Good thinking : a knowledge first virtue epistemology 9781138317697, 1138317691

592 45 3MB

English Pages 126 [145] Year 2019

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Good thinking : a knowledge first virtue epistemology
 9781138317697, 1138317691

Table of contents :
Process reliabilism --
Virtue reliabilism: justified belief --
Virtue reliabilism: knowledge --
Knowledge first virtue reliabilism --
The competition --
The safety dilemma --
Lottery cases.

Citation preview

Good Thinking

This book combines virtue reliabilism with knowledge first epistemology to develop novel accounts of knowledge and justified belief. It is virtue reliabilist in that knowledge and justified belief are accounted for in terms of epistemic ability. It is knowledge first epistemological in that, unlike traditional virtue reliabilism, it does not unpack the notion of epistemic ability as an ability to form true beliefs but as an ability to know, thus offering a definition of justified belief in terms of knowledge. In addition, the book aims to show that this version of knowledge first virtue reliabilism serves to provide novel solutions to a number of core epistemological problems and, as a result, compares favourably with alternative versions of virtue reliabilism both in the traditionalist and in the knowledge first camp. This is the first ever book-length development of knowledge first virtue reliabilism, and it will contribute to recent debates in these two growing areas of epistemology. Christoph Kelp is a Lecturer in Philosophy at the University of Glasgow, UK. His work in epistemology has been published in Philosophy and Phenomenological Research, Noûs, Synthese and the Journal of Philosophy. He is the winner of the 2017 Young Epistemologist Prize.

Routledge Studies in Contemporary Philosophy

106 New Directions in the Philosophy of Memory Edited by Kourken Michaelian, Dorothea Debus, and Denis Perrin 107 A Pragmatic Approach to Libertarian Free Will John Lemos 108 Consciousness and Physicalism A Defense of a Research Program Andreas Elpidorou and Guy Dove 109 The Value and Limits of Academic Speech Philosophical, Political, and Legal Perspectives Edited by Donald Alexander Downs and Chris W. Surprenant 110 The Significance of Interdeterminacy Perspectives from Asian and Continental Philosophy Edited by Robert H. Scott and Gregory S. Moss 111 Questions of Practice in Philosophy and Social Theory Edited by Anders Buch and Theodore R. Schatzki 112 The Phenomenology of Real and Virtual Places Edited by Erik Malcolm Champion 113 Context, Truth, and Objectivity Essays on Radical Contextualism Edited by Eduardo Marchesan and David Zapero 114 Good Thinking A Knowledge First Virtue Epistemology Christoph Kelp

For a full list of titles in this series, please visit www.routledge.com.

Good Thinking A Knowledge First Virtue Epistemology

Christoph Kelp

First published 2019 by Routledge 711 Third Avenue, New York, NY 10017 and by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN Routledge is an imprint of the Taylor & Francis Group, an informa business c 2019 Taylor & Francis

The right of Christoph Kelp to be identified as author of this work has been asserted by him in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging-in-Publication Data Names: Kelp, Christoph, author. Title: Good thinking: a knowledge first virtue epistemology / by Christoph Kelp. Description: 1 [edition]. | New York: Taylor & Francis, 2018. | Series: Routledge studies in contemporary philosophy; 114 | Includes bibliographical references and index. Identifiers: LCCN 2018031821 Subjects: LCSH: Virtue epistemology. | Knowledge, Theory of. Classification: LCC BD176 .K45 2018 | DDC 121—dc23 LC record available at https://lccn.loc.gov/2018031821 ISBN: 978-1-138-31769-7 (hbk) ISBN: 978-0-429-45506-3 (ebk) Typeset in Sabon by codeMantra

To my parents

Contents

Preface Introduction 1 Process Reliabilism

ix xii 1

2 Virtue Reliabilism: Justified Belief

17

3 Virtue Reliabilism: Knowledge

37

4 Knowledge First Virtue Reliabilism

65

5 The Competition

82

Appendix The Safety Dilemma Appendix Lottery Cases Bibliography

100 110 117

Preface

I have been attracted to the virtue reliabilist account of knowledge— according to which knowledge is an achievement, a success because of competence—ever since I first came into contact with them during my PhD. While I thought the view carries a considerable amount of promise, I was struck by the fact that it seemed to struggle with fake barn cases and wasn’t satisfied with extant attempt to solve it. All of them took for granted that the success at issue in knowledge is truth and ventured to fiddle either with the competence condition or with the because relation in order to get these cases right. At some point in the late 2000s, it occurred to me that the right way to go might just be to abandon truth as the relevant success and go for something stronger. My first attempt at this was in a paper entitled ‘In defence of virtue epistemology’ which argues that, at least for perceptual knowledge, the success at issue in knowledge is discrimination. However, I ultimately remained unsatisfied with the proposal mainly because it didn’t seem to generalise beyond perceptual knowledge in the right way. I then toyed with the idea that the success might be safety instead but also abandoned the idea pretty quickly. This was in part because it had its own difficulties but mostly because I found the idea of exploring a knowledge first version of virtue reliabilism very attractive. I started thinking about this in the early 10s and first presented a paper that developed this kind of view in Bologna in 2012. A year later, I was awarded two generous research grants for a project on the issue which helped tremendously in giving me the opportunity to work out the view in more detail. Even so, it took what appeared to me to be a long time and an entire series of lengthy and sometimes frustrating reviewing processes before the first paper on the topic actually appeared in print in 2016. But by then the spell appeared to have been broken and a number of pieces appeared in fairly quick succession. While the material from these papers provides the backbone of this book, I also draw on earlier work at various junctures. More specifically, Chapter 2 is based on ‘How to Be a Reliabilist,’ forthcoming in Philosophy and Phenomenological Research (DOI: 10.1111/phpr.12438). Chapters 3–5 draw on material from ‘Justified belief: knowledge first-style,’ Philosophy and Phenomenological Research 93: 79–100 (2016), ‘Knowledge

x Preface first virtue epistemology,’ in Carter, A., Gordon, E. and Jarvis, B., editors. Knowledge First. Approaches to Epistemology and Mind, Oxford: OUP (2017), ‘Lotteries and justification,’ Synthese 195: 1233–44 (2017), and, to a lesser extent, on ‘Two for the knowledge goal of inquiry’ American Philosophical Quarterly 51: 227–32 (2014) and ‘Unreflective epistemology’ Episteme 11: 411–22. (Note, however, that there is no one-toone correspondence between the contents of book chapters and papers. Sometimes chapters expand on papers or develop relevant ideas further. At the same time, in some cases, papers go beyond what is covered by the book.) Furthermore, Appendix ‘The Safety Dilemma’ is based on ‘Epistemic Frankfurt cases revisited,’ American Philosophical Quarterly 53: 27–37 (2016). I thank the relevant publishing houses and editors for granting me permission to use this material. I have no doubt in my mind that this book would not have seen the light of day without the input and support from colleagues, friends and family. In this regard, I’d like to especially thank Harmen Ghijsen and Mona Simion for offering not only extensive comments on the entire manuscript as well as on the papers on which it is based, but also for their unabated enthusiasm in discussing this book’s topics. Duncan Pritchard and Ernest Sosa have influenced my philosophical mind more than anyone. I have benefitted from their input profoundly and in numerous ways, including via discussion of the topics covered in this book and comments on specific work, and more indirectly in their role as my mentors. Thanks for invaluable input also to Kristoffer Ahlstrom-Vij, Matt Benton, Sven Bernecker, Michael Blome-Tillman, Cameron Boult, Elke Brendel, Fernando Broncano-Berrocal, Adam Carter, Annalisa Coliva, Juan Comesaña, Igor Douven, Philip Ebert, Pascal Engel, Markus Eronen, Mikkel Gerken, Sandy Goldberg, Peter Graham, Martin Grajner, Patrick Greenough, John Greco, Thomas Grundmann, Frank Hofmann, Jepser Kallestrup, Klemens Kappel, Dirk Kindermann, Clayton Littlejohn, Jack Lyons, Conor McHugh, Alan Millar, Lisa Miracchi, Anne Meylan, Erik Olsson, Nikolaj Jang Lee Linding Pederson, Wayne Riggs, Blake Roeber, Pedro Schmechtig, Martin Smith, Matthias Steup, Kurt Sylvan and Tim Williamson. (I am sure that I have missed people who deserve to be acknowledged by name and hope they will accept my apologies.) Elements of this book have been presented on a number of occasions, including: 2012 European Epistemology Network Meeting, Universities of Bologna and Modena; First Conference on Contemporary Philosophy in East Asia, Academia Sinica Taipei 2012; Epistemology Workshop, University of Groningen 2013; EPSA 2013, University of Helsinki; Epistemic Justification and Reasons Conference, University of Luxembourg 2013; Bled Philosophical Conferences, Bled 2013 and 2015; Saving Safety? Problems and Prospects of Safety-Based Accounts of Knowledge, University of Bonn 2013; The Virtue Turn: Virtue Ethics, Virtue Epistemology

Preface xi and Chinese Philosophy, Soochow University Taipei 2014; Yonsei Summer Philosophy Conference, Yonsei University Seoul 2014; BELUX2: Perceptual Normative Reasons, University of Luxembourg 2014; Philosophy Seminar Series, Lingnan University Hong Kong 2014; 5th International Meeting on Scepticism: The Value of Understanding, Federal University of Bahia; BELUX4: The Norm of Belief, University of Luxembourg 2015; Virtue Epistemology Conference, KU Leuven 2015; Departmental Colloquium, University of Groningen 2016; Departmental Colloquium, Institut Jean-Nicod 2016. Thanks to the audiences for their very helpful comments on the material in this book. Finally, I’d like to thank my family for their continuous support across all walks of life.

Introduction

Try for a moment to stop thinking; try, in particular, to stop forming beliefs. This is hard. For instance, you very likely can’t help but form the belief that I just asked you to stop forming beliefs. And even if we give you some time before starting to count your attempt, as it were, you very likely can’t help but form the belief that this book is still in your hand, that the screen is still in front of you and so on. You might try to close your eyes. But now you form the belief that your eyes are now closed and you may still form beliefs about things you hear and smell. You might try to put your fingers in your ears and nostrils. But even if you manage to pull this off, you will likely form the belief that you must be making quite an appearance. And if you are not that attuned to facts about your looks to take note of this, you will, at the very least, form the belief that your fingers are in your ears and nostrils. Perhaps you seek out a sensory deprivation tank. But now you might form beliefs about how curious it is to not experience anything. You might also form beliefs about the contents of your stream of consciousness. Perhaps you are singing a song in your head or perhaps you are just feeling a panic attack coming on and form the corresponding beliefs. I could go on for at least a bit more. But I do hope that you will now be convinced that no longer forming beliefs is a tall order indeed. Say that you actually do manage to no longer form beliefs. Try next to stop having beliefs altogether. In comparison, not forming them will look easy. Once you have a belief, you are pretty much stuck with it. Don’t get me wrong. I don’t mean to suggest that it’s impossible to lose a belief once you have it. You might figure out that what you believed was not true after all or that, contrary to how things appeared at the time, you don’t have good reason for thinking it to be true. All I am saying is that there is nothing in the order of what closing your eyes will do to forming visual perceptual beliefs that will allow you to get rid of beliefs that you already have. Perhaps therapy will do the job for at least some beliefs. But even that is an arduous process by comparison. There is good thinking and bad thinking. Since believing is a way of thinking, it is unsurprising that there is also good believing and bad believing. It matters to have good beliefs. Given that we are believing

Introduction xiii at all, we might as well believe well rather than badly. Or so some might think. But having good beliefs matters even to those who are not driven to perform well in all walks of life. After all, we act on our beliefs, and our actions will be more likely to be successful if the beliefs on which they are based are good and less likely if they are bad. Suppose, for instance, that you hold the bad belief that Paris is the capital of Scotland. Suppose furthermore that you have recently moved to Scotland and want to register with your consulate in the capital. If you now arrange for transportation to Paris to do this, you will not only fail in your project of registering with the consulate but you will also waste a lot of time and money. In this way, believing well is of direct practical importance to us. And one key philosophical question that immediately arises is what exactly believing well amounts to. It is just this question that will take centre stage in this book. To be perfectly clear, despite the fact that this book is entitled Good Thinking I do not aim to offer a full account of what it is to think well. In fact, I do not even aim to offer a full account of what it is to believe well. Rather, I will rest content with making progress on these questions. More specifically, I will here focus on two widely recognised forms that believing well may take, to wit, knowing and believing justifiably. My goal is to develop novel accounts of both knowledge and justified belief as well as the relation between the two. Aiming to make progress towards answering an important question may sound modest. It is not, at least not in this particular case. The reason for this is that knowledge and justified belief are not only two widely recognised forms of believing well, they are arguably widely recognised as the two most central forms of believing well. It will come as no surprise, then, that the questions of what knowledge and justified belief are have been at the very heart of epistemological theorising throughout its history. Any self-respecting epistemologist will have to address them sooner or later. In view of just how important the issue of believing well is to us and in view of just how central the questions about the nature of knowledge and justified belief are to the epistemological enterprise, it will be very hard indeed to underestimate just how ambitious the project is and just how significant its results promise to be. Before getting down to the nitty-gritty, I’d first like to survey the state of play, clarify some key allegiances, flag some core assumptions and sketch a game plan for the remainder of the book.

