibility of a mistake; I don’t form a belief one way or another.53 If HAPPY is a typical close oven-on-world, then (Sensitive Method-b) might not hold; if, in contrast, WORRY is a typical close oven-on-world, then (Sensitive Method-b) might hold. In one case I would, given everything else goes right, meet the condition of reliability as formulated by (Sensitive Method), in the other case I wouldn’t. Different closeness rankings can thus lead to different results about reliability. 53   There is one complication here: Perhaps one has to hold the method used by the subject fixed across the different possible worlds? One problem with that idea is the generality problem for methods. Another question is whether one should really make such a demand or whether this would be too strong in many cases (how, for instance, should one think about skeptical scenarios then if the methods aren’t available there that we typically use in non-skeptical scenarios?).

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

62  THE ARGUMENT FROM RELIABILITY The same kind of point can be made for (Sensitive Method-a). (Safe Method) does not differ either: What are the close worlds in which m does produce my target belief? Are these the worlds in which I am as careful as I like to think I am or rather worlds where my reckless tendencies win? Depending on the judgment about closeness we get ­different results for reliability, according to (Safe Method). Similar points can also be made in the case of (Safe Belief) and (Sensitive Belief); I won’t go through all the different variations here. For all four versions of a modal principle of reliability, we get the result that “reliability” is context-sensitive because “closeness” is. I have assumed one kind of view about the truth conditions of subjunctive conditionals like the ones we are considering; nothing will change here if one accepts one of the other views on that subject but I will not show this here. Not much is left to add here. If “reliability” is context-sensitive according to modal accounts of reliability, too, then “knowledge” will, again, “inherit” its own ­context-sensitivity from “reliability”—if we adhere to (1) in (RA), that is, to the claim that knowledge requires reliability, and if we also acknowledge that variations in the truth conditions of reliability attributions underlie variations in the truth conditions of knowledge attributions. Again, contextualism comes to the rescue against the threat of the indeterminacies of closeness, reliability, and knowledge.54 Even though both modal and probabilistic accounts of reliability support the central argument in this chapter, the argument from reliability for contextualism (RA), there are reasons to prefer the probabilistic account to the modal account. First, modal accounts can give ordinal closeness rankings but it is hard to see how they could even get close to cardinal rankings; probabilistic accounts have an advantage of substantial informativeness here because the probability calculus is quite powerful and allows for cardinal information about probabilities (see also Zalabardo  2012, 111–14 here). Second, the notion of conditional probability allows it to make context-sensitivity quite explicit: given C-1, the probability of a true belief is quite high whereas given C-2, the probability is quite low; in modal accounts the context-sensitivity is more implicit. With probabilistic accounts one can indicate where the context-sensitivity lies just by explaining what one is conditionalizing on; with modal accounts, one would have to indicate whole closeness rankings of lots of worlds. Third, if knowledge requires reliability but does not require safe or sensitive belief, then the different biconditionals linking reliability with safety or sensitivity conditions cannot be correct; safety and sensitivity can then also not be necessary for reliability (only, perhaps, sufficient). I think there are good arguments against the claims that safety or sensitivity are necessary for knowledge (see Baumann 2008a, 2012a); this also speaks against modal views of reliability. Fourth, probabilistic accounts might be closer to a naturalist outlook (if that is a good thing) than modal accounts; the sciences have a lot of use for probability theory but, so far, little interest in modal logic and modal metaphysics. As a frequentist 54   The context-sensitivity of “reliability” modally interpreted has the advantage of illustrating in still another way that it is very problematic to assume that there is an objective hierarchy of contexts.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

THE ARGUMENT FROM RELIABILITY  63 about probability in particular, it is easier to remain within the confines of a Humean restriction to what we find out about this world. All this is no decisive argument against modal accounts and for probabilistic accounts of reliability but I think it goes some way into that direction.

2.5 Conclusion What determines the different reference class parameters: topics, methods, spatial and temporal reference classes? Lewis’ rule of attention (see Lewis 1996, 559) can explain something here. What has come up in conversation or is on the attributor’s mind for whatever other reason has a plausible influence on the typing of topics (“What is Mary thinking about?”), the typing of methods (“How did she figure that out?”), and the spatial and temporal reference classes (“Isn’t she perceptive these days, even given the less than optimal circumstances around here?”). This should not be confused with the role the salience of chances of error plays in many (other) contextualist views (see Chapter 1). Similarly, the purposes and intentions of the attributor: they don’t only determine topics but also the types of topics. We can take Kurt’s court case as an illustration of the influence of this factor on the choice of temporal reference class: the defense will rather want to raise questions about Kurt’s reliability between 4:00 p.m. and 4:05 p.m. while the prosecution will rather focus on Kurt’s general reliability. Similar for spatial reference classes: the driving instructor will want to see how the person does in different parts of the city while the student is only interested in the few parts of the city where she would go by car. It is a bit harder to see an influence of attributors’ purposes and intentions on the typing of methods. Finally, there is some influence of contextual norms and conventions. The court room is, again, a good illustration for the normative fixing of types of topics (“But how well can the witness recognize these shades of colors at dawn?”), of types of methods (“Did you see it yourself or just hear it from neighbors?”), and spatial and temporal reference classes (“You said you were quite tired when you thought you saw the defendant?”). The other two kinds of parameter fixing determinants discussed in Chapter 1, stakes and salience of chances of error, don’t seem to have much application here. We can put parameters and determinants into Table 2. Table 2.  Fixing the reference class Determinants: →

Attention

Purposes and intentions

Norms and conventions

√ √ √ √

√ − √ √

√ √ √ √

Parameters: ↓ Topics Methods Time frame Spatial frame

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

64  THE ARGUMENT FROM RELIABILITY This concludes the argument from reliability from contextualism. It is not and is not meant as a valid deductive argument but it constitutes a good non-deductive argument and explanation. It constitutes a more “theoretical” argument than the argument from cases. Such arguments have an advantage over mere referrals to ordinary usage and our “intuitions” about what to say when. Apart from that, more parameters have been brought up, in addition to the different standard parameters in Chapter 1. In Chapter 3, I will make still another “theoretical” argument for contextualism—and add more aspects to the view.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

3 The Argument from Luck The Role of Descriptions There is another, also more theoretical argument for contextualism which adds another dimension to (standards) contextualism. This argument concerns the relation between knowledge and luck. One might wonder whether it is parallel to the argument from Reliability (RA) above: The Luck Argument for Contextualism (LA) (1) If “S knows that p” is true in C, then “S’s belief that p is not lucky” is true in C;1 (2) The truth conditions of sentences of the form “S’s belief that p is not lucky” can vary with the context of the speaker; (C) Hence, the truth conditions of sentences of the form “S knows that p” can vary with the context of the speaker.2 Even though (2) and (C) will stand up here, (1) won’t. The argument from luck defended here will go differently from (LA). Against current orthodoxy which treats (1) or the assumption that knowledge excludes luck (of a certain sort) as a truism I will argue that knowledge is indeed compatible with luck of that sort. It will turn out that the notion of luck is relative to description and for that reason context-sensitive, and that very similar considerations apply to the notion of knowledge, too. Like Chapter 2, this one does not just offer an argument for contextualism but also a view about knowledge and other important epistemic notions, like the notion of luck. I will first make an argument from cases for the compatibility of knowledge and luck (Sections 3.2 and 3.3) and then move on to more theoretical views about luck and knowledge (Sections 3.4 and 3.5); in the end I will propose a probabilistic notion of luck and knowledge which also explains in more detail in what sense knowledge is compatible with luck and in which sense it isn’t. But first some remarks on the ­relevant notion of luck.

  (1) would hold necessarily.   This would be so because and insofar as there are cases where nothing else apart from luck would stand in the way to S’s belief constituting knowledge. 1 2

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

66  THE ARGUMENT FROM LUCK

3.1  Luck, Its Varieties, and an Exclusion Claim The relation between knowledge and mere true belief which does not amount to knowledge has long kept philosophers busy. A number of authors hold that non-­ accidental true belief is knowledge (see, e.g., Unger  1970, 114–15). Whatever one might think about the relation between accidentality and luck (could someone ­compulsively and in that sense non-accidentally come to acquire a luckily true belief that p?), there is widespread consensus that there cannot be any lucky knowledge: knowledge is incompatible with luck. More precisely, a belief which is true only by luck cannot constitute knowledge. Let us call this exclusion claim the “No-Luck-Thesis” (see amongst many: Unger 1970, 114–15; see for one of the few dissenters Hetherington 2001, 84, and Hetherington 2011, ch.3): (NLT) If S knows that p, then S’s belief that p is not true by luck.3 Before this claim can be made more precise, some broader remarks about luck in general—whether epistemic or other (e.g., moral)—seem in place. The general notion of luck applies primarily to events and states.4 A state or event is lucky not in itself but for someone: in the case of good luck it is good for that being, in the case of bad luck bad.5 Some states or events are neither good nor bad for someone, they lack evaluative significance for that being. Unexpected rain might constitute good luck for farmers, bad luck for picnickers, and neither good nor bad luck for distant observers. Luck requires that the relevant event or state was not to be expected, is surprising, could easily not have happened or come about, or was improbable (much more on this below; see for the general conditions of luck: Rescher 1995, 32, passim; Riggs 1998; Pritchard 2005c, ch.5; Coffman 2007, 2015). People sometimes say that luck involves the lack of control (see on this: Rescher 1990; Katzer 1996, 106–11; Vahid 2001; Pritchard 2005c, 127; Riggs 2007; Coffman 2007; Lackey 2008, 256–60; Broncano-Berrocal 2015). If “lack of control” means total or nearly total lack of control, then we certainly have to say that lack of control is not sufficient for luck: that I don’t have any control over the tides does not make me lucky in that respect. Lack of control in this sense is also not necessary for luck because many things are to some degree but not completely under my control and still lucky, like my lucky shot in a basketball game. If, by contrast, we mean lack of total or almost total control by “lack of control,” then we also don’t get quite the right connection between   (NLT), like its variants (NLT*) and (NLT**), is meant to hold with necessity.   One may also want to include facts here. Persons are lucky in a secondary way, namely insofar as certain kinds of things happen to them or they find themselves in certain kinds of circumstances; we call such persons “lucky” because of the nature of those events and circumstances, not because of other, “deeper” traits of the person. This might seem controversial but I don’t have to argue for it here because the kind of luck relevant here only applies to (epistemic) events and states. 5   It is not necessary here to speculate about what kinds of beings or entities are susceptible to luck and what determines whether something constitutes good or bad luck for it: the being’s views about their own good or, rather, some more “objective” factor? 3 4

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  67 luck and control: that I am a brilliant but still far from perfect basketball player does not mean that all my shots are lucky (but cf. Engel 1992, 60–4 for the idea that limited control entails luck in epistemic cases). However, some degree of control (something between total lack of control and total control) seems necessary for luck: a lucky basketball shot requires a certain degree of skill but not too much. But how much control and how much lack of control is necessary or sufficient for luck? It is hard to see how one should answer these questions in a principled way. Apart from that, the notion of control relevant here is either a probabilistic or a modal notion (for a more negative evaluation of the control thesis see Lackey 2008, 256–60). Hence, we should turn to straightforwardly modal or probabilistic accounts of luck. I will argue below that the best explanation of all this can be given in probabilistic terms: if an event is lucky, then the probability that it would happen, given certain conditions, is low enough or below a certain value (but cf. Morillo 1984). In discussions of the relation between knowledge and luck a broad notion of luck should be used: whether one thinks that having true beliefs is always or usually a good thing or not does not matter here. What matters is the accidentality or non-accidentality of a true belief, not its evaluative significance. I am using “accidental” as a synonym of “lucky” here (see also Yamada 2011).6 The question then is whether knowledge is compatible with accidentally true belief, and for that it does not matter what the evaluative significance of true beliefs is (but see Ballantyne 2011 and 2012). This is an advantage because it means that we can leave the complicated issue of epistemic values aside here. In the literature, the term “luck” is also used in this broader sense of accidentality (though the difference between the broader and the more specific notion of luck is not often noticed or mentioned). A true belief can be called “lucky” in several different ways (see Engel  1992; Harper 1996; Pritchard 2005c, chs.5, 6).7 (NLT) is not specific enough as it stands: some kinds of luck are uncontroversially compatible with knowledge whereas others aren’t. A subject could be lucky to have certain rare abilities (“ability luck”) which alone enables her to acquire certain kinds of true belief and knowledge (a mathematician might be lucky to be talented enough to prove that Fermat was right). Or a subject could be lucky to find herself in circumstances (“circumstantial luck”) which enable her to gain true beliefs and knowledge of a certain kind (e.g., a witness of a crime who just happens to be at the scene of the crime when it happens). These kinds of luck are unproblematically compatible with knowledge. What matters here is a different kind of luck. Following Thomas Nagel’s classification of types of moral luck further 6   Should this raise concerns (see Riggs 2014), I’d be happy to replace all talk about something being accidental by talk about it being lucky; nothing substantial hinges on these verbal matters here. 7   We are not dealing here with luck concerning other epistemic states or events like having good warrant for some proposition. One can also be lucky not to acquire a false belief, when acquisition of a false belief is to be expected under the circumstances (also: one can be lucky to suspend belief, when that is the right thing to do but not what is to be expected). These kinds of epistemic luck lie beyond the topic under consideration here.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

68  THE ARGUMENT FROM LUCK (see Nagel 1979 and also Williams 1981), one might want to call this “resultant epistemic luck.” A true belief could be lucky in this sense even given the situation of the subject at the time of the acquisition of the belief (including their abilities, circumstances, and available evidence), that is, the belief can be lucky even if the belief is not lucky in the other ways mentioned. Engel 1992 was one of the first authors to clearly distinguish between different forms of epistemic luck; the one relevant to (NLT) he calls “veritic luck” which he explains as luck that a given belief is true, given the evidence for it (see Engel 1992, 67); in contrast, “evidential” luck is the luck of having the evidence in the first place (see Engel 1992, 67). Pritchard 2005c also uses the term “veritic luck” for the kind of luck relevant here and explains it as luck that a given belief is true (see Pritchard 2005c, 146); he spells this out in terms of a safety account of luck and knowledge (see Pritchard 2005c, 163; more on this in Section 3.4; see also Hetherington 2011, 90–8). As we will see, there are several different ways of explaining what resultant ­epistemic luck is (for more on epistemic luck, apart from Engel and Pritchard, see: Ravitch 1976; Foley 1984, 1987, ch. 4; Latus 2000; Vahid 2001). Whenever I speak of luck here, I have in mind some notion of resultant epistemic luck, that is, not evidential, circumstantial, or “ability” luck.8 We already know that we ought to modify (NLT) in some way. Here is a better version: (NLT*)  S knows that p, then S’s belief that p is not true by resultant epistemic luck. I will start arguing against (NLT*) now, first by discussing some cases. To that purpose, I will use an intuitive, pretheoretic, and, I think, common notion of luck (as well as of knowledge).

3.2  Lucky Knowledge: Variations on a Case by Russell Consider Russell’s famous clock example (see Russell 1948, 98 as well as a brief hint in Hetherington 1998, 466): A wants to know what time it is, looks at a clock which indicates the correct time, say 12:15, and thus comes to believe that it is 12:15. Unbeknownst to A, however, the clock has stopped some time ago and, as a matter of sheer accidence, just happened to indicate the correct time when A consulted it. A does not know the time, adherents of the No-Luck-Thesis would argue, because her belief was true only by luck.9 This appears quite plausible. Consider the following follow-up scenario: (a)  A’s watch has also stopped (though at a different time). After consulting the clock, A immediately sets her watch which from now on indicates the correct time. 8   One might doubt that one can draw this kind of distinction between different kinds of luck in any principled way. This is not the case but this is—as the remarks below will show—no problem for the view defended here (on the contrary, this will even support it). 9   The word “because” carries a lot of weight here. The defender of a No-Luck-Thesis would insist on it. My argument here implies, however, that even if one characterizes a case like the one above as one of luck and lack of knowledge, one should not say that there is the latter because of the former.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  69 Two minutes later, A consults her watch and comes to believe (correctly, again) that it is 12:17.10 Does A know this? Whoever denies knowledge in the original case will also tend to deny it in (a) because (a) is similar in all relevant respects. Again, A was just lucky in her belief turning out true. Could one argue that while there is luck in Russell’s original case, there is no luck in (a)? I don’t think so because in both cases we are dealing with a subject who is around a stopped clock and acquires a true belief about the time in close causal proximity to the presence of the stopped clock. However, if we change (a) slightly or change the description of (a) slightly, for instance by leaving out that the watch was set after a stopped clock and by stressing that it is a very reliable watch set after a clock indicating the right time, then we might well get the opposite result that the resultant belief was not lucky. Given the changed description, it was quite improbable that the subject would acquire a false belief; not easily could the subject have acquired a false belief. I will argue in much more detail below that an adequate notion of luck must incorporate such an element of relativity of luck. However, given the way (a) was described and constructed, it is plausible to say that the subject is lucky in that case, too, not just in Russell’s original case. Time passes: (b)  The next morning, someone, B, asks A what time it is. A looks at her watch and tells B correctly what time it is. Do A and B know what time it is? The only interesting difference between (a) and (b) is that more time has passed. The mere passage of time, it would seem, cannot turn non-knowers into knowers or lucky beliefs into non-lucky ones (see Reed 2000, 66–7). Here is an analogy with non-epistemic luck. Suppose you were lucky enough to win the lottery. Now you’re rich and you’re lucky to be rich. Time passes and ten years later you’re still rich, not having benefited from any other sources of wealth. Even given that no later events or circumstances were threatening your wealth, you’re still lucky to be rich. The passage of time does not make the luck go away. Also, it seems, if we denied knowledge in (a), then we should also deny it in (b). Let us assume that A keeps consulting her watch for the next year. It still indicates the correct time (within the boundaries set by common sense—A has no use for atomic clocks): One year after the clock incident—having consulted her watch many times—A checks the time again and comes to truly believe that it is 1:19 pm. 10   John Hawthorne uses this very example to a very similar conclusion (see 2000, 202–3). See also DeRose 2009, 17–18 who shortly mentions this case and agrees with Hawthorne; however, he doesn’t discuss it in detail and neither does Hawthorne. See also Hawthorne  2004a, 69, and Gendler and Hawthorne 2005.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

70  THE ARGUMENT FROM LUCK Now let us add some further details to the description of this case: (c)  A’s watch is an exquisite and expensive Swiss watch. It is the kind of watch which need not be set for decades after the first correct setting. A does not wear it but keeps it, for safety reasons, in a drawer. The watch is very dear to her because it was a wedding gift. Quite often she takes it out of the drawer to show it to guests. Whenever she does that, she consults it for the time. This kind of watch, she was told, is super-reliable and she has no reason to doubt it. At this moment, one year after the clock incident, having consulted her watch many times, she takes the watch out again and looks at it. It shows the correct time: 1:19 p.m. Neither A nor anyone else has any clue that the watch was set after a “Russellian” clock, exactly one year ago. A has no doubt and truly comes to believe that it is 1:19 p.m. now.11 Does she know that? My impression is that most of us would now grant A knowledge of the time. The original epistemic deficiency (lack of knowledge) in the clock incident has been removed. Or so it seems. However, no factors relevant to luck have changed between (a), (b), and (c). Hence, A would, I think, still count as lucky in (c). Hence, (c) is a case of lucky knowledge. One could object that A is also not lucky in (c). If the watch were not giving the correct time but rather, say, a much later or earlier time, then A would notice that something is not right and that the watch does not give the correct time. Hence, the fact that A does not notice anything like that justifies A in the (more or less implicit) assumption that the watch is correct and reliable. But under such circumstances, A is not lucky. As a reply, I would like to stress first that it is not crucial for the example that A acts on the information she gets from the watch or checks it (implicitly or explicitly) against other sources of information; she might even forget a few moments later what the watch indicated. She would still be lucky but also be a knower of the time. More important is another point. Even if we assume that A gets rough (implicit or explicit) confirmation that the watch is correct and reliable, she would only get confirmation that the watch is not wildly off the beam but we can easily assume that she is getting no confirmation that the watch is also correct and reliable with respect to giving the exact time, say in minutes of the hour. But that is exactly what her belief is about. Without such more specific confirmation about the correctness and reliability of the watch in indicating the minutes of the hour and not just the rough time of the day, she is still lucky in her correct belief in the time but also still a knower. But how is that possible? Does time heal such epistemic wounds? How could the passage of time turn non-knowers into knowers? And how could it do that without turning a lucky belief into a non-lucky one? Perhaps I have overlooked further relevant 11   In this case as well as in cases (d) and (e) we are assuming that the watch did not stop at any time before but only that the first time it was set it was set after a stopped clock indicating the correct time. Hence, there is nothing wrong with the assumption that the watch is extremely reliable. Thanks to Darrell Rowbottom for bringing this aspect of the cases to my attention.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  71 factors? Perhaps the fact that A has consulted her watch many times is relevant? But consider this case as a control case for (c): (d)  A’s watch is an exquisite and expensive Swiss watch. It is the kind of watch which need not be set for decades after the first correct setting. A does not wear it but keeps it, for safety reasons, in a drawer. The watch is very dear to her because it was a wedding gift. Quite often she takes it out of the drawer to show it to guests. Whenever she does that, she looks at it but curiously never consults it for the time. She admires the beauty of the watch so much that it never occurs to her to use it for such prosaic purposes. At this moment, one year after the clock incident, she takes the watch out again and looks at it. She wonders what time it is and for the first time consults it for the time. The watch shows the correct time: 1:19 p.m. This kind of watch, she was told, is super-reliable and she has no reason to doubt it. Neither A nor anyone else has any clue that the watch was set after a “Russellian” clock, exactly one year ago. A has no doubt and truly comes to believe that it is 1:19 p.m. now. Does A know the time? One might have to run this question by a representative sample of competent speakers of English. Even though (d) is perhaps not as clear a case as (c), I still suspect that many people would attribute knowledge in (d). I certainly would. But how could a source of information which does not give us knowledge the first time give us knowledge simply because of repeated use (as in case (c); more on this below)? One thing seems clear: the original luck has stayed. Hence, we have a case of lucky knowledge. Let us add one more variation of Russell’s clock example. This one is close in some respects to (a) and reminiscent of Alvin Goldman’s (1986, 45) thermometer example (see below): (e)  A is a watchmaker and has just finished another one of his famous handmade Fregeant-watches. To set the watch, the unsuspecting A looks at the “Russellian” clock and sets the watch after it. From now on, the watch indicates the correct time. These kinds of watches need not be set again for a long time. A then puts the watch into her display box containing eight other watches of the same series. The other watches have been set earlier, after the same clock but when it was still running OK. One hour later, B enters the store, wanting to buy a watch from the Fregeant series. B picks the one A had just put into the display. To B, they all look the same. B doesn’t set the watch because A tells him that it has just been set. Later in the day, B gives the watch to C as a birthday present. The next morning, C consults the watch for the first time. Given the great reputation of Fregeant watches, C has no doubts and comes to truly believe that it is 7:30 a.m. Neither C nor B, A, or anyone else has any clue that the watch was set after a “Russellian” clock, many hours ago. So, C simply comes to believe that it is 7:30 now. Again, does the subject know what time it is? I think, again, that the answer is positive. At the same time, it is pretty clear that the subject’s true belief is lucky.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

72  THE ARGUMENT FROM LUCK All these different variations on a Russellian theme suggest that there can be lucky knowledge. To be sure, we have to be careful with the intuitive and unanalyzed notion of luck that we used here. I will say a bit more on different notions of luck below. But we can already see a good reason to suspect that the absence of luck is no necessary condition of knowledge.12 But how can one explain the compatibility of knowledge and luck? The different cases above and their descriptions differ in several ways. One important difference concerns the kinds of factors that are at the center of the description or rather in the background (if explicitly mentioned at all). Case (a) differs in that respect from cases (c) or (e) (to pick just three of them). For instance, one could say that when judging (a) and (c) different kinds of factors are in the focus of the description: in (a) rather the questionable source of information when the watch was set and in (c) rather the reliability of the watch and the long history of (actually or potentially) giving the correct time. Or take (a) in comparison with (e). In the first case we have a focus on the initial (lucky) coincidence while in the second case (e) we focus much more on the reliability of the watch and its similarity to non-luck exemplars of the same kind and thus on the many other comparable and unproblematic watches. In all cases or in all descriptions above the “original Russellian sin” was present enough to make us judge that there is a case of epistemic luck. In contrast, in only some of them, the great reliability of the watch was central enough to make us judge that there is a case of knowledge. Some factors are more relevant to luck, some more to knowledge; and there is different focus on these kinds of factors in the different cases above. This explains the divergence of our luck-verdicts and our no-knowledge-verdicts in some of the cases. One can get different judgments about the same case, depending on the description. Let us assume that (c), (c*), and (c**) are different descriptions of the same case. Compare (c)  A owns an exquisite and expensive Swiss watch. It is the kind of watch which need not be set for decades. A does not wear it but keeps it, for safety reasons, in a drawer. The watch is very dear to her because it was a wedding gift. Quite often she takes it out of the drawer to show it to guests. Whenever she does that, she consults it for the time. This kind of watch, she was told, is super-reliable and she has no reason to doubt it. At this moment, one year after the clock incident, having consulted her watch many times, she takes the watch out again and looks at it. It shows the correct time: 1:19 p.m. Neither A nor anyone else has any clue that the watch was set after a “Russellian” clock, exactly one year ago. A has no doubt and truly comes to believe that it is 1:19 p.m. now. with (c*)  A owns a watch. At this moment, one year after the clock incident, she takes the watch out again and looks at it. It shows the correct time: 1:19 p.m. Neither A nor 12   The path from (a) to (c), (d), or even (e) does not constitute a sorites or slippery slope. I am not exploiting any relevant vagueness and am not making an argument via small steps to an absurd conclusion.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  73 anyone else has any clue that the watch was set after a “Russellian” clock, exactly one year ago. A has no doubt and truly comes to believe that it is 1:19 p.m. now. In (c*) we would be much less inclined to ascribe knowledge and much more inclined to ascribe luck than in (c); the relevant factors that would make us do that are absent from the description. Now compare (c) with (c**)  A owns an exquisite and expensive Swiss watch. It is the kind of watch which need not be set for decades. A does not wear it but keeps it, for safety reasons, in a drawer. The watch is very dear to her because it was a wedding gift. Quite often she takes it out of the drawer to show it to guests. Whenever she does that, she consults it for the time. This kind of watch, she was told, is super-reliable and she has no reason to doubt it. At this moment, one year after she first set it, having consulted her watch many times, she takes the watch out again and looks at it. It shows the correct time: 1:19 p.m. Neither A nor anyone else has any clue that there was something special about the way the watch was initially set. A has no doubt and truly comes to believe that it is 1:19 p.m. now. In (c**) we would be much less inclined to ascribe luck and much more inclined to ascribe knowledge; again, the relevant factors (for a different verdict) are mostly absent from the latter description. So, (c*) suggests that there is luck without knowledge, (c**) suggests that there is knowledge without luck. This is certainly not surprising as such but what is interesting is that a variation of the description of the same case gives us an intermediate case (c) of luck with knowledge. Suffice it to add briefly here that it does not help at all to demand that we only use (“absolutely”) complete descriptions of cases, not incomplete ones. First of all, our practice of attributing knowledge and luck never (or almost never) relies on anything that even approximates a complete description (assuming for the sake of the argument that we can make sense of that very idea). Furthermore, by “completeness” one can then only reasonably understand “completeness in the relevant respects.” But what are the relevant respects? What is relevant depends, for instance, on varying interests of attributors and there is nothing right or wrong about these interests (given that the interests are not misguided in some other way—but that is a different issue). Hence, it also does not make sense to demand that one should only consider descriptions of cases that have the uniquely correct focus on the truly relevant factors. Still, it is very tempting to expect that as knowledge appears as we move from (a) to (c) or (d) or even (e), so does luck disappear. However, as indicated, there is an important difference in the influence of certain factors (like the passage of time) on our judgments about knowledge on the one hand and about luck on the other hand. What could explain why knowledge appears after time has passed while luck still does not go away? Attributing knowledge to a subject involves the (more or less implicit) acknowledgment that the subject has certain cognitive abilities (taken in a broad sense, involving the use of instruments), the use of which helps explain that the subject acquired knowledge.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

74  THE ARGUMENT FROM LUCK When we’re thinking about a subject’s knowledge or lack thereof, we’re very much thinking about what the subject has been doing and is doing over time (again, including the use of more or less reliable instruments). This idea has affinities with certain agent-reliabilist forms of virtue epistemology (see, e.g., Sosa 1992, 2007 and Greco 2006 and 2010). In contrast, attributions of luck are less “track-record”-related and more “origin”-related: whether some state or event is lucky is very much a question of how it originated (in a “lucky” way or not) while our attributions of knowledge are not that closely tied to questions of origin and more to matters like performance over time. Therefore, we’re ready to “rehabilitate” the epistemic subject after a certain amount of time while we stick with our attributions of luck even after time has passed. We are happy enough, for instance, to use the lucky subject as a source of information and call him a “knower” while retaining the initial judgment about luck. All this suggests contextualism. The truth conditions of attributions of knowledge and of luck depend on contextually variable descriptions of the case under consideration. Depending on which description the attributor picks and focuses his attention upon, different kinds of factors are at the center of his attention or at the margins (or even absent). Relevantly different descriptions can then lead to different truth conditions of the corresponding knowledge- or luck-attributions. More will be said about these contextualist implications below (see especially Section 3.5).