Knowledge Whatever else may be true of knowledge, it is overwhelmingly plausible that knowledge entails true belief. However, it is also widely agreed that true belief alone is not sufficient for knowledge. Additionally, true belief needs to be “tethered”, to put it in Plato’s words. For the longest time

xiv Introduction it was taken for granted that the tether that knowledge requires is justification. That is to say that knowledge requires justified true belief. And conversely, it was also taken for granted that if one believes not only truly but also justifiably then one knows what one believes. Knowledge, then, is justified true belief. Let us call this ‘the received view’. The received view had long been taken to be the answer to the question as to what knowledge is. Enter Edmund Gettier. In a short 1963 paper, Gettier provided an argument that showed, to the satisfaction of nearly all researchers in the field, that the received view is false. To be more precise, he offered a couple of examples that are widely taken to show that the conditions of the received view are not sufficient for knowledge. Here is one prominent such example, albeit not one from Gettier’s original paper: Stopped Clock. You take a reading from a clock and form a belief about the time, say that it is 8:22. You know that the clock has been working reliably in the past and have no reason to believe that it is currently inaccurate or not working properly. Finally, you have the ability to tell the time by taking readings from this kind of clock. Your belief that it is 8:22 is clearly justified here. Now suppose that, unbeknownst to you, the clock has stopped. Luckily, however, it stopped exactly twelve hours earlier with the result that the reading is actually accurate: it actually is 8:22.1 Your belief is not only justified but also true. However, it is also clear that your belief does not qualify as knowledge. Hence, contrary to the received view, justified true belief is not sufficient for knowledge. Gettier’s paper was the source of a major upheaval in the theory of knowledge. Once it was clear that the received view is false, epistemologists have tried to find out what knowledge consists of instead. The task that epistemologists set for themselves was no less ambitious than to provide a reductive analysis of knowledge in terms of belief, that is, a non-circular set of individually necessary and jointly sufficient conditions for knowledge involving belief. This task turned out to be rather difficult to accomplish. While a set of conditions that would deal with Gettier’s original examples was not hard to come by, it has proved surprisingly hard to find a set of conditions that would deal with the phenomenon in general. Evidence for this claim can be gleaned from the way in which the debate over the analysis of knowledge has developed since 1963: In the run for the correct analysis of knowledge, a huge number of proposals have been advanced. However, these proposals have tended to fall prey to Gettier-style counterexamples, that is, cases in which the agent satisfies the proposed set of individually necessary and jointly sufficient conditions but does not know (henceforth ‘Gettier cases’). It is fair to say that over 50 years of research have not produced an account of knowledge that is even widely agreed among researchers in the field.

Introduction xv In fact, as a result of this development, some philosophers have become suspicious of the very prospects for the project of providing a reductive analysis of knowledge. Most notably, Timothy Williamson (2000) takes the upshot of the post-Gettier debate on the nature of knowledge to be that the search for a reductive analysis is irredeemably flawed and has proposed to scrap this point from the epistemological agenda. Instead, he has offered an alternative ‘knowledge first’ approach to epistemology. According to knowledge first epistemology, knowledge is epistemologically fundamental in the sense that it does not admit of reductive analysis in terms of belief. Instead, knowledge first epistemology aims to explain various core phenomena in epistemology in terms of knowledge. I’d like to take the opportunity to flag that, in this book, I will join Williamson in opting for a knowledge first approach to epistemology.

Justified Belief Gettier’s paper did not only initiate the quest for a condition that would deal with Gettier cases. It also spurred a debate over the nature of the justification condition that played the role of the tether in the received view. Perhaps the most widely discussed issue here is the internalism/externalism divide. Internalism is the thesis that whether a belief is justified supervenes only on factors that are, in some sense to be specified, internal to the agent. Externalism is simply the denial of internalism. That is to say, according to externalist theories of justification, whether a belief is justified supervenes at least in part on external factors. It is fair to say that internalism was the predominant view throughout the history of philosophy before Gettier. The classical version of internalism offers an epistemic account of the internal. A factor is internal to an agent if and only if the agent is in a position to know that it obtains by reflection alone. This version of internalism is known as accessibilism.2 Accessibilism can be motivated by appealing to a plausible idea about the nature of justification in general. On the face of it, justification has to do with the ability to justify. In order to justify something, one needs to be able to offer some reason in support of it. And, of course, one cannot offer anything as a reason unless one is in a position to identify it as such. What’s more, in order to have justification for something at a certain time, one must be able to justify it at that time. In that case, however, it simply won’t do if one still needs to launch an investigation into the supporting reasons. Rather, one must be in a position to know the relevant reasons at the time at which one is to offer the justification. What one is in a position to know at that time, it would seem, is limited to what one is in a position to know by reflection alone. Given that justification has to do with the ability to justify, then, it would seem that justification will feature exactly the kind of epistemic condition accessibilism countenances.

xvi Introduction In contrast, externalism has made its breakthrough in epistemology only fairly recently, with a paper by Alvin Goldman in 1979. Goldman defends an account of justified belief entitled process reliabilism. Roughly, process reliabilism claims that a belief is justified if and only if it is produced by a cognitive process that is reliable in the sense that it tends to produce beliefs with a favourable truth to falsity ratio. Externalism has been on the rise in epistemology ever since and by now there is a considerable variety of different externalist accounts of justification on the epistemological market. One of the strongest motivations for externalist accounts of justification is that they can make sense of the idea that justification is, by its very nature, connected to truth. Goldman offers one way of fleshing out this connection to truth. Justified beliefs are beliefs that are produced by processes that are connected to the truth in the sense that they tend to produce true rather than false beliefs. It may also be worth noting that internalist accounts of justification will struggle to accommodate the truth connection. After all, as cases of radical deception indicate, internal factors are not by their very nature connected to truth. Any connection between internal factors and truth is at best contingent. While the debate between internalism and externalism is ongoing, I would like to flag that I will not provide any detailed critical discussion of internalism in this book. What I will do is take a closer look at a range of what I take to be familiar motivations for the kind of externalism I favour (in Chapter 1). I believe that these considerations in support of externalism make a sufficiently strong case in support of the view. Other than that, I do not have much to add to the case against internalism. So, in what follows, I will set internalism to one side.3

The Relation between Knowledge and Justified Belief Gettier’s paper has drawn attention to the relation between knowledge and justified belief. According to the received view, knowledge entails justified belief. However, some, including some internalists, have since rejected the idea that knowledge entails justified belief.4 That said, that knowledge entails justified belief is still the dominant view in epistemology. What’s more, for externalists, there appears to be little reason to reject the idea that knowledge entails justified belief. Consider, once more, process reliabilism: why should someone who maintains that believing justifiably has to do with believing via reliable processes give up on the idea that knowledge entails justified belief? After all, it seems equally plausible that knowledge requires believing via reliable processes. For the purposes of this book, I will side with those who agree that knowledge entails justified belief. Another question one may ask is why knowledge should entail justified belief. Or, more generally speaking, why knowledge should feature

Introduction xvii some specific combination of epistemic conditions rather than some other. Consider, for instance, someone championing an account of knowledge featuring a reliabilist justification condition and a causal anti-Gettier condition. Why, one might ask, should knowledge feature precisely these conditions? One answer, of course, might be that it gets the cases right. But even so, one might wonder whether there isn’t a deeper theoretical explanation of why knowledge is the combination of just these two epistemic conditions. This question is somewhat less frequently addressed. One account that does offer an answer to this question is virtue reliabilism which is arguably the most prominent alternative to process reliabilism in the reliabilist camp.5 According to one version of virtue reliabilism, perhaps the most widely discussed one, epistemic normativity is but an instance of a more general kind of normativity, to wit, performance normativity. Performances with an aim, we are told, can be assessed along three dimensions: success, competence and achievement, where a performance qualifies as an achievement if and only if it is successful because competent. Belief is then viewed as one kind performance with an aim, to wit, a kind of cognitive performance that aims at truth. Given that this is so, the account of performance normativity can be applied to the case of belief. This means that beliefs can be assessed as successful (i.e. true), competent and as qualifying as achievements (i.e. true because competent). Virtue reliabilism identifies knowledge with belief that qualifies as achievement and justified belief with competent belief. Since achievements entail competent performances, it follows that knowledge entails justified belief. In this way, virtue reliabilism can offer exactly the kind of explanation of why knowledge entails justified belief that we were looking for. Moreover, the specific combination of epistemic conditions on knowledge, competence and the instantiation of the because relation between truth and competence, also receives the sought after theoretical motivation. It will not come as a great surprise, then, that the accounts of knowledge and justified belief virtue reliabilism offers are among the most popular in recent literature. The accounts of knowledge and justified belief that this book will develop are also virtue reliabilist. However, they differ from its most prominent cousins in the literature in that they are developed within a knowledge first approach to epistemology. In this way, the main aim of the book is to develop a knowledge first virtue reliabilism.

Game Plan The remainder of this book is divided into five chapters. Chapter 1 introduces the most widely discussed version of reliabilism, process reliabilism, and looks at some of the motivations and problems for the view, including the infamous new evil demon problem, clairvoyant cases and the generality problem. By the same token, the chapter indicates some

xviii Introduction places where there is opportunity for alternative forms of reliabilism to improve on their process reliabilist rival. Chapter 2 takes a closer look at virtue reliabilist accounts of justified belief. The standard virtue reliabilist framework is introduced and found wanting. An improved version of this framework is then developed and it is argued that it serves to solve a number of the problems that beset process reliabilism, including the new evil demon problem, the problem of clairvoyant cases and the generality problem. Crucially, all of this can be done without abandoning the traditionalist view in favour of a knowledge first alternative. Chapter 3 turns to virtue reliabilist accounts of knowledge. It is argued that a certain class of Gettier cases and lottery cases continue to constitute a stumbling stone for the view. Chapter 4 develops my preferred knowledge first version of virtue reliabilism. Virtue reliabilist accounts of knowledge and justified belief as, respectively, apt and competent belief, are offered. Crucially, however, aptness and competence are unpacked in terms of knowledge and so the account falls squarely within the knowledge first programme. It is shown that there are two sources of support for knowledge first virtue reliabilism over its traditionalist rival. The first is direct. Roughly, it is argued that the knowledge first thesis that knowledge is success in inquiry is preferable to the traditionalist alternative, according to which it is true belief. The second is indirect. Here, the strategy is to show that knowledge first virtue reliabilism can avoid the problems Chapter 3 outlined for its traditional virtue reliabilist rival. Finally, Chapter 5 looks at alternative knowledge first versions of virtue reliabilism that can be found in the literature and argues that my account compares favourably with them.

Notes 1 Perhaps somewhat surprisingly, it dates back even further than Gettier’s paper, to Russell’s 1948 book Human Knowledge: Its Scope and Limits. Crucially, Russell did not use the example to target the received view. 2 Roderick Chisholm (1977) and Laurence BonJour (1985, 2003) are perhaps the most prominent champions of accessibilism in the 20th century. However, the view has a number of further supporters including Robert Audi (2001), Carl Ginet (1975), Matthias Steup (1999) and Declan Smithies (2012). Another popular stripe of internalism is mentalism, which offers a metaphysical account of the internal. According to mentalism, internal factors are mental factors. The most prominent champions of mentalism are Earl Conee and Richard Feldman (e.g. 2001, 2004). 3 This is not to say, however, that this book is of no interest at all to internalists as it will offer responses to a number of objections that have been levelled against externalism from the internalist camp. 4 The justification condition on knowledge has been rejected, e.g. in Alston (1988), Audi (2001), Foley (2012) on the internalist side and e.g. in Plantinga (1993), Sartwell (1991) on the externalist side.