3.3  More Cases More examples similar to the ones above can be brought up. Consider inferences and the following argument for the incompatibility of luck and (inferential) knowledge. Suppose A (competently) infers q from p and comes to believe q. However, A does not know p but only has a lucky true belief that p. One could then argue that A’s belief that q is also lucky and for that reason does not constitute knowledge.13 First, why should A’s belief that q be lucky, too? Perhaps because it has been inferred from the content of another lucky true belief? This would suggest some principle like the following: (TL)  If S’s true belief that p is lucky and if S (competently) infers q from p, then S’s true belief that q is also lucky (if it was acquired on the basis of that inference alone). But is (TL) true? Is luck always transferred across competent inference (leading from the belief in the premises to the belief in the conclusion)? Consider this example. I come to believe that Joe is singing in the other room. And I am right. However, it is usually Jack that does the singing around here and I cannot 13   An inferred true belief can be lucky in more than one way. It can, for instance, be lucky insofar as the inference was not a competent one. In such a case, however, it is uncontroversial that the inferred belief does not constitute knowledge—no matter where one stands on the No-Luck-Thesis. The above case of inferential luck is more interesting here.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  75 distinguish Jack’s from Joe’s voice. So, my belief that Joe is singing in the other room is true but lucky. However, if I infer from this that someone is singing in the other room, then the belief in that conclusion seems not lucky. There are other interesting examples. Take any lucky true belief and assume that the subject correctly infers some truth of mathematics or logic from it. Joe is singing, hence 2 + 2 = 4. Or: Joe is singing, hence Joe is singing or Joe is not singing. Should we assume that belief in the conclusion is lucky in such cases? Or suppose that I am driving in my car and have no idea whether I’m above the speed limit of 35 mph. I look at the speedometer and it reads “72 mph.” And I am really going at 72 mph. However, my speedometer usually does not get the speed right exactly but diverges by a few miles from the true number. Thus, my belief that I am going at 72 mph is true by luck. Suppose I infer that I am way above the speed limit. Is belief in the conclusion really true by luck, too? I think these cases raise some doubt about (TL) but I also think that they are not decisive against (TL) or similar principles. With respect to the singing case one could reply that belief in the conclusion is as lucky as belief in the premise and only appears non-lucky because there is a different inferential path which the subject could have taken instead of the actual one, namely one leading from a non-lucky belief in a different premise (“Jack or Joe is singing”) to the (now) non-lucky belief in the conclusion that someone is singing. However, in the example above I did not take that path; hence, my belief in the conclusion only falsely appears to be non-lucky. So, this is a case of a lucky belief leading to another lucky belief which might falsely appear to be non-lucky. The other alleged counterexamples against (TL) are cases where belief in the conclusion can plausibly be taken to be non-lucky but there is, in addition, another non-lucky belief in another premise in the background doing relevant inferential work. In the cases of logical and mathematical truths it is hard to imagine how we could not also, in addition, believe the conclusion on the basis of other premises which we believe in a non-lucky way. Hence, our inference is overdetermined and involves non-lucky beliefs in some other premises. The speedometer case can be dealt with among similar lines: How can I not notice that I am going way too fast by looking outside the window (which I am presumably doing when driving my car)? So, even though there is, perhaps, some doubt about (TL) it is still pretty plausible. It might not hold in all cases and perhaps one will have to modify it in one way or other. But there are many cases where luck transfers across competent inference from belief in the premises to belief in the conclusion. This is all we need here. Even if luck (and also the absence of luck) transfers across competent inference (always or in a given type of case), lack of knowledge doesn’t (no matter whether knowledge does; see Baumann 2011a). The following principle of “reverse closure” is not correct: (RC)  If S does not know that p and infers q from p, then S does not thereby come to know that q. On the contrary, one can come to know something on the basis of an inference from something one does not know because it is false, as the case of an inference that someone

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

76  THE ARGUMENT FROM LUCK is singing from a false premise that Jack is singing suggests. Here is another, perhaps more convincing example involving a false premise. Assume that some theories or theoretical systems approximate the truth without being true in the strict sense. Suppose that Newtonian mechanics is such a case. Someone could use Newtonian assumptions to arrive at a true conclusion about the rough date of the next lunar eclipse. We would then, it seems, still credit the person with knowledge of the conclusion (see Warfield 2005; Klein 2008; and Hiller 2013, but also Luzzi 2009; Veber 2014; and Ball and Blome-Tillmann 2014). That (RC) is false has to do with the point mentioned above that the notion of luck is an origin-related notion whereas the notion of knowledge is a track-record-related notion: origin-related properties transfer more easily than track-record-related properties because the inference itself might be part of the track record while the origin always remains the same. All this makes room for the possibility of lucky inferential knowledge—where luck transfers across competent inference while lack of knowledge does not. Here is a case where both the belief in the premises and the belief in the conclusion are lucky while the latter but not the former constitute knowledge: (f)  Suppose that p and q are truths in number theory. Suppose further that A only has lucky true beliefs in p and in q and does not know them. Still, A starts to use p as well as q as premises for further inferences. The chain of inferences gets longer and more complicated. After a large number of (correct) steps and a lot of time thinking things through, A reaches an extremely interesting conclusion C. A has managed to prove an important theorem in number theory. In the meantime, A has forgotten about p and q even though the whole proof rests on them. A still truly believes the conclusion C.14 Should we then still say that A does not know the conclusion of his proof (and only knows what the conclusion is)? This is implausible. However, whether or not luck is always transmitted across inference, it seems in this case that the belief in the theorem is as lucky as belief in p or in q on which the whole chain of inferences rest. Again, it seems that we have to accept that there can be lucky knowledge. One might object that there are so many non-lucky steps in the chain of inferences that the subject is much less lucky (lucky to a much lesser degree) with respect to the conclusion than with respect to the premises p and q. Even if this is the case, it seems that the lucky reliance on the premises would still make the conclusion “lucky enough.” Here is another variation: (g)  Detective D makes a competent and correct inference from a great number of premises all of which she knows (that the deed was done after darkness, in the kitchen, with a knife, etc.) with the exception of one (only the psychoanalyst owns 14   If one thinks that a mathematical proof does not lead to knowledge (or even true belief) in the conclusion, then one can easily change the example without loss of substance. For a related but still significantly different case see Hetherington 2011, 91.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  77 these kinds of knives) which she only believes (truly) by luck. Nonetheless, D comes to truly believe that the shrink did it. D is lucky in her true belief but it seems very counterintuitive that such a bit of luck can destroy knowledge of the conclusion. One might object that a single lucky true belief in one of the many premises of an inference does not need to turn the true belief in the conclusion into a lucky one. Haven’t I myself raised at least some doubt about the idea above that luck always transfers across competent inference (TL)? This might be true but there are still many cases where luck transfers (see above). Take case (g) and add (if that helps to raise the plausibility of my point) that the detective is really extraordinarily able in noticing the evidence, in weighing it and putting it together into one coherent overall picture. Few others would have been able to make the relevant inference and arrive at the true conclusion. Given that our judgments about knowledge are focused on the ability and track record of the subject, we should say that the detective comes to know the conclusion. For the judgment about luck, however, the factor of origin is salient and decisive. Given that one belief in one of the many premises was lucky but without it there would not have been no inference to the true conclusion, we have to say that the belief in the conclusion is also lucky (even if it constitutes knowledge). Does it matter how much luck there is in the different beliefs in the different premises? Does a lucky belief in one unimportant premise necessarily make the belief in the conclusion of the whole inference lucky (against TL)? I don’t need to rule the possibility of such cases out categorically. What matters here is that case (g) is different and constitutes a case where belief in the conclusion is lucky: we may stress that the fact that only the shrink owns the relevant kinds of knives is quite important for the solution of the case and cannot be pushed into the background as unimportant. One might want to reply to all this in addition by introducing a notion of degrees of luck here and modify the No-Luck-Thesis along the following lines: (NLT**) If S knows that p, then S’s belief that p is not true by a degree of resultant epistemic luck greater than threshold value t. There are several problems with such a proposal. First, it is doubtful whether one can measure degrees of luck in a non-arbitrary way: Should one, in the above example, just go by relative numbers of premises? Or also weigh the importance of the different premises? But how? Second, how would one determine the threshold value t in a non-circular way, that is, without saying something along the lines of “t has to be just high enough to allow for knowledge”? Finally, the move to (NLT**) changes the nature of the thesis substantially. I myself am fine with the idea that knowledge is compatible with luck but as the remarks below show I don’t spell this out in terms of acceptable degrees of luck. The typical defenders of the No-Luck-Thesis, however, won’t want to have anything to do with the kind of “softening” of it that we see in (NLT**).15   Thanks to Darrell Rowbottom here.

15

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

78  THE ARGUMENT FROM LUCK Consider other kinds of knowledge and this version of Goldman’s well-known thermometer example (see Goldman 1986, 45): (h)  A wants to take B’s temperature and is lucky enough to pick the only reliable thermometer from a basket of 100 thermometers (99 of which are, unbeknownst to A, not working properly). A takes B’s temperature and discovers that B has got a temperature of 104°F. Goldman and almost everybody else would judge that A does not know but only has a lucky true belief that the temperature is 104°F. But is that true? Aren’t we invited to take as a given in (h) that A has picked a working thermometer and ignore the chance and possibility of picking a defect one? But if we do that, then we should rather say that A only suffers from the unproblematic circumstantial epistemic luck, not from the problematic (for knowledge) resultant epistemic luck. In other words, we should then rather say that A’s temperature-belief is not lucky. We might even want to count A as knowing the temperature. I don’t want to deny that one can indeed construct a case like this or describe our case in this way. The crucial point here is that nothing forces us to characterize A’s method of belief-­ acquisition as “measuring the temperature by using a working thermometer”; we might as well characterize it as “measuring the temperature by randomly choosing and using one of the thermometers in the basket” (see Chapter 2 on the generality problem). If we do the latter, then A must be characterized as suffering indeed from resultant epistemic luck, the kind of luck relevant here. What counts as circumstantial or resultant luck is thus also dependent on and relative to the description of the relevant case (here (h)). There is no fact of the matter that would determine whether a given case is a case of this or that type of luck (for more on this see below). In (h) we do have a lack of knowledge and a presence of luck. But what if we imagine the following continuation of the case? (i)  A is so concerned about B that he decides to get B to the hospital where they do further tests over the next few days and finally discover that B is suffering from a rare viral infection. It is still early enough to treat B successfully. Five days later, B starts to undergo extensive treatments. Two months later, B is released from the hospital. Another two weeks later, B feels normal again. The next weekend, he throws a party for his friends. He tells them, again and again, that “had A not known about my temperature, I would be dead now!” If we add this to the original story, the inclination to deny knowledge to A about B’s temperature weakens considerably; I think we would even attribute knowledge to A— even though the element of luck has not disappeared. What is the additional factor which makes the difference to our judgments?16 Whatever our answer will be here 16   Sure, A might infer from the fact that B has a viral infection that the temperature must have been high (perhaps even that it must have been 104°F). However, our example also works without this additional

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  79 (e.g., a focus on later important consequences of the lucky true belief or on later important actions of the subject and other agents) it is plausible to accept this as another case of lucky knowledge, against Goldman and many others (see Greco 2006 and 2010, 132–4 for a nice set of cases illustrating analogous points about moral luck). I won’t go into any detail here concerning the explanation of how in all the above cases (f)–(h) knowledge and the absence of luck can come apart. The explanation offered for the Russellian cases above also works here. Rather, I now want to move from the discussion of cases using an intuitive notion of luck to more theoretical approaches to luck. It is important to stress that the result of the discussion of cases is provisional to some degree. Semantic “intuitions” concerning the correct use of a term like “luck” and “knowledge” can only have the first but not the last word (as Austin used to stress with respect to ordinary language in general: see Austin 1970, 185). Apart from that, there is always the danger of bad “intuition pumping” (see Dennett 1984, ch. 1): of making one’s intuitions about luck and knowledge swing together or not, depending on antecedent epistemological views. For all these reasons, theoretical reflection needs to complement intuitive judgments. To this I turn now. It will turn out that our intuitive notion of luck needs to be clarified and reconstructed to a certain extent in order to be useful for judgments concerning the relation between knowledge and luck. As we will see, the incompatibility of knowledge and luck will be denied again. I will first go into modal views of luck.

3.4  Modal Luck Accounts of knowledge which explain the difference between mere true belief and knowledge will often (but certainly not necessarily) also suggest an explanation of what distinguishes knowledge from accidentally true belief and of what constitutes luck or accidentality. One example is the defeasibility account of knowledge: proponents of it could identify luck with the absence of undefeasible justification (see Lehrer and Paxson 1969). There will then be as many or almost as many different conceptions of luck as accounts of knowledge; typically, one is the shadow of the other (see, e.g., Goldberg 2015, sec.3). Few authors, though, have bothered to give an explicit account of luck in terms of their theory of knowledge. Modal accounts are an exception (see also Goldberg 2015, sec.1 for a quite different discussion of modal accounts). They explain luck in terms of what happens in close possible worlds: in terms of what could have easily happened. A true belief is lucky in virtue of how things stand in close possible worlds: whether it could have easily failed to be true. There are currently two main versions of this view. According to “sensitivity”

assumption. Even without this kind of inference and even before he finds out about the viral infection, we would still credit A with knowledge (in case (i)). Thanks to Darrell Rowbottom for pressing me on this point.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

80  THE ARGUMENT FROM LUCK views (modeled after sensitivity accounts of knowledge; see Nozick 1981, 172–96) the following holds: (Sensitivity-Luck)  A true belief is lucky just in case the person holds the belief in (all, most or many) close possible worlds in which it is false or the person does not hold the belief in (all, most, or many) close possible worlds in which the belief is true. According to a “safety” view (modeled after safety accounts of knowledge; see Sosa 1999 and, with much explicit reference to epistemic luck, Pritchard 2005c and also 2007) the following holds: (Safety-Luck)  A true belief is lucky just in case the person’s belief is false in (all, most, or many) close possible worlds in which the person holds the belief.17 Because counterfactuals don’t contrapose, the two views are not even partly equivalent. Further details would have to be added to both accounts (e.g., a relativization to ways of belief formation) but we can skip these complications here. One problem with both of these modal accounts is that there is no unique acceptable closeness ranking for possible worlds (see Chapter  2 for details). Given different and equally legitimate closeness rankings, we might end up with different but equally legitimate judgments (if we adhere to modal notions of luck) about the presence or absence of luck. And hence, if we adhere to the No-Luck-Thesis, we may also end up with different judgments about the presence or absence of knowledge. All this suggests a hidden relativity in the concept of luck (luck is relative to a given closeness ranking) or a hidden ­context-sensitivity (“luck” has application conditions which vary with the closeness ranking given by the speaker’s context). This context-sensitivity would then be “inherited” in many cases by “knowledge.” Most if not all of those who favor modal accounts of luck and knowledge (e.g., Nozick 1981; Sosa 1999; Williamson 2000; Pritchard 2005c) would resist this kind of move but I think considerations like this one give good support for contextualism. I won’t go into any details here because there are even more interesting problems for both accounts insofar as they are used in the defense of the No-Luck-Thesis. Here is a case which can be used to raise question about the relation between SafetyLuck and knowledge: (k)  It is 5 past midnight on January 1, 2001. Jack just finished his first letter ever to his old friend Jill. Jack knows the time and date and comes to believe that he finished his first letter to Jill in the 21st century. It seems clear that Jack knows that he finished the first letter to Jill in the 21st century. Does it matter that he also adheres to the very popular false belief that the 21st century began on January 1, 2000? If this were to show that Jack does not know this, 17   Here is Pritchard’s 2005c, 163 formulation of the safety condition for knowledge (or, mutatis mutandis, for the absence of luck): “For all agents, ϕ, if an agent knows a contingent proposition ϕ, then, in nearly all (if not all) nearby possible worlds in which she forms her belief about ϕ in the same way as she forms her belief in the actual world, that agent only believes that ϕ when ϕ is true.”

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  81 then many people, perhaps the majority, would turn out not knowing that they were in the 21st century when it began. This, however, seems clearly false (see Baumann 2015b and also Baumann 2008a; for more examples and objections against the idea that safety is necessary for knowledge see Gundersen  2003, 118–19; Neta/Rohrbaugh 2004, 399–400; Roush 2005, 118–26; and Comesaña 2005, 397; see also Sosa 2003a, 159 as well as Goldberg 2015, sec.1). However, if we assume (plausibly) that there is a close world in which Jack finishes his letter eight minutes earlier, then we also have to admit that Jack’s true belief in the actual world is lucky in the sense of (Safety-Luck). Given his calendaric confusion, Jack would have the false belief that he is already in the 21st century when he finishes the letter eight minutes earlier. Hence, if we adhere to a safety account of luck, we still have to admit that there are cases of lucky knowledge (given at least some admissible closeness rankings for possible worlds).18 It won’t help much to object that this won’t follow if one accepts a safety account of knowledge according to which knowledge just is true belief which is not lucky in the sense of (Safety-Luck). Even though case (k) could thus be characterized as a case of lack of knowledge and presence of luck and the No-Luck-Thesis thus be upheld (for the defense of a No-Luck-Thesis by a safety account of knowledge see, e.g., Pritchard 2005c), the damage done to the account would still be substantial: Case (k) is a good counterexample against the safety account of knowledge (for much more on this see Baumann 2008a) which would presumably take (Safety-Luck) down with it. (SafetyLuck) does thus have much better chances of survival if treated in isolation from a safety account of knowledge—as being done here. Case (k) also works against the use of (Sensitivity-Luck) here. A close world in which Jack did not finish his first letter to Jill in the 21st century is the world in which he finishes it eight minutes earlier. But in that world Jack would, given his calendaric confusion, still believe that he finished his first letter to Jill in the 21st century. So, Jack knows (at least according to ordinary judgments) that he finished his first letter to Jill in the 21st century but he is also lucky in the sense of (Sensitivity-Luck). Like in the case of (Safety-Luck) and for analogous reasons, it won’t help to tie (Sensitivity-Luck) to a sensitivity account of knowledge according to which knowledge just is true belief which is not lucky in the sense of (Sensitivity-Luck) (for problems of sensitivity accounts of knowledge see Luper-Foy  1987 or Becker and Black  2012; see also Baumann 2012a). By the way, one can easily modify case (k) to get a case which works against the second part of (Sensitivity-Luck): (k*)  It is 5 to midnight on December 31, 1999. Jack just finished his first letter ever to his old friend Jill. Jack knows the time and date and comes to believe that he finished his first letter to Jill in the 20th century. 18   (Safety-Luck) leaves it open whether the relevant belief of the subject has to be true in all close possible worlds. However, this technical issue can relatively easily be dealt with by certain modifications of case (k). I won’t go into this here.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

82  THE ARGUMENT FROM LUCK Eight minutes later he would still have been in the 20th century but believed to be in the 21st. Again, we have (sensitivity-)lucky knowledge in (k*). One might object to all this and point out that the relevant modal accounts of knowledge could be further developed so as to be able to accommodate the above difficulties. I have my doubts about the prospects but cannot go into this here; it is sufficient for me here to show that the basic version of a safety account of luck or of a sensitivity account of luck allows for lucky knowledge.19 Is all that really so bad for the modal theorist? One could give up on any sensitivityor safety-based defense of the No-Luck-Thesis, turn around, deny it, and explain how there can be lucky knowledge while holding onto (Sensitivity-Luck) or (SafetyLuck). One could argue that there can be different and equally acceptable closeness rankings some of which are more relevant to judgments about luck and some more relevant to judgments about knowledge. Hence, luck and knowledge can come apart. The defender of a modal view on luck would just have to turn contextualist and accept that closeness rankings vary with attributor contexts. Given that different closeness rankings determine different truth conditions for attributions of knowledge or luck, we are back with contextualism, namely modal contextualism. Not that this is a particularly popular view, on the contrary. But it makes a lot of sense of the data. In the end I prefer a different, probabilistic account of luck. The reasons mentioned in Chapter 2 for why a probabilistic account of reliability is preferable to a modal one also apply here: modal accounts are stuck with ordinal rankings, are not explicit about the relevant context-sensitivity, and are less agreeable to naturalist preferences. Finally, one might wonder whether (Safety-Luck) or (Sensitivity-Luck) correctly identify necessary or sufficient conditions of luck. Do they confirm our “intuitive” judgments about luck? The answer does not matter that much, given that we’re looking at theoretical reconstructions of the notion of luck here where it is not that important whether they match with an ordinary notion of luck. Suffice it to point at cases here which at least prima facie suggest negative answers. Jack’s belief in the letter case above is neither safe nor sensitive but, according to ordinary intuitions (though not according to the safety account or the sensitivity account), might not count as lucky and might count as a case of knowledge: this is absence of safety combined with the absence of luck or the absence of sensitivity combined with the absence of luck. Conversely, it seems that one can be both safe and lucky in the intuitive sense or sensitive and lucky in the intuitive sense: suppose I wonder who the winner of the national poetry contest will be. Unbeknownst to me, the jury always makes sure it will be the president. I take a lucky but safe and sensitive guess that it will be the president. More could be said about 19   Would it help to replace the safety or sensitivity requirement for belief by one for methods (see, e.g., Luper 2012a and Pritchard 2012)? One would have to find out more about what a safe or sensitive method is. Even given such an account, I doubt that the move from belief to methods will help much and won’t raise the same type of worries as the ones above. However, this is not the place to go into this.

OUP CORRECTED PROOF – FINAL, 09/15/2016, SPi

THE ARGUMENT FROM LUCK  83 such cases (see Lackey 2008, 260–6 as well as Lackey 2006 for a nice discussion along similar lines) from both sides but I won’t go into it here since, as pointed out before, complete congruence of our theoretic notion of luck with the ordinary one isn’t the goal here.

3.5  Probabilistic Luck Probabilistic accounts of luck do not rely on modal notions (presupposing a non-modal notion of probability, that is, a notion of probability which does not rely on modal logic but rather on the probability calculus). An event is lucky, one could say, just in case it is highly improbable that it would happen (where “improbable” and other terms are not understood in the subjective or the epistemic (see Stoutenberg 2015) sense). In the case of the acquisition of a true belief about some topic (t), this gives us the following formula: P(T) < a for a threshold value a typically if not always < 1.20 This, however, is not very satisfying as it stands: it is not easy to make sense of unconditional probabilities like the probability that someone acquired a true belief about something. What, for instance is the probability of rain here and now? It is hard to make sense of this if we don’t indicate conditions: given that it is August and we’re in Sicily, the probability is very low; given that it is December and we’re in Glasgow, the probability is much higher. If one is not convinced by this general skepticism about unconditional probabilities (but cf. Hájek 2007), then one can just restrict the discussion to the epistemic case. Is there an unconditional probability that I acquire a true belief about the presence of butter in the fridge? We should rather say that there is a probability that I acquire such true beliefs, given, say, that I look, etc. Only conditional probabilities seem to make sense here. What we need then is the notion of conditional probability, also for a probabilistic notion of lucky true belief. A true belief that p is lucky just in case the probability that the person would acquire a true belief on the question whetherp, given certain circumstances and ways of inquiry, pieces of evidence, etc., is below a threshold value a (again taking “probability” in a non-subjective or epistemic sense here). For instance, Julie’s true belief that there is an Air France airplane flying by is lucky insofar as the probability of getting this right, given that she just looks at the sky, is very low. At the same time, the probability might be quite high, given that she has just taken her FarSight eyedrops ( see also Chapter 2). We can say then that (Probability-Luck)  A true belief that p is lucky just in case the conditional probability that the person acquires a true belief about whether p(t), given that she is in   Here and elsewhere below I am not committed to “ 3) analysis of “knowledge” that there is also a plurality of standards, reference classes, and descriptions. A ternary analysis avoids the inflation problem. If it arises at all, it seems to concern rather a binary version of contextualism which assumes a ­plurality of different knowledge relations. However, I don’t think that the inflation problem is a serious one even for these latter views (but won’t go into this here).28 But suppose we consider all contexts C1, C2, . . . Cn (assuming there is a finite number29). Consider the corresponding epistemic relations knowledge-c1, knowledge-c2, . . . knowledge-cn where some relation knowledge-cx is what the word “knowledge” denotes in Cx. Take my relation to the fact that three apples are on the table. According to the view proposed here, I would, for instance, stand (and fail to stand) in lots of such epistemic relations with respect to such a proposition. I know by very low standards that there are three apples on the table, I also know it by lay standards, I know it by the standards of someone who’s just looked up “apple” on Wikipedia, finding a picture of an apple that looks like each of my three apples, and so on. I don’t know it by expert standards or even by the standards of someone who’s asked the seller relevant questions about the apples. Isn’t that a problem? Again, this worry is intuitively appealing but not easy to hold up. Why should it be a problem about standing in all these kinds of relations at the same time? Suppose I have 100,000 hairs on my head. This makes 100,000 different relations between me and particular hairs: having one here, having one there, having one over there to the left, and so on. There is no chance that I would ever think or talk about even a small fraction of all these different relations. But that does not constitute a problem. Similarly, people do really stand in lots of epistemic relations but only few of them are focused upon by us in thought and speech. Again, it is hard to see why this should bother us at the end of the day.30 28   On the related question whether contextualism changes the topic from knowledge to “knowledge” and thus loses epistemological relevance see Sosa 2011, ch.5 and Blome-Tillmann 2014, 49–52. 29   Can we make that assumption? If not, things would just get a bit more complicated but nothing substantial would change. 30   One might wonder whether the inflation problem also creates a problem for any closure principle: if there are always undemanding contexts in which the subject can correctly be said to “know” any proposition truly believed by her, then closure principles are true only in a trivial way; the subject “knows” the entailed proposition anyway. However, the closure principle used here is about what one comes to know on the basis

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

OBJECTIONS  193 Closely related to the inflation-problem is another issue. Keith DeRose raised the question how low standards can get (see DeRose 2009, 13–18) and suggested that in some contexts mere true belief might count as falling under “knowledge.” Contextual variability seems to stop with the truth condition and the belief condition. It is very hard to imagine how there could be a context in which a false belief could be truly called “knowledge” or a context in which a subject could be correctly said not to “believe” a given proposition but to “know” it. So, perhaps true belief is the non-­ contextual minimum basis for contextualists. On the other side, there are, according to contextualists, contexts (e.g., the famous “skeptical” ones) in which all (or almost all) knowledge attributions are false. To put it roughly: according to contextualism, knowledge can be extremely cheap (when true belief is sufficient) and ignorance, too, can be extremely cheap (when nothing is enough). One could call this the problem of “cheap states” (cheap knowledge and cheap ignorance).31 Each true belief falls in the rubric “knowledge” in some context, and no true belief falls in it in some (other) context. Again, this is a correct observation but it does not constitute a worrisome but rather an interesting discovery. That both knowledge and ignorance are or can be cheap (what “knowledge” and what “ignorance” denote can be “cheap”) is perhaps only surprising if one already holds invariantist views. On reflection it is hard to see why cheapness should constitute a problem.

7.5.2  Normativity and Arbitrariness According to the epistemic contextualist, there are many different (not just two) contexts of the use of a word like “knowledge” and they are all created equal. A word like “knowledge” can be used to mean many different things (though not wildly different things). One might want to raise two questions about this: Aren’t some contexts just “crazy”? And isn’t there really something like “the right context” (or at least a group of “right contexts”)? The first question is in need of further clarification: How could a context of use of a word (or a concept) be crazy? What could be meant here is, rather, that some uses of the word “knowledge” are “crazy” in the sense that they are not acceptable. This is correct and uncontroversial. For instance, to call a false belief “knowledge” is incorrect and there is no context in which a false belief could correctly be called “knowledge,” given the relevant facts about the English language. Contextualists have no problem at all with this. But in what other sense could a contextually sanctioned use of the word “knowledge” be crazy? A speaker might pick unusual parameters—surprising types of standards, reference classes, or descriptions. But this can be “crazy” only in the sense that others might have a hard time understanding him of competent inference. So, it is not trivially true. Apart from that, it can also be developed more fully. For instance, one could add that the context in which it is true to say of the subject that they “know” the entailed proposition need not be a minimally demanding one. 31   One can also call it the problem of “impossible” (overly “expensive”) knowledge and “almost impossible” (overly “inexpensive”) ignorance.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

194  OBJECTIONS or that it might be pragmatically more advisable to choose different parameters. Someone who answers 50 percent of the yes-no questions of a test correctly might count as “knowing” the correct ones of his answers (in a very relaxed context) but it would be unwise to let him pass a driving test on such a basis. It is our practical interests, purposes, and intentions (apart from other determinants like norms and conventions, foci of attention, or even character traits) that not only determine parameters but also give us reasons, pragmatic reasons, to choose the parameters we pick. This, however, does not mean that there is something semantically “crazy” in such cases of use of a word like “knowledge” or that the corresponding contexts are themselves “crazy” in some interesting sense. This leads to the second question: Isn’t there a right context?32 Isn’t there a good reason—say, based on the meaning of the word “knowledge”—to pick one context or one way of using the word? Some authors argue that contextualism is problematic because it does not or cannot account for the normative dimension of some standards or parameters or contexts being the right ones (see Davis 2004, sec.6, 2015, sec.11; Fogelin 2000, 44). The quarrel between the skeptic and the anti-skeptic, for instance, can be described as a quarrel about what the correct context is: the demanding or the relaxed one. It is interesting to see how hard it is even for classical invariantists to come up with good normative reasons (apart from pragmatic reasons) for selecting one or the other context. The contextualist would explain this by saying that asking for (non-pragmatic) reasons here is a mistake (see for this kind of move more generally: Wittgenstein 1969). The only reasons that there are are pragmatic ones (in a wide sense of the word): given our interests and plans (plus some other factors) it can be more reasonable or well advised to choose one context or use of a word over others. But more there isn’t. There is thus an irreducible but also unproblematic arbitrariness, if one wants to call it that. The semantics of “know” does not specific a unique use of the word. Non-contextualists and classical invariantists will find this unsatisfying but that might also be due to the fact that they have antecedent non-contextualist expectations. Apart from that, as long as there is such deep and radical disagreement amongst invariantists about “the right context,” the lack of normativity in contextualism has some plausibility (this does not mean that the disagreement of others could be used as an argument for one’s own position). Can this essential lack of “principled” reasons and the corresponding presence of arbitrariness survive awareness and reflection? Suppose someone notices the (alleged) context-sensitivity of epistemic terms. Such a person could, for instance, then say something like I know that these are apples but under different circumstances I would not call it “knowledge” 32   For the sake of simplicity I am leaving aside the possibility that there could be a group of right contexts (it is not so clear anyway how we are supposed to count contexts).

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

OBJECTIONS  195 or I don’t know that these are apples but under different circumstances I would call it “knowledge.” Such sentences may sound odd and somewhat resemble Moore-paradoxical utterances. In contrast to the latter, however, there is nothing specific about the first person singular indicative present here (as one can easily see by making the relevant substitutions of terms). Furthermore and more importantly, all this does not constitute a problem; the remarks above (see, e.g., Section 7.3) describe the resources to explain why there is nothing puzzling here. One just has to choose less misleading formulations and make the different variations of the knowledge relation explicit: “As a lay person I can say that I know that these are apples but don’t ask me as the expert which I’m not.”33 What if the speaker said something like I know that these are apples but there is no reason to call this “knowledge” or I don’t know that these are apples but there is no reason not to call this “knowledge,” perhaps even adding “and all that is completely arbitrary!”? There is even less of a puzzle here. It is a truism that what words we use is arbitrary, in a completely innocent way. Some philosophers hold that there are relatively straightforward epistemic, knowledge-related norms about what one can properly assert, believe, or act upon, like the rule that one should only assert what one knows, or that what one knows is what one ought to believe and act upon (see Unger 1975, ch.6 and passim; Williamson 2000, ch.11; Williamson 2005c, 231; Stanley 2005, 9 and passim; Hawthorne 2004a, passim). I have doubts about any such rule but won’t go into it here. Suffice it to mention that acceptance of such rules is compatible with contextualism, as long as one also holds that what is acceptable for assertion, belief, and practical reasoning and action varies across contexts with the applicability of the knowledge predicate (see, e.g., for the case of assertion in particular DeRose 2009, chs.3, 7).

7.5.3  Error, Blindness, and the Possibility of Communication One of the most widely discussed objections against contextualism in general says that contextualists have to attribute systematic and broad error to ordinary speakers and thinkers, given that they are not aware of any context-sensitivity of “knowledge” and related terms (see Schiffer  1996; Hofweber  1999; Neta  2003b; Bach  2005, sec.1; Brown 2005b; Cohen 2005a, 60–2; Conee 2005a, 54–5; DeRose 2006; Blome-Tillmann 2008, sec.3–5, 2014, 96–111; Montminy 2009a, 2009b; and Abath 2012 in response; 33   There is much more of a problem for SSI (if there is a problem at all; see Stanley 2005, 113–14). Both “I know that these are apples but if more was at stake I wouldn’t know it” and “I don’t know that these are apples but if less was at stake I would know it” sound odd.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

196  OBJECTIONS Adler 2012, sec.7; Davis 2015, sec.9; see also Schaffer and Szabó 2014, 533–5). Ordinary speakers just don’t talk like contextualists. This objection weighs particularly heavily against those contextualists who make their case centrally on the basis of data about what we would say about cases (see, e.g., DeRose 2009, ch.2). As I have pointed out repeatedly above, ordinary attributors are indeed pretty much aware of the relevant context-sensitivities. They just don’t express it in epistemological vocabulary (which is certainly not surprising). Rather, they use a variety of other linguistic tools to that purpose (see Ludlow 2005, sec.5.2). It would be too much to ask that ordinary attributors are very explicit and clear about the context-sensitivity of epistemic terms; it suffices that their thought and talk expresses the relevant sensitivity to context. That one doesn’t have perfect sight or even problems seeing certain things does not mean that one is blind. The alternative between blindness to and explicit acceptance of context-sensitivity is overly simplistic; there is much in between (see for this also Horgan and Potrč 2013). Apart from all that there is also a tu quoque here (see also DeRose 2009, 159–60, 174–9). The classical invariantist who uses WAMs to explain ordinary verdicts on contextualist cases helps himself to a distinction between truth and warranted assertibility. If ordinary thinkers and speakers are not aware of the context-sensitivity of “knowledge,” then they will be even more unaware of the difference between truth and warranted assertibility. Adherents of SSI claim that knowledge depends on and varies with practical interests. I don’t think that this claim is more at home in ordinary ways of thinking and speaking than contextualist ideas. This just as an aside—one should not put much weight on tu quoque-arguments.34 In ordinary communication there is certainly a lot of misunderstanding of and ­perhaps even some blindness to what other speakers have in mind when they attribute or deny knowledge to some subject. That two or more people share a conversational context does not mean that they also share an epistemic context (but see, e.g., DeRose 2004a). It would be surprising if they did; it can happen way too easily that the different participants of a conversation pick different descriptions, reference classes, and standards because they’re under the influence of different constellations of determinants. Only when the divergence becomes too great or too salient might a conversation about their conversation begin. Most cases of ordinary communication (say, about “knowledge”) will be such that the conversants differ a bit, usually without noticing it, in their contextual parameters but not radically; if the latter should happen, then the chances of discovery are often good. That communication across different contexts is impossible, given contextualism, (see, e.g., Cappelen and Lepore 2005, ch.8) is not plausible if one takes ordinary people’s ability into account to flag up context-sensitivity to each other—and if one also acknowledges that in normal cases of communication there is 34   One might wonder why the need for an error theory is never seen as an objection against skepticism. If skepticism is true, then almost all speakers are systematically wrong in their knowledge attributions. If that is not an objection against skepticism, then how can it be an objection against, say, contextualism that it needs an error theory?

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

OBJECTIONS  197 always a certain amount of content lost in transition (see, e.g., Davidson 1986). Since I have dealt with these kinds of questions concerning error and blindness above, I’ll leave it at that.

7.5.4  One Last Smaller Problem Suppose Al is in a low stakes context and that he can truly say in this context that “Frank knows that the deal is going through.” Suddenly, the situation develops in such a way that the deal is affecting Al importantly just in case it is true to say in Al’s context (whatever it is) that Frank knows about it (Frank will then make sure that Al gets involved). So, given that it is true for Al to say “Frank knows that the deal is going through,” the stakes are high for Al and it can thus be false for Al in his context to say “Frank knows that the deal is going through” and true to say “Frank doesn’t know that the deal is going through.” But then the stakes are low for Al and it is true to say for Al “Frank knows the deal is going through.” In other words, if Al is in a low context regarding The deal is going through, then Al’s utterance of “Frank knows that the deal is going through” is true: but if it is true, then Al is in a high context regarding The deal is going through and his utterance is false (or not true). But if it is false (or not true), then Al is in a low context and his utterance is true. “Frank knows the deal is going through” is true iff it is false. Or: it is true iff it is not true. Either there is no stable truth value for Al’s attribution or we have an outright contradiction. Similar problems arise if we use other contextual parameters than stakes. There does not seem to be a way out of this difficulty. The big consolation is that this is a rare, deviant, convoluted kind of case (also see Section 8.1). So much on different objections to and problems for contextualism. It comes out pretty well. However, it is also important to look at alternative theories. If some of them do even better, then one should move away from contextualism and rather accept such an alternative view.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

8 Alternatives? The version of contextualism proposed here (see, e.g., the beginning of Chapter 7) is supported to some degree by considerations of cases and, more importantly, by more “theoretical” considerations (see Chapters 1–3). It allows for promising responses to philosophical problems like the lottery problem (see Chapter 4), meets adequacy conditions for theories of knowledge (see Chapter  5), and can also be extended to non-epistemic, e.g., moral notions (see Chapter 6). It also survives a range of objections pretty well (see Chapter 7). I think that the form of contextualism defended here is more attractive than alternative contextualist views. However, couldn’t there be non-contextualist alternatives that do (even) better? The main alternative to think about here is not so much classical invariantism: the arguments for contextualism presented here are at the same time arguments against classical invariantism. Rather, the more interesting alternatives are “closer to home.” There are three important ones that shall be discussed here. One is contrastivism which is very close to contextualism, especially to the relationalist version proposed here; it allows both for contextualist and non-contextualist versions (more on this in Section 8.2). Another one is relativism (Section 8.3). The third view is SSI. Let us start with it and see how well it (and any of the other views) does in comparison with contextualism.

8.1  Subject-Sensitive Invariantism Subject-sensitive invariantism, SSI (see Fantl and McGrath 2002 and 2009; Hawthorne 2004a; Stanley 2005; also see Weatherson 2011, 2012b), agrees with contextualism that the truth value of knowledge ascriptions varies with factors other than the epistemic situation of the subject (whatever this is taken to involve—possession of evidence, reliable belief formation, etc.); however, in contrast to contextualism SSI locates this variation in subject factors and not in attributor factors; the variation of truth values would thus not be due to a variation of truth conditions. Adherents of SSI mainly but not exclusively focus on what is practically at stake for the subject (Schaffer 2006 focuses on this aspect very much in his critique of SSI). High stakes produce high standards for knowledge and low stakes low standards. It is harder (easier) to know a particular proposition (e.g., that it has not started to rain) when the stakes for the subject are higher (lower), that is, when the expected costs of acting on that proposition if

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  199 it is false are greater (smaller) (e.g., my precious new silk socks being ruined in the rain). Stanley (2005) proposes to call this view “Interest-Relative Invariantism” but SSI seems more appropriate since other factors than practical stakes or interests may also play a role. More generally, one can characterize SSI as the view that whether a subject S knows some proposition p depends upon whether S’s true belief that p has some epistemic property (possession of evidence, reliable belief formation, etc.) where the threshold for how much of that property is necessary for knowledge depends on and varies with non-epistemic factors, like the practical stakes for the subject.

8.1.1  Some Questions about Stakes How should one understand the notion of stakes or practical stakes? Here is the first step towards a plausible and hopefully uncontroversial proposal. What is at stake for a subject S with respect to a proposition p depends on how much worse off the subject would be if they acted on p when p is false than if they acted on p when p is true, and on how great the probability that not p is.1 One possible measure for the stakes of a proposition p for a subject S is the following one (taking “U(x)” for “the utility of x” with 0 ≤ U (x) ≤ 1, and “Stakes(p)” for “the stakes of p”): Stakes(p) = [U(acting on p when p is true)—U(acting on p when p is false)] x P(not p). This measure gives us values ≥ 0 for the kinds of cases just mentioned. However, we should not neglect the possibility that the subject could also be better off if they acted on p when p is false than if they acted on p when p is true. One can call this the case of “positive stakes” in contrast to the above one of “negative stakes” (this has nothing to do with the stakes having positive or negative values). Suppose Humphrey believes truly that his old love Isa won’t come today; a lot of what he knows speaks against that. Does he know she won’t come today, according to SSI? Suppose it were really great for Humphrey if she came today and suppose there is indeed a remarkable probability that she will come today. Wouldn’t an adherent of SSI then typically want to say that Humphrey doesn’t know that Isa won’t come today because the positive stakes of the relevant proposition are so high that the standards for knowledge are too high for Humphrey (he hasn’t checked who’s reserved rooms in the local hotels, etc.)? It is remarkable that it is or seems much easier to think of cases of high negative stakes than of cases of high positive stakes. However, both are possible—even if the discussion so far has very much if not exclusively focused on negative stakes. It seems very plausible on reflection that there is a relevant symmetry here between positive and negative stakes. 1   I trust that it is not problematic to rely on an intuitive understanding of “acting on” here. One acts on a proposition if one uses it as a premise in one’s practical reasoning. We are assuming here that nothing goes wrong with the agent’s practical reasoning. For the sake of simplicity, I am neglecting here that acting on some proposition when it is true (or false) can still lead to different outcomes with different utilities; one could take this into account by using the notion of the expected or average utility of an act, given the truth (or falsehood) of the proposition. However, this would unnecessarily complicate things here.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

200  ALTERNATIVES? We should thus include the possibility of positive stakes and also the possibility of neutral stakes or 0-stakes (where U(acting on p when p is true) = U(acting on p when p is false). We should thus extend the above informal explanation of the notion of stakes and say something like the following. What is at stake for a subject S with respect to a proposition p depends on how much worse or better off the subject would be if they acted on p when p is false than if they acted on p when p is true, and on how great the probability that not p is (again neglecting the possibility of different outcomes when acting on some proposition). A plausible measure for the stakes of a proposition p for a subject S which would give us values ≥ 0 for all cases would then be the following one (taking “|x|” for “the absolute value of x”): (Stakes) Stakes(p) = |U(acting on p when p is true)—U(acting on p when p is false)| x P(not p). Both “utility” and “probability” can be interpreted in more than one way: there are more “objective” and more “subjective” interpretations. Depending on these interpretations one gets different notions of stakes resulting from them. At this point, we don’t need to worry about these aspects. This explanation of stakes covers two kinds of cases: (a) cases where the act chosen when acting on p differs from the act chosen when acting on not p (because the agent realizes that the act chosen does not even weakly dominate each alternative act); (b) cases where the act chosen when acting on p is the same as the act chosen when acting on not p (because the act chosen at least weakly dominates each alternative act). Intuitively, the stakes don’t matter for knowledge in the latter kind of case. If Humphrey were to do the exact same thing when he assumes that Isa won’t come today than when he assumes that she will come today, then the stakes won’t be relevant—no matter how great they are—for the question whether Humphrey knows that Isa won’t come today. Hence, the defender of SSI should restrict the claim that stakes are relevant to knowledge to cases of type (a). The relation between the notion of stakes proposed here and the notion of expected utility is interesting. Considering two circumstances (p, not p) and two available acts (A, B), with exactly one outcome for each act in a circumstance (O1, O2, O3, O4; these represent utilities of outcomes), we can construct the following outcome matrix: Circumstances:→

p

not p

Acts:↓ A B

O1 O3

O2 O4

Suppose A is the act the agent would choose when acting on p (obviously, nothing depends on whether it is A or B). Then, according to (Stakes), Stakes(p) = |O1 – O2| x P(not p).

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  201 Given our assumptions, the expected utility EU of acting on p is the expected utility of A: EU(acting on p) = EU(A) = [O1 x (1 – P(not p))] + [O2 x P(not p)] = O1 – [O1 x P(not p)] + [O2 x P(not p)] If the stakes are negative (O1 > O2), then EU(acting on p) + Stakes(p) = O1 = the outcome of acting on p when p is true. If the stakes are positive (O2 > O1), then EU(acting on p)—Stakes(p) = O1 = the outcome of acting on p when p is true. If the stakes are neutral (O1 = O2), then Stakes(p) = 0 but EU(acting on p) = O1 = the outcome of acting on p when p is true (= O2 = the outcome of acting on p when not p is true). The same things hold, mutatis mutandis, for cases with more than two acts or more than two circumstances. So, the notion of stakes captured in (Stakes) is different from but also interestingly related to the notion or expected utility (and they might only be equal in very special cases).2 So far we have left open how to take the notion of practical interests or utility on the one hand and the notion of probability on the other hand. Starting with the first, one can distinguish between (i) a subjectivist notion of practical interests according to which a subject’s interests consist in what they want or think they want; (ii) an enlightened subjective notion of practical interests according to which a subject’s interests consist in what they come to want or think they want after due reflection; 2   Consider in contrast the brief hint in Stanley 2005, 94: “If whether or not a proposition is true has no effect (or only a minimal effect) on the warranted expected utilities of the actions at one’s disposal, then it is not a serious practical question whether or not that proposition is true.” Stanley specifies the idea a bit in the following: “That is, on this account, a proposition p is practically irrelevant if and only if, where a1 . . . an are the actions at my disposal, the differences between the warranted expected utilities of a1 . . . an relative to the nearest states of the world in which p are not meaningfully different from the differences between the warranted expected utilities of a1 to an relative to the nearest states of the world in which ~p” (Stanley 2005, 94–5). All this only gives a criterion for practical relevance or irrelevance, that is, for neutrality or non-neutrality (positive or negative) of stakes; it does not give a measure for non-neutral stakes like the proposal above and is thus not that useful. Apart from that, it is not clear why the other actions (those that the agent would not choose, given p) are relevant here (whether, as in the first quote, the utilities of their outcomes are concerned, or, as in the second quote, the differences between the utility differences of the outcomes of different acts in one and in the other circumstance). What is at stake concerning p seems determined only by what could happen if one acted as if p; this, however, selects exactly one act as the relevant one (A in our example in the text above). If one considers alternative acts, then one doesn’t consider only the stakes regarding p anymore but also the stakes of other propositions (those which would, if accepted, suggest an alternative act). One can, of course, compare the stakes of different propositions but that is a different topic. Finally, that there should be “no effect” (or only a small one) also seems too strong even for Stanley’s purposes: one can have neutral stakes at very different levels of utility.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

202  ALTERNATIVES? (iii) an objective notion of practical interests according to which a subject’s interests are determined by some matter of fact and is independent of what the subject wants or thinks they want.3 Interestingly, adherents of SSI have spent little time discussing these different options.4 However, the differences matter because different notions of practical interests lead to different judgments about knowledge for the adherent of stakes-based SSI. Consider Party Teenagers A, B, and C have heard that there will be a party later in the evening at Jay’s house but they are not certain about it. A has had a rough day and it would really (“objectively”) be great for her to go out even though she hasn’t thought at all about whether she would like to go and feels indifferent. In contrast, drop the idea of an objective interest in the case of B or C and assume the following. B has engaged in some good reasoning and come to want to go to the party (and think that she ought to) while C hasn’t thought much about it at all but just feels the intense wish to go and proclaims that it would be crucial for her not to miss this party.

Who, according to SSI, knows and who doesn’t know that there will be a party at Jay’s house later in the evening? It depends on what the subjects’ practical interests are. To be sure: the question is not what the correct conception of practical interests in general is but rather which conception is relevant to knowledge. I find it hard to tell. And it thus also seems that one can argue both ways (for or against knowledge) in all three cases. For the adherent of SSI the question is very complicated and seems genuinely open. Alternatively, the adherent of SSI could argue that all the different kinds of “interests” matter (perhaps equally, perhaps not equally). In that case one should expect the stakes to be higher when both objective and subjective interests are strong than when only one of them is strong. Again, I find it hard to choose one strategy for SSI over the other one. However, if the answers to all these questions are indeterminate or not well motivated, then one wonders what kind of work the notion of practical interests and of practical stakes can play for the adherent of SSI. There are also questions about the interpretation of “probability” (see Gillies 2000; Kyburg 1970, chs.3–7); and Resnik 1987, 61–79). Should the adherent of SSI have some kind of objective (classical, frequentist, logical, propensity, etc.) probability in mind or rather epistemic or subjective probability? Let us focus on subjective probabilities (degrees of belief) and objective chances. Consider the following two cases: High Credence-Low Chances Anne is hiking in the countryside of Faravaia. She just drank water from a little river. Suddenly she is getting concerned and her credence that the water is polluted goes up significantly 3   One can have doubts about the notion of an objective interest or of due reflection, etc. At this point, I can leave these doubts aside. I will also leave aside complications having to do with the relation between wanting that p and thinking that one wants that p. 4   One might hold that our judgment about cases like Ignorant High Stakes (see Section 7.4) presupposes an objective notion of practical interests but this is not so. What subjects are ignorant of in such cases is some relevant information but that is compatible with a subjective identification of interests.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  203 though her conviction that the water is fine is still stronger. However, the water is fine and there is only a very small chance that this river could carry polluted water. Low Credence–High Chances Kurt is hiking in the countryside of Outovaia. He just drank water from a little river. He is not concerned at all about the quality of the water and his expectation that the water might be polluted is very slim. However, the water in this river is often polluted and there is a high chance that he drank some polluted water. But Kurt is lucky: the water that he drank is fine.

How high are the stakes here? We may assume that the disutility of drinking polluted water is equally considerable for Anne and Kurt. But whose stakes are higher? And who knows that the water is safe and who doesn’t? It depends on the notion of probability relevant here. The question is not what the correct or adequate interpretation of probability is but rather what type of probability is relevant to stakes and knowledge. Again, I find it hard to tell whether Anne’s or Kurt’s stakes are higher. Perhaps both subjective credence and objective chance matter (equally or not). Then one should expect that the stakes for someone with a high degree of belief that the water is ­polluted when the objective chances of pollution are high should be higher (ceteris paribus) than the stakes of Anne or Kurt. Again, I find it hard to tell whether this is so. If the answers to all these questions are indeterminate or not well motivated, too, then again one should wonder what kind of role the notion of stakes can play for the adherent of SSI.5 It can, of course, also happen that a person is seriously conflicted about what matters to her. Consider George who has been encouraged to apply and just did apply for a promotion at work. George is seriously divided about this. Sometimes he badly wants to have a great professional career and sometimes he just couldn’t care less about “all the fuzz”—and sometimes he feels both pulled (or pushed) at the same time. What is his practical interest then? Is the question whether he handed his application in in time a “serious practical question” for him (see Stanley 2005, ch.5)? It seems incorrect to say (taking the average over George’s two tendencies) that it is a mildly serious practical question for him. He is not mildly interested at all. Rather, it seems that it is all either indifference or strong interest. But how can it be any of these? Rather, it seems that there is a genuine indeterminacy here as to whether the relevant question is a serious practical question for George. Accordingly, it would then also be indeterminate whether George knows or doesn’t know that he handed his application in in time. Sure, the notion of knowledge is vague in many ways and one might hold a view on vagueness according to which it is indeterminate in borderline cases whether the relevant predicate applies or not. However, the indeterminacy here would not be one of vagueness. This raises the question whether knowledge can be indeterminate in this way. If it can, then stakes-based SSI helps us to discover an interesting fact about knowledge. If it cannot, then we should have serious doubts about SSI.6   What matters for the contextualist are the subjective probabilities of the attributor.   The issue is, again, much less serious for the contextualist. The attributor goes with whatever attitude is prevalent at the moment of the attribution. 5 6

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

204  ALTERNATIVES? Finally, one might wonder how exactly stakes and epistemic standards are correlated, according to SSI. Part of the idea of stakes is that they are graded on some scale; they can be higher or lower. Assuming that both utilities or interests and probabilities can be measured somehow (not an uncontroversial assumption), we may assume that stakes can also be measured in some way and located on a scale of higher or lower stakes. But how much of a difference do increased or decreased stakes make for the epistemic standards? What degrees of stakes determine what thresholds for the degree to which epistemic standards must be met in order to make knowledge possible? What amounts of increases of stakes are sufficient in a given case for the loss of knowledge and what kinds of decreases are necessary in a given case for knowledge? It seems that nobody has a clue and perhaps one should not ask questions of that generality. But the problem remains if one takes on case by case or type of case by type of case. However, it might turn out that there is no principled answer to these questions. We simply might not have any way of measuring stakes to a degree of precision that would allow for the establishment of an informative enough correlation between stakes and standards. Perhaps the idea of a precise correlation between stakes and standards does not even make sense. Whatever the source of the problem, if there is no principled answer to the above questions, then it is plausible to assume that it is the attributor who determines the nature of the correlation between stakes and standards. But then we’re back with contextualism, with an additional argument in favor of it (and against SSI).

8.1.2  SSI on Factors and Cases (Again) According to SSI, there are non-epistemic factors which determine the thresholds of the epistemic standards, that is, which determine how much of the relevant epistemic property a true belief must have in order to qualify as knowledge. Practical stakes are one such factor (much emphasized by, e.g., Stanley  2005) but SSI should not be restricted to stakes. As the consideration of cases suggests (see Chapter 1) there are other factors, too. Hawthorne 2004a, 158, ch.4, sec.2–3 mentions salience of or attention to alternative possibilities in addition. However, it was argued above that the salience of chances of error is no plausible factor (and we can leave the factor of attention aside here). But there are still others, namely purposes and intentions on the one hand and norms and conventions on the other hand. Consider the following case again: Two Allergies Mary and John are on a plane. Mary has a nut allergy and John has an egg allergy. Mary asked the flight attendant twice and got confirmation that the food contains no nuts; John asked another flight attendant about eggs and got confirmation that there is no problem for him either. Both also then look at the flight information leaflet and take it from there, correctly, that the food contains nothing problematic for people with allergies. They both say to themselves, “Now I know that the food won’t cause any problems for people with allergies.”

If one objects that this is rather a case involving different subject’s stakes than different subject’s intentions and purposes, then one can easily change the example and assume

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  205 that Mary and John have no allergies but just intend for rather trivial reasons not to have certain food items (nuts, eggs) on the flight (see Chapter 1). It seems very plausible for the adherent of SSI to say that in Two Allergies Mary knows that the food served on board is not problematic for people with allergies because she passes her epistemic standards’ thresholds determined by her intentions and purposes. However, had she asked about eggs instead of nuts, then (given the same intentions and purposes) she would not know that the food served on board is fine for people with allergies; her purposes and intentions fixed thresholds which she would then not have passed. Similar things hold for John. Cases like Two Allergies suggest that the subject’s intentions and purposes would have to count as another factor for SSI, apart from the subject’s stakes. What about norms and conventions? Consider, again, this case: Amateur Genealogists Lineage and Genealogy (LaG) is a pretty ambitious association of amateur genealogists. At their meetings they discuss topics like “Who was Jack the Ripper’s maternal grandfather?” U. N. B. Daft is not a member but visiting today because he might be interested in becoming a member. He volunteers his true opinion that so and so was Jack the Ripper’s maternal grandfather and mentions that he’s read that in a popular biography. The members of LaG frown upon this, politely shake their heads and whisper amongst themselves that “This guy doesn’t know a thing about the Ripper family.” They then proceed to explain to Daft that they themselves make trips to different archives and check on different hypotheses using different kinds of sources before they even participate in a general discussion of cases at LaG: “We can’t help it but we just take research very seriously!”

It seems very plausible for the adherent of SSI to say that Daft does not know the relevant proposition because the norms and conventions that apply to him in his situation determine higher thresholds for the epistemic standards relevant to knowledge of that proposition. Had he done some research, he might have met the standards fixed by the situational norms and conventions. So, cases like this one suggest that one should accept norms and conventions as a third possible subject factor apart from stakes and purposes and intentions. An additional advantage of this wider set of f­ actors is that not all of them are tied to the idea that standards are graded on a hierarchical scale; with the additional factors one can account for standards that differ ­without being rankable on a hierarch­ ical scale. That there are such situations and standards was argued for in Chapter 1.7 However, for each of these three factors the contextualist explanation seems more plausible than the one suggested by SSI. Consider, again Weather Forecast Mary works as a weather forecaster for the local weather station. While on her lunch break with her friend Charlotte and her colleague Frank, she is discussing with them the possibility of 7   Like in the case of stakes and practical interests, there is an issue about conflicting purposes and intentions as well as about conflicting norms and conventions. The exact correlation between purposes and intentions on the one side and epistemic standards on the other side is an issue, too, like in the case of stakes. To avoid repetition, I won’t go into these parallel issues here.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

206  ALTERNATIVES? having dinner outside together later in the day. At one point, Mary looks at the sky and notices some dark clouds approaching quickly. She says “Oops, I think we can forget about dinner in the garden today: it is going to rain within the hour. I can see dark clouds approaching rapidly.” Mary is pretty good at making predictions like that. And, indeed, within the hour rain is falling where they are. Just half a minute after Mary spoke, both Charlotte’s and Frank’s cell phones ring. Charlotte receives a call from her husband who is asking about the dinner plans. Charlotte tells him that they have to call the whole thing off because of incoming rain. “Really?” asks the disappointed husband, “Does Mary know it’s going to rain?” “Yes,” answers Charlotte, “she knows that.” Frank’s call is by the station manager who wants to know whether Mary or he already know whether it is going to rain later in the day. Frank replies “No, we don’t know yet. We haven’t seen the most recent data yet. But we’ll get to it first thing after lunch.”

What are the norms and conventions that apply to Mary and her situation here? Is Mary in a work situation or in a lay situation (or in some intermediate kind of situation)? It seems that there is nothing in Mary’s situation as such that would determine an answer here. Rather, it seems very plausible to say that it all depends on the norms and conventions brought into play by the attributor (Charlotte or Frank). This would favor contextualism over SSI. It is not hard to think of other cases like this one. The dilemma for the adherent of SSI is that they have to accept norms and conventions as a threshold-fixing factor but that the way this factor works in at least many cases suggests contextualism rather than SSI. The situation is similar for purposes and intentions. Consider, again Cheating Al has a strong hunch that his boss has been cheating on his wife for some time. And he is right. However, Al can’t prove anything and does not possess any real evidence. Carl is a colleague of Al and a very close ally of the boss; he introduced the boss to his current mistress. Carl has found out that Al has this hunch and is seriously thinking about informing the boss’s wife. The boss has recently noticed a behavior change in Al. Over lunch with Carl he asks him, “Does Al know anything about me and, you know?” Carl bluntly replies, “Yes, Al knows that you’re cheating on your wife. But he can’t prove anything.” The next day the boss discusses the case with lawyer friend Lou. After explaining everything to Lou, she replies “Don’t you worry. Al might have a hunch but he doesn’t know anything; he cannot prove anything. You’re safe!”

The adherent of SSI would try to offer an explanation that differs from the contextualist one. Al knows that his boss is cheating on his wife because Al’s situational purposes and intentions only fix a threshold for his epistemic standards that he can pass easily. In contrast, Lou might not know that Al’s boss is cheating on his wife because Lou’s legal purposes and intentions fix a threshold for epistemic standards that is much harder to pass. However, in a case like this we also have Carl’s and Lou’s remarks about Al and here it seems that it is the attributor’s and not the subject’s purposes and intentions that call the shots. Again, contextualism seems more convincing in this and other cases. The adherent of SSI has to open the door to a further factor, purposes and intentions, that then turn out to be the attributor’s determinants rather than the subject’s factors.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  207 Finally, stakes again. This has already been discussed in Chapter 7 with at least one eye on SSI (see also Chapter 1). Here is the kind of case which adherents of SSI worry about a lot and which others think leads to a “killer objection” to SSI (DeRose 2009, 233, and ch.7, sec.2–4; see also Schaffer 2006, sec. 2 and Kim 2016, sec.1; for the case see Stanley 2005, 5): High Attributor-Low Subject Stakes. Hannah and her wife Sarah are driving home on a Friday afternoon. They plan to stop at the bank on the way home to deposit their paychecks. Since they have an impending bill coming due, and very little in their account, it is very important that they deposit their paychecks by Saturday. Hannah calls up Bill on her cell phone, and asks Bill whether the bank will be open on Saturday. Bill replies by telling Hannah, “Well, I was there two weeks ago on a Saturday, and it was open.” After reporting the discussion to Sarah, Hannah concludes that, since banks occasionally change their hours, “Bill doesn’t really know that the bank will be open on Saturday.”

The stakes relevant here seem to be the attributor’s stakes, not the subject’s stakes. This suggests contextualism over SSI. Given that similar cases can be constructed relatively easily, it appears again that the adherent of SSI has to face a dilemma: basing the case for SSI on a factor, stakes, that rather turn out to be contextualist determinants. Jason Stanley has proposed a different explanation of our intuitions about High Attributor-Low Subject Stakes, one that is compatible with SSI (see Stanley 2005, 101–4; see also 114–19): Hannah is really making a counterfactual statement to the effect that if Bill were in her own practical situation, then he would not know that the bank will be open on Saturday. The motive behind the counterfactual statement is that Hannah is interested in using Bill as a source of information and that she can only properly do so if Bill knows the relevant propositions—which has to be judged from the perspective of her own practical situation. This explanation seems ad hoc and designed specifically to solve one particular problem. There seems nothing even implicitly counterfactual about Hannah’s statement. Couldn’t she simply be saying something about Bill and his epistemic situation? One would have to come up with some independent evidence that Hannah’s utterance is really meant as a counterfactual. Apart from that, Hannah need not be in a high-stakes situation; she could be in a situation where her stakes are low but the norms and conventions applying to her situation or her intentions and purposes establish demanding standards. She could, for instance, be a member of the “Skeptical Society” trying out the society’s standards on innocent subjects. Under such circumstances it would seem particularly implausible to attribute to her a counterfactual statement along the lines of “If Bill were a member of the Skeptical Society, then he would not know that the bank will be open on Saturday.” Sure, the adherent of SSI could just claim that in cases like High Attributor-Low Subject Stakes the attributor always makes a counterfactual statement along the needed lines. But again, there appears to be no independent reason for this theoretical

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

208  ALTERNATIVES? move.8 Stanley’s proposed solution fares better than the idea that in such cases the attributor just projects her own worries to the subject (see Hawthorne 2004a, 162–6; DeRose  2005, 186–7 and 2009, 234–8; and convincingly critically Stanley  2005, 98–101) but it is still not convincing. But perhaps one should not overestimate the importance of what we would say about certain cases. I have argued myself for a less case-based case for contextualism— even though contextualism does quite well in that dimension. Interestingly, Stanley 2005, 97–8, 114–15 argues that a view like SSI that supports a knowledge-action ­principle (according to which what one may act on just is what one knows) but might lead to counter-intuitive judgments about certain cases is preferable to a view like ­contextualism that presumably has problems to account for knowledge-action principles but is more in line with our intuitive judgments about those cases. It is, however, controversial whether any such knowledge-action principle is true (see against it: Baumann 2012c). I cannot go more into that here; this would lead us too far away from our topic and the discussion of contextualism.

8.1.3  Other Problems In Chapters 2 and 3 I argued that there is a context-sensitivity of “knowledge” due to the contextual variability of relevant reference classes and descriptions. If these arguments are valid, then the question arises whether SSI can account for this variability in a non-contextualist way. It is hard to imagine how it could. It seems quite natural to assume that these parameters are tied to the attributor and not to the subject. The version of contextualism proposed here also has the advantage of allowing for explanations why other, non-epistemic notions (e.g., moral notions of responsibility) show a similar context-sensitivity; the notion of knowledge is not special in this way (see Chapter 6). There is an issue here for SSI and it is an open question whether it can say anything convincing (but see Stanley 2005). Defendants of SSI have, however, a lot to say about problems a view on knowledge might help us solve, like, e.g., the lottery problem (see Hawthorne 2004a; see also Chapter 4). In this respect the comparison between SSI and contextualism does not seem to favor one of the two views in any obvious way (I am not going into details here). Some of the objections against contextualism (see Chapter  7) can also be made against SSI. WAMs, for instance, can be made against anyone who is not a classical invariantist. Consider, for instance, Stanley’s cases Low Stakes and High Stakes: it is easy to see how one could construct a WAM against SSI here (it is surprising that WAMs have mainly if not exclusively been discussed as an objection against contextualism). The response proposed in Chapter 7 would also work for SSI. SSI also seems to license odd sentences. This is perhaps the most popular objection against the view. Suppose a subject S is in a high-stakes situation with respect to some 8   The parallel move of the contextualist concerning cases involving a low-standards attributor and a high-standards subject is different and well-motivated: see Section 7.4.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  209 proposition p. The stakes drive up the epistemic standards relevant to S’s true belief that p. They drive them up so much that S does not know that p. Stanley’s High Stakes is an example (see Stanley 2005, 4). Suddenly, something happens that makes p a low-stakes proposition again (in Stanley’s example: Hannah and Sarah discover that the bill has been cancelled because it was based on a mistake). Suddenly, the subject knows the proposition they did not know a minute ago, and only because the stakes have decreased. SSI thus seems to license the following odd sentences (see also Davis 2007, 398) Before they cancelled the bill, Hannah did not know that the bank will be open on Saturday, but now that the bill has been cancelled she knows it,

or (for other versions of the case) Hannah does not know that the bank will be open on Saturday, but if the stakes were lower she would know it,

or Hannah knows that the bank will be open on Saturday, but if the stakes were higher she would not know it.