Introduction xix 5 The most prominent champions of this kind of view are Ernest Sosa (e.g. 1980, 1991, 2007, 2011, 2015), John Greco (e.g. 1999, 2000, 2003b, 2007, 2010, 2012) and Duncan Pritchard (e.g. Pritchard et al. 2010, Pritchard 2012a, 2015). Versions of the view have also been defended by Wayne Riggs (e.g. 2002, 2007, 2009), John Turri (e.g. 2011, 2016) and Adam Carter (e.g. 2015, 2015). I have also defended versions of virtue reliabilism e.g. in (Kelp 2011a, 2013).

1

Process Reliabilism

This chapter’s main focus will be on process reliabilism (henceforth ‘PR’). In the reliabilist camp, PR is arguably the most prominent competitor of virtue reliabilism (‘VR’), the kind of view I favour. More specifically, I will first introduce the view and then look at some of the ups and downs of PR. The ups promise to also constitute motivations for VR. The downs are points at which VR has the opportunity to prove its superiority over PR.

Process Reliabilism in Outline The core idea of PR is that justified beliefs are beliefs that are produced by processes that tend to produce beliefs with a favourable truth to falsity ratio. In other words, justified beliefs are beliefs produced by processes that reliably produce true beliefs (henceforth also ‘reliable processes’).1 With the core idea in play, let’s turn to some important qualifications. Note that not all belief-forming processes are created equal. Some beliefforming processes, most notably perception, do not have beliefs among their inputs. For instance, your perceptual belief that you are looking at a book has no beliefs among its inputs. It is not as if you believe that you are looking at a book based on a prior belief that you have an experience as of looking at a book. Rather, the only inputs to the process are non-doxastic. They may comprise an experience as of a book, certain retinal stimuli or perhaps something entirely different. Other belief-forming processes do have beliefs among their inputs. The most prominent example here is inference. When you form a belief that q by inference from a belief that p and a belief that if p, then q, the inferential process that outputs your belief that q has your belief that p and your belief that if p, then q among its inputs. Following Goldman, I will call the former kinds of process ‘belief-independent’ and the latter ‘belief-dependent’. While it makes sense to require belief-independent processes to be reliable in the sense that in order to deliver justification they must produce beliefs with a favourable truth to falsity ratio, the same does not hold for belief-dependent processes. To see this notice that belief-dependent processes need not be expected to produce beliefs with a favourable truth

2 Process Reliabilism to falsity ratio when the input beliefs happen to be false. Moreover, they need not be expected to ensure that the input beliefs be true. Process reliabilists will do well, then, to weaken the reliability condition on beliefdependent processes. Rather than requiring that they produce beliefs with a favourable truth to falsity ratio unconditionally, these processes need only be conditionally reliable in the sense that they must produce belief with a favourable truth to falsity ratio given that the input beliefs are true.2 Finally since conditionally reliable processes can only transmit justification but not generate it, belief-dependent processes will produce justified beliefs only if the input beliefs are themselves justified. This gives us PR in Goldman’s classic formulation: Process Reliabilist Justification (PRJ). S’s belief that p is justified if and only if (i) it “results (‘immediately’) from a belief-independent process that is (unconditionally) reliable” or (ii) it “results (‘immediately’) from a belief-dependent process that is (at least) conditionally reliable, and . . . the beliefs (if any) on which this process operates in producing S’s belief that p . . . are themselves justified.” (Goldman 1979: 13–14) Even this cannot be the whole story. After all, compatibly with a given belief’s being formed reliably, the justification of the belief can be defeated. For instance, you may have a reliably produced visual perceptual belief that the painting you are looking at is red. According to PRJ, your belief is justified. However, when you are told that you have been given a drug that makes green objects appear red to you, your perceptual belief will no longer be justified. The justification for your belief is defeated. Goldman proposes a PR-friendly account of defeat along the following lines: Process Reliabilist Defeat (PRD). S’s belief in p is defeated if and only if “there is an alternative reliable or conditionally reliable process available to S which, had it been used by S in addition to the process actually used, would have resulted in S’s not believing p . . . ” (Goldman 1979: 20) Whether a belief is genuinely justified depends on whether it is justified in the sense of PRJ and not defeated in the sense specified in PRD. While defeat is an important phenomenon in epistemology, for the purposes of this book, I would like to set it aside, unless otherwise noted. Accordingly, in what follows, I will use ‘PR’ to refer to process reliabilism’s account of justification minus its account of defeat, i.e. PRJ.

Process Reliabilism 3

Motivations Properties of Justified Belief One initial benefit of PR is that justified belief turns out to have a number of properties it is widely believed to have. To begin with, PR allows for the non-factivity of justified belief. That is to say, it is possible to have justified false beliefs. To see how PR can achieve this, note that a process can be reliable—and thus satisfy PR’s demands—without being perfectly reliable. In fact, nearly all, if not all, of our belief-forming processes are not perfectly reliable. They are fallible. When their fallibility manifests itself, they produce false beliefs. However, since they are nonetheless reliable, according to PR, these false beliefs will turn out justified. While justified belief is not factive, it is widely agreed that it is not entirely unconnected to truth. On the contrary, it is widely believed that justified belief is truth conducive in the sense that a justified belief is, in some sense to be specified, more likely to be true than not. One crucial advantage of PR is that it can make sense of the truth conduciveness of justified belief while holding on to non-factivity. To see this, note that beliefs produced by reliable processes are more likely true than beliefs that aren’t produced by reliable processes. More specifically, the conditional probability of the belief that p’s being true given that it has been produced by a reliable process is higher than the probability of the belief that p’s being true alone. In this way, PR does establish the desired connection between justification and truth. Paradigm Cases PR also makes correct predictions about paradigm cases of justified and unjustified beliefs. Consider paradigm cases of justified beliefs: they include beliefs produced by standard perceptual processes, introspection, memory and good reasoning. Now contrast this with paradigm cases of unjustified beliefs: they include beliefs produced by confused reasoning, guesswork and wishful thinking. What distinguishes beliefs in the first camp from beliefs in the second camp is that the processes producing the former are all reliable or conditionally reliable, while the ones producing the latter are all unreliable or conditionally unreliable. Given that, according to PR, a belief is justified if and only if it is produced by a reliable or, in the case of non-basic beliefs, conditionally reliable process, PR makes the right predictions for both paradigm cases of justified beliefs and of unjustified beliefs (Goldman 1979: 9–10). Agrippa’s Trilemma Another noteworthy benefit of PR is that the view promises to offer attractive solutions to a number of key problems in epistemology,

4 Process Reliabilism including the problem of basic beliefs, the related regress problem and the problem of scepticism. Let’s look at the regress problem and basic beliefs first. Agrippa’s trilemma famously describes three accounts of the structure of justification: (i) there is an infinite chain of justified beliefs (infinitism), (ii) the chain of justified beliefs circles back to the original belief (coherentism) or (iii) the chain of belief ends with basic beliefs that are not themselves justified by other beliefs (foundationalism). While infinitism and coherentism have their adherents in the literature, it is fair to say that foundationalism is the most popular view.3 In fact, many take the trilemma to constitute an argument for foundationalism. Even if we agree that the trilemma makes a strong case for foundationalism, foundationalists still encounter a difficult problem. To see how it arises, note that not all basic beliefs are justified. Consider, for instance, an expert and a novice doctor who are currently inspecting an X-ray image. The expert can tell from the picture that the patient has a torn ligament, whereas the novice can’t. Suppose both expert and novice arrive at a true basic belief about the patient’s ailment. Say that the novice did so based on mere guesswork. In this case, the expert’s belief is justified, while the novice’s belief isn’t. While some basic beliefs are justified, other’s aren’t. Given that this is so, it is incumbent on foundationalists to offer an account of when exactly a basic belief is justified and when it is unjustified. This account should enable us to explain why, in the above case, the expert’s belief is justified, whereas the novice’s isn’t. Reliabilism can offer an attractive answer here. Basic beliefs are justified if they are produced by (unconditionally) reliable processes and unjustified otherwise. The expert’s belief about the torn ligament is justified because it is produced by a reliable belief-forming process, the novice’s is unjustified because the belief-forming process that produced it is not reliable. In this way, PR offers an appealing account of basic justification and an attractive foundationalist solution to the regress problem (Kornblith 1980; Goldman 2008). Scepticism Let’s move on to the problem of scepticism, which is one of epistemology’s most venerable problems. A particularly virulent and perhaps the most unsettling version of the problem threatens to show that it is simply impossible for us to have much in the way of justified belief at all. Here is the argument: SK1. We cannot justifiably believe the denials of various sceptical hypotheses. For instance, you cannot justifiably believe that you are not the victim of a mad scientist who has taken your brain out of your

Process Reliabilism 5 skull, placed it in a vat with nutrients and hooked it up to a supercomputer that stimulates it in such a way as to produce deceptive experiences as if you were a normal human being with hands. SK2. If we cannot justifiably believe the denials of various sceptical hypotheses, then we cannot justifiably believe many ordinary empirical propositions. For instance, if you cannot justifiably believe that you are not a handless brain in a vat that is deceived into believing it has hands, then you cannot justifiably believe that you have hands. SK3. Hence, we cannot justifiably believe many ordinary empirical propositions. For instance, you cannot justifiably believe that you have hands. PR allows for an attractive solution to this problem. According to PR, what matters to whether or not a given belief is justified is whether it is reliably produced. And whether or not our beliefs, including beliefs about whether we are radically deceived, are actually produced by reliable processes, it is surely possible that they are. In this way, reliabilists have excellent cause for resisting the key premise in the above sceptical argument, to wit, SK1, as well as its conclusion, SK3. In this way, PR can resist one prominent sceptical argument (Pritchard 2005). Even if PR escapes the above sceptical argument, it might be thought that it does not serve to solve the sceptical problem fully adequately. After all, to establish an unsettling sceptical conclusion, the sceptic need not make so strong a claim as to say that it is impossible for us to have justified beliefs. Rather, it will be bad enough if he can show that, whether possible or not, we simply do not have much in the way of justified beliefs. And all that is needed for this is the weaker premise that we do not justifiably believe the denials of various sceptical hypotheses. One issue the sceptic now faces is that, once he has conceded that, on PR, it is possible to justifiably believe both the denials of sceptical hypotheses and ordinary empirical propositions, then the weaker premise that we do not justifiably believe the denials of various sceptical hypotheses affords separate support. To make a plausible case for this, however, sceptics will now have to convince us that the world we inhabit is not one of the possible worlds in which the possibility of justified belief in denials of sceptical hypotheses and ordinary empirical propositions, which PR opens up, is actualised. In order to achieve this, they have to argue that the processes that lead us to beliefs in the denials of sceptical hypotheses are not reliable in the sense at issue in PR. However, sceptics have done precious little to achieve this. (In a way, this is unsurprising because the issue of whether the relevant belief-forming processes are reliable is just the kind of empirical matter that sceptics think we cannot know or have justified beliefs about.) So, while the weaker sceptical argument is in principle still available to sceptics, it comes to light that one of

6 Process Reliabilism its key premises remains unsupported. As a result, the weaker argument will not be cogent against champions of PR. These are some of the main considerations that move champions of PR to adopt their view. If PR can be made to work, it offers an attractive and powerful account of justified belief. That said, PR also faces a number of well-known problems. One major aim of this book is to offer an account of justified belief that retains the benefits of PR, while avoiding these problems. In what follows, then, I will outline these problems.