DeRose 2009, ch.6 seems right when he notes that contextualists have an answer to such issues whereas it is far from clear whether SSI can respond convincingly to such worries. SSI cannot use a ternary analysis of knowledge in a way parallel to the way contextualists can use it. This is also a case where SSI would have to attribute systematic error to ordinary speakers and thinkers concerned with knowledge—insofar as they would systematically find sentences like the ones above odd. More generally, insofar as the variability postulated by SSI is not clear to ordinary speakers and thinkers about knowledge, SSI would have to come up with an error theory. And this could be seen as a disadvantage for the theory. It is not obvious whether adherents of SSI could argue convincingly that ordinary speakers and thinkers do indeed acknowledge the variability of knowledge with non-epistemic factors like practical stakes. But even if SSI should have not that much to offer here, this should not necessarily count as a decisive objection, also because every view will be in some mismatch with the judgments of at least some ordinary speakers and thinkers. More interesting is the fact that SSI faces its own version of the knowability problem (see the contextualist version in Chapter 5). Having gone through the details of the knowability problem for contextualism, it would be superfluous and even tedious to go through the knowability problem for SSI in the same detail; the parallels are too strong for that (see Baumann forthcoming; Brueckner and Buford 2009). If there is a problem for contextualism here, then there is also a problem for SSI (but cf. Stanley 2005, 115 and 54, fn.5). However, the solution proposed for contextualism—a ternary analysis of knowledge plus a modified closure principle—won’t work for SSI. The former is not

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

210  ALTERNATIVES? available to adherents of SSI. For that reason nothing like a contextualist downwards closure principle (see Chapter 5) is available to the adherent of SSI: (DC) If S knows-x that p, and if S knows-y that (If p, then q), then there is a knowledge relation know-z such that S knows-z that q. While the contextualist knowability problem has a solution, it is hard to see how SSI could have the resources to solve its own version of the knowability problem. It seems desperate and ad hoc to propose a revised closure principle along the lines of the following: (Revised Closure) If S knows that p, and if S knows that p entails q, then S knows that q, but only if: if “p” contains the knowledge operator, then S’s stakes are not significantly higher than the stakes of the subject referred to in “p.” Speaking of closure: everyone or almost everyone (including Dretske and Nozick) accepts some principle of closure. Can SSI afford one? Consider, for the sake of simplicity, a rough and simple version (the simplification won’t matter here): (Simple Closure) If S knows that p, and if S knows that p entails q, then S knows that q. Suppose both the entailing proposition p and the conditional proposition concerning the entailment of q by p are of low stakes for the subject so that S knows p and also knows the entailment of q by p. What if the entailed proposition q is of such high stakes for the subject that it does not know it? In that case we would have a violation of closure (or (Simple Closure) for that matter; see Stanley 2005, 93–4 for the problem). This is a problem for every adherent of SSI who wants to hold onto closure. The problem can also be formulated for other factors, like the purposes and intentions of the subject, and perhaps also for the norms and conventions in force in the subjects situation.9 One reply to this problem would say that as soon as the subject notices the entailment or infers q from p, the entailing proposition p turns as much into a high stakes proposition as the entailed proposition q (see Stanley 2005, 94; see also the related discussion of Neta  2007 in Fantl and McGrath  2009, 201–7 and another round in Neta 2012, 459–60 and Fantl and McGrath 2012c, 482–4). Let us put aside the worry that the high stakes of the conclusion would have to raise the stakes for the premises so much that knowledge of the premises is lost. Even though this is not a trivial assumption at all, there is a more important problem. What underwrites this kind of reply is a principle like (Inferential Stakes) If q is a high-stakes proposition for S and if S can deduce q from p (together with some r), then p is itself a high-stakes proposition for S, 9   But cf. Hawthorne 2004a, 160–1 on salience and how there is, according to him, no problem with closure.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  211 or (Entailed Stakes) If q is a high-stakes proposition for S and if p (together with some r) entails q, then p is itself a high-stakes proposition for S. The main problem is that such principles are very hard to defend (see also the brief remarks in Schiffer 2007). Here are three reasons why this is so. First, every proposition p entails some high-stakes proposition q if combined with certain other appropriate premises (whether of the form “If p then q” or not). Hence, according to (Entailed Stakes) every proposition is a high-stakes proposition if some are. Since some are, every proposition is a high-stakes proposition. This, however, seems clearly false: clearly, stakes differ from proposition to proposition and some are not high while some others are not low (for a subject at a time). Defenders of SSI should try not to get into a situation where they have to deny this. A similar argument can (with slight modifications) be made against (Inferential Stakes). Second, consider an example involving a great number of premises from which a high-stakes conclusion can be inferred. Suppose a detective is working on some case. There are 1,000 different propositions describing different pieces of evidence. Only taken together do they imply a proposition which is of high stakes for the detective (namely that the chauffeur did it). It is very implausible that a particular single, isolated piece of evidence as such (e.g., that the profile of the tires of the chauffeur’s private car is medium) should count as being of high stakes for the detective. Again, similar things hold (mutatis mutandis) for (Entailed Stakes). Finally, a special case having to do with the knowability problem. Consider the following triple of propositions: p: B knows that the train to X will leave from track 3; r: If B knows that the train to X will leave from track 3, then the train to X will leave from track 3; q: The train will leave from track 3. Consider a possible case where a subject A knows the proposition p which is of low stakes for her and also knows the proposition r which is also of low stakes for her. Suppose A can infer from p and the instantiation of the factivity-principle for knowledge r the proposition q which happens to be of high stakes for her. If this turns p into a high-stakes proposition for A, then presumably it also turns r or the factivity principle for knowledge which r instantiates into a high-stakes proposition for A. The latter, however, is very implausible. But then it seems arbitrary to say that p is high stakes for A—given that the factivity principle for knowledge or r is not. Why should one premise be high stakes and the other not? Even if one were to restrict (Inferential Stakes) and (Entailed Stakes) to a small number of premises and not to take the impression of this being an ad-hoc move very seriously, one would still have to explain which propositions would be high stakes given that they help infer or entail a high-stakes proposition and which propositions wouldn’t be high stakes for that reason. There are serious

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

212  ALTERNATIVES? doubts that this can be done in any non-arbitrary way. We should reject the likes of (Inferential Stakes) and (Entailed Stakes) and not bet much on them. So, the problem that stakes are not closed under known entailment while knowledge is remains a ser­ ious one for SSI. Finally, a smaller problem (corresponding to the problem for contextualism mentioned in Section 7.5). Consider the alleged stake-dependency of knowledge. A problem arises if the subject’s stakes also depend on what they know—which seems possible in principle. Here is an example. Suppose Zoe knows where Jack was yesterday evening—but only as long as the stakes concerning that proposition are low. Now, the state secret service would be strongly interested in doing some interrogation with Zoe just in case Zoe knows where Jack was. Zoe, however, fears few things more than being of interest to the state secret service. If the state secret service is interested in talking to Zoe, then everything they want to know becomes high stakes for Zoe: both where Jack was (since Zoe’s “punishment” by the state secret service depends on where exactly Jack was—in her neighborhood or somewhere else), and also whether Zoe knows it (because the state secret service hates citizens who “know too much”). Hence, if the stakes are low and Zoe knows where Jack was yesterday, then (because the state secret service is interested in Zoe under such circumstances) Zoe’s stakes are high such that Zoe does not know where Jack was yesterday. If, however, Zoe does not know this, then the stakes are low such that Zoe knows where Jack was yesterday (see also the jogger example in Neta 2007). If S knows that p, S is in a high-stakes situation with respect to p. However, if S is in such a high-stakes situation, she does not know that p. However, if she does not know that p, then she is in a low-stakes situation— which makes her a knower of p, again. We seem caught in an inconsistent circle where the knowledge of p is not grounded in anything: whatever determines whether S knows that p is determined by those very facts about S’s knowledge. Such cases are, admittedly, not very common but they raise a problem worth attention and treatment (see also Reed 2012, 470). The upshot of all these remarks on SSI as an alleged alternative to contextualism is that SSI faces serious problems itself, obviously more serious than any problems for contextualism. But we do not even have to go this far. It is even sufficient to claim that SSI is not a better alternative to contextualism. What now about the second alternative to be discussed in this chapter, contrastivism?

8.2 Contrastivism According to contrastivism, knowledge is a ternary relation between a subject, a proposition known, and a contrast proposition—which is incompatible with the known proposition and which can consist of a conjunction of propositions (see SinnottArmstrong 2004 and 2008a; Schaffer 2004a, 2005a, 2007a, 2007b, 2008; Karjalainen and Morton 2003; Morton 2013; Johnsen 2001; an early precursor or related view can be found in Dretske 1972; see also a short passage in Swinburne 2001, 34; see further:

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  213 Morton 2011; Schaffer 2012a; for the incompatibility of the contrast proposition with the “target” proposition see Rourke 2013, sec.2). According to contrastivism, knowledge attributions have the form “S knows that p, rather than q” (where “rather than q” is, of course, not part of what is claimed to be known here10). Here is an example. Jack is eating cheddar cheese which he can distinguish from marmalade but not from gruyère cheese. Hence, he might know that he is eating (cheddar) cheese rather than marmalade but he does not know and is not in a position to know that he is eating cheddar rather than gruyère. Contrastivism has certain affinities with relevant alternatives views of knowledge and justification (see, e.g., Dretske 1970); the latter, however, are not tied to a ternary account of knowledge or justification.11 Contrastivism does not entail contextualism but can be and, as I will argue, should be combined with context­ ualism (see for a non-contextualist version of contrastivism Sinnott-Armstrong 2004, 2002 and also Schaffer 2004a and in reply Stalnaker 2004; for a contextualist one see Schaffer 2005a). The type of contextualism proposed here also has it that knowledge is ternary—though the nature of the third relatum is seen in a different way (see also Woudenberg 2008 here).

8.2.1  Knowledge without Contrasts There are many cases which seem to give strong support to a contrastivist account of knowledge (see, e.g., Schaffer 2005a). Perhaps the best case for contrastivism is perceptual knowledge based on discriminatory abilities. Jack can distinguish between dogs and cats but not between dogs and wolves. He sees a dog in front of him (under normal conditions) and thus comes to know that there is a dog in front of him. According to the contrastivist, “Jack knows that there is a dog in front of him” would in this case be elliptical for something like “Jack knows that there is a dog in front of him rather than a cat.” One of the greatest advantages of contrastivism is that it can account for the lack of knowledge in cases like Jack’s inability to tell dogs from wolves: “Jack does not know that there is a dog in front of him rather than a wolf ” is also true, according to the ­contrastivist. A binary view of knowledge has problems accounting for our opposing tendencies to attribute and deny knowledge that there is a dog to Jack; contrastivism easily resolves the puzzle. Crucial to all this is the idea that a subject typically knows some proposition p with respect to some but not all potential contrast propositions (but cf. Buenting 2010). To be sure, the contrastivist does not need to deny the possibility of cases where a subject knows some proposition with respect to all possible contrast propositions. But this would be an extreme case. Contrastivism has its real point when analyzing normal cases where a subject knows that p rather than q but does not know that p rather than r 10   But cf. Blaauw 2008a, sec.2 and Jespersen 2008 who toys with the idea that the content of contrastivist knowledge ascriptions is of the form “S knows that (p and not q).” 11   Talk about contrast propositions would have to be reformulated as talk about relevant alternatives and about the possibilities needed to be ruled out in order for the subject’s belief to be correctly called “knowledge.”

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

214  ALTERNATIVES? (for some propositions p, q, and r). If that were not the case, contrastivism would lose its point and attractiveness (see, e.g., Schaffer 2005a, sec.2; 2005b). This leads to the following plausibility constraint for any contrastivist analysis of knowledge: (Specificity) A contrastivist analysis of knowledge is plausible only if there are for any given subject S a lot of triples of propositions p, q, and r such that S knows that p rather than q but S does not know p rather than r.

This condition is vague insofar as it does not indicate a minimal number of such triples of propositions for a given person. However, we do not need to be more specific here. One can see that only some but not all types of knowledge meet the Specificity-condition: perception for instance, seems to meet it but other types of knowledge don’t meet it. Take, for example, knowledge of obvious mathematical truths, like the simple one that 2 + 2 = 4. Does anyone who knows this know it in contrast to something else? In contrast to what, then? To 2 + 2 = 5 (Or 2 + 2 = −.7? Or 3 + 3 = 4? Or 12 x 12 = 1212?)? There simply does not seem to be a plausible contrast proposition around. (Specificity) does not seem to hold for this kind of knowledge. It is hard to imagine, for instance, how there could be two numbers x and y (not equal to 4) such that S knows that 2 + 2 = 4 rather than 2 + 2 = x but that S does not know that 2 + 2 = 4 rather than 2 + 2 = y. It would be very interesting to find someone who, e.g., knows that 2 + 2= 4 rather than 2 + 2 = 40 but who does not know that 2 + 2 = 4 rather than 2 + 2 = 5 (“too close!”). It would be very hard to make sense of such a subject or even to attribute knowledge to her in the first place. It seems that whoever knows that 2 + 2 = 4 knows that this is the case rather than 2 + 2 = z, for any z not equal to 4.12 The contrastivist might reply here that all this is still compatible with his view: such subjects know some proposition p rather than not-p (or rather than “any alternative”). This, however, is at best compatible with the letter but certainly not with the spirit of contrastivism. The prize for this kind of maneuver is the rejection of (Specificity) and trivialization. Nobody would deny that one can talk like that but this way of talking would be like an additional wheel spinning in the void. We might as well go back to the binary view of knowledge. The problem here is not restricted to mathematical knowledge. Similar things can be said about basic forms of logical knowledge. To be sure, there are very difficult questions in mathematics or logic with respect to which it might make sense to say that someone 12   To be sure, few, if any, subjects are able to grasp all possible contrast propositions to a given, known proposition. For instance, someone might understand that 2 + 2 = 5 but not understand that 2 + 2 = 1/4, due to a lack of understanding of fractions. However, that there is such a difference between “accessible” (for the subject) contrast propositions and non-accessible contrast propositions does not at all support the claim that knowledge is contrastive in the first place or that contrast propositions or a subset of them play the role the contrastivist thinks they play. Referring to non-accessible contrast propositions here in order to defend contrastivism saves (Specificity) at the prize of making it trivially true. The problem above remains even if we focus only on accessible contrast propositions and restrict (Specificity) to them (I will leave this detail aside here). Similar things hold in the case of contrast propositions which either lack a truth value or are false for non-mathematical reasons, like, e.g., “2 + 2 = Julius Caesar.” See for this also Schaffer 2012b, sec.1.1 which responds to my 2008c, sec.1; see also my reply to Schaffer in Baumann 2012d.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  215 knew or knows at a time that this formula is valid rather than that one. However, this maneuver won’t work for basic and simple cases like knowledge that if p then p, or that everything is identical with itself (what would be the contrast here?). I will not repeat the argument from (Specificity) for these kinds of cases but rather shortly mention still another case, namely linguistic or conceptual knowledge. What would be a plausible contrast proposition for knowledge that vixen are female foxes? Taking all these problem cases together it seems that at least some of the traditional candidates for aprioricity create this kind of trouble for contrastivism because they don’t meet (Specificity). All this suggests that not all knowledge is contrastive (see also the brief remarks in Becker 2007, 83, and 80–6). More cautiously: nobody so far has shown (and it does not seem likely) that all types of knowledge admit of (non-trivial) contrastivist analyses. To put it more positively: it seems that contrastivism is most plausible and probably only plausible with respect to a particular kind of knowledge: knowledge of propositions which involves the use of discriminatory cognitive abilities. The use of such abilities need not be restricted to one’s external environment: discrimination is also needed when it comes, for instance, to the identification of a particular emotion as sympathy rather than pity. So, why not explicitly restrict contrastivism to cases of discriminatory knowledge? One might, however, want to propose more radical consequences. Starting with the assumption that (A) If some kinds of knowledge are contrastive, then all kinds of knowledge are contrastive one could argue that since some kinds of knowledge (e.g., knowledge based on discriminatory abilities) are contrastive, all kinds of knowledge are contrastive (see, e.g., Schaffer  2012b, sec.1.2, 2.2). However, given the above point about mathematics, things also cut the other way around: since not all kinds of knowledge are contrastive, none are. This conclusion would, of course, be very bad for the contrastivist. However, since I do not see any good reason to accept (A), and since it cuts both ways, I would rather propose to restrict contrastivist analyses of knowledge in the way indicated above (see also the brief remark in Luper 2012b, 56). But doesn’t this threaten the unity of knowledge (see Schaffer 2012b, sec.1, 2)? It does not threaten the unity of knowledge (or “knowledge”) in general but rather a very specific unity and the idea expressed in (A) above. But (A) is not plausible. There are all kinds of differences between the kinds of epistemic states we call “knowledge”: e.g., some kinds of “knowledge” preserve information (memory), others involve the acquisition of new information (perception). Why shouldn’t there be some kinds of the epistemic states we call “knowledge” which are contrastive while others aren’t? This does not turn the notion of knowledge into a disunified one.13 13   Even if one denies that all knowledge is contrastive one can still hold onto a view of knowledge in general as ternary. The latter is supported by the version of contextualism proposed here.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

216  ALTERNATIVES?

8.2.2  More Relata for the Third Slot The third relatum in the contrastivist’s ternary analysis of knowledge is a contrast proposition. But why should only contrast propositions be able to fill the third argument place? Consider, again Weather Forecast Mary works as a weather forecaster for the local weather station. While on her lunch break with her friend Charlotte and her colleague Frank, she is discussing with them the possibility of having dinner outside together later in the day. At one point, Mary looks at the sky and notices some dark clouds approaching quickly. She says, “Oops, I think we can forget about dinner in the garden today: it is going to rain within the hour. I can see dark clouds approaching rapidly.” Mary is pretty good at making predictions like that. And, indeed, within the hour rain is falling where they are. Just half a minute after Mary spoke, both Charlotte’s and Frank’s cell phones ring. Charlotte receives a call from her husband who is asking about the dinner plans. Charlotte tells him that they have to call the whole thing off because of incoming rain. “Really?” asks the disappointed husband, “Does Mary know it’s going to rain?” “Yes,” answers Charlotte, “she knows that.” Frank’s call is by the station manager who wants to know whether Mary or he already know whether it is going to rain later in the day. Frank replies “No, we don’t know yet. We haven’t seen the most recent data yet. But we’ll get to it first thing after lunch.”

As argued above there is a relativity to epistemic standards here. So, the third slot in the knowledge relation is filled by epistemic standards. How could one analyze this in terms of contrast propositions? To be sure, Mary might know that it is going to rain rather than snow but not know that it is going to rain rather than drizzle. However, she knows that it is going to rain rather than drizzle only given lab standards while she knows that it is going to rain rather than snow also given lay standards. More import­ antly, all these contrasts are beside the point of our example. What varies in the example above is not the contrast proposition but something else: standards of evidence, etc. relevant in particular contexts (meteorological contexts, non-work contexts). There is something different or at least more going on in Weather Forecast than what can be captured by an exclusive focus on contrast propositions. To be sure, differences between standards can concern what needs to be ruled out and constitutes a relevant alternative. However, there is much more to epistemic standards than that: there are also standards for the quality of evidence, the degree of reliability, the degree of belief required for knowledge, or the type of epistemic position required for knowledge (as pointed out in Chapter 1; see also Schaffer 2012b, sec. 2.1 and Baumann 2012d, sec.3). Changes of standards do not just consist in the expansion or contraction of the contrast set (or the set of relevant alternatives). All this suggests that it is not just contrast propositions (if at all) but also all kinds of different epistemic standards that can fill the third slot of the knowledge relation. Apart from the different kinds of standards, there is also a relativity to different kinds of reference classes and to descriptions, as argued above. So, what goes into the third slot is a

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  217 complex set of different (contextual) parameters. It is an undue simplification to just focus on contrast propositions. It does not matter whether one thinks of the knowledge relation as ternary relation where the third slot is filled by n-tuples of different kinds of entities (standards, reference classes, descriptions) or whether one adds a 4th, 5th, etc. relatum; for the sake of simplicity of exposition I prefer the former way of p ­ resenting things. Again, the charge (see, e.g., Schaffer 2012b, sec.2) that all this gives a disunified picture of knowledge (or “knowledge”) seems unjustified: there is still enough unity in all the complexity.

8.2.3  More Relativization Consider Sue who cannot distinguish between a terrier and a dachshund even though she clearly sees a difference between a terrier and a cat as well as between a dachshund and a cat. She also sees a difference between a dachshund and a Labrador or between a terrier and a collie; it is just the small dogs which give her epistemic trouble. According to the standard contrastivist view, Sue might well know on one occasion that there is a dachshund in front of her rather than a cat and she might also know on another occasion that there is a terrier in front of her rather than a cat—even though she cannot tell dachshunds from terriers and thus cannot know whether there is a dachshund in front of her rather than a terrier. This, however, sounds abominable and incorrect.14 It seems false to say that she knows out of the contrast class {There is a dachshund; There is a cat} that there is a dachshund. To say that Sue knows that there is a dachshund rather than a cat seems to imply that she can identify dachshunds as such (at least in contrast to, say, terriers) and refer to dachshunds as such. It seems to imply (in our case) that she also knows that there is a dachshund rather than a terrier. If she does not know the latter, then it would rather be correct to say something like “She knows that there is a small dog rather than a cat” but not “She knows that there is a dachshund rather than a cat.” This suggests that for the contrastivist there has to be a further condition concerning the propositions involved in the knowledge relation. Suppose S cannot distinguish between Fs and Gs (e.g., dogs and wolves). Then there are pairs of propositions p containing F (“F-propositions”; e.g., There is a dog) and r containing G (“G-propositions”: e.g., There is a wolf ) such that

14   Schaffer 2012b, 421 brings up the question “Is the beast a dachshund or a cat?” and continues: “it is plausible to say that she does know the answer to the question—she knows whether the beast is a dachshund or a cat.” I don’t agree. Sure, if we assume that Sue has been explicitly asked “Is this a dachshund or a cat?” then it would be reasonable for her to assume that by asking this question the questioner conveys (by implicature, for instance) the information that it is either a dachshund or a cat. So, the asking of the question gives Sue more information than she would have had otherwise (given that she can trust the questioner as a source of information). Under these circumstances, she does know that the beast is a dachshund and not (rather than) a cat. The point, however, is that it is still as odd as before to say this kind of thing if we don’t include the asking of such an explicit question in the description of the scenario. See on this also Becker 2009.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

218  ALTERNATIVES? replacing F by G at one or more places in p gives us r and S does not know that p rather than r, and only because S cannot distinguish between Fs and Gs. Let us call each member of such a pair of propositions a “defeating proposition” of the other member of the pair. We can now generalize the point made above in a very preliminary way: (Distinguish-a) If someone knows that p rather than q, then there is no defeating proposition r for p. In our case above, Sue cannot distinguish between dachshunds and terriers and thus does not know that there is a dachshund in front of her rather than a terrier; thus she does not know that there is a dachshund in front of her rather than a cat. There is a defeating proposition for her (that there is a terrier). What we can still say about Sue is this: (1) Sue knows that there is a dachshund or another kind of small dog rather than a cat. For reasons analogous to the ones above, we also have to introduce a further condition concerning the contrast proposition. Suppose Sue cannot distinguish (under any circumstances) between cats and mountain lions. We should then say something like this about her: (2) Sue knows that there is a dachshund or another kind of small dog rather than a cat or another four-legged furry animal indistinguishable from it. We thus have to add a complement to (Distinguish-a) relating to the contrast proposition instead of the known proposition. Since the contrast proposition is not known, we have to add a slightly different kind of condition here. Suppose that S cannot distinguish between Hs and Ks (e.g., cats and mountain lions). Then there are pairs of contrast propositions q containing H (“H-propositions”; e.g., There is a cat) and s containing K (“K-propositions”: e.g., There is a mountain lion) such that replacing H by K at one or more places in q gives us s and S cannot tell whether q or s, and only because S cannot distinguish between Hs and Ks. Let us call each member of such a pair of propositions an “undermining proposition” of the other member of the pair. Again, we can say in a very preliminary way: (Distinguish-b) If someone knows that p rather than q, then there is no undermining proposition s for q. If Sue cannot distinguish between cats and mountain lions and thus cannot tell whether there is a cat or a mountain lion in front of her, then she does not know that there is a dog in front of her rather than a cat. There is an undermining proposition for her (that it is a mountain lion).

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  219 Putting (Distinguish-a) and (Distinguish-b) together, we get (Distinguish*) If S knows that p rather than q, then there are no defeating propositions r for p and no undermining propositions s for q.15 There is, however, a big fly in the ointment: What if r is a skeptical hypothesis? This includes traditional skeptical hypotheses like the evil demon-hypothesis as well as more ordinary hypotheses, like the hypothesis that it is not real cheddar one is eating but something that just looks and smells like it (for a further worry see below). By the very nature of a skeptical hypothesis, we can in principle not distinguish between an ordinary scenario (p) and a skeptical one (r). The contrastivist wants to say that we don’t know that p rather than r. Given (Distinguish*), we would then have to accept that we don’t know that p rather than q—for any ordinary propositions p and q. This is unacceptable to the contrastivist. An analogous point can be made about the contrast proposition and some skeptical undermining proposition. What then can the contrastivist do about this? Insofar as (Distinguish*) seems to go in the right direction, we should not give it up but rather introduce a further restriction. Not only do we have to think about defeating or undermining propositions (r, s) but also about what to admit into the contextually relevant group of potentially defeating or undermining propositions. In a skeptical “mood” we let skeptical propositions into this group, in a non-skeptical mood we don’t. What belongs into the classes R and S of potentially defeating or undermining propositions can vary from situation to situation. Knowledge is relative not just to contrast propositions or classes thereof but also to classes of potentially defeating or undermining propositions. This does not only concern traditional skeptical propositions but all kinds of propositions the subject cannot rule out. Hence, I propose to modify (Distinguish*) in the following way: (Distinguish) If S knows that p rather than q, then there are no defeating propositions r for p in R and no undermining propositions s for q in S. This condition forces the contrastivist also to relativize to classes R and S, not just to contrast propositions. That this further relativization is quite valuable and important becomes clearer if we look at cases. Consider Sue, again, who cannot tell dachshunds from terriers but can tell them from Labradors and collies. Contrast her with Jack who is unable to tell any kind of dog from any other kind of dog. Both can tell dogs from cats. Suppose they are both confronted with a terrier. We might then, given certain circumstances, be entitled to say about both that they know there is a terrier in front of them rather than a cat. However, while “Jack knows that there is a terrier rather than a cat” is true only if propositions like It is a Labrador are excluded from R, “Sue knows 15   Since this is not a definition of “knowledge,” there is no bad infinite regress looming here. One could also put the point like this: if there is such a proposition r or s, then the correct knowledge ascription has the form of “S knows that (p or r) rather than q” or “S knows that p rather than (q or s)” or “S knows that (p or r) rather than (q or s).”

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

220  ALTERNATIVES? that there is a terrier rather than a cat” is still true even if It is a Labrador is included in R (but not if, say, It is a dachshund is included). And if It is a Labrador is included in R both in Sue’s and in Jack’s case, then it might true to say of Sue but it is false to say of Jack that they know that there is a terrier in front of them rather than a cat. Hence, the further relativization to the classes R (and similarly to S) in Sue’s and in Jack’s case is needed for a full explanation of the truth conditions of the corresponding knowledge attributions.16 What determines R and S? It is tempting to give a contextualist answer here (see also in favor of a contextualist version of contrastivism Montminy 2008c, sec.4). In that case, contrastivism would have to be combined with contextualism. This also goes some way towards explaining the initial phenomenon: that it seems false to attribute knowledge to Sue that there is a dachshund in front of her rather than a cat. If There is a terrier is included in R, then Sue violates (Distinguish) and the corresponding knowledge attribution is false. What if only There is a dachshund is included in R? Isn’t it then correct to say that she knows that there is a dachshund rather than a cat? Yes, but in many if not most or all contexts we would not want to restrict R in this way. In these contexts, such a restriction seems inadmissible for pragmatic reasons (perhaps for other reasons, too); the context puts a constraint on admissible restrictions of R (similar things hold for S, of course).17 But doesn’t (Distinguish) clash with (Specificity) above? Doesn’t (Specificity) require for knowledge that p that there are defeating propositions for p while (Distinguish) requires for knowledge that p that there are no defeating propositions for p? No. (Distinguish) only requires that there are no defeating propositions for p in R. It allows that there are defeating propositions for p in the set of all contrast propositions C. The contrastivist can and should assume that R is a proper subset of C (otherwise there is indeed a clash between the two principles).18 When we say that S knows that p rather than q, we are excluding certain propositions from R but they are still in the set of contrast propositions. The appearance of a clash between (Distinguish) and (Specificity) might well be due to not explicitly relativizing to the restricted class R. However, one should also not forget that there is a delicate balance between the two principles. It indicates that contrastivism is characterized by a tension between the attempt to account for the essential limitations of our knowledge ((Specificity)) and the attempt to account for the substantial scope and extent of our knowledge ((Distinguish)).

8.2.4 Conclusion Contrastivism also lends itself to the application to other epistemic and certain non-­ epistemic terms, like “justification,” “reason” (see Snedegar 2013), or “moral obligation” 16   Given that R is not empty, all this also amounts to a holistic condition on knowledge: in order to know that p rather than q, S also needs to know that p rather than r (for a number of substitutions for “r”). 17   All this should also address Kelp’s  2011 worries that Schaffer would have to claim knowledge of anti-skeptical propositions. See also Montminy 2008c on this. 18   S would also have to be a proper subset of C.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  221 (on the latter see in particular Sinnott-Armstrong 2006; see also the debate between Sinnott-Armstrong 2008b; Hough 2008; Blaauw 2008b; Baumann 2008d; and SinnottArmstrong 2008c). This is an advantage in itself; but this is not the place to discuss contrastivist theories of, say, moral reasons. The potential to come up with a good response to serious philosophical problems, like the lottery problem or the traditional problems of epistemic skepticism, is another important issue. It is not clear whether contrastivism has anything specific to contribute to the lottery problem (we will have to wait and see). It is, in contrast, often mentioned that contrastivism has an interesting reply to the traditional philosophical skeptic, a reply that resembles a common contextualist strategy (see Section 4.1; see also the exchange between Luper 2012b and Blaauw 2012): one can concede something but not too much to both the skeptic and the anti-skeptic by pointing out that a subject can know one ordinary proposition rather than another ordinary proposition but cannot know an ordinary proposition rather than the denial of a skeptical hypothesis. For instance, a subject can know that she has hands rather than stumps even if no subject can know that she has hands rather than that she is suffering from a Cartesian hand illusion. This is not the place to return to the discussion of such anti-skeptical strategies. Suffice it to add that I think Jason Stanley is right when he points out that neither SSI nor contextualism should hope to be able to refute the radical skeptic (see Stanley 2005, 125–30); we may include contrastivism here. I have argued that the contrastivist focus on contrast propositions is too narrow (see Sections 8.2.1 and 8.2.2). At the same time, contrastivism needs more development and additional relativization along the lines presented in Section 8.2.3. The latter considerations also suggest that contrastivism should take a contextualist turn. Contrastivism is, especially in its contextualist version, quite close to the type of contextualism proposed here. However, I think it turns out that the latter still has a clear advantage over contrastivism, given the respective problems and advantages.19

8.3 Relativism There is a great variety of quite different forms of relativism. According to the relativism relevant here, “truth relativism,” the truth values of sentences (or of the propositions expressed by sentences) vary with context. However, they don’t do so because their truth conditions varied with the speaker’s (or thinker’s) context; rather one and the same proposition can have variable truth values in different contexts in which it is assessed for truth (in different contexts of assessment). Relativism that is not restricted to certain types of discourse or certain kinds of sentences is problematic and implausible while relativism might be a plausible view about certain domains and expressions 19   For further critical discussion of contrastivism see, e.g., DeRose 2009, 38–41, and 2011, 107; Rickless 2014 on Schaffer’s linguistic evidence for contrastivism; Schaffer and Knobe 2012, DeRose 2011, and Gerken and Beebe 2016 on experimental data concerning contrastivism. Contrastivists haven’t really put much weight on the considerations of cases, like adherents of SSI; therefore I am leaving this topic aside here, too.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

222  ALTERNATIVES? (apart from “knowledge,” e.g., epistemic modals, moral and aesthetic expressions, taste predicates, future contingents, certain gradable adjectives (like “rich”), and so on). Let us focus here on relativism about knowledge—though some of what follows also holds more generally for relativism as such. In contrast to contextualists, truth relativists20 don’t hold that the truth values of knowledge propositions vary because the truth conditions or the meaning of the corresponding knowledge sentences vary. Rather, given one and the same subject context and one and the same context of attribution—given one and the same proposition21—the truth value of the relevant proposition can still vary with the context of assessment.22 Using Lewis’ notions of index and context (see Lewis 1998) one can also characterize relativism in the following way (and it is often done this way). According to the relativist, a sentence S expressing an assessment-sensitive proposition p has the following type of truth conditions (leaving aside assignments of values): S as used at c1 and assessed at c2 (which can but need not be identical with c1) is true just in case S as used at c1 is true at the index (where w1 and t1 are the world and time of c1 and a2 is the assessment parameter of c2). Leaving aside worlds and times, we can simplify when convenient and appropriate and just talk about S as used in c1 being true at a2. Since differences of the context of use won’t make a difference here, we can simplify further and put it like this: S (expressing the proposition that p) is true at a2. Applied to knowledge, we can say that according to the epistemic relativist “S knows that p” as used at c1 and assessed at c2 is true just in case that sentence as used at c1 is true at a2. There are a number of considerations typically adduced in favor of relativism (for instance data concerning agreement and disagreement; there are also quite a number of typical objections against relativism (see the overview in MacFarlane 2012). Here I want to focus on those points that to me seem particularly problematic for truth relativism about knowledge.

8.3.1  Relative Truth, Monadic Truth, and Direct Expressibility A first problem has to do with the relation between relativist and non-relativist, “ordinary” notions of truth. Obviously, a predicate of relative truth does not obey anything like the disquotational equivalence schema that holds for ordinary monadic truth: The proposition that p is true iff p.   Or, briefly (from now on): relativists.   It does not matter here what one takes to be the primary truth bearer. We can therefore just go with propositions (see MacFarlane 2014, ch.3.1, 49) and allow for substitution by other truth bearers. 22   The characterization of relativism used here is close to John MacFarlane’s. One could characterize it in a different way but nothing much depends on such differences. For truth relativism in general see Kölbel 2002; Richard 2008; Wright 2008; Cappelen and Hawthorne 2009; MacFarlane 2014. For knowledge relativism see, e.g., MacFarlane 2005a; Stanley 2005, ch.7; MacFarlane 2014, ch.8; Richard 2008, appendix II. For some detail on a relativist semantics see MacFarlane 2014, ch.7. 20 21

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  223 If it did, a contradiction would be very easy to derive. Using a schema like The proposition that p is true at a iff p, one could derive p from S is true at a1 and not-p from not-S is true at a2. Since relativism allows for the truth of a sentence at one context and the truth of its negation at another context, it would also allow for the truth of the contradiction that p and not p. Ignoring dialetheism (even though the marriage of truth relativism and dialetheism would be quite something; see Beall 2006), this would seem very bad. So, the relativist should and must claim that the relative truth predicate lacks the disquotational property. Is this lack a problem?23 Why should it be? One could argue that something important is being lost without disquotation, even if inconsistency is thus avoided. But what would be lost? It is useful here to compare assessment sensitivity with the apparently parallel cases of times- and worlds-index-relativity. If there is no loss in the case of times and worlds, then why, one could ask on behalf of the relativist, should one think there would be a problem in the case of the assessment parameter? According to the relativist, times and worlds are, like the parameter of assessment, part of the index (see Lewis 1998). There is, of course, a crucial difference here: while the time and the world of the index are determined by the context of use, the assessment parameter is determined by the context of assessment. However, as far as the role of disquotation is concerned this difference is irrelevant here. Now, in the temporal case, a temporalist would allow for the possibility that a sentence S (expressing a non-eternal proposition like the proposition that Descartes is running) can be true at one time (t1) and its negation true at a different time (t2): S is true at t1 and Not-S is true at t2. We avoid inconsistency because we cannot go from “Descartes is running is true at 10 a.m. on March, 31, 1606” to “Descartes is running,” or from “Descartes is not running is true at 11 p.m. on March, 31, 1606” to “Descartes is not running.” But isn’t the price for that very high because we cannot express the true content directly (in a non-relativized and first-order kind of way)? No, this is not a problem because there is something else that we can easily say instead: “Descartes is running at 10 a.m. on March, 31, 1606 but he is not running at 11 p.m. on March, 31, 1606.” There is a content-function f mapping pairs of some non-eternal, index-sensitive content p- (Descartes is running) and a time of evaluation t (10 a.m.) to some eternal, index-insensitive content p+ (Descartes is running at 10 a.m.): p+ = f(p-, t). This is enough for our ordinary purposes; it does not require disquotational schemas for the index-sensitive propositions and avoids the brute contradiction mentioned above. At the same time it ensures direct (or direct enough) expressibility of true contents. 23   To be sure, the relativist can allow for disquotation as applied to a monadic truth predicate and within a given context of assessment (see, e.g., MacFarlane 2014, 38, 93); however, that is not the point here.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

224  ALTERNATIVES? The same things hold, mutatis mutandis, for worlds. We should allow for the possibility that a sentence S (expressing a proposition like the proposition that dogs don’t sing opera) can be true at one world (w1) and its negation true at a different world (w2): S is true at w1 and Not-S is true at w2. Again: it is, of course, a good thing that we cannot disquote here and go from, say, “Dogs don’t sing opera is true at this world” to “Dogs don’t sing opera” or from “Dogs sing opera is true at some other world” to “Dogs sing opera.” There is also no loss of direct expressibility because we can say something different (we don’t have to worry about the details): “Actually, dogs don’t sing opera, but if things were quite different in the right way with their vocal chords and their brains, they would.” We do have expressions like “as things stand,” “possibly,” etc. There is a function g mapping pairs of some world w and some index-sensitive “worldless content” p- to some index-insensitive “worldly” content p+: p+ = g(p-, w). This is sufficient for ordinary talk and thought: no disquotation for the index-sensitive propositions necessary and no contradiction threatening. The lack of a disquotational property for the notion of truth at a time-index or at a world-index does not do any damage to the direct expressibility of content because there is a temporal content function as well as a modal content function such that the function produces a sufficiently close content to which we can apply a monadic notion of truth, including a disquotational equivalence schema (I don’t see why the relativist semanticist should not also keep a monadic notion of truth at hand, or why one should think that notions of relative truth are completely absent from ordinary first-order ways of thinking and talking; see MacFarlane 2011, 442). In the case of assessment-sensitivity—S as used at c1 and assessed at c2 is true just in case S as used at c1 is true at a2—there is no disquotation either. But here it hurts, in contrast to the cases of times and worlds, because there is no content function m ­ apping pairs of some assessment-sensitive proposition p- and some parameter of assessment a to some assessment-insensitive content p+. We can go from “That Descartes is ­running is true now” to “Descartes is running now” and from “That dogs don’t sing is true in this world” to “As things stand, dogs don’t sing,” but it is nonsensical to go from “That licorice is tasty is true for Emily’s standards” to “Licorice is tasty emilyly.” (To be sure, one can say things like “Licorice is tasty for Emily” but the ­relativist, in contrast to the contextualist, will see a big d ­ ifference between this and “Licorice is tasty” (see, e.g., Kölbel 2002, ch.3); the difference is too big, according to the relativist, to see a parallel with the pair “Socrates is sneezing” (uttered at 1:58 p.m.) and “Socrates is sneezing at 1:58 p.m.”) This is not just a contingent limitation of our given ways of thinking and speaking. We cannot just remove the limitation by introducing terms like “emilyly” or “PeterPaul-and-Mary’s-standards-wise.” The terms would still lack intelligible content. If there are assessment-sensitive contents, then we still cannot make sense of the idea of

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  225 some assessment-insensitive cousins; we cannot make sense of that because it makes no sense. Or so it seems. At least to my knowledge, no one has shown how to make sense of this. One could also express this by saying that while talk of temporal or modal facts makes sense, talk of perspectival facts doesn’t (but cf. Einheuser 2008; Brogaard 2008a; Torrengo 2010; and Rovane 2011, 2012, and 2013; see also Nozick 2001, 34–6 and Almér 2005, 150–1). So, in the case of the indices of times and worlds, there is a harmless and “uncostly” lack of the disquotational property of the index-sensitive truth predicate because there is a content function which saves a close enough connection with a disquotational monadic truth predicate for sufficiently closely related contents. In the case of assessment-sensitivity, there is no disquotation to have at all, not even ersatz disquotation for ersatz propositions. This difference has to do with the fact that the index parameters time and world are part of the circumstances of evaluation whereas the assessment parameter isn’t. Why is this bad? When we call something “true” (in whatever sense), then we should, it seems, also be able to express directly (that is, in a non-relativized, first-order kind of way) what we call “true.” Disquotation is necessary for such directly expressed content and the latter can be taken as a core feature of ordinary talk (and thought, mutatis mutandis). Concerning index-sensitive cases we can say this: (A) For any index-sensitive content there is another index-insensitive content that allows for the application of a monadic truth predicate (including disquotation) such that the latter content is close enough to the former. For assessment-sensitivity in particular we should claim that: (Contents) For any content that allows for the application of a relative truth predicate there is another content that allows for the application of a monadic truth predicate to it, and such that the latter content is close enough to the former. Relativists have to deny (Contents). There are doubts that they can do so without incurring serious costs for their view. Without something like (Contents), the relation between the ordinary monadic notion of truth and the notion of relative truth remains obscure.24

8.3.2  But Is It Truth? All this also raises a related question. It is hard to understand how and in what sense a non-disquotational notion of “relative truth” is a notion of truth at all. Isn’t disquotation essential to truth (see, e.g., Harth 2014)? It is no coincidence that one of the main protagonists of current relativism, Max Kölbel, while acknowledging the lack of

24   Relativists might complain that in the context of the remarks above (Contents) begs the question against them.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

226  ALTERNATIVES? disquotational properties for relative truth (see also Richard  2008, 106–8, 135–6) replaces talk of “truth” by talk of “correctness” (2008, 250–2), claiming that correctness is close enough to truth in non-relativist cases. But why should that also be so in relativist cases? And wouldn’t that be necessary? Aren’t many other things closely related to but not identical with truth, like, say, ideal warranted assertibility? To be sure, one can make a “close-enough” type of argument in the cases of truth at the indices of time or world because there is a close enough content satisfying disquotational schemas; the problem, however, is that this route is blocked in the case of assessment-sensitivity (see above). MacFarlane (2005a, 2014, ch.5) puts great weight on the claim that certain rules about assertion, especially about their retraction, favor relativism over its competitors. The role the notion of relative truth plays in our practice of assertion and the fact that there are certain rules of assertion supports the claim that the notion of relative truth really is a notion of truth. However, it can be doubted that these rules have to be seen as rules about truth instead of, say, ideal warranted assertibility or something else. Here is what I have in mind. MacFarlane 2005a argues that a given notion is a notion of truth if it plays a specific role in our practice of assertion. Taking the notion of monadic truth first and focusing on the rule for retraction we get the commitment “to withdraw the assertion if and when it is shown to have been untrue” (2005a, 318). That is, (Role) A notion F is a notion of monadic truth if the following rule of retraction holds: withdraw the assertion if and when it is shown to have been not-F.25 The problem is that at least prima facie this schema seems to hold for other notions different from the notion of truth, too, like, e.g., for the notion of justification. The rule to withdraw an assertion if and when it is shown to have been unjustified seems very plausible. So, the notion of justification satisfies the rule of retraction in the right-hand side of (Role) but it is not a notion of truth. Hence, (Role) seems incorrect and MacFarlane’s general strategy to show that the notion of relative truth really deserves to be called a notion of “truth” is in trouble. One could, of course, argue against the claim that justification or other non-truth notions play this particular kind of role for our practice of assertion—that, e.g., the justification rule for assertion is incorrect. However, even if that should be made plausible, this strategy has the disadvantage of making relativism crucially dependent on one’s views about the rule for assertion (if there are any such rules). We get similar results if we consider MacFarlane’s relativist version of the rule for retraction: (W*) In asserting that p at C1, one commits oneself to withdrawing the assertion (in any future context C2) if p is shown to be untrue relative to context of use C1, and context of assessment C2. (2005a, 320) 25   Note that Macfarlane only talks about a sufficient condition for retraction; it is not clear that switching to necessary or to necessary and sufficient conditions will change anything here.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  227 or Retraction Rule. An agent in context c2 is required to retract an (unretracted) assertion of p made at c1 if p is not true as used at c1 and assessed from c2. (2014, 108)

Just replace “not true” or “untrue” by “unjustified” and you get some at least prima facie plausible cousins of (W*) or Retraction Rule. Again, one could deny the corres­ ponding justification rules for assertion but the question is whether the relativist really should be willing to make relativism dependent on some very non-trivial and controversial account of assertion according to which only the notion of truth meets the following schema: (S) An agent in context c2 is required to retract an (unretracted) assertion of p made at c1 if p is not F as used at c1 and assessed from c2. There is thus some reason to worry that truth relativism faces a dilemma: with disquotation it becomes inconsistent, without disquotation it’s not about truth anymore. “Nonindexical contextualism” is a new term for a recent view which differs from relativism insofar as the assessment parameter is fixed by the context of use (see Recanati 2008; Brogaard 2008a, 2008b; MacFarlane 2009, 2014, ch.8.4.1; Kompa 2012; Davis 2013b): S as used at c1 is true just in case S as used at c1 is true at the index (where w1 and t1 are the world and time of c1 and a1 is the assessment parameter of c1). This view faces the same kind of problems discussed in this section and in the one above (but cf. Greenough 2011, 200–1 who denies that relativism shares the lack of an equivalence schema with non-indexical contextualism).

8.3.3  Other Problems? Jason Stanley has argued (2005, 145–7) that relativism about knowledge cannot account for the factivity of knowledge, that is, for the validity of a schema like (taking a non-relativist version) (F) If S knows that p, then p. Stanley distinguishes between three different candidates for a relativist factivity principle. The first one which makes truth relative to all judges (see Stanley 2005, 145–6) is obviously too strong and it is hard to imagine any relativist holding this view. The ­second option (see Stanley 2005, 146–7) seems to be the truly relativist one. That John knows that p might be true as assessed by John but not true as assessed by Hannah. Stanley seems to have no direct quarrel with that. His objection is rather based on the idea that John and Hannah also know something else that is true at their respective contexts of assessment: namely that John knows that p (John’s context) and that John does not know that p (Hannah’s context). Stanley takes the relativist to avoid this

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

228  ALTERNATIVES? contradiction by being more precise: John knows that John knows that p is true as assessed at John’s context of assessment while Hannah knows that John doesn’t know that p is true as assessed at Hannah’s context of assessment.26 But with the latter step, Stanley argues, the disagreement is gone. This is probably right but beside the point: that A and B disagree about a not explicitly relativized knowledge claim, does not entail that they also disagree about the relativized knowledge claim. At least, Stanley would have to say much more here and argue for the substantial claim that if one disagrees about the unrelativized claim, then one also disagrees about the relativized claim. The notion of disagreement is very tricky and it is very hard to avoid question-begging maneuvers here. Stanley’s third option (2005, 147) seems to identify the judge with the subject which is not a relativist idea at all. Apart from that, Stanley objects that the relativist would have to accept that Kxp could be true at y’s context even if p is not true at y’s context (see also Brogaard 2008a, sec.3, and 2008b, 447–50 for this). The critical passage is very brief and Stanley does not make the presumed assessment-sensitivity explicit. However, if one does, the problem disappears: at every context of assessment factivity holds even if it does not hold across contexts. If Kp is true at ay, then so will p; that p might not be true at ax does not matter and is beside the point (see also Cappelen and Hawthorne 2009, 130–1; Montminy 2009a, sec.4; Richard 2008, 168–9; MacFarlane 2014, 195–6). But back to this question: What could relativist factivity look like? One good starting point would be something like (FR-1) If “S knows that p” is true as assessed in a, then the proposition that p is true as assessed in a. One might want something stronger since at least in many cases the relevant prop­ o­sition p will also be true in other contexts of assessment though perhaps not in all contexts of assessment. One could propose (FR-2) If “S knows that p” is true as assessed in a, then the proposition that p is true as assessed in a and some other contexts but not in all contexts of assessment. But things are still a bit more complicated. Consider mixed cases where “knowledge” is taken to be assessment-sensitive while the proposition presumably “known” or not “known” is assessment-insensitive. It is plausible to hold that the proposition that Ann has hands is not assessment-sensitive (if you have doubts, consider the proposition that 2 + 3 = 5 or your favorite candidate for assessment-insensitivity). Allowing for such mixed cases, the relativist would have to accept a principle like the following: 26   This is already more precise than Stanley’s quite informal presentation but in principle one should be even more explicit and say something like: “John knows that John knows that p is true as assessed at John’s context of assessment” is true as assessed at, say, the relativist’s context and “Hannah knows that John doesn’t know that p is true as assessed at Hannah’s context of assessment” is also true as assessed at the relativist’s context. For simplicity’s sake we can disregard this further complexity but it deserves to be noted briefly.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

ALTERNATIVES?  229 (FR-3) If “S knows that p” is true as assessed in a, then (a) if the proposition that p is itself assessment-insensitive, then it is true as assessed in all contexts of assessment whatsoever whereas (b) if the proposition that p is assessment-sensitive, then it is true as assessed in a and some other contexts but not in all contexts of assessment. More would need to be added but here we can leave it at that. Relativism does not have a problem with the factivity of knowledge. It also does not suffer from knowability problems (see Chapter 5 but cf. also Brendel 2014); the lack of the disquotational property saves relativism from that problem (though the price is high: see above). I want to end by briefly mentioning two other issues which might pose real problems for relativism. A relativist should still be able to use a monadic notion of truth and say things like “Jack knows that the earth is not flat and when you say that he doesn’t know it, then you’re just wrong!” As a relativist, they should at the same time be entitled to use a relative notion of truth and say things like “Jack knows that the earth is not flat but, strictly speaking, that he doesn’t know it is true at your context of assessment while false at mine!” The point is not so much that such statements sound odd. That in itself does not constitute a strong objection. But there is a serious worry underlying this oddity: Can a relative and a monadic notion of truth peacefully coexist in the relativist’s view (with both being acknowledged as notions of truth)? Isn’t there a potential for self-defeat here insofar as the use of the relative notion of truth undermines the use of the monadic notion of truth? Counterfactual and temporal embeddings raise similar questions (for this kind of problem see also Lawlor 2005, 218, and Boghossian 2011). The second issue I want to bring up has to do with the question what exactly a context of assessment is and how it differs from a context of attribution. What does it mean to assess a proposition for truth? At least in standard cases27 one has to think a proposition in order to assess it for truth. Furthermore, one can only come to attribute truth to a given proposition if one assents to it. On the basis of what could one attribute truth if not on the basis of assent? Similarly, one can only come to attribute falsehood to a given proposition if one denies it. What else could be the basis for that attribution? However, if all this is correct, then at least in the typical case assessment of knowledge claims for truth involves attribution (or denial) of knowledge. But then it is not clear anymore what the difference between relativism and contextualism (of the “non-indexical” or the non-non-indexical variant) could consist in.28 To put it in a more critical way: the relativist needs to propose a notion of context of assessment that is interestingly relativist. So far, it seems, this notion has rather been taken as an unproblematic given.29 27   Where one grasps the proposition directly. Cases where one assesses someone’s statement for truth without knowing its content (“Whatever Charles says, it’s true”) seem secondary here. 28   For a quite different debate about the substance or lack thereof of the differences between relativism and some alternatives see Stojanovic 2007, sec.4, 2012, 623; Dreier 2009, 90–2; and Lasersohn 2008, 317–18. 29   These seem to constitute problems for the non-indexical contextualist, too.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

230  ALTERNATIVES?

8.4 Conclusion We can thus conclude here that relativism does not seem to offer a better alternative to contextualism (at least of the kind proposed here). The overall conclusion of this chapter is quite straightforward: if one does not want to embrace classical invariantism (see the arguments against that especially in Chapters 1–3), then the version of context­ ualism defended here does better than the three main “variantist” alternatives, SSI, contrastivism, and relativism. This also completes my attempt to explain and defend a version of epistemic contextualism. I take it that “more theoretical” arguments are more important than arguments from cases. I also think that the view defended here can deal with import­ ant problems and has a clear advantage over alternative views. On a couple of topics I disagree with the current mainstream (e.g., on knowledge and lotteries, or on closure). I hope this will be counted as an advantage in times where there might not be quite enough explicit discussion of the presuppositions and allegedly “truistic” assumptions underlying a given debate.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography Abath, André J. 2012, Epistemic Contextualism, Semantic Blindness and Content Unawareness, Australasian Journal of Philosophy 90, 593–7. Adler, Jonathan E. 1986, Knowing, Betting and Cohering, Philosophical Topics 14, 243–57. Adler, Jonathan E. 2006, Withdrawal and Contextualism, Analysis 66, 280–5. Adler, Jonathan E. 2012, Contextualism and Fallibility: Pragmatic Encroachment, Possibility and Strength of Epistemic Position, Synthese 188, 247–72. Adler, Jonathan and Levin, Michael 2002, Is the Generality Problem Too General? Philosophy and Phenomenological Research 65, 87–97. Almér, Alexander 2005, Truth Relativism and Genuine Disagreement, in Dag Westerståhl and Torbjörn Tännsjö (eds), Lectures on Relativism, Göteborg: Department of Philosophy, Göteborg University, 123–57. Alston, William P. 1986, Epistemic Circularity, Philosophy and Phenomenological Research 47, 1–30. Alston, William P. 1989, Internalism and Externalism in Epistemology, in William P. Alston, Epistemic Justification: Essays in the Theory of Knowledge, Ithaca, NY: Cornell University Press, 185–226. Alston, William P. 1995, How to Think about Reliability, Philosophical Topics 23, 1–29. Alston, William P. 2005, Beyond “Justification”: Dimensions of Epistemic Evaluation, Ithaca, NY: Cornell University Press. Annis, David B. 1978, A Contextualist Theory of Epistemic Justification, American Philosophical Quarterly 15, 213–19. Armstrong, David M. 1973, Belief, Truth and Knowledge, Cambridge: Cambridge University Press. Audi, Robert 1995, Deductive Closure, Defeasibility, and Scepticism: A Reply to Feldman, Philosophical Quarterly 45, 494–9. Audi, Robert 1998, Epistemology: A Contemporary Introduction to the Theory of Knowledge, London and New York: Routledge. Austin, John L. 1970, A Plea for Excuses, John L. Austin, Philosophical Papers (2nd edition), Oxford: Clarendon, 175–204. Bach, Kent 2005, The Emperor’s New “Knows,” in Gerhard Preyer and Georg Peter (eds), Contextualism in Philosophy: Knowledge, Meaning, and Truth, Oxford: Clarendon, 51–89. Bach, Kent 2008, Applying Pragmatics to Epistemology, Philosophical Issues 18, 68–88. Ball, Brian and Blome-Tillmann, Michael 2014, Counter Closure and Knowledge despite Falsehood, Philosophical Quarterly 64, 552–68. Ballantyne, Nathan 2011, Anti-Luck Epistemology, Pragmatic Encroachment, and True Belief, Canadian Journal of Philosophy 41, 485–504. Ballantyne, Nathan 2012, Luck and Interests, Synthese 185, 319–34. Barke, Antonia 2002, The Closure of Knowledge in Context, Paderborn: mentis. Baumann, Peter 2000, Knowledge and Context, in Berit Brogaard (ed.), Rationality and Irrationality, Kirchberg: Austrian Ludwig Wittgenstein Society, 53–7.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

232  Bibliography Baumann, Peter 2001, Im Auge des Betrachters. Über Wissen, Rechtfertigung und Kontext, in Thomas Grundmann (ed.), Erkenntnistheorie. Positionen zwischen Tradition und Gegenwart, Paderborn: mentis, 72–89. Baumann, Peter 2004, Lotteries and Contexts, Erkenntnis 61, 415–28. Baumann, Peter 2005a, Three Doors, Two Players, and Single Case Probabilities, American Philosophical Quarterly 42, 71–9. Baumann, Peter 2005b, Varieties of Contextualism: Standards and Descriptions, Grazer Philosophische Studien 69, 229–45. Baumann, Peter 2006, Information, Closure, and Knowledge: On Jäger’s Objection to Dretske, Erkenntnis 77, 403–8. Baumann, Peter 2008a, Is Knowledge Safe? American Philosophical Quarterly 45, 19–31. Baumann, Peter 2008b, Contextualism and the Factivity Problem, Philosophy and Phenomenological Research 76, 580–602. Baumann, Peter 2008c, Contrastivism Rather than Something Else? On the Limits of Epistemic Contrastivism, Erkenntnis 69, 189–200. Baumann, Peter 2008d, Problems for Sinnott-Armstrong’s Moral Contrastivism, Philosophical Quarterly 58, 463–70. Baumann, Peter 2009a, Reliabilism: Modal, Probabilistic or Contextualist, Grazer Philosophische Studien 79, 77–89. Baumann, Peter 2009b, Was Moore a Moorean? On Moore and Scepticism, European Journal of Philosophy 17, 181–200. Baumann, Peter 2010, Factivity and Contextualism, Analysis 70, 82–9. Baumann, Peter 2011a, Epistemic Closure, in Sven Bernecker and Duncan Pritchard (eds), The Routledge Companion to Epistemology, London: Routledge, 597–608. Baumann, Peter 2011b, A Puzzle about Responsibility, Erkenntnis 74, 207–24. Baumann, Peter 2011c, WAMs: Why Worry? Philosophical Papers 40, 155–77. Baumann, Peter 2012a, Nozick’s Defense of Closure, in Kelly Becker and Tim Black (eds), The Sensitivity Principle in Epistemology, Cambridge: Cambridge University Press, 11–27. Baumann, Peter 2012b, Knowing about Other Contexts, in Christoph Jaeger and Winfried Loeffler (eds), Epistemology: Context, Values, Disagreement, Frankfurt: Ontos, 63–79. Baumann, Peter 2012c, Knowledge, Practical Reasoning and Action, Logos and Episteme 3, 7–26. Baumann, Peter 2012d, PS: Response to Schaffer’s Reply, in Stefan Tolksdorf (ed.), Conceptions of Knowledge, Berlin: de Gruyter, 425–31. Baumann, Peter 2013, Gettier, Wissen, Zufall, in Gerhard Ernst and Lisa Marani (eds), Das Gettierproblem. Eine Bilanz nach 50 Jahren, Paderborn: mentis, 9–27. Baumann, Peter 2014a, No Luck with Knowledge? On a Dogma of Epistemology, Philosophy and Phenomenological Research 89, 523–51. Baumann, Peter 2014b, A Contradiction for Contextualism, in Franck Lihoreau and Manuel Rebuschi (eds), Epistemology, Context, and Formalism, Heidelberg: Springer, 49–57. Baumann, Peter 2014c, Knowledge, Assertion, and Inference, Acta Analytica 29, 487–90. Baumann, Peter 2015a, Begriffe Analysieren? in Dirk Koppelberg and Stefan Tolksdorf (eds), Erkenntnistheorie: wie und wozu? Paderborn: mentis, 133–51. Baumann, Peter 2015b, Safety, Virtue, Scepticism: Remarks on Sosa, Croatian Journal of Philosophy XV(45), 295–306.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  233 Baumann, Peter forthcoming, Knowledge across Contexts. A Problem for Subject-Sensitive Invariantism, Dialogue. Beall, J. C. 2006, Modelling the “Ordinary View,” in Patrick Greenough and Michael P. Lynch (eds), Truth and Realism, Oxford: Oxford University Press, 61–74. Becker, Kelly 2007, Epistemology Modalized, New York and London: Routledge. Becker, Kelly 2008, Epistemic Luck and the Generality Problem, Philosophical Studies 139, 353–66. Becker, Kelly 2009, Contrastivism and Lucky Questions, Philosophia 37, 245–60. Becker, Kelly and Black, Tim (eds) 2012, The Sensitivity Principle in Epistemology, Cambridge: Cambridge University Press. Bedke, Matthew S. 2010, Developmental Process Reliabilism: On Justification, Defeat, and Evidence, Erkenntnis 73, 1–17. Beebe, James R. 2004, The Generality Problem, Statistical Relevance and the Tri-Level Hypothesis, Noûs 38, 177–95. Bianchi, Claudia 2005, How to Be a Contextualist, Facta Philosophica 7, 261–72. Bishop, Michael A. 2010, Why the Generality Problem Is Everybody’s Problem, Philosophical Studies 151, 285–98. Björnsson, Gunnar and Finlay, Stephen 2010, Metaethical Contextualism Defended, Ethics 121, 7–36. Blaauw, Martijn 2005, Challenging Contextualism, Grazer Philosophische Studien 69, 127–46. Blaauw, Martijn 2008a, Contra Contrastivism, Philosophical Issues 18, 2–34. Blaauw, Martijn 2008b, Contesting Pyrrhonian Contrastivism, Philosophical Quarterly 58, 471–7. Blaauw, Martijn 2008c, Subject Sensitive Invariantism: In Memoriam, Philosophical Quarterly 58, 318–25. Blaauw, Martijn 2012, Defending Contrastivism, International Journal for the Study of Skepticism 2, 59–64. Black, Tim 2005, Classic Invariantism, Relevance and Warranted Assertibility Manœuvres, Philosophical Quarterly 55, 328–36. Black, Tim Ms., Contextualism, Closure, and Contingent A Priori Knowledge. Blome-Tillmann, Michael 2007, Contextualism and the Epistemological Enterprise, Proceedings of the Aristotelian Society 107, 387–94. Blome-Tillmann, Michael 2008, The Indexicality of “Knowledge,” Philosophical Studies 138, 29–53. Blome-Tillmann, Michael 2009a, Knowledge and Presuppositions, Mind 118, 241–94. Blome-Tillmann, Michael 2009b, Contextualism, Subject-Sensitive Invariantism, and the Interaction of “Knowledge”-Ascriptions with Modal and Temporal Operators, Philosophy and Phenomenological Research 79, 315–31. Blome-Tillmann, Michael 2012, Presuppositional Epistemic Contextualism and the Problem of Known Presuppositions, in Jessica Brown and Mikkel Gerken (eds), Knowledge Ascriptions, Oxford: Oxford University Press, 104–19. Blome-Tillmann, Michael 2014, Knowledge and Presuppositions, Oxford: Oxford University Press. Bogdan, Radu J. 1985, Cognition and Epistemic Closure, American Philosophical Quarterly 22, 55–63. Boghossian, Paul 2011, Three Kinds of Relativism, in Steven D. Hales (ed.), A Companion to Relativism, Malden, MA: Wiley-Blackwell, 53–69.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

234  Bibliography Bowie, G. Lee 1979, The Similarity Approach to Counterfactuals, Noûs 13, 477–98. Brandom, Robert B. 1994, Making It Explicit: Reasoning, Representing and Discursive Commitment, Cambridge, MA: Harvard University Press. Brandom, Robert B. 1998, Insights and Blindspots of Reliabilism, Monist 81, 371–92. Brendel, Elke 2003, Was Kontextualisten nicht wissen, Deutsche Zeitschrift für Philosophie 51, 1015–32. Brendel, Elke 2005, Why Contextualists Cannot Know They Are Right: Self-Refuting Implications of Contextualism, Acta Analytica 20–2 (35), 38–55. Brendel, Elke 2007, Kontextualismus oder Invariantismus? Zur Semantik epistemischer Aussagen, in Adolf Rami and Heinrich Wansing (eds), Referenz und Realität, Paderborn: mentis, 11–37. Brendel, Elke 2009, Contextualism, Relativism, and Factivity, in Hannes Leitgeb and Alexander Hieke (eds), Reduction and Elimination in Philosophy and the Sciences, Frankfurt am Main: Ontos, 403–16. Brendel, Elke 2012, Knowledge, Contextualism, and Moorean Paradox, in Christoph Jaeger and Winfried Loeffler (eds), Epistemology: Context, Values, Disagreement, Frankfurt: Ontos, 15–40. Brendel, Elke 2014, Contextualism, Relativism and the Semantics of Knowledge Ascriptions, Philosophical Studies 168, 101–17. Brogaard, Berit 2003, Epistemological Contextualism and the Problem of Moral Luck, Pacific Philosophical Quarterly 84, 351–70. Brogaard, Berit 2008a, In Defence of a Perspectival Semantics for “Know,” Australasian Journal of Philosophy 86, 439–59. Brogaard, Berit 2008b, Moral Contextualism and Moral Relativism, Philosophical Quarterly 58, 385–409. Brokes, Audre Jean 2001, What Does the Generality Problem Show? Pacific Philosophical Quarterly 82, 145–56. Broncano-Berrocal, Fernando 2014, Is Safety in Danger? Philosophia 42, 63–81. Broncano-Berrocal, Fernando 2015, Luck as Risk and the Lack of Control Account of Luck, Metaphilosophy 46, 1–25. Brown, Jessica 2004, Anti-Individualism and Knowledge, Cambridge, MA: MIT Press. Brown, Jessica 2005a, Adapt or Die: The Death of Invariantism? Philosophical Quarterly 55, 263–85. Brown, Jessica 2005b, Comparing Contextualism and Invariantism on the Correctness of Contextualist Intuitions, Grazer Philosophische Studien 69, 71–99. Brown, Jessica 2005c, Williamson on Luminosity and Contextualism, Philosophical Quarterly 55, 319–27. Brown, Jessica 2013, Experimental Philosophy, Contextualism and SSI, Philosophy and Phenomenological Research 86, 233–61. Brown, Jessica 2014, Shifty Talk: Knowledge and Causation, Philosophical Studies 167, 183–99. Brueckner, Anthony 1994a, The Shifting Content of Knowledge Attributions, Philosophy and Phenomenological Research 54, 123–6. Brueckner, Anthony 1994b, The Structure of the Skeptical Argument, Philosophy and Phenomenological Research 54, 827–35.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  235 Brueckner, Anthony 2004, The Elusive Virtues of Contextualism, Philosophical Studies 118, 401–5. Brueckner, Anthony 2005, Contextualism, Hawthorne’s Invariantism and Third-Person Cases, Philosophical Quarterly 55, 315–18. Brueckner, Anthony 2010, SSI Disinterred, Philosophical Quarterly 60, 160–1. Brueckner, Anthony 2012, Roush on Knowledge: Tracking Redux? in Kelly Becker and Tim Black (eds), The Sensitivity Principle in Epistemology, Cambridge: Cambridge University Press, 229–41. Brueckner, Anthony and Buford, Christopher T. 2009, Contextualism, SSI and the Factivity Problem, Analysis 69, 431–8. Brueckner, Anthony and Buford, Christopher T. 2010, Reply to Baumann on Factivity and Contextualism, Analysis 70, 486–9. Buckwalter, Wesley 2010, Knowledge Isn’t Closed on Saturday: A Study in Ordinary Language, Review of Philosophy and Psychology 1, 395–406. Buckwalter, Wesley 2014, The Mystery of Stakes and Error in Ascriber Intuitions, in James R. Beebe (ed.), Advances in Experimental Philosophy, London: Bloomsbury, 145–73. Buckwalter, Wesley and Schaffer, Jonathan 2015, Knowledge, Stakes, and Mistakes, Noûs 49, 201–34. Buenting, Joel 2010, An Epistemic Reduction of Contrastive Knowledge Claims, Social Epistemology 24, 99–104. Cappelen, Herman and Hawthorne, John 2009, Relativism and Monadic Truth, Oxford: Oxford University Press. Cappelen, Herman and Lepore, Ernie 2003, Context Shifting Arguments, Philosophical Perspectives 17, 25–50. Cappelen, Herman and Lepore, Ernie 2005, Insensitive Semantics: A Defense of Semantic Minimalism and Speech Act Pluralism, Oxford: Blackwell. Carnap, Rudolf 1958, Meaning and Necessity: A Study in Semantics and Modal Logic (2nd edition), Chicago: University of Chicago Press. Carrier, L. S. 1971, An Analysis of Empirical Knowledge, Southern Journal of Philosophy 9, 3–11. Christensen, David 2007, Three Questions about Leplin’s Reliabilism, Philosophical Studies 134, 43–50. Coffman, E. J. 2007, Thinking about Luck, Synthese 158, 385–98. Coffman, E. J. 2015, Luck: Its Nature and Significance for Human Knowledge and Agency, Basingstoke: Palgrave Macmillan. Cohen, Stewart 1986, Knowledge and Context, Journal of Philosophy 83, 574–83. Cohen, Stewart 1987, Knowledge, Context, and Social Standards, Synthese 73, 3–26. Cohen, Stewart 1988, How to Be a Fallibilist, Philosophical Perspectives 2, 91–123. Cohen, Stewart 1990, Scepticism and Everyday Knowledge Attributions, in Michael D. Roth and Glenn Ross (eds), Doubting: Contemporary Perspectives on Skepticism, Dordrecht: Kluwer, 161–9. Cohen, Stewart 1991, Skepticism, Relevance, and Relativity, in Brian McLaughlin (ed.), Dretske and His Critics, Oxford: Blackwell, 17–37. Cohen, Stewart 1998a, Contextualist Solutions to Epistemological Problems: Scepticism, Gettier, and the Lottery, Australasian Journal of Philosophy 76, 289–306.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