Problems The New Evil Demon Problem First, consider once more cases of radical deception. Here is one famous example: Envatted. You have become the victim of a mad scientist who has turned you into a brain in a vat (henceforth also ‘BIV’ for short) that is being fed deceptive experiences as if everything were normal. For instance, just now you appear to be waking up in your bed in the morning, appear to see that your alarm clock reads 7:45 and form the corresponding false beliefs. The beliefs you form based on your experiences in this case are justified. Or, at any rate, they are no less justified than they were before you were turned into a BIV. The problem for PR is that since you are subject to radical deception, the processes that produce your beliefs are highly unreliable. You are a disembodied BIV. You are not lying and surely not in your bed. Your belief about the time is mistaken also. Moreover, it is not as if your beliefs about your bodily orientation and the time are false only on this occasion. No, whenever you form such beliefs, they are going to be false (or so we may assume). What’s worse, the problem is not restricted to beliefs about bodily orientation and time. Rather, your perceptual beliefs in general will be affected. As a result, the processes that produce these beliefs will do so with a highly unfavourable truth to falsity ratio. PR predicts, mistakenly, that the beliefs you form are not justified.4 Clairvoyance Cases PR meets with a similar fate when it comes to cases like the following. Clairvoyant. As a result of exposure to radiation, you come to be gifted with a ‘clairvoyance’ belief-forming process that produces true beliefs about distant events with a high degree of reliability. You do not know that you are gifted with such a process. In fact, you have no evidence that there exists a process of this kind or that it should be

Process Reliabilism 7 so much as possible for it to exist. On the other hand, you also do not have evidence that such a process does not exist/is not possible. From your point of view, you spontaneously form beliefs about distant places. Just now, while being on vacation in a faraway country, your clairvoyance process has produced a belief that your house is on fire. In this case, your belief is not justified. At the same time, the process that produced it is highly reliable and so PR predicts, again mistakenly, that your belief is justified.5 These two problems are especially worrisome for PR in tandem. Cases like Envatted provide reason to believe that process reliability is not necessary for justified belief, while cases like Clairvoyant suggest that it is not sufficient either. If process reliability is neither necessary nor sufficient for justified belief, it looks as though PR is heading down the wrong epistemological track entirely. The Generality Problem And this is not even the end of the difficulties PR encounters. There are a couple of further formidable problems that I would like to briefly outline. The first one is the infamous generality problem. The core idea of PR is that a belief is justified if and only if it is produced by a reliable process. To be more precise, the core idea here is that a belief is justified if and only if it is produced by a token of a reliable process type. Individual beliefs are produced by token processes. Token processes are particulars, which, among other things, are not repeatable. As a result, it’s not clear that the notion of reliability even applies to token processes which is why champions of PR typically take justified belief to require production via a reliable process type.6 To see the difficulty for PR, notice that every token process instantiates indefinitely many types. The process that produced my belief that I am sitting at my desk, for instance, instantiates all of the following types: cognitive process, belief-forming process, perceptual belief-forming process, and visual belief-forming process, to name but a few. Each process type has a different degree of reliability. Of course, unless it is specified which process types exactly are the ones at issue in PR, PR simply does not make determinate predictions about whether or not individual beliefs are justified. For instance, before PR can make a determinate prediction on whether my belief that I am sitting at my desk is justified, we need a specification of which of the various process types that the token process instantiates is the one at issue in PR. In order to work as an account of justified belief at all, PR must be supplemented with an account that specifies the relevant type of process at issue in PR. The problem of specifying the relevant type of process has become known as the generality problem for PR.7

8 Process Reliabilism World-Bound Reliability We still haven’t reached the end of the difficulties PR faces. Notice that, according to PR, process reliability is evaluated relative to worlds. That is to say, whether a given process type is reliable depends on the ratio of true to false beliefs it attains at a set of worlds.8 But now consider a process type, T, that takes a certain look of a certain mushroom as input and outputs a classification of the mushroom as edible. Suppose T is a highly unreliable process type because, at the set of worlds at which it is evaluated for reliability, both edible chanterelles and poisonous jack-o’-lantern mushrooms have this look and throughout this set of worlds nearly everywhere where there are chanterelles, there are also jack-o’-lanterns. But now suppose, compatibly with this, that throughout this set of worlds there also exists a remote island with only a handful inhabitants where there are no jack-o’-lanterns. None of the inhabitants have ever left the island and they never will. Suppose you are an inhabitant of this island and form a belief of the mushroom before you that it is edible via a token of T. Your belief qualifies as knowledge and so is justified. Since T is highly unreliable due to its tendency to produce too many false beliefs at the set of worlds at which it is evaluated for reliability, however, PR predicts, incorrectly, that your belief is unjustified. Consider also the converse situation in which T is highly reliable partly because, throughout the set of worlds at which it is evaluated for reliability, nearly only chanterelles have the look in question. Jack-o’-lanterns are nowhere to be found, with the exception of the remote island you inhabit where they grow as abundantly as chanterelles. When you form the belief of the mushroom before you that it is edible via T, your belief is very plausibly not justified. Since T is highly reliable as it tends to produce true beliefs at the set of worlds at which it is evaluated for reliability, PR makes yet another incorrect prediction, this time of the presence of justified belief.

Solutions There is no shortage of proposed solutions for the classical problems PR faces. In what follows, I will briefly survey some of the main avenues that have been pursued by champions of PR and flag some issues that arise here. It may be worth noting that I am not trying to mount a conclusive case that PR can’t solve these problems. Rather, my main ambition here is to generate a feel for how champions of PR have ventured to deal especially with the new evil demon problem and the problem posed by clairvoyant cases. In particular, what the discussion below will bring to light is that although process reliabilists have been very creative in tweaking their view to deal with these problems, it remains unclear exactly why the proposed conditions that do the tweaking should hold in the first place.

Process Reliabilism 9 In this way, process reliabilist solutions to the classical problems face a motivational problem, even if they can get the cases right. And the reason I am interested in developing this motivational problem, in turn, is that it lays the groundwork for giving the view I will develop in the next chapter a chance to shine as it compares favourably with PR. As will become clear in due course, my view offers a rationale for the conditions that do the relevant work in the sense that these conditions drop out of an independently motivated general normative framework. If so, my view can improve on process reliabilism in that it steers clear of the motivational problem and offers a more unified and elegant solution to these problems. In addition, the process reliabilist proposals I am about to look into run into some difficulties when it comes to the fine details. Since my view can avoid these difficulties, there is reason to think that it compares favourably with these proposals on yet another count. However, to repeat, my main interest lies with the motivational problem. After all, this problem remains even if the process reliabilist proposals can be modified further to get the fine details right. With these preliminary points in play, let’s get down to business. I’ll go through the classical problems for process reliabilism in turn, starting once again with the new evil demon problem. The New Evil Demon Problem When dealing with the new evil demon problem, process reliabilists have to make a choice between two options: they can either go steadfast and/or concessive. Steadfast approaches hold that radically deceived agents do not satisfy the reliability condition on justified belief and, as a result, do not have justified beliefs. Instead, they venture to explain the appearance of justification away. In contrast, concessive approaches grant that radically deceived agents have justified beliefs and venture to modify the view to accommodate this claim. Since my own view falls squarely in the concessive camp and since going steadfast always threatens to leave a lingering worry that the solution isn’t fully satisfactory after all, I will focus here on concessive approaches. That said, some of the theoretical tools that process reliabilists have developed here can be put to use for a steadfast approach also. I’ll indicate how as I go along. One important type of solution that process reliabilists have developed is to index the notion of reliability required for justified belief to a set of privileged worlds. For instance, Goldman himself has defended views called ‘actual worlds reliabilism’ (1979) and ‘normal worlds reliabilism’ (1986) at various stages of his career. Actual world reliabilism takes the privileged world to be the actual world: whether a belief is justified depends on whether it is produced by a process that is reliable at the actual world. Normal worlds reliabilism takes the privileged worlds to be the set of normal worlds. Normal worlds are worlds consistent with our

10 Process Reliabilism general beliefs about the sorts of objects, events and changes that occur in the actual world. It is easy to see that both actual and normal worlds reliabilism promise to solve the new evil demon problem. After all, the perceptual processes that agents in radical sceptical scenarios use to arrive at perceptual beliefs are reliable at both at the actual world and at normal worlds. One key problem with this kind of approach is that it leads to trouble elsewhere, as Goldman also acknowledges at a later point. Here is a case of an alien cognisers that causes trouble for both actual and normal world reliabilism: Consider a possible non-normal world W, significantly different from ours. In W people commonly form beliefs by a process that has a very high truth ratio in W, but would not have a high truth ratio in normal worlds. Couldn’t the beliefs formed by the process in W count as justified? To be concrete, let the process be that of forming beliefs in accord with feelings of clairvoyance. Such a process presumably does not have a high truth ratio in the actual world; nor would it have a high truth ratio in normal worlds. But suppose W contains clairvoyance waves analogous to sound or light waves. By means of clairvoyance waves people in W accurately detect features of their environments just as we detect features of our environment by light and sound. Surely, the clairvoyance belief-forming processes of people in W can yield justified beliefs. (Goldman 1988: 62) The problem is, of course, that the clairvoyance process is justification conferring at W, even though it is reliable neither at the actual world nor at normal worlds. Even if actual and normal worlds reliabilism can get cases of radically deceived agents right, it falters on other cases, including the one of the alien cognisers. By the same token, there is reason to think that these approaches can’t be correct after all. A different kind of approach starts by giving an account of justification attributions. According to the perhaps most prominent incarnation of the approach (Goldman 1993), we generate ‘approved lists’ of processes that are justification conferring in accordance with whether we believe them to be reliable. Crucially, this list is then deployed in attributions about justification. To deal with the new evil demon cases, process reliabilists favouring this line may offer a contextualist or relativist semantics for attributions of justified belief according to which an attribution of a justified belief is true if and only if the belief is produced by a process that’s on the attributor’s/assessor’s list. Once again, this proposal promises to solve the new evil demon problem. After all, the justified beliefs of radically deceived agents are formed

Process Reliabilism 11 in just the same way as our justified belief. Since the processes that produce our justified beliefs are on the approved list, so are the processes that produce justified beliefs in radically deceived agents. At the same time, it is not hard to see that problems are again looming. First, suppose that a process like wishful thinking or hasty generalisation somehow makes it on someone’s list. The proposed view predicts that attributions of justified belief by this person will come out true for beliefs that are formed by wishful thinking. However, that doesn’t seem to be correct. Second, there is reason to think that, for one and the same process type, we can be brought to think that it is justification conferring and that it isn’t. To see this, note that in the above case of alien cognition, we judge clairvoyance to be a justification conferring process, while in standard clairvoyant cases, we judge it not to be justification conferring. On the present proposal, however, it is hard to see how this could be the case. After all, either clairvoyance is on the approved list or it isn’t.9 Clairvoyants As with the new evil demon problem earlier, process reliabilists have to choose between going steadfast or concessive on clairvoyant cases. Again, I will here focus on concessive approaches, for the reasons I already mentioned. First, one might venture to extend at least some of the solutions to the new evil demon problem to handle clairvoyant cases as well. Since I have already raised a number of issues for these approaches, I will not explore these routes in any more detail here. One distinctive approach to clairvoyant cases is in terms of an antidefeat condition. In a nutshell, the thought is that one has reason to believe that beliefs that simply pop up in one’s head, as it were, are unreliably formed. But if one has reason to believe that a given belief is unreliably formed, then one has a defeater for one’s belief. In this way, clairvoyants have defeaters for their clairvoyant beliefs. Since a belief is justified only if one doesn’t have defeaters for it, clairvoyant beliefs are correctly classified as not justified. Even if this line works as envisaged, one obvious problem is that it is not clear that it is available to process reliabilists. After all, what they need to do, ultimately, is unpack the notion of defeat in process reliabilist terms. And while we have seen that Goldman himself has made a proposal in this direction, his account has faced severe criticism (Kvanvig 2007; Beddor 2015). In fact, whether the notion of defeat can adequately cashed out in process reliabilist terms remains an open question. As a result, the present approach to clairvoyant cases is not safe home. What’s worse, there is reason to think that the defeat approach remains unsatisfactory in any case. If the idea of having reason to think that clairvoyant beliefs are unreliably formed can be cashed out in process

12 Process Reliabilism reliabilist terms at all, it will have to be in terms of higher-order processes by means of which we think about our first-order processes. However, it looks as though the relevant clairvoyant agents may be cognitively too unsophisticated to have such higher-order processes. For instance, it may be that the clairvoyant is a small child who hasn’t developed the relevant capacity for higher-order thought yet (Graham 2012). It is hard to deny that the anti-defeat approach carries little promise when it comes to cases like this one. By the same token, there is reason to believe that it does not succeed at the end of the day. Another line that process reliabilists have pursued to deal with clairvoyant cases is to place further conditions on the relevant belief-forming processes. In this vein, Jack Lyons has argued that non-inferentially justified beliefs must be produced by what he calls a primal system, which is inferentially opaque and developed as a result of learning and/or innate conditions (Lyons 2009: 166). The reason why, in standard clairvoyant cases, beliefs produced by clairvoyant cases are not justified is that the systems that produce them are not primal. After all, while they are inferentially opaque, they are the result of neither learning nor innate conditions. Rather, the relevant clairvoyants simply stumble into them. Unfortunately, there are problems even for Lyons’s account. Consider the following case, which is due to Harmen Ghijsen: Norbert is the son of a mother and father who both have reliable clairvoyant abilities and have been able to reproduce because of the benefit these clairvoyant abilities have provided for them. However, there are not (yet) many people who have these clairvoyant abilities, and their existence is kept secret. The abilities are due to specialized internal organs that pick up on special energy waves in the environment, and then output brief visual images which represent that such-and-so is currently the case at some distant place. Furthermore, the abilities usually become active quite suddenly some time after puberty. Norbert’s parents have decided not to tell him about the existence of his clairvoyant abilities, and Norbert has no evidence for or against their existence in general or his own having them. Some time after puberty, Norbert suddenly experiences a brief visual image of the President being in New York City and on that basis believes that the president is in New York City. (Ghijsen 2016: 121) Ghijsen points out that Norbert’s belief is unjustified, just as the original clairvoyant’s belief. At the same time, the system that produced this belief qualifies as a primal system in Lyons’s sense as it is developed as a result of innate conditions. What comes to light is that none of the process reliabilist attempts to deal with new evil demon and clairvoyant cases I have surveyed here are fully satisfactory. But perhaps this shortcoming can be remedied, either