236  Bibliography Cohen, Stewart 1998b, Two Kinds of Skeptical Argument, Philosophy and Phenomenological Research 58, 143–59. Cohen, Stewart 1999, Contextualism, Skepticism, and the Structure of Reasons, Philosophical Perspectives 13, 57–89. Cohen, Stewart 2000, Contextualism and Skepticism, Philosophical Issues 10, 94–107. Cohen, Stewart 2001, Contextualism Defended: Comments on Richard Feldman’s Skeptical Problems, Contextualist Solutions, Philosophical Studies 103, 87–98. Cohen, Stewart 2002, Basic Knowledge and the Problem of Easy Knowledge, Philosophy and Phenomenological Research 65, 309–29. Cohen, Stewart 2004a, Contextualism and Unhappy-Face Solutions: Reply to Schiffer, Philosophical Studies 119, 185–97. Cohen, Stewart 2004b, Reply to Baumann, Erkenntnis 61, 429–33. Cohen, Stewart 2004c, Knowledge, Assertion, and Practical Reasoning, Philosophical Issues 14, 482–91. Cohen, Stewart 2005a, Contextualism Defended, in Mathias Steup and Ernest Sosa (eds), Contemporary Debates in Epistemology, Oxford: Blackwell, 56–62. Cohen, Stewart 2005b, Contextualism Defended Some More, in Mathias Steup and Ernest Sosa (eds), Contemporary Debates in Epistemology, Oxford: Blackwell, 67–71. Cohen, Stewart 2005c, Knowledge, Speaker and Subject, Philosophical Quarterly 55, 199–212. Collins, John 2006, Lotteries and the Close Shave Principle, in Stephen Hetherington (ed.), Aspects of Knowing: Epistemological Essays, Amsterdam: Elsevier, 83–96. Colyvan, Mark, Regan, Helen M., and Ferson, Scott 2001, Is It a Crime to Belong to a Reference Class? Journal of Political Philosophy 9, 168–81. Comesaña, Juan 2005, Unsafe Knowledge, Synthese 146, 395–404. Comesaña, Juan 2006, A Well-Founded Solution to the Generality Problem, Philosophical Studies 129, 27–47. Conee, Earl 2005a, Contextualism Contested, in Mathias Steup and Ernest Sosa (eds), Contemporary Debates in Epistemology, Oxford: Blackwell, 47–56. Conee, Earl 2005b, Contextualism Contested Some More, in Mathias Steup and Ernest Sosa (eds), Contemporary Debates in Epistemology, Oxford: Blackwell, 62–6. Conee, Earl 2013, The Specificity of the Generality Problem, Philosophical Studies 163, 751–62. Conee, Earl and Feldman, Richard 1998, The Generality Problem for Reliabilism, Philosophical Studies 89, 1–29. Daukas, Nancy 2002, Skepticism, Contextualism, and the Epistemic “Ordinary,” Philosophical Forum 33, 63–79. David, Marian and Warfield, Ted A. 2008, Knowledge-Closure and Scepticism, in Quentin Smith (ed.), Epistemology: New Essays, Oxford: Oxford University Press, 137–87. David, Marian and Warfield, Ted A. Ms., Six Possible Counterexamples to One or Two Epistemic Closure Principles. Davidson, Donald 1984, Truth and Meaning, in Donald Davidson, Inquiries into Truth and Interpretation, Oxford: Clarendon, 17–36. Davidson, Donald 1986, A Nice Derangement of Epitaths, in Ernest Lepore (ed.), Truth and Interpretation: Perspectives on the Philosophy of Donald Davidson, Oxford: Blackwell, 433–46.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  237 Davies, Martin 1998, Externalism, Architecturalism and Epistemic Warrant, in Crispin Wright, Barry Smith, and Cynthia Macdonald (eds), Knowing Our Own Minds, Oxford: Clarendon, 321–61. Davies, Martin 2000, Externalism and Armchair Knowledge, in Paul Boghossian and Christopher Peacocke (eds), New Essays on the Apriori, Oxford: Clarendon, 384–414. Davies, Martin 2003a, The Problem of Armchair Knowledge, in Susana Nuccetelli (ed.), New Essays on Semantic Externalism and Self-Knowledge, Cambridge, MA and London: MIT Press, 23–55. Davies, Martin 2003b, Externalism, Self-Knowledge and Transmission of Warrant, in Maria J. Frápolli and Esther Romero (eds), Meaning, Basic Self-Knowledge, and Mind: Essays on Tyler Burger, Stanford, CA: CSLI Publications, 105–30. Davies, Martin 2004, On Epistemic Entitlement: Epistemic Entitlement, Warrant Transmission and Easy Knowledge, Proceedings of the Aristotelian Society, Supplement 78, 213–45. Davis, Wayne A. 2004, Are Knowledge Claims Indexical? Erkenntnis 61, 257–81. Davis, Wayne A. 2005, Contextualist Theories of Knowledge, Acta Analytica 20, 29–42. Davis, Wayne A. 2007, Knowledge Claims and Context: Loose Use, Philosophical Studies 132, 395–438. Davis, Wayne A. 2013a, Grice’s Razor and Epistemic Invariantism, Journal of Philosophical Research 38, 147–76. Davis, Wayne A. 2013b, On Nonindexical Contextualism, Philosophical Studies 163, 561–74. Davis, Wayne A. 2015, Knowledge Claims and Context: Belief, Philosophical Studies 172, 399–432. de Almeida, Claudio 2012, Epistemic Closure, Skepticism and Defeasibility, Synthese 188, 197–215. Dennett, Daniel 1984, Elbow Room: The Varieties of Free Will Worth Wanting, Cambridge, MA: MIT Press. Derksen, A. A. 1978, The Alleged Lottery Paradox Revisited, American Philosophical Quarterly 15, 67–74. DeRose, Keith 1992, Contextualism and Knowledge Attributions, Philosophy and Phenomenological Research 52, 913–29. DeRose, Keith 1995, Solving the Skeptical Problem, Philosophical Review 104, 1–52. DeRose, Keith 1996a, Relevant Alternatives and the Content of Knowledge Attributions, Philosophy and Phenomenological Research 56, 193–7. DeRose, Keith 1996b, Knowledge, Assertion and Lotteries, Australasian Journal of Philosophy 74, 568–80. DeRose, Keith 1998, Simple “Might”s, Indicative Possibilities and the Open Future, Philosophical Quarterly 48, 67–82. DeRose, Keith 1999, Contextualism: An Explanation and Defense, in John Greco and Ernest Sosa (eds), The Blackwell Guide to Epistemology, Oxford: Blackwell, 187–205. DeRose, Keith 2000, Now You Know it, Now You Don’t, Proceedings of the Twentieth World Congress of Philosophy, Bowling Green/OH: Philosophy Documentation Center, vol. V (Epistemology), 91–106. DeRose, Keith 2002, Assertion, Knowledge, and Context, Philosophical Review 111, 167–203. DeRose, Keith 2004a, Single Scoreboard Semantics, Philosophical Studies 119, 1–21.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

238  Bibliography DeRose, Keith 2004b, The Problem with Context-Sensitive Invariantism, Philosophy and Phenomenological Research 68, 346–50. DeRose, Keith 2005, The Ordinary Language Basis for Contextualism and the New Invariantism, Philosophical Quarterly 55, 172–98. DeRose, Keith 2006, “Bamboozled by Our Own Words”: Semantic Blindness and Some Arguments against Contextualism, Philosophy and Phenomenological Research 73, 316–38. DeRose, Keith 2009, The Case for Contextualism: Knowledge, Skepticism, and Context, vol. 1, Oxford: Clarendon. DeRose, Keith 2011, Contextualism, Contrastivism, and X-Phi Surveys, Philosophical Studies 156, 81–110. Descartes, René 1907–13, Meditationes de prima philosophia, Oeuvres de Descartes (eds Charles Adam and Paul Tannery), Paris: Vrin (reprint Paris 1964–1976), vol. VII. Di Bello, Marcello 2014, Epistemic Closure, Assumptions and Topics of Inquiry, Synthese 191, 3977–4002. Dinges, Alexander forthcoming, Epistemic Invariantism and Contextualist Intuitions, Episteme. Douven, Igor 2004, The Context-Insensitivity of “Knowing More” and “Knowing Better,” Canadian Journal of Philosophy 34, 313–26. Douven, Igor 2007, A Pragmatic Dissolution of Harman’s Paradox, Philosophy and Phenomenological Research 74, 326–45. Douven, Igor 2008, The Lottery Paradox and Our Epistemic Goal, Pacific Philosophical Quarterly 89, 204–25. Dreier, James 1990, Internalism and Speaker Relativism, Ethics 101, 6–26. Dreier, James 2009, Relativism (and Expressivism) and the Problem of Disagreement, Philosophical Perspectives 23, 79–110. Dretske, Fred I. 1970, Epistemic Operators, Journal of Philosophy 67, 1007–23. Dretske, Fred I. 1971, Conclusive Reasons, Australasian Journal of Philosophy 49, 1–22. Dretske, Fred 1972, Contrastive Statements, Philosophical Review 81, 411–37. Dretske, Fred I. 1981a, Knowledge and the Flow of Information, Cambridge, MA: MIT Press. Dretske, Fred I. 1981b, The Pragmatic Dimension of Knowledge, Philosophical Studies 40, 363–78. Dretske, Fred I. 1982, A Cognitive Cul-de-sac, Mind 91, 109–11. Dretske, Fred I. 2003, Skepticism: What Perception Teaches, in Steven Luper (ed.), The Skeptics: Contemporary Essays, Aldershot: Ashgate, 105–18. Dretske, Fred I. 2004, Externalism and Modest Contextualism, Erkenntnis 61, 173–86. Dretske, Fred I. 2005a, The Case against Closure, in Matthias Steup and Ernest Sosa (eds), Contemporary Debates in Epistemology, Malden, MA: Blackwell, 13–26. Dretske, Fred I. 2005b, Reply to Hawthorne, in Matthias Steup and Ernest Sosa (eds), Contemporary Debates in Epistemology, Malden, MA: Blackwell, 43–6. Dretske, Fred I. 2006, Information and Closure, Erkenntnis 77, 409–13. Dudman, V. H. 1992, Probability and Assertion, Analysis 52, 204–11. Dutant, Julien and Olsson, Erik J. 2013, Is There a Statistical Solution to the Generality Problem? Erkenntnis 78, 1347–65. Einheuser, Iris 2008, Three Forms of Truth Relativism, in Max Kölbel and Manuel GarcíaCarpintero (eds), Relative Truth, Oxford: Oxford University Press, 187–203.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  239 Engel, Mylan, Jr. 1992, Is Epistemic Luck Compatible with Knowledge? Southern Journal of Philosophy 30 (2), 59–75. Engel, Mylan, Jr. 2004, What’s Wrong with Contextualism, and a Noncontextualist Solution to the Skeptical Paradox, Erkenntnis 61, 203–31. Engel, Mylan, Jr. 2005, A Noncontextualist Account of Contextualist Linguistic Data, Acta Analytica 20–2 (35), 57–79. Ernst, Gerhard 2004, In Defense of Indexicalism: Comments on Davis, Erkenntnis 61, 283–93. Fantl, Jeremy and McGrath, Matthew 2002, Evidence, Pragmatics, and Justification, Philosophical Review 111, 67–94. Fantl, Jeremy and McGrath, Matthew 2009, Knowledge in an Uncertain World, Oxford: Oxford University Press. Fantl, Jeremy and McGrath, Matthew 2012a, Arguing for Shifty Epistemology, in Jessica Brown and Mikkel Gerken (eds), Knowledge Ascriptions, Oxford: Oxford University Press, 55–74. Fantl, Jeremy and McGrath, Matthew 2012b, Contextualism and Subject-Sensitivity, Philosophy and Phenomenological Research 84, 693–702. Fantl, Jeremy and McGrath, Matthew 2012c, Replies to Cohen, Neta and Reed, Philosophy and Phenomenological Research 85, 473–90. Feldman, Richard 1985a, Reliability and Justification, Monist 68, 159–74. Feldman, Richard 1985b, Schmitt on Reliability, Objectivity, and Justification, Australasian Journal of Philosophy 63, 354–60. Feldman, Richard 1995, In Defence of Closure, Philosophical Quarterly 45, 487–94. Feldman, Richard 1999, Contextualism and Skepticism, Philosophical Perspectives 13, 91–114. Feldman, Richard 2001, Skeptical Problems, Contextualist Solutions, Philosophical Studies 103, 61–85. Feldman, Richard and Conee, Earl 1985, Evidentialism, Philosophical Studies 48, 15–34. Feldman, Richard and Conee, Earl 2002, Typing Problems, Philosophy and Phenomenological Research 65, 98–105. Feltz, Adam and Zarpentine, Chris 2010, Do You Know More When It Matters Less? Philosophical Psychology 23, 687–706. Fetzer, James H. 1977, Reichenbach, Reference Classes, and Single Case Probabilities, Synthese 34, 185–217. Fine, Kit 1975, Critical Notice [on Lewis 1973], Mind 84, 451–8. Fodor, Jerry A. 1975, The Language of Thought, Cambridge, MA: Harvard University Press. Fogelin, Robert J. 2000, Contextualism and Externalism: Trading in One Form of Skepticism for Another, Philosophical Issues 10, 43–57. Foley, Richard 1984, Epistemic Luck and the Purely Epistemic, American Philosophical Quarterly 21, 113–24. Foley, Richard 1985, What’s Wrong with Reliabilism? Monist 68, 188–202. Foley, Richard 1987, The Theory of Epistemic Rationality, Cambridge, MA: Harvard University Press. Foley, Richard 2012, When Is True Belief Knowledge? Princeton, NJ: Princeton University Press. Forbes, Graeme 1984, Nozick on Scepticism, Philosophical Quarterly 34, 43–52. Frege, Gottlob 1884, Die Grundlagen der Arithmetik: Eine logisch-mathematische Untersuchung über den Begriff der Zahl, Breslau: W. Koebner.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

240  Bibliography Freitag, Wolfgang 2011, Epistemic Contextualism and the Knowability Problem, Acta Analytica 26, 273–84. Freitag, Wolfgang 2012, Epistemic Variantism and the Factivity of Knowledge, in Christoph Jaeger and Winfried Loeffler (eds), Epistemology: Context, Values, Disagreement, Frankfurt: Ontos, 81–94. Freitag, Wolfgang 2013, I Know: Modal Epistemology and Scepticism, Münster: mentis. Fumerton, Richard 1988, The Internalism/Externalism Controversy, Philosophical Perspectives 2 (James E. Tomberlin (ed.), Epistemology, Atascadero, CA: Ridgeview), 443–59. Fumerton, Richard 2010, Fencing Out Pragmatic Encroachment, Philosophical Perspectives 24, 243–53. Gendler, Tamar Szabó and Hawthorne, John 2005, The Real Guide to Fake Barns: A Catalogue of Gifts for Your Epistemic Enemies, Philosophical Studies 124, 331–52. Gerken, Mikkel and Beebe, James R. 2016, Knowledge in and out of Contrast, Noûs 50, 133–64. Gettier, Edmund L. 1963, Is Justified True Belief Knowledge? Analysis 23, 121–3. Gillies, Anthony S. Ms., Shifty Epistemology. Gillies, Donald 2000, Philosophical Theories of Probability, London: Routledge. Goldberg, Sandy 2015, Epistemic Entitlement and Luck, Philosophy and Phenomenological Research 91, 273–302. Goldman, Alan 2008, Knowledge, Explanation, and Lotteries, Noûs 42, 466–81. Goldman, Alvin I. 1986, Epistemology and Cognition, Cambridge, MA and London: Harvard University Press. Goldman, Alvin I. 1988, Strong and Weak Justification, Philosophical Perspectives 2, 51–69. Goldman, Alvin I. 1992a, Discrimination and Perceptual Knowledge, in Alvin I. Goldman, Liaisons: Philosophy Meets the Cognitive and Social Sciences, Cambridge, MA and London: MIT Press, 85–103. Goldman, Alvin I. 1992b, What Is Justified Belief? in Alvin I. Goldman, Liaisons: Philosophy Meets the Cognitive and Social Sciences, Cambridge, MA and London: MIT Press, 105–26. Goodman, Nelson 1972, Seven Strictures on Similarity, in Nelson Goodman, Problems and Projects, New York: Bobbs Merrill, 337–47. Greco, John 2003, Knowledge as Credit for True Belief, in Michael DePaul and Linda Zagzebski (eds), Intellectual Virtue: Perspectives from Ethics and Epistemology, Oxford: Oxford University Press, 111–34. Greco, John 2004, A Different Sort of Contextualism, Erkenntnis 61, 383–400. Greco, John 2006, Virtue, Luck and the Pyrrhonian Problematic, Philosophical Studies 130, 9–34. Greco, John 2008, What’s Wrong with Contextualism, Philosophical Quarterly 58, 416–36. Greco, John 2010, Achieving Knowledge: A Virtue-Theoretic Account of Epistemic Normativity, Cambridge: Cambridge University Press. Greene, Richard 2001, A Rejection of the Epistemic Closure Principle, Southwest Philosophy Review 17, 59–73. Greenough, Patrick 2011, Truth-Relativism, Norm-Relativism, and Assertion, in Jessica Brown and Herman Cappelen (eds), Assertion: New Philosophical Essays, Oxford: Oxford University Press, 197–231. Grice, Paul 1989, Logic and Conversation, in Paul Grice, Studies in the Ways of Words, Cambridge, MA: Harvard University Press, 22–40.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  241 Grimm, Stephen 2011, On Intellectualism in Epistemology, Mind 120, 705–33. Grimm, Stephen 2015, Knowledge, Practical Interests, and Rising Tides, in John Greco and David Henderson (eds), Epistemic Evaluation: Point and Purpose in Epistemology, Oxford: Oxford University Press, 117–37. Grobler, Adam 2001, Truth, Knowledge, and Presupposition, Logique et Analyse 44 (173–5), 291–305. Gundersen, Lars Bo 2003, Dispositional Theories of Knowledge, Aldershot and Burlington, VT: Ashgate. Gundersen, Lars Bo 2009, Disjunctivism, Contextualism and the Sceptical Aporia, Synthese 171, 387–97. Hájek, Alan 2003, Conditional Probability Is the Very Guide to Life, in Henry Kyburg and Mariam Thalos (eds), Probability Is the Very Guide to Life: The Philosophical Uses of Chance, Chicago, IL: Open Court, 183–203. Hájek, Alan 2007, The Reference Class Problem Is Your Problem Too, Synthese 156, 563–85. Hájek, Alan and Smithson, Michael 2012, Rationality and Indeterminate Probabilities, Synthese 187, 33–48. Hales, Steven D. 1995, Epistemic Closure Principles, Southern Journal of Philosophy 33, 185–201. Halliday, Daniel 2005, What Explains Our Intuitions about Knowledge Ascriptions? Proceedings of the Aristotelian Society 105, 393–402. Halliday, Daniel 2007, Contextualism, Comparatives and Gradability, Philosophical Studies 132, 381–93. Hambourger, Robert 1987, Justified Assertion and the Relativity of Knowledge, Philosophical Studies 51, 241–69. Hannon, Michael 2013, The Practical Origins of Epistemic Contextualism, Erkenntnis 78, 899–919. Hannon, Michael 2015, Stabilizing Knowledge, Pacific Philosophical Quarterly 96, 116–39. Hansen, Nat and Chemla, Emmanuel 2013, Experimenting on Contextualism, Mind and Language 28, 286–321. Harman, Gilbert 1968, Knowledge, Inference, and Explanation, American Philosophical Quarterly 5, 164–73. Harman, Gilbert 1973, Thought, Princeton, NJ: Princeton University Press. Harman, Gilbert 1986, Change in View: Principles of Reasoning, Cambridge, MA: MIT Press. Harman, Gilbert 2007, Epistemic Contextualism as a Theory of Primary Speaker Meaning, Philosophy and Phenomenological Research 75, 173–9. Harman, Gilbert and Sherman, Brett 2004, Knowledge, Assumptions, Lotteries, Philosophical Issues 14, 492–500. Harper, William 1996, Knowledge and Luck, Southern Journal of Philosophy 34 (3), 273–83. Harth, Manfred 2014, Is Relative Truth Really Truth? Dialectica 68, 409–28. Hawthorne, John 2000, Implicit Belief and Apriori Knowledge, Southern Journal of Philosophy 38, supplement (Spindel conference 1999; David Henderson and Terry Horgan (eds), The Role of the Empirical (and of the Apriori) in Epistemology), 191–210. Hawthorne, John 2001, Freedom in Context, Philosophical Studies 104, 63–79. Hawthorne, John 2002, Lewis, the Lottery, and the Preface, Analysis 62, 242–51. Hawthorne, John 2004a, Knowledge and Lotteries, Oxford: Clarendon.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

242  Bibliography Hawthorne, John 2004b, Replies, Philosophical Issues 14, 510–23. Hawthorne, John and Stanley, Jason 2008, Knowledge and Action, Journal of Philosophy 105, 571–90. Hazlett, Allan 2009, Knowledge and Conversation, Philosophy and Phenomenological Research 78, 591–620. Heil, John 1987, Doubts about Skepticism, Philosophical Studies 51, 1–17. Heller, Mark 1995, The Simple Solution to the Problem of Generality, Noûs 29, 501–15. Heller, Mark 1999a, The Proper Role for Contextualism in an Anti-Luck Epistemology, Philosophical Perspectives 13, 115–30. Heller, Mark 1999b, Relevant Alternatives and Closure, Australasian Journal of Philosophy 77, 196–208. Hempel, Carl G. 1965, Aspects of Scientific Explanation and Other Essays in the Philosophy of Science, New York and London: Free Press and Collier. Hetherington, Stephen 1998, Actually Knowing, Philosophical Quarterly 48, 453–69. Hetherington, Stephen 2001, Good Knowledge, Bad Knowledge: On Two Dogmas of Epistemology, Oxford: Clarendon. Hetherington, Stephen 2011, How to Know: A Practicalist Conception of Knowledge, Malden, MA: Wiley-Blackwell. Hiller, Avram 2013, Knowledge Essentially Based upon False Belief, Logos and Episteme 4, 7–19. Hitchcock, Christopher 2011, Probabilistic Causation, Stanford Encyclopedia of Philosophy (ed. Edward N. Zalta; Winter), available at . Hofweber, Thomas 1999, Contextualism and the Meaning-Intention Problem, in Kepa Korta, Ernest Sosa, and Xabier Arrazola (eds), Cognition, Agency and Rationality, Dordrecht: Kluwer, 93–104. Holton, Richard 2003, David Lewis’ Philosophy of Language, Mind and Language 18, 286–95. Hookway, Christopher 1989/90, Scepticism and Autonomy, Proceedings of the Aristotelian Society 90, 103–18. Horgan, Terry and Potrč, Matjaž 2013, Epistemological Skepticism, Semantic Blindness, and Competence-Based Performance Errors, Acta Analytica 28, 161–77. Hough, Gerry 2008, A Dilemma for Sinnott-Armstrong’s Moderate Pyrrhonian Moral Scepticism, Philosophical Quarterly 58, 457–62. Hudson, Robert G. 2004, The Generality Problem, Southern Journal of Philosophy 42, 193–211. Hume, David 1975, An Enquiry Concerning Human Understanding (ed. Peter H. Nidditch), Oxford: Clarendon. Hume, David 1978, A Treatise of Human Nature (ed. Peter H. Nidditch), Oxford: Clarendon. Hunter, David 2009, Contextualism, Skepticism and Objectivity, in Robert J. Stainton and Christopher Viger (eds), Compositionality, Context and Semantic Values, Dordrecht: Springer, 105–28. Jackson, Frank 1977, A Causal Theory of Counterfactuals, Australasian Journal of Philosophy 55, 3–21. Jäger, Christoph 2004, Skepticism, Information, and Closure: Dretske’s Theory of Knowledge, Erkenntnis 61, 187–201.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  243 Jäger, Christoph 2012, Contextualism and the Knowledge Norm of Assertion, Analysis 72, 491–8. Jenkins, Carrie S. and Nolan, Daniel 2010, Maximising, Satisficing and Context, Noûs 44, 451–68. Jespersen, Bjørn 2008, Knowing that p Rather than q, Sorites 20, 125–34. Johnsen, Bredo C. 2001, Contextualist Swords, Skeptical Plowshares, Philosophy and Phenomenological Research 62, 385–406. Johnson-Laird, Philip N. 1983, Mental Models: Toward a Cognitive Science of Language, Inference and Consciousness, Cambridge, MA: Harvard University Press. Johnson-Laird, Philip N. and Byrne, Ruth M. J. 1991, Deduction, Hillsdale, NJ: Erlbaum. Jönsson, Martin L. 2013, A Reliabilism Built on Cognitive Convergence: An Empirically Grounded Solution to the Generality Problem, Episteme 10, 241–68. Kahneman, Daniel, Slovic, Paul, and Tversky, Amos (eds) 1982, Judgment under Uncertainty: Heuristics and Biases, Cambridge: Cambridge University Press. Kallestrup, Jesper 2005, Contextualism between Scepticism and Common-Sense, Grazer Philosophische Studien, 247–53. Kaplan, David 1989, Demonstratives, in Joseph Almog, John Perry, and Howard Wettstein (eds), Themes from Kaplan, New York and Oxford: Oxford University Press, 481–563. Kappel, Klemens 2006, A Diagnosis and Resolution to the Generality Problem, Philosophical Studies 127, 525–60. Karjalainen, Antti and Morton, Adam 2003, Contrastive Knowledge, Philosophical Explorations 6(2), 74–89. Katzer, Charlotte 1996, When Is Knowledge a Matter of Luck? Grazer Philosophische Studien 51, 105–20. Kelp, Christoph 2011, A Problem for Contrastivist Accounts of Knowledge, Philosophical Studies 152, 287–92. Kelp, Christoph 2012, Do “Contextualist Cases” Support Contextualism? Erkenntnis 76, 115–20. Kim, Brian 2016, In Defense of Subject-Sensitive Invariantism, Episteme 13, 233–51. Kim, Jaegwon 1993, Internalism and Externalism in Epistemology, American Philosophical Quarterly 30, 303–16. Klein, Peter D. 1981, Certainty: A Refutation of Scepticism, Brighton: Harvester Press. Klein, Peter D. 2000, Contextualism and the Real Nature of Academic Skepticism, Philosophical Issues 10, 108–16. Klein, Peter D. 2008, Useful False Beliefs, in Quentin Smith (ed.), Epistemology: New Essays, Oxford: Oxford University Press, 25–61. Kölbel, Max 2002, Truth without Objectivity, London and New York: Routledge. Kölbel, Max 2008, Truth in Semantics, Midwest Studies in Philosophy 32, 242–57. König, Josef 2004, Probleme der Erkenntnistheorie. Göttinger Kolleg im WS 1958/59 (ed. Günter Dahms), Norderstedt: Books on Demand. Kompa, Nikola 2002, The Context Sensitivity of Knowledge Ascriptions, Grazer Philosophische Studien 64, 1–18. Kompa, Nikola 2005, The Semantics of Knowledge Attributions, Acta Analytica 20–1 (34), 16–28. Kompa, Nikola 2012, Nonindexical Contextualism: An Explication and Defense, in Stefan Tollksdorf (ed.), Conceptions of Knowledge, Berlin: De Gruyter, 329–49.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

244  Bibliography Korcz, Keith Allen 2015, The Epistemic Basing Relation, Stanford Encyclopedia of Philosophy (ed. Edward N. Zalta; Fall), available at . Kornblith, Hilary 2000, The Contextualist Evasion of Epistemology, Philosophical Issues 10, 24–32. Krasner, Daniel and Heller, Mark 1994, The Miracle of Counterfactuals: Counterexamples to Lewis’s World Ordering, Philosophical Studies 76, 27–43. Kvanvig, Jonathan L. 2006, Closure Principles, Philosophy Compass 1/3, 256–67. Kvart, Igal 2006, A Probabilistic Theory of Knowledge, Philosophy and Phenomenological Research 72, 1–43. Kyburg, Henry E., Jr. 1961, Probability and the Logic of Rational Belief, Middletown, CT: Wesleyan University Press. Kyburg, Henry E., Jr. 1970, Probability and Inductive Logic, London, Macmillan. Kyburg, Henry E., Jr. 1977, Randomness and the Right Reference Class, Journal of Philosophy 74, 501–21. Lackey, Jennifer 2006, Pritchard’s Epistemic Luck, Philosophical Quarterly 56, 284–9. Lackey, Jennifer 2008, What Luck Is Not, Australasian Journal of Philosophy 86, 255–67. Lasersohn, Peter 2008, Quantification and Perspective in Relativist Semantics, Philosophical Perspectives 22, 305–37. Lasonen-Aarnio, Maria 2008, Single Premise Deduction and Risk, Philosophical Studies 141, 157–73. Latus, Andrew 2000, Moral and Epistemic Luck, Journal of Philosophical Research 25, 149–72. Lawlor, Krista 2005, Enough Is Enough: Pretense and Invariance in the Semantics of “Knows That,” Philosophical Perspectives 19, 211–36. Lawlor, Krista 2013, Assurance: An Austinian View of Knowledge and Knowledge Claims, Oxford: Oxford University Press. Lehrer, Keith and Paxson, Thomas, Jr. 1969, Knowledge: Undefeated Justified True Belief, Journal of Philosophy 66, 225–37. Leite, Adam 2005, Some Worries for Would-Be WAMmers, Grazer Philosophische Studien 69, 205–29. Leplin, Jarrett 2007a, In Defense of Reliabilism, Philosophical Studies 134, 31–42. Leplin, Jarrett 2007b, Reply to Christensen, Philosophical Studies 134, 51–2. Lepock, Christopher 2009, How to Make the Generality Problem Work for You, Acta Analytica 24, 275–86. Levi, Isaac 1977, Direct Inference, Journal of Philosophy 74, 5–29. Lewis, David 1973, Counterfactuals, Oxford: Blackwell. Lewis, David 1979a, Scorekeeping in a Language Game, Journal of Philosophical Logic 8, 339–59. Lewis, David 1979b, Counterfactual Dependence and Time’s Arrow, Noûs 13, 455–76. Lewis, David 1986, On the Plurality of Worlds, Oxford: Blackwell. Lewis, David 1996, Elusive Knowledge, Australasian Journal of Philosophy 74, 549–67. Lewis, David 1998, Index, Context and Content, in David Lewis, Papers in Philosophical Logic, Cambridge: Cambridge University Press, 21–44. Lihoreau, Franck and Rebuschi, Manuel 2009, Contextualism and the Factivity of Knowledge, in Dariusz Lukasiewicz and Roger Pouivet (eds), Scientific Knowledge and Common

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  245 Knowledge, Bydgoszcz: Publishing House Epigram and University of Kazimierz Wielki Press, 209–32. Lowe, E. Jonathan 2002, A Survey of Metaphysics, Oxford: Oxford University Press. Ludlow, Peter 2005, Contextualism and the New Linguistic Turn in Epistemology, in Gerhard Preyer and Georg Peter (eds), Contextualism in Philosophy: Knowledge, Meaning, and Truth, Oxford: Clarendon, 11–50. Ludlow, Peter 2012, Contextualism, Multi-Tasking, and Third-Person Knowledge Reports: A  Note on Keith DeRose’s The Case for Contextualism, Philosophy and Phenomenological Research 84, 686–92. Luper, Steven 2003, Indiscernibility Skepticism, in Steven Luper (ed.), The Skeptics: Contemporary Essays, Aldershot: Ashgate, 183–202. Luper, Steven 2012a, False Negatives, in Kelly Becker and Tim Black (eds), The Sensitivity Principle in Epistemology, Cambridge: Cambridge University Press, 207–26. Luper, Steven 2012b, Contrastivism and Skepticism, International Journal for the Study of Skepticism 2, 51–8. Luper-Foy, Steven (ed.) 1987, The Possibility of Knowledge: Nozick and His Critics, Totowa, NJ: Rowman and Littlefield. Luzzi, Federico 2009, Counter-Closure, Australasian Journal of Philosophy 88, 673–83. MacFarlane, John 2005a, The Assessment Sensitivity of Knowledge Attributions, in Tamar Szabó Gendler and John Hawthorne (eds), Oxford Studies in Epistemology, vol. 1, 197–233. MacFarlane, John 2005b, Knowledge Laundering: Testimony and Sensitive Invariantism, Analysis 65, 132–8. MacFarlane, John 2009, Nonindexical Contextualism, Synthese 166, 231–50. MacFarlane, John 2011, Simplicity Made Difficult, Philosophical Studies 156, 441–8. MacFarlane, John 2012, Relativism, in Delia Graff Fara and Gillian Russell (eds), The Routledge Companion to the Philosophy of Language, New York: Routledge, 132–42. MacFarlane, John 2014, Assessment Sensitivity: Relative Truth and Its Applications, Oxford: Oxford University Press. Madison, B. J. C. 2011, Combating Anti Anti-Luck Epistemology, Australasian Journal of Philosophy 89, 47–58. Maitzen, Stephen 1998, The Knower Paradox and Epistemic Closure, Synthese 114, 337–54. Makinson, D. C. 1965, The Paradox of the Preface, Analysis 25, 205–7. Marr, David 1982, Vision: A Computational Investigation into the Human Representation and Processing of Visual Information, New York: Freeman. May, Joshua, Sinnott-Armstrong, Walter, Hull, Jay G., and Zimmerman, Aaron 2010, Practical Interests, Relevant Alternatives, and Knowledge Attributions: An Empirical Study, Review of Philosophy and Psychology 1, 265–73. McEvoy, Mark 2005, Belief-Independent Processes and the Generality Problem for Reliabilism, Dialectica 59, 19–35. McGinn, Colin 1984, The Concept of Knowledge, in Peter French, Theodore Uehling, Jr., and Howard Wettstein (eds), Midwest Studies in Philosophy 9 (Causation and Causal Theories), Minneapolis, MN: University of Minnesota Press, 529–54. McKenna, Robin 2011, Interests Contextualism, Philosophia 39, 741–50. McKenna, Robin 2013, Epistemic Contextualism: A Normative Approach, Pacific Philosophical Quarterly 94, 101–23.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