Process Reliabilism 13 by tweaking some of the above proposals further or by a opting for different approach altogether. Even so, recall that the main purpose of my discussion was not to mount a conclusive case against PR, but rather to convey a feel for how process reliabilists have ventured to deal with these problems. In particular, the motivational problem I mentioned at the start of the section should by now have come into clear view. Suppose some combination of the above proposals can be made to work; say some version of a normal worlds reliabilism in combination with a Lyons-style approach to clairvoyant cases can effectively make the right predictions across all cases. The question remains why we should think that justification features these conditions in the first place. Now, I don’t mean to suggest that getting the cases right can’t be enough to motivate the relevant version of PR. All I am saying is any such version of PR will be at a disadvantage when compared to a view that not only gets the cases right but can also address the motivational issue. Since the view I will develop in the next chapter does just this, it compares favourably with PR, whether or not its champions can successfully tweak their view to get the cases right. The Generality Problem It will come as no surprise that process reliabilists have also worked hard to come up with a solution to the generality problem. Again, there are at least two prominent general strategies. The first tackles the generality problem head-on and attempts to specify a way of identifying the type of process at issue in process reliabilist accounts of justified belief. Prominent instances of this kind of approach construe the relevant process types as natural kinds (Alston 1995), as information-processing algorithms (Beebe 2004), or as determined by context (Heller 1995), among others. The other proposal tries to defuse the problem by arguing that it is a problem that every account of justified belief faces. For instance, Juan Comesaña (2006) shows that any adequate account of the basing relation will run into a version of the generality problem. And since every workable account of justified belief needs an adequate account of the basing relation, every workable account of justified belief will face a version of the generality problem. In a similar vein, Michael Bishop (2010) argues that any epistemological theory that wants to allow for a certain form of justification, i.e. reflective justification, will encounter the generality problem. Since every respectable epistemology will make room for reflective justification, once again, the generality problem is a problem that we all have to deal with. As will become clear in the next chapter, a version of the generality problem arises for my own version of virtue reliabilism as well. I will defend a defusing line, arguing that the generality problem is not only a problem that any account of justified belief encounters but also one that every account of competent performance encounters. In this way, the

14 Process Reliabilism virtue epistemological approach of seeing justified belief as an instance of a general theory of the normativity of performances offers a distinctive and attractive way of dealing with the generality problem, or so I will argue. This approach is not only compatible with but can also strengthen the case made by extant attempts to defuse the generality problem. For that reason, I will not discuss the ups and downs of the relevant proposals in any further detail here. My view is also compatible with a head-on approach to the issue, at least in principle. And while, as a result, the question as to whether a head-on approach can be made to work at all and if so how remains an open question, I will not get into it here. One reason for this is that if the approach I develop in the next chapter is correct, it may also serve to problematise the more head-on approaches to the generality problem, at least potentially. After all, if the generality problem generalises beyond the epistemic case to performances in general, then we may expect the right solution to the generality problem to generalise likewise. The potential trouble that extant head-on solutions to the generality problem encounter is that it is just not clear that these proposals generalise beyond the epistemic case in the way they would need to. Since settling these issues here would take me to far afield, I will set them aside for the time being.

Conclusion It comes to light, then, that PR’s account of justified belief has a number of virtues. It accommodates some key properties of justified belief, issues correct predictions in paradigm cases of justified and unjustified belief, and offers promising solutions to some of the most difficult problems in epistemology such as the regress problem and the problem of scepticism. At the same time, it also encounters a number of serious problems, including problematic cases, the generality problem, and the problem arising from assessing reliability of processes relative to worlds. Champions of PR have offered a number of ways to solve these problems, some of which are quite promising. Even so, I hope my discussion of the process reliabilist proposals has also made clear that there remains room for improvement: the view continues to struggle with the fine details, and it remains unclear exactly why we should think that the conditions that do the work should hold. Fortunately, the next chapter will develop a view that can deliver just these improvements.

Notes 1 PR has been defended extensively by Alvin Goldman (e.g. 1979, 1986, 2012). Other champions include Hilary Kornblith (e.g. 2002, 2004, 2009) and Jack Lyons (e.g. 2009). 2 It may be worth noting that the usefulness of the distinction between beliefdependent and belief-independent processes does not go undisputed. For

Process Reliabilism 15 instance, Lyons (2009) argues that the key distinction for process reliabilists is the distinction between basic and non-basic beliefs. While it might be thought that the distinction between belief-dependent and belief-independent processes unpacks just this distinction in process reliabilist terms, there is excellent reason to think that this cannot be the case after all. For instance, the resulting account will categorise introspective beliefs about other beliefs as non-basic, whereas they should come out as basic. Another way in which the distinction may be problematic is if the processes that produce perceptual beliefs turn out to have perceptual experiences with propositional contents as input. In this case, an argument parallel to the above argument should provide reason to think that the reliability condition for perceptual processes should be weakened also. Since perceptual experiences are not beliefs, Goldman’s distinction is problematic on yet another count. While these are important problems for process reliabilism, I will not take issue with them within the confines of this chapter. Rather, I will rest content with simply following Goldman’s exposition here. 3 Infinitism has been defended by Peter Klein (e.g. 1999, 2012, 2014), Scott Aikin (2005, 2008, 2011) and Jeanne Peijnenburg (e.g. 2007). See BonJour (1985) and Lehrer (1974, 1990) for prominent defences of coherentism. 4 This problem, which is also known as the new evil demon problem, was first stated in (Lehrer and Cohen 1983 and Cohen 1984) as a problem for reliabilist accounts of justified belief. Wedgwood (2002) argues that it generalises to all externalist accounts of justified belief. 5 Clairvoyant cases were first adduced by BonJour (1980, 1985) to argue against reliabilist accounts of justified belief. The related case of Mr. Truetemp is due to Lehrer (1990). 6 Conee and Feldman (1998: 2). It may be worth noting that if ‘tendency’ is given a propensity (rather than frequentist) interpretation, then it will be possible for token processes to be reliable and unreliable. Note, however, that ‘token versions’ of process reliabilism will face analogues of all the classical problems for ‘type versions’. To see this, note first that tokens of the wildest belief-forming processes can be reliable, e.g. when one has a helper in the wings who has the power to make the belief true and is determined (in a modally robust manner) to do so. Suppose, for instance, that I believe that I will be the world’s richest man based on wishful thinking. My belief is not justified. However, with a powerful and committed helper in the wings who is ready to (modally robustly) see to it that I become the world’s richest man, the token process that produced this belief may well be reliable. On the other hand, even tokens of normally excellent belief-forming processes can be unreliable, e.g. when one is up against an opponent who has the power to make the belief come out false and is determined (in a modally robust manner) to do so. Suppose, for instance, that I believe that I am looking at an apple based on normally highly reliable visual perceptual processes. My belief is justified. However, when up against a powerful and committed opponent, who on this particular occasion (modally robustly) made me look at an indistinguishable fake, the token process that produced this belief is unreliable. Finally, as Comesaña (2006) convincingly argues, token versions of process reliabilism also encounter a version of the generality problem. So, not much will be gained by abandoning the classical type version of process reliabilism in any case. 7 The generality problem was already noted in Goldman’s original 1979 paper. It was developed into a serious challenge for process reliabilism by Conee and Feldman (1998).

16 Process Reliabilism 8 This is clear especially from various responses to problems that sceptical cases pose for PR. Goldman himself considers ‘actual world’ and ‘normal worlds’ versions of PR according to which whether a belief is justified according to PR depends on whether the relevant process type is reliable at the actual world or at normal worlds (for more on this, see the Section entitled ‘Solutions’ below). Both of these solutions presuppose that the domain of evaluation are worlds of some denomination. 9 Note that this kind of approach can also be put to use is within the steadfast framework. Here, the idea is that we can use this account about justification attributions to explain how our intuitions go awry. While agents in sceptical scenarios really don’t have justified beliefs, we mistakenly think that they do. This is because the relevant perceptual processes that produce their beliefs are on the approved list, and as a result, we are misled into judging that these beliefs are justified. That said, at least the second problem outlined above arises for this kind of approach also.

2

Virtue Reliabilism Justified Belief

The Standard Framework The following two chapters will focus on virtue reliabilism’s (VR’s) accounts of justified belief and knowledge and whether they can make progress with respect to the problems of PR that we looked at in the last chapter. First things first, however, I will first introduce the standard virtue reliabilist framework, including the account of performance normativity its champions have appealed to and a number of ways of interpreting its core terms. Performance Normativity Recent statements of VR start from a general account of performance normativity. The core idea is that performances with a goal can be assessed along the following three dimensions: Success. Does the performance reach its goal? In other words, is it successful? Competence. Does the agent perform competently? In other words, is the performance produced by an ability to attain the performance’s success? Achievement. Does the performance qualify as an achievement? In other words, is it successful because competent? That is to say, besides being successful and competent, does the right kind of (explanatory) relation obtain between competence and success? By way of illustration, consider a simple version of target archery in which shots are taken at a disc-shaped target with no further structure, call it ARCH.1 The goal of ARCH is to hit the target. Now suppose you are an archer who takes a shot in ARCH. When you do, you produce a performance that has the goal of hitting the target. Once a performance’s goal is specified, we can easily see what it takes for the performance to be successful: the performance must attain its goal. In our toy case, your shot will be successful if and only if the arrow fired hits the target.

18 Virtue Reliabilism What matters to whether a performance is competent is whether it is the product of an exercise of an ability to produce the performance’s success. Agents may or may not have the ability to produce successful performances of a certain kind. For instance, you may or may not possess the ability to hit the target in ARCH. Moreover, a certain performance may or may not be produced by an exercise of that ability. For instance, you may or may not exercise your ability to hit the target when taking a shot. True, in many cases you will. However, you may be distracted, drunk or simply choose to miss. To perform competently, your performance must have been produced by an exercise of an ability to produce the performance’s success. In the case in which you are practicing ARCH, a competent performance is a shot that is produced by an ability to hit the target in ARCH. Achievements require success and ability to be related in the right way. The success must be attained because of the exercise of ability. Suppose you take a shot in ARCH. You are at peak concentration and in good practice, the orientation of bow and arrow are just right as are the conditions in which the shot is taken. You release the arrow, which goes on to hit the target right in the centre. In this case, your shot qualifies as an achievement: it is successful because competent. (I will henceforth also use Sosa’s expression ‘is apt’ as a shorthand for ‘qualifies as an achievement’.) In contrast, consider a case that is very much like the one just described with the exception that, unbeknownst to you, there is a magnet in the area that will cause shots that would normally hit the target to miss. At the same time, you are lucky enough to have me as a helper in the wings: I am operating a wind machine that counterbalances the influence of the magnet with the result that, when you take a shot, it hits the target just as it does in the first case. While, in this case, your shot is successful, it is not successful because competent. Rather, it is successful because I lent you a helping hand. Unlike in the first case, your shot is thus not apt. In consequence, while achievement entails both success and ability, the converse does not hold. Note also that performances can be successful without being competent and competent without being successful. To see the former, suppose you take a shot while severely intoxicated and blindfolded that, thanks to an incredible stroke of luck, hits the target nonetheless. Your shot will be successful without being competent. Concerning the latter, suppose you take a shot that would have hit the target, had your arrow not hit a bird that was flying by. In this case, the shot is competent but unsuccessful. Beliefs as Epistemic Performances In addition, VR assumes that beliefs are a type of performance with a goal. More specifically, they are a type of epistemic performance. This means that the general account of performance normativity can be