246  Bibliography McKenna, Robin 2014, Normative Scorekeeping, Synthese 191, 607–25. McKenna, Robin 2015, Epistemic Contextualism Defended, Synthese 192, 363–83. McKinnon, Rachel 2013, Lotteries, Knowledge, and Irrelevant Alternatives, Dialogue 52, 523–49. McLaughlin, Brian P. 2003, McKinsey’s Challenge, Warrant Transmission, and Skepticism, in Susana Nuccetelli (ed.), New Essays on Semantic Externalism and Self-Knowledge, Cambridge, MA and London: MIT Press, 79–96. Mills, Eugene 2012, Lotteries, Quasi-Lotteries, and Scepticism, Australasian Journal of Philosophy 90, 335–52. Moeller, Emil Frederik Lundbjerg 2015, Consuming Knowledge Claims across Contexts, Synthese 192, 4057–70. Moghaddam, Ahmad Reza Hemmati 2015, Epistemic Contextualism, Unarticulated Constituents and Hidden Variables, Dialogue 54, 225–46. Montminy, Martin 2007a, Epistemic Contextualism and the Semantics-Pragmatics Distinction, Synthese 155, 99–125. Montminy, Martin 2007b, Moral Contextualism and the Norms for Moral Conduct, American Philosophical Quarterly 44, 1–13. Montminy, Martin 2008a, Contextualist Responses to Philosophical Debates, Metaphilosophy 39, 571–90. Montminy, Martin 2008b, Can Contextualists Maintain Neutrality, Philosophers’ Imprint 8 (7), 1–13. Montminy, Martin 2008c, Cheap Knowledge and Easy Questions, Grazer Philosophische Studien 77, 127–46. Montminy, Martin 2009a, Contextualism, Relativism and Ordinary Speakers’ Judgments, Philosophical Studies 143, 341–56. Montminy, Martin 2009b, Contextualism, Invariantism and Semantic Blindness, Australasian Journal of Philosophy 87, 639–57. Montminy, Martin 2013, The Role of Context in Contextualism, Synthese 190, 2341–66. Montminy, Martin and Skolits, Wes 2014, Defending the Coherence of Contextualism, Episteme 11, 319–33. Morillo, Carolyn R. 1984, Epistemic Luck, Naturalistic Epistemology and the Ecology of Knowledge, or, What the Frog Should Have Told Dretske, Philosophical Studies 46, 109–29. Morreau, Michael 2010, It Simply Does Not Add Up: Trouble with Overall Similarity, Journal of Philosophy 107, 469–90. Morton, Adam 2011, Contrastivism, in Sven Bernecker and Duncan Pritchard (eds), The Routledge Companion to Epistemology, London: Routledge, 513–22. Morton, Adam 2013, Contrastive Knowledge, in Martijn Blaauw (ed.), Contrastivism in Philosophy, New York: Routledge, 101–15. Murphy, Peter 2005, A Sceptical Rejoinder to Sensitivity Contextualism, Dialogue 44, 693–706. Nagel, Jennifer 2008, Knowledge Ascriptions and the Psychological Consequences of Changing Stakes, Australasian Journal of Philosophy 86, 279–94. Nagel, Jennifer 2010, Knowledge Ascriptions and the Psychological Consequences of Thinking about Error, Philosophical Quarterly 60, 286–306. Nagel, Jennifer 2012, The Attitude of Knowledge, Philosophy and Phenomenological Research 84, 678–85.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  247 Nagel, Thomas 1979, Moral Luck, in Thomas Nagel, Mortal Questions, Cambridge: Cambridge University Press, 24–38. Nelkin, Dana K. 2000, The Lottery Paradox, Knowledge, and Rationality, Philosophical Review 109, 373–409. Neta, Ram 2002, S Knows that P, Noûs 36, 663–81. Neta, Ram 2003a, Contextualism and the Problem of the External World, Philosophy and Phenomenological Research 66, 1–31. Neta, Ram 2003b, Skepticism, Contextualism, and Semantic Self-Knowledge, Philosophy and Phenomenological Research 67, 396–411. Neta, Ram 2007, Anti-Intellectualism and the Knowledge-Action Principle, Philosophy and Phenomenological Research 75, 180–7. Neta, Ram 2012, The Case against Purity, Philosophy and Phenomenological Research 85, 456–64. Neta, Ram and Rohrbaugh, Guy 2004, Luminosity and the Safety of Knowledge, Pacific Philosophical Quarterly 85, 396–406. Newhard, Jay 2012, The Argument from Skepticism for Contextualism, Philosophia 40, 563–75. Nichols, Shaun, Stich, Stephen, and Weinberg, Jonathan M. 2003, Meta-Skepticism: Meditations in Ethno-Epistemology, in Steven Luper (ed.), The Skeptics: Contemporary Essays, Aldershot and Burlington, VT: Ashgate, 227–47. Norcross, Alastair 2005a, Contextualism for Consequentialists, Acta Analytica 20–2 (35), 80–90. Norcross, Alastair 2005b, Harming in Context, Philosophical Studies 123, 149–73. Nozick, Robert 1981, Philosophical Explanations, Cambridge, MA: Harvard University Press. Nozick, Robert 2001, Invariances. The Structure of the Objective World, Cambridge, MA: Harvard University Press. Okasha, Samir 2004, Wright on the Transmission of Support: A Bayesian Analysis, Analysis 64, 139–46. Olen, Jeffrey 1977, Knowledge, Probability, and Nomic Connections, Southern Journal of Philosophy 15, 521–6. Olin, Doris 2003, Paradox, Chesham Bucks: Acumen. Olin, Doris 2005, A Case against Closure, Veritas 50, 235–47. Partee, Barbara H. 2004, Comments on Jason Stanley’s “On the Linguistic Basis for Contextualism,” Philosophical Studies 119, 147–59. Perry, John 1979, The Problem of the Essential Indexical, Noûs 13, 3–21. Phelan, Mark 2014, Evidence that Stakes Don’t Matter for Evidence, Philosophical Psychology 27, 488–512. Pinillos, N. Ángel 2011, Some Recent Work in Experimental Epistemology, Philosophy Compass 6, 675–88. Pinillos, N. Ángel 2012, Knowledge, Experiments and Practical Interests, in Jessica Brown and Mikkel Gerken (eds), New Essays on Knowledge Attributions, Oxford: Oxford University Press, 192–219. Pinillos, N. Ángel and Simpson, Shawn 2014, Experimental Evidence Supporting AntiIntellectualism about Knowledge, in James R. Beebe (ed.), Advances in Experimental Epistemology, London: Bloomsbury, 9–43.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

248  Bibliography Pollock, John 1984, Reliability and Justified Belief, Canadian Journal of Philosophy 14, 103–14. Prades, Josep L. 2000, Skepticism, Contextualism and Closure, Philosophical Issues 10, 121–31. Prijic-Samarzija, Snjezana 2007, Trust and Contextualism, Acta Analytica 22, 125–38. Prinz, Jesse J. 2007, The Emotional Construction of Morals, Oxford: Oxford University Press. Pritchard, Duncan 2001, Contextualism, Scepticism, and the Problem of Epistemic Descent, Dialectica 55, 327–49. Pritchard, Duncan 2004, Epistemic Luck, Journal of Philosophical Research 29, 193–222. Pritchard, Duncan 2005a, Neo-Mooreanism, Contextualism, and the Evidential Basis of Scepticism, Acta Analytica 20, 3–25. Pritchard, Duncan 2005b, Neo-Mooreanism versus Contextualism, Grazer Philosophische Studien 69, 1–24. Pritchard, Duncan 2005c, Epistemic Luck, Oxford: Oxford University Press. Pritchard, Duncan 2007, Anti-Luck Epistemology, Synthese 158, 277–97. Pritchard, Duncan 2010, Contextualism, Skepticism and Warranted Assertibility Manœuvres, in Joseph Keim Campbell, Michael O’Rourke, and Harry Silverstein (eds), Knowledge and Skepticism, Cambridge, MA: MIT Press, 85–103. Pritchard, Duncan 2012, In Defence of Modest Anti-Luck Epistemology, in Kelly Becker and Tim Black (eds), The Sensitivity Principle in Epistemology, Cambridge: Cambridge University Press, 173–92. Pryor, James 2004, What’s Wrong with Moore’s Argument? Philosophical Issues 14, 349–78. Pryor, James 2012, When Warrant Transmits, in Annalisa Coliva (ed.), Mind, Meaning, and Knowledge: Themes from the Philosophy of Crispin Wright, Oxford: Oxford University Press, 269–303. Putnam, Hilary 1981, Brains in a Vat, in Hilary Putnam, Reason, Truth and History, Cambridge: Cambridge University Press, 1–22. Pynn, Geoff 2015, Pragmatic Contextualism, Metaphilosophy 46, 26–51. Radford, Colin 1966, Knowledge—By Examples, Analysis 27, 1–11. Ramsey, Frank Plumpton 1990a, Truth and Probability, in Frank Plumpton Ramsey, Philosophical Papers (ed. D. H. Mellor), Cambridge: Cambridge University Press, 52–94. Ramsey, Frank Plumpton 1990b, Knowledge, in Frank Plumpton Ramsey, Philosophical Papers (ed. D. H. Mellor), Cambridge: Cambridge University Press, 110–11. Rast, Erich 2014, Context as Assumptions, in Manuel Rebuschi and Franck Lihoreau (eds), Epistemology, Context, Formalism, New York: Springer, 9–31. Ravitch, Harold 1976, Knowledge and the Principle of Luck, Philosophical Studies 30, 347–9. Recanati, François 2008, Moderate Relativism, in Manuel García-Carpintero and Max Kölbel (eds), Relative Truth, Oxford: Oxford University Press, 41–62. Reed, Baron 2000, Accidental Truth and Accidental Justification, Philosophical Quarterly 50, 57–67. Reed, Baron 2010, A Defense of Stable Invariantism, Noûs 44, 224–44. Reed, Baron 2012, Resisting Encroachment, Philosophy and Phenomenological Research 85, 465–72. Reichenbach, Hans 1949, The Theory of Probability: An Inquiry into the Logical and Mathematical Foundations of the Calculus of Probability, Berkeley and Los Angeles, CA: University of California Press.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  249 Reid, Thomas 1863a, An Inquiry into the Human Mind, on the Principles of Common Sense, The Works of Thomas Reid (ed. William Hamilton, 6th edition, vol. 1), Edinburgh and London: MacLachlan and Stewart/Longman, 93–211. Reid, Thomas 1863b, Essays on the Intellectual Powers of Man, The Works of Thomas Reid (ed. William Hamilton, 6th edition, vol. 1), Edinburgh/London: MacLachlan and Stewart/ Longman, 213–508. Rescher, Nicholas 1990, Luck, Proceedings and Addresses of the American Philosophical Association 64 (3), 5–19. Rescher, Nicholas 1995, Luck: The Brilliant Randomness of Everyday Life, New York: Farrar Straus Giroux. Rescher, Nicholas 2014, The Machinations of Luck, Metaphilosophy 45, 620–6. Resnik, Michael D. 1987, Choices: An Introduction to Decision Theory, Minneapolis/MN: University of Minnesota Press. Richard, Mark 2004, Contextualism and Relativism, Philosophical Studies 119, 215–42. Richard, Mark 2008, When Truth Gives Out, Oxford: Oxford University Press. Rickless, Samuel C. 2014, The Contrast-Insensitivity of Knowledge Ascriptions, Philosophy and Phenomenological Research 88, 533–55. Rieber, Steven 1998, Skepticism and Contrastive Explanation, Noûs 32, 189–204. Rieber, Steven 2006, Free Will and Contextualism, Philosophical Studies 129, 223–52. Riggs, Wayne 1998, What Are the “Chances” of Being Justified? Monist 81, 453–72. Riggs, Wayne 2007, Why Epistemologists Are So Down on Their Luck, Synthese 158, 329–44. Riggs, Wayne 2014, Luck, Knowledge, and “Mere” Coincidence, Metaphilosophy 45, 627–39. Roeber, Blake 2014, Minimalism and the Limits of Warranted Assertibility Maneuvers, Episteme 11, 245–60. Roland, Jeffey and Cogburn, Jon 2011, Anti-Luck Epistemologies and Necessary Truths, Philosophia 39, 547–61. Rourke, Jason 2013, A Counterexample to the Contrastive Account of Knowledge, Philosophical Studies 162, 637–43. Roush, Sherrilyn 2005, Tracking Truth: Knowledge, Evidence, and Science, Oxford: Clarendon. Roush, Sherrilyn 2012, Sensitivity and Closure, in Kelly Becker and Tim Black (eds), The Sensitivity Principle in Epistemology, Cambridge: Cambridge University Press, 242–68. Rovane, Carol 2011, Relativism Requires Alternatives, Not Disagreement or Relative Truth, in Steven D. Hales (ed.), A Companion to Relativism, Malden, MA: Wiley-Blackwell, 31–52. Rovane, Carol 2012, How to Formulate Relativism, in Annalisa Coliva (ed.), Mind, Meaning, and Knowledge: Themes from the Philosophy of Crispin Wright, Oxford: Oxford University Press, 238–66. Rovane, Carol 2013, The Metaphysics and Ethics of Relativism, Cambridge, MA: Harvard University Press. Russell, Bertrand 1948, Human Knowledge: Its Scope and Limits, New York: Simon and Schuster. Russell, Bruce 2004, How to Be an Anti-Skeptic and a Noncontextualist, Erkenntnis 61, 245–55. Ryan, Sharon 1996, The Epistemic Virtue of Consistency, Synthese 109, 121–40. Rysiew, Patrick 2001, The Context-Sensitivity of Knowledge Attributions, Noûs 35, 477–514. Rysiew, Patrick 2005, Contesting Contextualism, Grazer Philosophische Studien 69, 51–69.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

250  Bibliography Rysiew, Patrick 2011, Surveys, Intuitions, Knowledge Attributions: Comments on Keith DeRose’s “Contextualism, Contrastivism, and X-Phi Surveys,” Philosophical Studies 156, 111–20. Salmon, Nathan 1989, Illogical Belief, Philosophical Perspectives 3, 243–85. Salmon, Wesley C. 1966, The Foundations of Scientific Inference, Pittsburgh, PA: University of Pittsburgh Press. Salmon, Wesley C. 1970, Statistical Explanation and Statistical Relevance, Pittsburgh, PA: University of Pittsburgh Press. Sartwell, Crispin 1991, Knowledge Is Merely True Belief, American Philosophical Quarterly 28, 157–65. Sartwell, Crispin 1992, Why Knowledge Is Merely True Belief, Journal of Philosophy 89, 167–80. Schaffer, Jonathan 2004a, From Contextualism to Contrastivism, Philosophical Studies 119, 73–103. Schaffer, Jonathan 2004b, Skepticism, Contextualism, and Discrimination, Philosophy and Phenomenological Research 69, 138–55. Schaffer, Jonathan 2005a, Contrastive Knowledge, in Tamar Szabó Gendler and John Hawthorne (eds), Oxford Studies in Epistemology, vol. 1, 235–71. Schaffer, Jonathan 2005b, What Shifts? Thresholds, Standards, or Alternatives? in Gerhard Preyer and Georg Peter (eds), Contextualism in Philosophy, Knowledge, Meaning, and Truth, Oxford: Clarendon, 115–30. Schaffer, Jonathan 2006, The Irrelevance of the Subject: Against Subject-Sensitive Invariantism, Philosophical Studies 127, 87–107. Schaffer, Jonathan 2007a, Closure, Contrast, and Answer, Philosophical Studies 133, 233–55. Schaffer, Jonathan 2007b, Knowing the Answer, Philosophy and Phenomenological Research 75, 383–403. Schaffer, Jonathan 2008, The Contrast-Sensitivity of Knowledge Ascriptions, Social Epistemology 22, 235–45. Schaffer, Jonathan 2012a, What Is Contrastivism? in Stefan Tolksdorf (ed.), Conceptions of Knowledge, Berlin: de Gruyter, 353–6. Schaffer, Jonathan 2012b, Contrastive Knowledge: Reply to Baumann, in Stefan Tolksdorf (ed.), Conceptions of Knowledge, Berlin: de Gruyter, 411–24. Schaffer, Jonathan and Knobe, Joshua 2012, Contrastivism Surveyed, Noûs 46, 675–708. Schaffer, Jonathan and Szabó, Zoltán Gendler 2014, Epistemic Comparativism: A Contextualist Semantics for Knowledge Ascriptions, Philosophical Studies 168, 491–543. Scharifi, Gilbert 2004, Contextualism and the Skeptic: Comments on Engel, Erkenntnis 61, 233–44. Schiffer, Stephen 1996, Contextualist Solutions to Scepticism, Proceedings of the Aristotelian Society 96, 317–33. Schiffer, Stephen 2004, Skepticism and the Vagaries of Justified Belief, Philosophical Studies 119, 161–84. Schiffer, Stephen 2007, Interest-Relative Invariantism, Philosophy and Phenomenological Research 75, 188–95. Schmitt, Frederick F. 1984, Reliability, Objectivity and the Background of Justification, Australasian Journal of Philosophy 62, 1–15.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  251 Schmitt, Frederick F. 1992, Knowledge and Belief, London and New York: Routledge. Schmoranzer, Sebastian 2012, The Contextualist Promise, in Christoph Jäger and Winfried Löffler (eds), Epistemology: Contexts, Values, Disagreement, Frankfurt: Ontos, 103–13. Schroeder, Mark 2009, Hybrid Expressivism: Virtues and Vices, Ethics 119, 257–309. Schroeder, Mark 2012, Stakes, Withholding, and Pragmatic Encroachment on Knowledge, Philosophical Studies 160, 265–85. Sextus Empiricus 2000, Outlines of Scepticism (trans. Julia Annas and Jonathan Barnes), Cambridge: Cambridge University Press. Shackel, Nicholas 2006, Shutting Dretske’s Door, Erkenntnis 77, 393–401. Silins, Nicholas 2005, Transmission Failure Failure, Philosophical Studies 126, 71–102. Sinnott-Armstrong, Walter 1996, Moral Skepticism and Justification, in Walter SinnottArmstrong and Mark Timmons (eds), Moral Knowledge? New Readings in Moral Epistemology, Oxford: Oxford University Press, 3–48. Sinnott-Armstrong, Walter 2002, What’s in a Contrast Class? Analysis 62, 75–84. Sinnott-Armstrong, Walter 2004, Classy Pyrrhonism, in Walter Sinnott-Armstrong (ed.), Pyrrhonian Skepticism, Oxford: Oxford University Press, 188–207. Sinnott-Armstrong, Walter 2006, Moral Skepticisms, Oxford: Oxford University Press. Sinnott-Armstrong, Walter 2008a, A Contrastivist Manifesto, Social Epistemology 22, 257–70. Sinnott-Armstrong, Walter 2008b, Moderate Classy Pyrrhonian Moral Scepticism, Philosophical Quarterly 58, 448–56. Sinnott-Armstrong, Walter 2008c, Replies to Hough, Baumann and Blaauw, Philosophical Quarterly 58, 478–88. Slote, Michael A. 1978, Time in Counterfactuals, Philosophical Review 87, 3–27. Snedegar, Justin 2013, Reason Claims and Contrastivism about Reasons, Philosophical Studies 166, 231–42. Sosa, Ernest 1986, On Knowledge and Context, Journal of Philosophy 83, 574–85. Sosa, Ernest 1988, Knowledge in Context, Skepticism in Doubt: Virtue of Our Faculties, Philosophical Perspectives 2, 139–55. Sosa, Ernest 1992, Generic Reliabilism and Virtue Epistemology, Philosophical Issues 2 (Enrique Villanueva (ed.), Rationality in Epistemology, Ridgeview: Atascadero, CA), 79–92. Sosa, Ernest 1999, How to Defeat Opposition to Moore, Philosophical Perspectives 13, 141–54. Sosa, Ernest 2000, Skepticism and Contextualism, Philosophical Issues 10, 1–18. Sosa, Ernest 2003a, The Place of Truth in Epistemology, in Michael DePaul and Linda Zagzebski (eds), Intellectual Virtue: Perspectives from Ethics and Epistemology, Oxford: Clarendon, 155–79. Sosa, Ernest 2003b, Neither Contextualism nor Skepticism, in Steven Luper (ed.), The Skeptics: Contemporary Essays, Aldershot: Ashgate, 165–82. Sosa, Ernest 2004, Relevant Alternatives, Contextualism Included, Philosophical Studies 119, 39–65. Sosa, Ernest 2007, A Virtue Epistemology: Apt Belief and Reflective Knowledge, vol. I, Oxford: Oxford University Press. Sosa, Ernest 2009, Reflective Knowledge: Apt Belief and Reflective Knowledge, vol. II, Oxford: Oxford University Press. Sosa, Ernest 2011, Knowing Full Well, Princeton, NJ: Princeton University Press.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

252  Bibliography Speaks, Jeff 2008, Conversational Implicatures, Thought, and Communication, Mind and Language 23, 107–22. Sripada, Chandra Sekhar and Stanley, Jason 2012, Empirical Tests of Interest-Relative Invariantism, Episteme 9, 3–26. Stainton, Robert J. 2010, Contextualism in Epistemology and the Context Sensitivity of “Knows,” in Joseph Keim Campbell, Michael O’Rourke, and Harry S. Silverstein (eds), Knowledge and Skepticism, Cambridge, MA: MIT Press, 137–63. Stalnaker, Robert 2004, Comments on “From Contextualism to Contrastivism,” Philosophical Studies 119, 105–17. Stanley, Jason 2004, On the Linguistic Basis for Contextualism, Philosophical Studies 119, 119–46. Stanley, Jason 2005, Knowledge and Practical Interests, Oxford: Clarendon. Stemmer, Nathan 1982, A Solution to the Lottery Paradox, Synthese 51, 339–53. Steup, Matthias 2005, Contextualism and Conceptual Disambiguation, Acta Analytica 20–1 (34), 3–15. Stine, Gail 1976, Skepticism, Relevant Alternatives, and Deductive Closure, Philosophical Studies 29, 249–61. Stojanović, Isidora 2007, Talking about Taste: Disagreement, Implicit Arguments, and Relative Truth, Linguistics and Philosophy 30, 691–706. Stojanovic, Isidora 2008, The Scope and the Subtleties of the Contextualism-LiteralismRelativism Debate, Language and Linguistics Compass 2 (6), 1171–88. Stojanović, Isidora 2012, On Value-Attributions: Semantics and Beyond, Southern Journal of Philosophy 50, 621–38. Stone, Jim 2007, Contextualism and Warranted Assertion, Pacific Philosophical Quarterly 88, 92–113. Stoutenburg, Gregory 2015, The Epistemic Analysis of Luck, Episteme 12, 319–34. Stroud, Barry 1984, The Significance of Philosophical Scepticism, Oxford: Clarendon. Stroud, Barry 2000, Understanding Human Knowledge in General, in Sven Bernecker and Fred Dretske (eds), Knowledge: Readings in Contemporary Epistemology, Oxford: Oxford University Press, 307–23. Swinburne, Richard 2001, Epistemic Justification, Oxford: Clarendon. Thomas, Alan 2006a, Value and Context: The Nature of Moral and Political Knowledge, Oxford: Oxford University Press. Thomas, Alan 2006b, Inferential Contextualism and Moral Cognitivism, Ms. Timmerman, Travis 2013, The Persistent Problem of the Lottery Paradox and Its Unwelcome Consequences for Contextualism, Logos and Episteme 4, 85–100. Timmons, Mark 1993, Moral Justification in Context, Monist 76, 360–78. Timmons, Mark 1996, Outline of a Contextualist Moral Epistemology, in Walter SinnottArmstrong and Mark Timmons (eds), Moral Knowledge? New Readings in Moral Epistemology, Oxford: Oxford University Press, 293–325. Timmons, Mark 1999, Morality without Foundations: A Defense of Ethical Contextualism, Oxford: Oxford University Press. Torrengo, Giuliano 2010, Perspectival Truth and Perspectival Realism, in François Recanati, Isidora Stojanović, and Neftali Villanueva (eds), Context-Dependence, Perspective and Relativity, Berlin and New York: De Gruyter Mouton, 333–48.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  253 Turri, John 2015a, Unreliable Knowledge, Philosophy and Phenomenological Research 90, 529–45. Turri, John 2015b, An Open and Shut Case: Epistemic Closure in the Manifest Image, Philosophers’ Imprint 15 (2), 1–18. Turri, John and Friedman, Ori 2014, Winners and Losers in the Folk Epistemology of Lotteries, in James R. Beebe (ed.), Advances in Experimental Epistemology, London: Bloomsbury, 45–69. Tversky, Amos 1977, Features of Similarity, Psychological Review 84, 327–52. Unger, Peter 1970, An Analysis of Factual Knowledge, in Michael Roth and Leon Galis (eds), Knowing: Essays in the Analysis of Knowledge, New York: Random House, 113–30. Unger, Peter 1971, A Defense of Skepticism, Philosophical Review 80, 198–219. Unger, Peter 1975, Ignorance: A Case for Scepticism, Oxford: Clarendon. Unger, Peter 1984, Philosophical Relativity, Oxford: Blackwell. Unger, Peter 1986, The Cone Model of Knowledge, Philosophical Topics 14, 125–78. Unger, Peter 1995, Contextual Analysis in Ethics, Philosophy and Phenomenological Research 55, 1–26. Vahid, Hamid 2001, Knowledge and Varieties of Epistemic Luck, Dialectica 55, 351–62. van Woudenberg, René 2005, Contextualism and the Many Senses of Knowledge, Grazer Philosophische Studien 69, 147–64. van Woudenberg, René 2008, The Knowledge Relation: Binary or Ternary? Social Epistemology 22, 281–8. Veber, Michael 2004, Contextualism and Semantic Ascent, Southern Journal of Philosophy 42, 261–72. Veber, Michael 2014, Knowledge with and without Belief, Metaphilosophy 45, 10–132. Venn, John 1876, The Logic of Chance, London: Macmillan. Vogel, Jonathan 1990, Are There Counterexamples to the Closure Principle? in Michael D. Roth and Glenn Ross (eds), Doubting. Contemporary Perspectives on Skepticism, Dordrecht: Kluwer, 13–27. Vogel, Jonathan 1999, The New Relevant Alternatives Theory, Philosophical Perspectives 13, 155–80. Wallis, Charles 1994, Truth-Ratios, Process, Task, and Knowledge, Synthese 98, 243–69. Warfield, Ted A. 2004, When Epistemic Closure Does and Does Not Fail: A Lesson from the History of Epistemology, Analysis 64, 35–41. Warfield, Ted A. 2005, Knowledge from Falsehood, Philosophical Perspectives 19, 405–16. Weatherson, Brian 2005, Questioning Contextualism, in Stephen Hetherington (ed.), Aspects of Knowing: Epistemological Essays, Amsterdam: Elsevier, 133–47. Weatherson, Brian 2008, Attitudes and Relativism, Philosophical Perspectives 22, 527–44. Weatherson, Brian 2011, Defending Interest-Relative Invariantism, Logos and Episteme 2, 591–609. Weatherson, Brian 2012a, The Temporal Generality Problem, Logos and Episteme 3, 117–22. Weatherson, Brian 2012b, Knowledge, Bets, and Interests, in Jessica Brown and Mikkel Gerken (eds), Knowledge Ascriptions, Oxford: Oxford University Press, 75–103. Wedgwood, Ralph 2006a, The Meaning of “Ought,” in Russ Shafer-Landau (ed.), Oxford Studies in Metaethics, vol. 1, Oxford: Oxford University Press, 127–60. Wedgwood, Ralph 2006b, The Context-Sensitivity of “Ought,” and the Normativity of Rationality, Ms.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

254  Bibliography Wedgwood, Ralph 2008, Contextualism about Justified Belief, Philosophers’ Imprint 8–9, 1–20. Weintraub, Ruth 2001, The Lottery: A Paradox Regained and Resolved, Synthese 129, 439–49. Wilburn, Ron 2010, Possible Words of Doubt, Acta Analytica 25, 259–77. Willaschek, Marcus 2009, Non-Relativist Contextualism about Free Will, European Journal of Philosophy 18, 567–87. Williams, Bernard 1973, Deciding to Believe, in Bernard Williams, Problems of the Self, Cambridge: Cambridge University Press, 136–51. Williams, Bernard 1981, Moral Luck, in Bernard Williams, Moral Luck: Philosophical Papers 1973–1980, Cambridge: Cambridge University Press, 20–39. Williams, Michael 1996, Unnatural Doubts: Epistemological Realism and the Basis of Scepticism, Princeton, NJ: Princeton University Press. Williams, Michael 2000, Is Contextualism Statable? Philosophical Issues 10, 80–5. Williams, Michael 2004, Scepticism and the Context of Philosophy, Philosophical Issues 14, 456–75. Williamson, Timothy 2000, Knowledge and Its Limits, Oxford: Oxford University Press. Williamson, Timothy 2001, Comments on Michael Williams’ Contextualism, Externalism and Epistemic Standards, Philosophical Studies 103, 25–33. Williamson, Timothy 2005a, Knowledge, Context, and the Agent’s Point of View, in Gerhard Preyer and Georg Peter (eds), Contextualism in Philosophy: Knowledge, Meaning, and Truth, Oxford: Clarendon, 91–114. Williamson, Timothy 2005b, Knowledge and Scepticism, in Frank Jackson and Michael Smith (eds), The Oxford Handbook of Contemporary Philosophy, Oxford: Oxford University Press 2005, 681–700. Williamson, Timothy 2005c, Contextualism, Subject-Sensitive Invariantism, and Knowledge of Knowledge, Philosophical Quarterly 55, 213–35. Wittgenstein, Ludwig 1969, On Certainty, Oxford: Blackwell. Wright, Crispin 1985, Facts and Certainty, Proceedings of the British Academy 71, 429–72. Wright, Crispin 2000, Cogency and Question-Begging: Some Reflections on McKinsey’s Paradox and Putnam’s Proof, Philosophical Perspectives 10, 140–63. Wright Crispin 2002, (Anti-)Sceptics Simple and Subtle: G. E. Moore and John McDowell, Philosophy and Phenomenological Research 65, 330–48. Wright, Crispin 2003, Some Reflections on the Acquisition of Warrant by Inference, in Susana Nuccetelli (ed.), New Essays on Semantic Externalism and Self-Knowledge, Cambridge, MA and London: MIT Press, 57–77. Wright, Crispin 2004, Warrant for Nothing (and Foundations for Free?), Proceedings of the Aristotelian Society, Supplement 78, 167–212. Wright, Crispin 2005, Contextualism and Scepticism: Even-Handedness, Factivity and Surreptitiously Raising Standards, Philosophical Quarterly 55, 236–62. Wright, Crispin 2007, The Perils of Dogmatism, in Susana Nuccetelli and Gary Seay (eds), Themes from G. E. Moore: New Essays in Epistemology and Ethics, Oxford: Oxford University Press, 25–48. Wright, Crispin 2008, Relativism about Truth Itself: Haphazard Thoughts about the Very Idea, in Max Kölbel and Manuel García-Carpintero (eds), Relative Truth, Oxford: Oxford University Press, 157–85. Wright, Sarah 2010, Virtues, Social Roles, and Contextualism, Metaphilosophy 41, 95–114.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

Bibliography  255 Wunderlich, Mark 2003, Vector Reliability: A New Approach to Epistemic Justification, Synthese 136, 237–62. Yamada, Masahiro 2011, Getting It Right by Accident, Philosophy and Phenomenological Research 83, 72–105. Yourgrau, Palle 1983, Knowledge and Relevant Alternatives, Synthese 55, 175–90. Zalabardo, José L. 2012, Scepticism and Reliable Belief, Oxford: Oxford University Press.