Virtue Reliabilism 19 applied to the special case of belief, delivering an account of the normativity of belief. A key question then concerns the nature of the goal of belief understood as a kind of epistemic performance. According to traditionalist versions of VR, the goal of belief is truth. Given that this is so, we get the following: Successful Belief. A belief is successful if and only if true. Competent Belief. A belief is competent if and only if it is produced by an exercise of an ability to form true beliefs. Apt Belief. A belief is apt if and only if it is true because competent. Crucially, champions of VR identify knowledge and justified belief with different normative properties of beliefs as epistemic performances. In particular, according to VR: VR-JB. One justifiably believes that p if and only if one competently believes that p. VR-K. One knows that p if and only if one aptly believes that p. Note that the resulting virtue reliabilism has a number of highly attractive features. First, it offers an attractive account of the normativity of belief. Second, it can offer a theoretical motivation for its accounts of justified belief and knowledge. After all, the accounts are backed by a general framework for the normativity of performances. Knowledge and justified belief are instances of familiar normative categories, to wit, that of apt and competent performance. Third, a number of plausible theses about knowledge and justified belief and their relations drop right out of the account. First, according to VR, knowledge is a kind of achievement. Second, since achievements entail both success and competence, knowledge entails both true belief and justified belief. In contrast, third, since success and competence do not entail aptness, justified true belief does not entail knowledge. Fourth, since success does not entail competence and vice versa, justified belief and true belief turn out to be logically independent. We can have true but unjustified beliefs and justified false beliefs. Of course, in order to give precise accounts of competent and apt belief, we need a more detailed account of the relevant abilities and of what it takes for a belief to be true because competent. Virtue reliabilists venture to derive the specific accounts of ability to form true belief and of a belief’s being true because competent from general accounts of ability to attain success and of a performance’s being successful because competent. Given that this is so and given that the main focus of this chapter are virtue reliabilist accounts of justified belief, in what follows I will take a closer

20 Virtue Reliabilism look at what virtue reliabilists have said about the nature of abilities, leaving the discussion of the because relation for the next chapter. Abilities It is widely agreed among champions of VR that abilities are agent dispositions.2 More specifically, they are agent dispositions to produce successful performances. For instance, in order to have the ability to hit the target in target archery, you must have the disposition to hit the target. Since abilities are dispositions, a closer look at dispositions may help illuminate the nature of abilities. First, it is plausible that dispositions are relative to conditions. Take the disposition of water to boil when heated to 100◦ C. While water has this disposition in certain conditions, i.e. at sea level, it does not have this disposition in other conditions. For instance, it doesn’t have it at altitudes below sea level. Second, dispositions have trigger conditions (T) and manifestation conditions (M). In the case of water’s disposition to boil when heated to 100◦ C, the trigger condition is heating to 100◦ C and the manifestation condition is boiling. Third, dispositions correspond to trigger-manifestation conditionals. To say that something, x, has the disposition to M when T is to say that were T to obtain in C, x would (likely enough) exhibit M. For instance, to say that water has the disposition to boil when heated to 100◦ C at sea level is to say that were water heated to 100◦ C at sea level, it would (likely enough) boil. If abilities are dispositions and dispositions have these properties, we may expect that abilities have these properties as well. And this is exactly what we find. First, abilities are relative to conditions. Your ability to hit the target in ARCH is relative to conditions. You do not have it when drunk beyond comprehension, when strong winds are blowing, etc. Why not? Because you don’t have the disposition to hit the target in those conditions. Second, abilities have trigger and manifestation conditions. According to Sosa, for instance, the trigger conditions for abilities are tryings and the manifestation conditions successes (Sosa 2015: 96). In the case of your ability to hit the target, the trigger conditions is trying to hit the target and the manifestation condition is hitting the target. Third, abilities correspond to trigger-manifestation conditionals. Your ability to hit the target in conditions C (sufficiently sober, not too strong winds, etc.) corresponds to the conditional: if you were to try to hit the target in C, you would (likely enough) succeed.3 It may be worth noting that the trigger-manifestation conditionals effectively impose a reliability condition on abilities. Abilities are, by nature, reliably connected to the relevant successes in the sense at is specified by the trigger-manifestation conditional.

Virtue Reliabilism 21 Improvements on PR Besides the fact that VR can rationalise PR’s key reliability condition on justified belief, the view can improve on PR in the sense that it avoids some of the problems PR encounters. In particular, as I am about to argue, there is reason to believe that it can avoid the problem of world-bound reliability and solve the new evil demon problem. The Problem of World-Bound Reliability Revisited To see how VR can solve the problem of world-bound reliability, note that what matters to justification according to VR is not the reliability of a process type at a set of worlds, but the reliability of an agent in a particular set of conditions. In the case in which you are one of a few inhabitants of a remote island on which there are no jack-o’-lanterns, your belief that the mushroom you are holding is edible will be justified according to VR. After all, you have a highly reliable disposition to form true beliefs about edible mushrooms in your conditions. Were you to form a belief that a mushroom with a certain look is edible in your conditions, your belief would very likely be true. In the converse case in which jack-o’lanterns exist nowhere except on the island you inhabit, your belief is unjustified. After all, you do not have a reliable disposition to produce true beliefs about edible mushrooms in your conditions. It is not the case that were you to form a belief that a mushroom with a certain look is edible in your conditions, your belief would very likely be true. Agent and condition relativity appear to give VR an important advantage here. We thus have some reason to believe that VR is on the right track. The New Evil Demon Problem Another front on which VR may hope to make progress is the new evil demon problem. At the very least it can do so given the additional assumption that one can exercise an ability even when one is not in suitable conditions. Fortunately, this assumption is independently plausible. Suppose you take a shot in ARCH. While on its way to the target, an earthquake strikes and your shot remains unsuccessful. Here you are not in conditions suitable for your ability to hit the target. Even so, it would seem that you did exercise your ability to hit the target and that you produced a competent shot. Given that this is so, champions of VR are well positioned to offer a promising solution to the new evil demon problem. What happens when you fall in the hands of the mad scientist is that you are moved to unsuitable conditions for your epistemic abilities. Just as it is plausible that moving you to unsuitable conditions does not prevent you from exercising your ability to hit the target in the archery case, here too it is plausible that this move does not prevent

22 Virtue Reliabilism you from exercising your epistemic abilities. If so, you continue forming beliefs via the same epistemic abilities as before. In this way, your beliefs continue to be competent. The new evil demon problem can be solved.

Remaining Problems Clairvoyants and the Generality Problem When I first introduced PR, I mentioned four problems the view encounters. We have now seen that VR compares favourably with PR in that it promises to avoid at least two of these problems. What about the other ones? Here it is much less clear that VR can do the trick. Concerning the problem clairvoyant cases pose for PR, it is easy to see that it arises for VR just as well. After all, in Clairvoyant, once you have been exposed to radiation, you not only acquire a belief-forming process that delivers true beliefs about distant places reliably, but you also acquire a very reliable agent disposition to form such true beliefs. As a result, there is reason to believe that VR, just like PR before, also predicts that the beliefs of agents in Clairvoyant and similar cases come out justified. The problem still stands. The same goes for the generality problem. Recall that PR must tell a story as to which of the indefinite number of process types instantiated by token processes are the ones at issue in PR. For instance, the token process that produced your belief that you are currently reading this book instantiates a variety of different process types, including cognitive process, belief-forming process, perceptual belief-forming process, visual belief-forming process, etc. Now, it might be thought that since VR abandons process types in favour of agent abilities, the generality problem does not arise for VR. On second thought, however, this is too quick. After all, agent abilities are taken to be agent dispositions and dispositions are relative to conditions. But now notice that there are indefinitely many ways of individuating the conditions to which a given disposition is relative. For instance, in the case of your belief that you are reading this book, the conditions may or may not include that it is a Thursday, that the object seen is obstructed by a solid object, etc. It is easy to see that whether or not you have the disposition to form true beliefs about whether you are reading a book will vary depending on how these conditions are individuated. Unless it is specified how the conditions to which VR’s key dispositions are relative are individuated, it would seem that VR simply does not make determinate predictions about whether or not individual beliefs are competent and hence justified. In order to work as an account of justified belief at all, VR must be supplemented with an account that specifies how these conditions are to be individuated. In this way, a version of generality problem also arises for VR.

Virtue Reliabilism 23 Abilities without Agent Dispositions The clairvoyant and generality problems arise equally for both PR and VR. As a result, while these are genuine problems for VR, the view does at least not lose ground compared to PR. Unfortunately, there is at least one problem for VR that does not equally endanger PR. To see how it arises, notice that an agent may have different ways of producing performances at his disposal. To keep things simple, let’s say he has two such ways. For instance, you may produce shots in ARCH with your left hand or with your right hand. Now, it may well be that a given agent who has two ways of producing performances of a certain kind at his disposal is disposed to perform successfully when producing performances of that kind in a certain set of conditions in one way but not the other. Call the former ‘the good way’ and the latter ‘the bad way’. You may be disposed to hit the target when shooting with your right hand, but not when shooting with your left hand. Suppose, finally, that the agent is disposed to produce performances of said kind in the bad way. His dominant way of producing performances is the bad way. You took a sacred vow never to shoot with your right again. However, you still love practising archery. That’s why you have taken to shooting with your left hand. In this situation, there is a clear sense in which the agent is not disposed to perform successfully. According to the standard virtue reliabilist account of abilities, the agent does not have the relevant ability. For instance, you do not have the ability to hit the target in ARCH. After all, since you sacredly vowed never to shoot with your right again, if you were to produce a shot, you would produce a shot with your left. Since there are no conditions relative to which producing a shot with your left makes success highly likely, the trigger-manifestation conditional that corresponds to the ability to hit the target is false of you no matter what conditions you may be in. As a result, you do not have the ability to hit the target. Crucially, this appears to be the wrong result. The mere fact that an agent is disposed to produce performances of a certain kind in a bad way does not entail that the agent no longer has the ability to produce successful performances of that kind, at least not so long as he still has the good way at his disposal. A vow never to shoot with your right again does not make your ability to hit the target when shooting with your right disappear. Given that this is so, it may look as though something is amiss with the standard virtue reliabilist account of abilities as agent dispositions.4

A New Framework According to VR, justified belief is competent belief, where the notion of competent belief is unpacked in terms of the notion of the exercise

24 Virtue Reliabilism of an epistemic ability, that is, an ability to form true beliefs. We have seen that VR compares favourably with its main reliabilist competitor, PR, in a number of respects. VR embeds the reliabilist account of belief in a general theory of performance normativity, thereby rationalising the key reliabilist idea. It also avoids a couple of important problems that PR encounters, to wit, the problem of world bound reliability and the new evil demon problem. At the same time, we also saw that a couple of PR’s problems could not be immediately resolved by opting for VR and that VR scores worse than PR on at least one count. In what follows, I will develop a novel version of VR that improves both on PR and on extant versions of VR. More specifically, I will develop a detailed account of a kind of competent performance that takes on board insights from both sides. I will then apply this general account to the case of belief and derive a detailed virtue reliabilist account of justified belief. Finally, I will argue that this account can either solve or defuse the remaining problems of process reliabilism I introduced in the last chapter.5

Simple Goal-Directed Practices Let’s start with a framework for simple goal-directed practices (henceforth ‘SGPs’). For a practice to be goal-directed is for it to have a success condition, a condition under which the practice’s goal is attained. One very simple kind of goal-directed practice involves two types of particular, targets and moves, and a designated property. The success condition of this kind of practice can be defined as obtaining if and only if a move has the designated property. In ARCH, for instance, the target is a disc with a set surface area, moves are shots taken from a set distance, and the designated property is the property of hitting the target. A success in ARCH is a shot that hits the target. Practitioners of SGPs are move-producers. They may attain success in a given SGP. They do so if and only if they produce a move that has the designated property. Practitioners of ARCH are shot-producers. A practitioner of ARCH attains a success in ARCH if and only if he produces a shot that hits the target.