OUP CORRECTED PROOF – FINAL, 09/14/2016, SPi

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

Index Note: Italic t after a page number indicates a Table. Abath, A. J.  195 accounts of knowledge: contextualist  1, 159 modal 3 probabilistic  3, 65, 83, 85 sensitivity 113 Adler, J. E.  47 n., 102 n., 103, 116, 190 n., 196 Agrippa’s Trilemma  92 Airport cases  11 n., 13, 15, 18–21, 24–7, 32, 160, 184 allergies, see Two Allergies case Almeida, C. de  96 n. Almér, A.  225 Alston, W. P.  35, 47, 97 Amateur Genealogists case  28–9, 205 ambiguity maneuvers  17–18 Annis, D. B.  92 n. arbitrariness 48 normativity and  4, 193–5 unproblematic 194 Armstrong, D. M.  34 attention  4, 63t, 155t, 204 change of  111 n. foci of  74, 87, 153, 190, 194 see also rule of attention attributions  22, 45 n., 152, 185, 197, 229 cross-context  120–39, 159, 160 distinction between  162 first-person 129 intercontextual 179 linguistic and non-linguistic  163 luck  74, 82, 86, 87 mental 163 modal 179 reliability 62 responsibility  143, 145 similarity 60 speechless thought  166 temporal 179 see also knowledge attribution Attributor Subject Stakes, see High Attributor Subject Stakes; Low Attributor Subject Stakes Audi, R.  96 n. Austin, J. L.  32, 79 Bach, K.  99, 122, 159, 162, 190 n., 195 Ball, B.   76 Ballantyne, N.  67

Bank cases  11 n., 13, 18–21, 23–6, 31 n., 32, 160–1, 163–7, 170–1, 172, 183–5, 188–9, 207, 209 Barke, A.  92 n., 97 Baumann, P.  5 n., 10, 54, 55 n., 60 n., 62, 75, 81, 86, 87, 96 n., 97 n., 112, 124, 126, 129, 148, 154, 162, 189 n., 208, 209, 214 n., 216, 221 Beall, J. C.  223 Becker, K.  42, 48, 81, 215, 217 n. Beebe, J. R.  221 n. belief: indexical 39 mathematical 39 metaphorical 169 moral  153 n. non-lucky 75 outright 16 probabilistic explanation of  103 reliability of  2, 33, 34, 37–9, 59 safe  59, 62 sensitive  59, 62 strength of  16, 29 truth value of  41 see also belief acquisition; degrees of belief; false beliefs; justified belief; target beliefs; true beliefs belief acquisition  44, 49, 85 methods of  47, 57, 58, 78, 84 processes of  34, 57, 58 Bianchi, C.  177 Bishop, M. A.  47 Björnsson, G.  153 n. Blaauw, M.  136, 177, 213 n., 221 Black, T.  81, 100 n., 162 blindness  196, 197 semantic  4, 167 Blome-Tillmann, M.  1, 18, 76, 93, 96, 136, 174, 178, 179, 192 n., 195 Bogdan, R. J.  95 Boghossian, P.  229 Bowie, G. L.  59 Brandom, R. B.  56 Brendel, E.  120, 124, 125, 126, 229 Brogaard, B.  153 n., 225, 227, 228 Brokes, A. J.  47 n., 52 n. Broncano-Berrocal, F.  66, 143 Brown, J.  96, 136, 162, 190, 195 Brueckner, A.  11 n., 94, 102 n., 121, 127, 128, 135, 136, 185 n., 209 Buckwalter, W.  190

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

258  index Buenting, J.  213 Buford, C. T.  127, 128, 209 Byrne, R. M. J.  36 Cappelen, H.  140, 173–7, 196, 222 n., 228 Carnap, R.  38 Cartesian skepticism  91, 92, 95, 98, 102, 115, 127, 221 lottery skepticism and  3, 102 n. case descriptions  10 n., 16, 70, 72, 73, 78, 124, 136, 149, 152 contextually variable  74 referring to the same case  142 causal determinism  86 n. causal operativeness/sufficiency  48, 49, 50 causality  69, 87 n., 151–2 probabilistic accounts of  41 n. chances of error  117 probabilistic 111 salience of  26–7, 63, 111–14 character traits  4, 155t, 190, 194 individual 145 stable  115 n. Cheating case  16, 30, 206 Chemla, E.  190 Christensen, D.  50 circularity  60 n., 92 epistemic 97 closeness 37 context-sensitivity of  60, 61, 62 remoteness and  19 n., 42, 59 closeness-rankings  60, 61, 82 acceptable  80, 82 admissible 81 ordinal 62 closure  99, 102, 160, 180, 192 n., 230 contextualist  96, 97, 101, 131–6, 137, 138 downwards  132–4, 136, 210 epistemic  3, 5 lotteries and  100, 109–11 modified  98, 109, 137, 138, 209–10 multi-premise  106, 107 reformed 110 reverse 75–6 revised 118 simple  94–8, 100, 118, 123–4, 127, 128, 131, 210 Coffman, E. J.  66 Cogburn, J.  59 Cohen, S.  5 n., 11 n., 12, 13, 15, 20–2, 24–7, 32, 51, 94, 99, 100 n., 111, 113, 114, 117, 118 n., 122, 123 n., 160, 177, 184, 190 n., 195 Collins, J.  101, 105 Colyvan, M.  153 n. Comesaña, J.  35, 45 n. 47 n., 81 competent inference  95, 96 n., 97, 109, 127, 137 luck transfers across  74, 75, 76, 77 subject comes to believe proposition on basis of 110

what one comes to know on basis of  192–3 nn. complexity  4, 21, 159, 173, 190–3, 217, 228 n. conditionals  100, 109 n., 132 subjunctive  58–9, 62 see also probability Conee, E.  11 n., 35, 36, 46, 47, 51, 168, 195 context: contradictions across  121–4 conversational  20, 23–5, 27, 29, 30 hierarchy of  83, 178 knowledge attributions content can vary with 12 lax 129 mathematical 17 ordinary  93, 94–5 professional 14 reliability and  43–58 “right” 194 skeptical  93, 94, 138 context-sensitivity  3, 11, 12, 15, 21, 32, 65, 82, 96 n., 112 alleged  17, 140, 194 closeness  60, 61 explicit  122, 196 knowledge  23, 34, 51, 57, 85, 86, 123, 140, 141, 159, 163, 165 n., 167, 176, 177, 182–3, 195, 208 linguistic tests for  173–7 luck 86 moral terms  153, 159 reliability  33, 34, 37, 39, 45, 51, 62 sentence  140, 164 uncontroversial expressions  175 contextual parameters  2, 3, 12, 45, 53, 56, 60, 152, 196, 197 complex set of  217 determinants differ from proposition to proposition 137 function mapping onto contents  164 hierarchy of  87, 93, 178 multidimensional space of  190 ternary relations  4, 122 contextualism: alternatives to  4–5, 155, 183, 212 cases for: see Airport; Amateur Genealogists; Bank; Courts; Exam; Fermat; Party; Stakes; Thelma and Louise; Train; Two Allergies; Weather Forecast consistency problem for  128 contradiction and  120–39 defenders of  2 direct application to knowledge attributions 130 epistemological  123 n., 154 indexical  138 n. interpropositional variability  190 non-indexical  227, 229 n.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

index  259 objections against  2, 4, 134, 155, 208; linguistic 177–82 practical 151–5 probabilistic version of  3 semantical version of  123 n. standard(s)  2, 9–32, 65, 87, 99–100, 118–19 typical  99, 100 virtues of  2, 3 contextualist solutions  46 pessimism about the prospects of  120 skepticism, lotteries, and  91–119 standard  93, 111–14 contextualization  93, 111, 146 contradiction  142, 152 brute 223 contextualism and  120–39 outright 197 contrastivism  152 n., 198, 212–21, 230 epistemic 5 linguistic evidence for  221 n. conventions, see norms and conventions Courts case  14–15, 28, 55 n. DAT (dogmatic argument/puzzle)  93, 94, 101 Daukas, N.  25, 99, 177 David, M.  94, 96 Davidson, D.  122, 197 Davies, M.  96 Davis, W. A.  117, 136, 161, 162, 168, 177, 179, 194, 209, 227 degrees of belief: required 2 standards for  14, 31t, 87t threshold for  16, 29 Dennett, D.  79 Derksen, A. A. 1978  107 DeRose, K.  9, 11, 13, 18–20, 21 n., 23, 25, 32, 61, 92, 102 n., 111, 113, 115 n., 136, 138, 153 n., 160, 161, 162, 167, 174, 177, 178, 179, 185 n., 186, 190, 193,195,196, 207, 208, 209, 221 n. Descartes, R.  3, 92, 223, 224 see also Cartesian skepticism descriptions  3, 47, 49, 56, 88t, 122, 154 n., 159, 217 n. collective  175, 176 contextual variability of  208 contextualist solution  152 different types of  190, 196, 216 intention crucial for  143 n. most specific  150 parameters of  87 plurality of  192 relevant  146 n., 208 role of  65–88 surprising types of  193 see also case descriptions

determinants  2, 20–31, 88t, 113, 137, 154, 155t, 159, 191, 194, 206 contextually variable  13 crucial roles played by  112 different possible kinds of  87 distinction between parameters and  4, 9, 13, 45 parameter fixing  63 plurality of  190 relevant 3 special focus on  32 stakes play double role as  145 Di Bello, M.  97 Dinges, A.  190 n. disquotation  223–4, 226, 229 contextualist  123, 177 ersatz 225 factivity and  122, 124 truth relativism faces a dilemma with  227 see also intercontextual disquotational indirect reports dogmatism  3, 92, 102, 114 see also DAT Douven, I.  102 n., 107, 173 Dreier, J.  153 n., 154 n., 229 n. Dretske, F. I.  35, 96 n., 97, 102, 103, 116, 210, 212 Dudman, V. H.  102 n. Dutant, J.  53 n. Einheuser, I.  225 Engel, M.  67, 68, 84 n., 99, 121, 189 n. Entailed Stakes  211–12 epistemic position  2, 216 better or worse  115, 116 fixed 115 good (enough) and bad (enough)  114 standards for  14, 16–18, 19, 29, 30, 31t, 87t epistemological theory  5 n., 47 n. epistemology  5, 18, 60, 79, 108, 119, 123 n., 138 n., 154, 162, 192 n., 196 basic tasks of  1 challenge to  3 contextualist 1 dogma of  3 folk  102 n. impact on  153 least plausible ideas to gain currency in  102 n. lottery skepticism a challenge to  3 virtue  74, 115 n. Ernst, G.  177 error  4, 197 broad 195 systematic  195, 209 see also chances of error error theory  167, 172, 196 n., 209 evidence  2, 10, 19, 30, 32, 45, 84 n., 140, 153, 161 anti-skeptical  100 n. available 68 believability of  51

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

260  index evidence (cont.) better or worse  114, 116 checking more carefully  170 n. contrary 18 different propositions describing different pieces of  211 gathering more  115 good  171, 191 incompatible with certain possibilities  18 n. independent  152, 207 indirect  185 n. kinds of  14, 15, 19 n., 27, 28, 57 n., 101 n., 137 linguistic  182, 221 n. more and better  27, 28 possession of  16, 30, 198, 199, 206 real  16, 30, 206 restricted 102 specific 104 standards for  14–16, 19 n., 27, 31t, 87t, 94, 111, 118, 216 threshold for sufficiency of  191 weighing and putting together  77 Exam case  15–16, 29–30 explication 38 extension  2, 44 n., 155 generalization and  51–7 factivity  121, 134, 211, 228 contextualist 123 disquotation and  122, 124 fallibilism/fallibility  105, 107, 117, 183 false beliefs  34, 47, 53 n. acquiring  38 n., 67 n., 69, 84 n., 85 n. good inference from  41 n. knowledge and  193 warranted  169, 172 Fantl, J.  5 n., 10 n., 11 n., 13, 20, 32, 126, 168, 170 n., 166 n., 173 n., 182–3, 186 n., 187 n., 198, 210 Feldman, R.  35, 36, 45 n., 46, 47 n., 51, 96 n., 99, 102 n. Feltz, A.  190 Fermat case  17, 26, 30–1, 67, 129, 175 Fetzer, J. H.  51, 145 Fine, K.  59 Finlay, S.  153 n. Fogelin, R. J.  194 Foley, R.  47 n., 68, 116 Frege, G.  36 n. Freitag, W.  60 n., 121, 124, 126 Friedman, O.  102, 113, 114 Fumerton, R.  35, 172 Gendler, T. S.  69 n. genealogists, see Amateur Genealogists case generalization  37, 58, 101, 105, 121, 132, 187 extension and  51–7

Gerken, M.  221 n. Gettier-problem  5 n., 35, 57 n., 84 n. Gillies, A. S.  178 Gillies, D.  38 n., 51, 145, 148, 202 Goldberg, S.  79, 81 Goldman, A.  34, 37, 38, 42, 47 n., 48, 53 n., 56, 71, 78, 79, 84 n., 97, 103, 104, 106 Goodman, N.  59 Greco, J.  11, 20, 51, 74, 79, 87 n., 102 n., 113, 115 n., 116 n., 146 n., 153 n., 186, 192 Greene, R.  97 Grice, P.  161, 162, 168 n. Grimm, S.  191 Grobler, A.  60 n. Gundersen, L. B.  81, 99, 103 Hájek, A.  51, 53 n., 83, 145 Hales, S. D.  95 Halliday, D.  162, 178 Hambourger, R.  92 n. Hannon, M.  20, 136, 191 Hansen, N.  190 Harman, G.  36 n., 97, 102, 103, 113, 116, 166–7, 168 Harper, W.  67 Hawthorne, J.  5 n., 10 n., 25, 69 n., 95, 96, 102 n., 103, 104, 106 n., 107, 111, 115 n., 116, 126, 153 n., 166 n., 170 n., 172, 173 n., 174, 176, 179, 182, 190 n., 195, 199, 204, 208, 210 n., 222 n., 228 Hazlett, A.  162 Heil, J.  100 Heller, M.  51, 59, 60, 96 n. Hempel, C. G.  51, 53 n., 145, 148 Hetherington, S.  65, 68, 86 n., 107 High Attributor-High Subject Stakes  188 High Attributor-Low Subject Stakes  184, 185, 187, 188, 189 n., 207–8 High Credence-Low Chances  202–3 Hiller, A.  76 Hitchcock, C.  41 n. Hofweber, T.  195 Holton, R.  25 Hookway, C.  96 n. Horgan, T.  122, 196 Hough, G.  221 Hudson, R. G.  50 Hume, D.  92, 139, 146 Hunter, D.  189 n. Ignorant High Stakes  25, 184, 185 n., 187, 188–9, 202 n. inference  3, 14, 15, 35, 42 n., 78 n., 106, 108 n., 111 n., 118, 130, 131 context might change during  136 false premise and  75–6 logical and psychological aspects of  36

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

index  261 Moorean anti-skeptical  96 non-deductive 41 overdetermined 75 subject only knows the premise of  97 see also competent inference Inferential Stakes  210–12 infinite regress  92, 219 n. intercontextual disquotational indirect reports  173, 174, 175, 176 invariantism  15, 111, 153, 161, 163, 165–6, 167, 172–3, 193 classical  126, 127, 182, 183, 191, 192, 194, 198, 208, 230 interest-relative 199 moderate 170 subject-sensitive, see SSI Jackson, F.  59 Jäger, C.  97 n., 124 Jenkins, C. S.  153 n. Jespersen, B.  213 n. Johnsen, B. C.  99, 212 Johnson-Laird, P. N.  36 Jönsson, M. L.  51 justification  1, 11, 13, 31, 99, 110, 126, 161–2, 220, 226, 227 argumentative step in need of  149 knowledge and  36, 115 n., 213 n. sufficient 35 undefeasible 79 justified belief  108, 143 n., 151, 153 n., 169, 170 n. direct reference in tension with closure principles for  96 n. process leading from justifier to  35 Kahneman, D.  36 Kallestrup, J.  121, 125 Kaplan, D.  17–18, 44, 164 Kappel, K.  45 n., 46, 47 n., 51 Karjalainen, A.  212 Katzer, C.  66 Kelp, C.  189 n. Kim, B.  207 Kim, J.  35 Klein, P. D.  76, 96 n., 99 Knobe, J.  190, 221 n. knowability problem  4, 117, 211 cross-context attributions and  120–39 SSI and  209–10 knowledge: conceptual 215 context-invariant 126 denial of  78, 188, 191, 196, 213, 229 epistemological theories of  5 n. factivity of  122 false belief and  193 fundamental characteristics  1

inferential  35, 74, 76, 95 judgments about  85 justification and  36, 115 n., 213 n. linguistic 215 lottery 104 lucky 68–74 necessary condition for  117 positive view of  5 possibility of  103 reliability and  34–7 semantic theories of  5 n. standards for  11, 14–15, 18–20, 30, 111–13, 117, 164, 183, 198 without contrasts  213–15 see also accounts of knowledge knowledge attribution  1, 2, 27, 31, 35, 87, 120 n., 168, 175, 229 complexity concerning  191 content can vary with context  12 context-sensitive  21, 23 contextualism supported for  163 contrastivism and  213 correct  187, 188 denials of  118 direct application of contextualism to  130 false  114, 117, 118, 185, 220 intercontextual  138, 179 linguistic aspect of  12, 163 made in lax contexts  129 mental aspect of  163 modal 179 ordinary  46, 100, 191 pragmatics of  161 semantics of  161, 162 sensitivity to track-record-related factors  86 systematically wrong  167, 196 n. temporal 179 ternary analysis of  174 truth conditions of  51, 57, 62, 74, 82, 86, 165, 172, 220 truth values of  4–5, 121, 130, 198 Kölbel, M.  222 n., 224 Kompa, N.  121, 177, 178, 227 König, J.  98 Korcz, K. A.  35 Kornblith, H.  99 Krasner, D.  59 Kvanvig, J. L.  96 Kvart, I.  38 n., 85 Kyburg, H.  38 n., 51, 102, 106, 107, 145, 202 Lackey, J.  66, 67, 83 language  9, 38, 193 hyperbolic 168 ordinary  21 n., 32, 79, 182 thought and  1, 12, 162–6, 168 n., 169 see also natural language

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

262  index Lasersohn, P.  229 n. Lasonen-Aarnio, M.  107 Latus, A.  68 Lawlor, K.  110, 121, 179, 229 Lehrer, K.  35 Leite, A.  162 Leplin, J.  50 Lepock, C.  45 n., 47 n. Lepore, E.  140, 173–7, 196 Levi, I.  51, 145 Levin, M.  47 n. Lewis, D.  1, 11 n., 12, 18 n., 19 n., 23, 24 n., 25, 59, 60, 63, 91, 93, 96 n., 103 n., 111, 163, 168, 186, 222, 223 Lihoreau, F.  124, 177 linguistic issues  164, 166, 173, 191, 196, 215 knowledge attributions  12, 163, 165 objections 177–82 see also metalinguistics lotteries  69, 112–16, 119, 159, 198, 221 closure and  109–11 knowing propositions  3, 85 n., 91, 102–8, 117–18 skepticism and  3, 91, 100–2, 103, 110 Low Attributor-High Subject Stakes  184, 187 ignorant 185 informed 186 mistaken 185 see also UILA-HSS Low Attributor-Low Subject Stakes  188, 189 Low Credence-High Chances  203 Lowe, E. J.  60 luck  21, 65–79, 87–8, 203 bad  66, 141–2, 144, 146, 152 circumstantial  67, 68, 84 epistemic  3, 5, 85 good 66 judgments about  85 modal  67, 79–83 moral  67–8, 79, 141–2, 146, 154 n. probabilistic  65, 67, 82, 83–6 see also NLT; Probability-Luck; Safety-Luck; Sensitivity-Luck Ludlow, P.  20, 174, 177, 178, 179, 181, 185 n., 196 Luper, S.  81, 82 n., 136, 179, 221 Luzzi, F.  76

McKenna, R.  20, 178, 181 McKinnon, R.  101 McLaughlin, B. P.  96 metalinguistics  58 n., 122, 124, 133, 137 Mills, E.  103 modality 58–63 Moeller, E. F. L.  136 Moghaddam, A. R. H.  5 n. Montminy, M.  121, 122, 129, 132, 139, 153 n., 162, 195, 228 Moore, G. E.  96, 125, 137 Moore-paradoxality  124–5, 130, 134, 135, 171 moral realism  152 n. Morillo, C. R.  67 Morreau, M.  59 Morton, A.  212, 213 Murphy, P.  19 n.

MacFarlane, J.  126, 136, 179, 222, 226, 227, 228 Madison, B. J. C.  86 n. Maitzen, S.  96 n. Makinson, D. C.  105 Marr, D.  46 May, J.  190 McEvoy, M.  45 n., 47 n. McGinn, C.  42, 97 n. McGrath, M.  5 n., 10 n., 11 n., 13, 20, 32, 126, 168, 170 n., 166 n., 173 n., 182–3, 186 n., 187 n., 198, 210

objections  2, 5, 9, 16, 36, 59 n., 81, 102 n., 104, 107, 109 n., 113, 124 n., 135, 143, 159–97, 198, 227 alternatives and defenses against  140 core 4 factivity 134 “killer” 207 Okasha, S.  96 Olen, J.  102 n., 103 Olin, D.  96, 102 n., 103 n., 107, 116 n. Olsson, E. J.  53 n.

Nagel, J.  20, 172, 185 n., 189 n., 190 n. Nagel, T.  67, 141 natural language  12, 18, 140, 168 n. attributions made in  1, 162 competence and performance of ordinary speakers 191 Nelkin, D. K.  102 n., 103, 104 Neta, R.  15, 60 n., 81, 93, 195, 212 neutrality  125, 134, 201 n. metalinguistic 124 Newhard, J.  99 Nichols, S.  190 NLT (No-Luck-Thesis)  66–8, 74 n., 77, 86, 87 defense of  80, 81, 82 Nolan, D.  153 n. Norcross, A.  153 n. normative realism  152 n. normativity  63, 99, 110, 151, 152 n., 153 n. and arbitrariness  4, 193–5 norms and conventions  2–4, 20, 27–31, 88t, 190, 204, 207, 210 contextual 63 situational  87, 119 social 145 Nozick, R.  35, 80, 96, 97, 102, 103, 210, 225

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

index  263 parameters  14–20, 27, 31t, 58, 63–4, 88t, 95, 136, 154, 159, 190, 191 contextually variable  45, 51 description 87 distinction between determinants and  4, 9, 13, 45 epistemic  14, 126 non-epistemic 126 plurality of  123 n. special focus on  32 see also contextual parameters Partee, B. H.  178, 181 Party case  202 Paxson, T.  35 Perry, J.  38 n., 159 Phelan, M.  190 philosophical skepticism  124, 125, 176, 221 Pinillos, N.  190 Pollock, J.  47 n. Potrč, M.  122, 196 Prades, J. L.  99 Prijic-Samarzija, S.  15 Prinz, J. J.  153 n. Pritchard, D.  66, 67, 68, 80, 82 n., 84 n., 86 n., 99, 162 probability  3, 46, 50, 53, 65, 67, 86–7, 101–3, 107, 111–14, 144–5, 147–54, 201, 204 conditional  52, 62, 83–4, 85, 142, 143 distribution of  150 epistemic 202 indeterminate  53 n. interpretation of  200, 203 reliability and  34, 37–43, 58, 62–3, 82 subjective 202 unconditional  83, 84 unique 84 Probability-Luck  83–4, 85, 86 proposition-specificity 137–8 Pryor, J.  96 purposes and intentions  2–4, 20, 22–3, 27–31, 63, 88t, 112, 113, 118, 119, 190, 194, 204–5, 210 demanding standards established  207 legal 206 situational 206 Putnam, H.  100 Pynn, G.  161 Pyrrhonian skepticism  92 Radford, C.  15 Ramsey, F. P.  34 Rast, E.  121 Ravitch, H.  68 realism  152 n. Rebuschi, M.  124, 177 Recanati, F.  227 Reed, B.  99

reference classes  33–64, 145–7 contextual variability of  208 different types of  190, 196, 216 plurality of  192 spatial  56, 57, 63, 149, 151 surprising types of  193 temporal  55, 56, 57, 63 Reichenbach, H.  51, 53 n., 145, 148 Reid, T.  92 n. relativism  5, 92, 154, 221–30 knowability problems for  126 relativization  45 n., 80, 84 n., 85 n., 146, 175, 217–20, 221, 228 reliability  1, 2, 5, 64, 72, 140 context and  43–58 degree of  216 knowledge and  33–7 limited 113 maximal 34 modality and  58–63 probability and  34, 37–43, 58, 62–3, 82 standards for  14, 15, 19, 31t, 87t, 118–19 threshold for sufficiency of  191 variable 114 Rescher, N.  66, 86 n. Resnik, M. D.  202 responsibility  4, 140–55, 159 retraction 180 rule for  226, 227 Richard, M.  138, 178, 222 n., 226, 228 Rickless, S. C.  221 n. Rieber, S.  93, 103, 153 n. Riggs, W.  66, 67 n. Rohrbaugh, G.  81 Roland, J.  59 Rourke, J. 2013  213 Roush, S.  81, 102 n. Rovane, C.  225 Rowbottom, D.  70 n ., 77 n., 79 n. rule of attention  23, 24 ruling out  25, 26, 104 n. standards for  14, 18–20, 22, 27, 31t, 87t Russell, Bertrand  18, 68–74, 79, 96 Russell, Bruce  99 Ryan, S.  102 n., 103 n., 116 Rysiew, P.  162, 168, 190 Safety-Luck  80, 81, 82 salience  20, 25, 29, 30, 77, 87, 103 n., 107, 119, 175, 176, 196, 210 n. attention to alternative possibilities  204 chance  26–7, 63, 111–14, 115 n. neglect of  177 psychological  2, 150 Salmon, N.  96 n. Salmon, W. C.  51, 53 n., 145, 148 Schaffer, J.  20, 122, 190, 198, 207, 212, 213, 214, 215, 217 n., 221 n.

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

264  index Scharifi, G.  99 Schiffer, S.  20, 99, 167, 177, 195, 211 Schmitt, F. F.  47 n. Schmoranzer, S.  99 Schroeder, M.  20, 153 n. score keeping  24 n. semantic descent  58 n., 122 Sensitivity-Luck  80, 81, 82 Sextus Empiricus  92 Shackel, N.  97 n. Sherman, B.  97 Silins, N.  96 Simpson, S.  190 Sinnott-Armstrong, W.  153 n., 212, 213, 221 skepticism  10, 61, 124, 125, 176, 221 epistemological  3, 18, 92, 154 lottery  3, 91, 100–2, 103, 110 radical  110, 139 n. traditional  92–100, 110, 221 see also Cartesian skepticism; Pyrrhonian skepticism Skolits, W.  121, 129, 132, 139 Slote, M. A.  59 Smithson, M.  53 n. Snedegar, J.  220 Sosa, E.  1, 11 n., 17, 74, 80, 81, 93, 94, 111 n., 118 n., 122 spatial frames  3, 63t, 88t Speaks, J.  168 specificity  26, 104, 116, 214–15, 220 lack of  23 see also proposition-specificity Sripada, C. S.  21–2, 190 SSI (subject-sensitive invariantism)  4–5, 126, 127, 173 n., 182–5, 186–7 nn., 188, 189, 196, 198–212, 221 n. Stainton, R. J.  174 Stakes case  20–3, 27–31, 63, 87, 112–13, 118, 119, 126, 137, 144–5, 155, 161, 170, 172, 183–90, 197–201 see also Entailed Stakes; High Attributor Subject Stakes; High Credence Chances; Ignorant High Stakes; Inferential Stakes; Low Attributor Subject Stakes; Low Credence Chances; SSI Stalnaker, R.  213 standards 14–20 different kinds of  190, 196 plurality of  192 see also degrees of belief; epistemic position; evidence; knowledge; reliability; ruling out Stanley, J.  5 n., 9, 10 n., 21–2, 99, 126, 129, 136, 168, 173 n., 174, 177, 178, 179, 180, 181, 185, 186, 187, 188, 189 n., 190, 195, 198, 199, 201 n., 207, 208, 209, 210, 221, 227, 228 n.

Stemmer, N.  102 n. Steup, M.  121, 122 Stine, G.  18 n., 96 n., 107 Stojanović, I.  5 n., 229 n. Stone, J.  189 n. Stoutenburg, G.  83 Stroud, B.  92, 100 subjectivism  154, 201 Swinburne, R.  212 target beliefs  38, 44, 45, 61 acquisition of  49, 50 production of  48, 50, 62 Thelma and Louise case  11 n., 185 n. Thomas, A.  153 n. thought  140, 172, 196, 225 assimilation to speech  167 language and  1, 12, 162–6, 168 n., 169 solitary  1, 23 n., 162 Timmerman, T.  115 n. Timmons, M.  153 n. Torrengo, G.  225 Train cases  13, 48, 176, 186–8, 211 true beliefs  69, 102, 204, 209 accidental  67, 79 acquired  38, 39, 74, 83–4, 85 n., 86 n. all reliable beliefs have to be  59 n. attributions of credit for  87 n. chances of arriving at  46 crediting to intellectual abilities and epistemic virtues  115 n. evaluative significance of  67 falsehood might lead to  168 inferred  74 n. knowledge and  16, 30, 35–6, 66, 67, 79, 81, 85, 86, 87 n., 116, 126, 191 lucky  67, 68, 71, 74–81, 84, 85 mere  17, 66, 79, 193 non-accidental  66, 67 non-lucky 86 probability of  34, 37, 40–1, 62 share of  42 standards of epistemic positions require less than 30 tendency of visual perception to produce 42–3 truth  27, 35, 36, 97, 109, 114, 116, 117, 137, 175, 182, 199 n. context-invariant 15 falsehood reached leads to  169 index-sensitive predicate  225 inferred 75 knowledge requires/entails  16 n., 38 n. logical 75 mathematical  75, 214 monadic  222–3, 224, 225, 226, 229 negation of  10 n., 223

OUP CORRECTED PROOF – FINAL, 09/16/2016, SPi

index  265 non-epistemic pragmatic warrant does not have link with  169–70 relative  222–3, 224, 225, 226, 227, 229 reliable method of finding out  54 speaker saying what might be close enough to 168 truth theoretical systems approximate  76 warranted assertibility and  167, 196 warranted believability and  166 truth conditions  13, 17, 87 changing 166 context-dependent 111 context-invariant 111 context-sensitivity of  164, 165 contextual variability seems to stop with  193 difference in  16, 18, 28, 30, 74, 93, 163 invariant 161 judgment about  188 knowledge attributions  51, 57, 62, 74, 82, 86, 172, 220 necessary 14 overall similarity judgments  59–60 responsibility attributions  145, 152 subjunctive conditionals  59, 62 varying  1, 11, 16, 19, 33, 57, 65, 145, 152–3, 161, 164, 167, 198, 221, 222 truth-in-a-context  94 n. truth values  22, 38 n., 41, 129, 164, 214 n., 222 contrast propositions which lack  214 n. gaps in  11 n., 24 knowledge attributions  4–5, 121, 130, 198 stable 197 variable 221 Turri, J.  102, 113, 114 Tversky, A.  59 Two Allergies case  22–3, 27, 31 n., 204–5 typings topics/methods  38, 43–51, 63

UILA-HSS (Uncaring Informed Low Attributor-High Subject Stakes)  186–7 Unger, P.  5 n., 10 n., 11 n., 65, 92, 153 n., 168 uniformity assumption  137 unnecessary concern  188 Vahid, H.  66, 68 Van Woudenberg, R.  122, 213 Veber, M.  76, 121, 136 Venn, J.  51 Vogel, J.  18 n., 96 n., 101, 102 n., 103, 104 n., 116 Wallis, C.  47 n. WAMs (warranted assertibility maneuvers)  4, 160–73, 182, 185, 196, 208 Warfield, T. A.  34 n., 76, 94, 96 warranted assertibility, see WAMs WBMs (warranted believability maneuvers)  166–7, 168, 170–3 Weather Forecast case  9–11, 13, 14, 27–8, 174–80, 205–6, 216–17 Weatherson, B.  153 n., 173, 198 Wedgwood, R.  153 n. Weintraub, R.  102 n. Wilburn, R.  60 n. Wiles, A.  17, 175 Willaschek, M.  153 n. Williams, B.  68, 141, 170 Williams, M.  18, 42, 92 n., 96, 99 n., 135 Williamson, T.  5 n., 10 n., 80, 92, 125, 136, 138, 172, 179, 190 n., 195 Wittgenstein, L.  31, 92 n., 97, 98, 194 Wright, C.  96, 99, 121, 124, 222 n. Wright, S.  153 n. Wunderlich, M.  47 n. Yourgrau, P.  136, 179 Zalabardo, J. L.  60, 62 Zarpentine, C.  190