Abilities Practitioners of SGPs may have the ability to attain success in a given SGP. You, the reigning world champion in ARCH, have the ability to hit the target, while I, a blind man, don’t. Let’s take a closer look at these abilities (henceforth ‘SGP abilities’). I agree with the general account of abilities offered by extant versions of VR that abilities, including SGP abilities, involve dispositions and that,

Virtue Reliabilism 25 as a result, they are relative to conditions, have trigger and manifestation conditions, and correspond to trigger-manifestation conditionals. Crucially, I want to suggest that what the problem of abilities without agent dispositions from the last section shows is that the general account of abilities that champions of VR have offered is too coarsegrained. What we need is something more fine-grained. My proposal is to relativise abilities to ways of producing performances, or, for SGP abilities, to ways of move production. In this way, my account of abilities differs from the ones provided by other champions of VR. Once we relativise SGP abilities in this way, we can allow that some ways of producing performances constitute abilities to succeed, while others don’t. As a result, agents may have abilities to attain a certain kind of success, while not being disposed to exercise them. In fact, they may be disposed to use ways of move production that do not qualify as abilities. In that case, agents will not be disposed to attain the relevant kind of success, even though they do have the ability to do so. In the toy case I used to illustrate the problem, this is exactly the situation you find yourself in after you have taken the vow never to shoot with your right again. Here you have the ability to hit the target: you have a way of producing shots, i.e. with your right, that disposes you to produce successful shots at least in suitable conditions. As a result, this way of shooting constitutes an ability to hit the target. However, thanks to your vow, you are no longer disposed to exercise this ability. Since you still love practising archery, you have taken to shooting with your left. Unfortunately, you are a terrible shot with your left: the way in which you now produce shots does not dispose you to hit the target, no matter what conditions you may find yourself in. You are in a situation in which you have the ability to hit the target, i.e. when shooting with your right. At the same time, thanks to your vow and love of archery, you are no longer disposed to hit the target when trying.6 Since SGP abilities are relative to ways of move production, so are the dispositions at issue in them. And, of course, the same applies to the conditions to which the dispositions are relative. Accordingly, my suggestion is that a practitioner’s way of move production, W, qualifies as an ability to attain success in a given SGP, S, relative to conditions C only if he is disposed to attain S’s success when using W in C. A given way of shooting you may have will qualify as an ability to hit the target in ARCH relative to C only if you are disposed to hit the target when using it in C. Not every way of move production that disposes one to attain success in a certain SGP and in certain conditions qualifies as an SGP ability. Suppose I am an extravagant archer. The only occasions on which I take a shot at all is when I see a sculpture by my favourite sculptor. When I do, I fire it right up in the air. As it so happens, the only remaining sculpture is located at a shooting range that is manipulated by an army of clandestine helpers who will see to it that all and only shots fired right up in the air

26 Virtue Reliabilism hit the target. Currently, I am at this very shooting range. I have noticed the sculpture and have fired arrows straight up into the air. I don’t stay in order to verify whether my shots actually hit the target. Why should I? There is no reason for this. However, my shots do hit the target. In fact, in this case I am strongly disposed to hit the target when using my way of shooting in the conditions that obtain at the range where I take these shots. Does my way of shooting qualify as an SGP ability, if only relative to those conditions? I take it to be clear that the answer to this question is no. Following Ruth Millikan (2000: 61) I would like to suggest that there is a distinction between mere dispositions and genuine abilities. Not all dispositions to attain success qualify as abilities. My disposition to hit the target in the above case is a case in point. As a result, the above picture of SGP abilities as ways of move production that dispose agents to attain success in certain conditions needs refinement. What makes the difference between a genuine ability and a mere disposition? This question has rarely if ever been taken up by champions of VR. In order to supply this gap, I’d like to take another leaf from Millikan (2000: ch. 4): it is etiology that matters. Here is Millikan’s proposal: In general, the conditions under which any ability will manifest itself are the conditions under which it was historically designed as an ability. These are conditions in which it was learned . . . They are conditions necessary to completing the mechanisms by which past successes were reached by the systems or programs responsible for the abilities. (Millikan 2000: 61) Here is a natural way of connecting Millikan’s proposal with the picture of SGP abilities sketched above: to qualify as an SGP ability, a way of move production must have led one to successful SGP moves in the conditions in which it was acquired by learning/selected for. If so, since my extravagant way of shooting arrows up in the air did not lead me to hits in the conditions in which I acquired it, it does not qualify as an ability. Unfortunately, however, this way of connecting Millikan’s proposal with the above picture of SGP abilities encounters a problem. As Millikan also acknowledges, abilities can be acquired by coming to know that a way of producing moves works, i.e. by coming to know that a way of move production leads to success in certain conditions (Millikan 2000: 64). The problem is that when one acquires an ability in this way, one need not acquire a new way of move production. Rather, one may simply learn something about an existing way of move production. For instance, suppose that, in the above case, I am told that, at the relevant shooting range, I will hit the target if and only if I fire my shots straight up in the air. I now possess the ability to hit the target at that shooting range. At the same time, I have not acquired a new way of move production.

Virtue Reliabilism 27 Rather, I learned something about an existing way of move production. The problem is that the way of move production at issue in my SGP ability never led me to success in the conditions in which it was acquired by learning. Here is a way of addressing this problem. In certain cases, agents acquire an SGP ability by acquiring a new way of move production. In the learning process, the way of move production gets its shape. The process involves interaction with the environment during which the way of move production becomes “tuned” (Millikan 2000: 63) to producing successful SGP moves in the conditions that obtain during the learning process. This is what happened during your training period, after you had first taken up archery. You practised shooting, adjusting your way of performing in the light of past successes, tips from your coach and so on. You underwent a learning process that shaped your way of shooting with the result that using this way of shooting now disposes you to produce hits in the conditions of learning. In this way, the learning process tuned your way of shooting to certain conditions. As a result, you now have the ability to hit targets in ARCH in those conditions. Let us say that, in this kind of case, the agent’s way of move production is ‘metaphysically grounded’. In contrast, in the case in which I am told that shooting arrows up in the air will lead to success at the relevant shooting range, I do not acquire a new way of shooting. Rather, I learn something about an existing way of shooting. That said, there is a clear sense in which my new knowledge has the same effect here as your training does. It also tunes my way of move production to producing successful SGP moves in certain conditions. Crucially, the tuning is of a rather different kind. In contrast with the metaphysical kind of tuning we find in your case, here the tuning is epistemic in nature. Accordingly, let us say that, in this kind of case, the agent’s way of move production is ‘epistemically grounded’. I want to suggest that in order to qualify as an SGP ability, a way of move production must have been tuned to some conditions, thereby grounding the way of move production, be it metaphysically or epistemically. This grounding condition on abilities enables my account to distinguish between genuine abilities and mere behavioural dispositions.7 I also want to allow that grounded ways of move production can qualify as SGP abilities for SGPs and conditions to which they have not been tuned. Suppose ARCH is practised only in strongly controlled conditions: indoors, under very specific artificial lighting conditions, while completely sober, etc. (henceforth C). Suppose you, the reigning world-champion of ARCH, are trained in C with the result that your way of shooting is tuned to C. Even so, it is hard to deny that you may have the ability to hit targets in ARCH in different conditions (e.g. outdoors, under different lighting conditions, after a beer = C0 ), at least so long as your way of shooting continues to dispose you to hit targets in ARCH in those conditions. Similarly, even if your ability is tuned specifically to ARCH, there can be no

28 Virtue Reliabilism question that you may have the ability to hit targets in certain other SGPs (e.g. in ARCH0 which is just like ARCH except that the target is a square rather than a disc), at least so long as your way of shooting continues to dispose you to hit targets in those SGPs. SGP abilities are thus relative to a range of SGPs and conditions that may differ from the SGP and conditions to which the underlying way of move production had been tuned. What matters is that the way of move production continues to dispose the agent to produce successful moves in the relevant SGPs and/or conditions. Finally, the SGPs and conditions to which an ability is tuned and the SGPs and conditions relative to which an agent may have an SGP ability may vary from one way of move production to another. One of your ways of move production—shooting with your left hand (W1 )—may be tuned to ARCH0 , slight intoxication and natural light, while another—shooting with your right (W2 )—is tuned to ARCH, sobriety and artificial light. It may also be that W2 disposes you to produce successful moves not only in ARCH but also in ARCH0 , not only in artificial light but also in natural light but only when entirely sober. In contrast, W1 may work only for ARCH0 and only in natural light, no matter whether slightly intoxicated or entirely sober. Further extensions and other combinations are of course possible. With these points in play, I would like to propose the following general account of SGP abilities: SGP Ability, General. One has an ability to attain success in a range, RS , of SGPs and relative to a range, RC , of conditions if and only if one has a grounded way of move production, W, such that, for any S ∈ RS , there is some C ∈ RC such that using W in C disposes one to attain success in S, and for any C ∈ RC , there is some S ∈ RS such that using W in C disposes one to attain success in S.8 While this offers a fully general account of SGP abilities, it has the disadvantage of being rather complex. At the same time, for present purposes, I rarely need the account in its full generality. For that reason, I will be working with the following slightly simplified version of the account: SGP Ability. One has an ability to attain success in a range, R, of SGPs and relative to conditions, C, if and only if one has a grounded way of move production, W, such that, for any S ∈ R, using W in C disposes one to attain success in S. Exercises of Abilities What does it take to exercise an ability? While virtue reliabilists have developed detailed accounts of ability, they have rarely stopped to consider this question. I would like to supply this lack. In particular, I want to suggest that exercises of SGP abilities are uses of ways of move production involved in SGP abilities. Or, more precisely,

Virtue Reliabilism 29 SGP Exercise. One exercises an ability, A, to attain success for a range, R, of SGPs and relative to conditions, C, if and only if one has A and produces a move via the way of move production at issue in A. It is important to note that placing the agent in conditions relative to which he does not have an SGP ability can have different effects on an agent’s performances. Some such conditions will result in preventing an agent from using his way of move production. For instance, being too drunk, distracted, nervous, shoved while releasing the arrow and so on will prevent you from using the way of shooting that qualifies as an SGP ability relative to some (albeit different) conditions. I will henceforth refer to conditions that, when not satisfied, prevent the agent from using his way of move production as conditions of shape (SH). According to SGP Exercise, then, exercising an SGP ability requires that SH be satisfied. In contrast, other such conditions do not prevent the agent from using his way of move production when not satisfied and so allow him to exercise his ability anyway. Suppose, for instance, that you fire a shot that would have hit the target had it not been for a jokester who destroys the target when the arrow is about to hit it. Even though your shot misses the target, you do get to produce a move via the way of shooting that qualifies as an SGP ability relative to some (albeit different) conditions. I will henceforth refer to conditions that, when not satisfied, do not prevent the agent from using his way of move production as situational conditions (SI). According to SGP Exercise exercising an SGP ability does not require that SI be satisfied.9 Competent Moves Competent moves in an SGP require the exercise of an SGP ability. When producing a shot in ARCH, your shot will be competent only if it is produced by an ability to hit the target. To the best of my knowledge, extant versions of VR don’t offer further conditions on competent performance, besides being produced by an ability to attain the relevant success. On reflection, there is reason to think that further conditions should be countenanced. A competent move requires more than the exercise of an SGP ability. To see this, let’s return to the case in which you are the reigning world champion of ARCH. Suppose that you are currently engaging in ARCHX in which the target changes its position discontinuously, randomly, and rapidly. Let’s assume, as is plausible anyway, that you do not have the ability to hit the target in ARCHX . You have no grounded way of shooting that disposes you to produce successful moves in ARCHX , no matter what conditions we may place you in. Suppose you take a shot using a grounded way of move production that disposes you to hit the target in a range R of SGPs and relative to conditions C. Here you exercise your SGP ability to hit the targets in range R and relative to C. However, that does not make your

30 Virtue Reliabilism shot competent. The ability you exercise is the wrong ability for the SGP you are engaging in. For a move to be competent, it must be a move in an SGP, S, that is within the range R for which your way of move production qualifies as an ability.10 Contrast the situation described above with one in which you engage in an SGP that, we may assume, is within the range, R, of your SGP ability, but in which a jokester prevents the shot from being successful. Here you not only exercise an ability to hit the target, your shot is also competent. There is thus reason to believe that, in order to produce a competent move in a given SGP, the SGP must be within the range (of SGPs) of the SGP ability exercised. The above considerations thus motivate the following account of competent moves: Competent SGP Moves. A move in a given SGP, S, is competent if and only if it is produced by an exercise of an SGP ability to attain success in a range, R, of SGPs and relative to conditions, C, such that S ∈ R. A Different Virtue Reliabilism The View With the general account of competent moves in SGPs in play, I will now move on to the application to the case of belief. To begin with, I would like to suggest that a relevant fragment of a key epistemic practice—viz. inquiry into specific whether questions (henceforth simply ‘inquiry’)—can be understood as an SGP. Or, to be more precise, it can be understood as a collection of SGPs, one for each question. More specifically, my suggestion is that the targets of inquiry are true answers. For instance, the target of an inquiry into whether p is the true member of the set including the proposition that p and the proposition that not-p. Moves in inquiry are beliefs. For instance, believing p constitutes a move in an inquiry into whether p, as does believing not-p. The designated property in inquiry is truth.11 For instance, a belief that p has the designated property if and only if p is true. It is easy to see that this gives us the standard virtue reliabilist account of successful belief according to which a belief is successful if and only if true. With the account of successful belief in play, we can now apply the above accounts of SGP abilities, their exercises and competent moves to the case of belief. This gives us: Epistemic Ability. One has an ability to form true beliefs about propositions in a range, R, and relative to conditions, C, if and only if one has a grounded way of belief formation, W, such that, for any p ∈ R, using W in C disposes one to form true beliefs that p. Exercises of Epistemic Abilities. One exercises an ability, A, to form true beliefs about propositions in range R and relative to conditions C

Virtue Reliabilism 31 if and only if one has A and forms a belief via the way of belief formation at issue in A. Competent Belief. One competently believes that p if and only if one’s belief that p is formed by an exercise of an ability to form true beliefs about propositions in range R and relative to conditions C such that p ∈ R. Recall that, according to extant virtue reliabilist accounts of justified belief, a belief is justified if and only if competent. I would like to adopt this account of justified belief as competent belief. In conjunction with Competent Belief, it entails the following detailed account of justified belief: VR-JB*. One justifiably believes that p if and only if competently believes that p. It is important to keep in mind that, while VR-JB* does not explicitly feature a reliable process condition, it is implicitly present in it. After all, the notions of ability and their exercise at issue in the right-hand side of VR-JB* are analysed in terms of Exercise of Epistemic Ability and Epistemic Ability. And, according to Epistemic Ability, epistemic abilities require reliable ways of belief formation. Since VR-JB* does feature a reliable process condition on justified belief, the question as to whether VR-JB* succumbs to the classical problems of PR remains. This question will be taken up in the remainder of this chapter. The Problems for Process Reliabilism Revisited Recall that PR encounters at least four problems: the new evil demon problem, the problem of clairvoyant cases, the generality problem and the problem of world-bound reliability. I have already shown how virtue reliabilism can avoid the problem of world-bound reliability and the new evil demon problem. Since the present version of virtue reliabilism has little to add on this front, I will not return to these problems. Rather, I will restrict my focus on the remaining two problems for PR, starting with the problem of clairvoyant cases. Clairvoyant Cases Let’s move on to clairvoyant cases. Recall that, in our toy case, Clairvoyant, you come to be gifted with a ‘clairvoyance’ belief-forming process that produces true beliefs about distant events with a high degree of reliability. Even so, the beliefs you go on to form via this clairvoyance process are intuitively not justified. The key to VR-JB*’s account of clairvoyant cases is the grounding condition on abilities. In particular, I want to suggest that even though your

32 Virtue Reliabilism clairvoyant process produces true beliefs with a high degree of reliability, it is not grounded. After all, it is not the case that you underwent a learning process involving interaction with the environment during which your clairvoyant process was tuned to producing true belief in the conditions that obtained during the learning process. This means that your clairvoyant process is not metaphysically grounded. Moreover, you also do not have any evidence that your clairvoyant process works and so it is not epistemically grounded either. Since your clairvoyant process is neither metaphysically nor epistemically grounded, it does not satisfy the grounding requirement on abilities. In consequence, it does not qualify as an ability to form true beliefs about any range of propositions. Since you cannot exercise an ability you do not have, you do not form your clairvoyance beliefs via the exercise of an ability to form true beliefs. By VR-JB*, your clairvoyance beliefs are not justified.12

The Generality Problem Recall that the generality problem for PR arises from the fact that every token process instantiates indefinitely many types. Many of these process types have different degrees of reliability, some differ rather dramatically in the degree of reliability they achieve. Unless it is specified which process types exactly are the ones at issue in PR, PR simply does not make determinate predictions about whether or not individual beliefs are justified. I’d like to flag that I do not mean to offer a head-on solution to the generality problem here. Rather, my strategy will be to try and defuse it. A large step in the direction of defusing the generality problem has already been made in Comesaña (2006) and Bishop (2010). Comesaña and Bishop both argue, convincingly to my mind, that the generality problem is a problem that affects not just process reliabilism but all theories of justified belief. If so, the generality problem is a general epistemological problem, rather than a problem for process reliabilism in particular. In addition, I will now argue that the generality problem is a problem in the general theory of performance normativity as it is a problem that arises for competent SGP moves in general. This is good news especially for VR-JB*. After all, if the generality problem arises already in the general theory of performance normativity, it will not be a specifically epistemological problem. As far as VR-JB* is concerned, then, the generality problem is not only a problem that every theory of justified belief encounters but also a problem that finds its proper home in the general theory of performance normativity. What’s more, I will provide reason to believe that, as a problem in the general theory of performance normativity, it is not particularly worrisome in the first place.

Virtue Reliabilism 33 To begin with, let’s take a look at how the generality problem arises for VR-JB*’s account of justified belief as competent belief: According to VR-JB* competent beliefs require exercises of abilities to form true beliefs and exercises of abilities to form true beliefs require uses of ways of belief formation. More specifically, they require uses of tokens of types of ways of belief formation. But, of course, each such token instantiates indefinitely many types. For instance, my belief that I am sitting at my desk is produced by a token way of belief formation that instantiates the types cognitive process, belief-forming process, etc. To make determinate predictions about whether a given belief is competent, we need a specification of which of these types is the one at issue in VR-JB*’s account of competent belief. This is the generality problem as it arises for VR-JB*. It is now fairly easy to see that the generality problem arises for competent SGP moves in general. To see this notice that competent SGP moves in general require exercises of abilities and exercises of abilities require uses of ways of move production. More specifically, they require uses of tokens of types of ways of move production. But, of course, each such token instantiates indefinitely many types. Consider, once more, a case in which you take a shot at a target in ARCH. Your token way of shooting instantiates all of the following types: shooting, shooting with a bow, shooting with arrows of make M, etc. To make determinate predictions about whether a given shot is competent, then, we need a specification of which of these types is the one at issue in the general account of competent performance. In this way, it comes to light that the generality problem arises for competent performances in general. The generality problem, insofar as it arises for VR-JB*, is a problem in the general theory of performance normativity rather than a problem that affects VR-JB*’s account of justified belief in particular. Finally, the generality problem as it arises for competent performances in general is not particularly worrisome. To see this, notice first that we typically have no special difficulties in discerning agents who have certain abilities and agents who do not have them. For instance, it’s not difficult to distinguish agents who have the ability to hit the target in ARCH from those who don’t. The same goes, once again, for the epistemic case. It is not too difficult to find out who has the ability to recognise apples, BMWs or Picassos and who doesn’t. Crucially, we can do so without being able to offer a precise account of how the process at issue in the ability is typed. For instance, we can discover that you have the ability to hit the target in ARCH without being able to pinpoint the exact process type at issue in your ability. And the same goes for your ability to recognise apples. As a result, VR-JB* can defuse the generality problem. According to VR-JB, justified belief does require reliable belief-forming processes of

34 Virtue Reliabilism sorts. However, it does so because justified beliefs are beliefs produced by epistemic abilities and epistemic abilities involve reliable belief-forming processes. Since we can identify abilities, including epistemic ones, without being able offer a general account of how the process at issue in the ability is typed, we do not need a solution to the generality problem in order to have a workable version of VR-JB.

Conclusion This chapter has taken a closer look at virtue reliabilist accounts of justified belief. VR derives its accounts of both justified belief and knowledge from an application of a general account of the normativity of performances with an aim. According to this account, any performance with an aim can be evaluated in terms of success, competence and achievement. Champions of VR argue that beliefs are performances with an aim. More specifically, beliefs are performances that aim at truth. VR identifies justified belief with competent belief. Standard versions of virtue reliabilism take abilities to be reliable agent dispositions. As a result, VR’s account of justified belief can be seen as a close rival to process reliabilism. Unsurprisingly, VR can retain the benefits of process reliabilism. At the same time, I also argued that already standard versions of VR improve on PR in a number of ways. Some central problems that process reliabilism encounters can be solved by VR, including the new evil demon problem and the problem that arises from evaluating reliability of processes relative to worlds. However, not all problems that process reliabilism encounters could be addressed in this way. In particular, we saw that the problem posed by clairvoyant cases and the generality problem remain unsolved. Moreover, standard versions of VR have some problems of their own: agent dispositions appear to be too coarse-grained for a workable account of abilities. In view of this, this chapter has introduced a novel version of VR. This version makes significant concessions in the direction of its process reliabilist competitor in that it grants that abilities in general are relative to ways of performing. As a result, the kinds of epistemic ability required for justified belief are relative to ways of belief formation. In addition, the account improves on standard versions of VR in that it offers detailed accounts not only of abilities but also of the exercise of ability and of competent performance. Moreover, I have argued for a grounding condition on abilities, which eventually serves to resolve the problem of clairvoyant cases, and have shown that this version of VR can defuse the generality problem. In this way, this chapter has developed a virtue reliabilist account of justified belief that carries considerable promise. The next chapter will move on to virtue reliabilist accounts of knowledge.

Virtue Reliabilism 35

Notes 1 I take this example from Sosa (e.g. 2007, 2015). 2 See e.g. Sosa (2010), Greco (2010), Kelp (2011b, 2013), Pritchard (2012a), and Turri (2016). 3 My presentation of VR’s account of abilities follows (Sosa 2015). Other champions of VR have offered slightly different accounts. Greco, for instance, holds that to have an ability to attain S in certain conditions, C, one must be such that one attains S with a high-rate success across nearby possible worlds at which C obtain (Greco 2010: 77). However, these differences are of little consequence for present purposes. 4 Could a champion of VR maintain that the disposition is simply masked by the vow here? Perhaps. That said, there is reason to think that masking will not solve all the problems for VR. For instance, it looks as though it is possible to have two abilities to attain a certain kind of success. For instance, you may have the ability to hit the target with your left and with your right. At the same time, it would seem that one ability may be stronger than the other. Even though you have the ability to hit the target with your left and with your right, you are a better shot with your right. Now, strength of ability very plausibly supervenes on the strength of the trigger-manifestation conditional associated with them, at least provided that all else is equal. In the archery case, all else equal, your ability to hit the target with your right is stronger if the probability of hitting the target conditional on taking a shot with your right is higher than with your left. It is not clear how this could be if abilities were agent dispositions such that trigger conditions are tryings. After all, you will be trying whether you are taking a shot with your right or with your left. On the standard virtue reliabilist picture, then, you can have at most one ability with one degree of strength. How strong the ability will be will depend at least in part on how you are disposed to try. For instance, in the aforementioned case, you have only one ability to hit the target and how strong it is will depend on whether you are disposed to take shots with your left and your right. Again, this seems to be the wrong result. What is going on here is that you have two abilities to hit the target, one with your left and another with your right, such that one is stronger than the other. That’s how you differ from a less skilled archer who only has one ability—with his left, say—but is as likely to hit the target upon trying as you are once your dispositions to take shots with, respectively, your left and your right are factored in. 5 I’d like to emphasise that I am indebted to extant work on VR in the literature. In particular, the account I am about to develop builds very much on Sosa’s work on the issue. That said, my account also departs from and adds further detail to other virtue reliabilist views in the literature, including Sosa’s. 6 We can also allow that one agent has two abilities to attain a certain kind of success in virtue of having two ways of producing performances that dispose them to do so. Moreover, one of these abilities may well be stronger than the other for instance, in virtue of the associated trigger-manifestation conditional being stronger. Again, this is exactly the situation you find yourself in when, even though you are disposed to hit the target no matter whether you take a shot with your left or your right, you are more likely to hit it when taking a shot with your right than with your left (see en.18 above). 7 It may be worth pointing out that, contrary to what Millikan suggests, these conditions need not be the condition in which the ability was acquired by learning. To see this, suppose I, the extravagant archer, am not at the shooting range when I am told that shooting arrows up in the air will produce successful shots there. I have now acquired the ability to hit the target at that range.

36 Virtue Reliabilism Evidently, in this case, the way of move production does not dispose me to produce successful moves in the conditions in which it was acquired (i.e. my present conditions). However, my way of move production disposes me to produce successful moves in the conditions for which it was acquired (i.e. the relevant shooting range). 8 One might wonder whether it is really the case that all abilities must be grounded in this way. After all, couldn’t there be innate abilities that are genuine abilities even though they don’t have a history of success? There are a number of responses available to champions of this view. First, what is on offer here is an account of a specific type of ability, to wit, SGP ability. Accordingly, it is compatible with this view that some abilities are not grounded. Second, those who think that the grounding requirement on abilities generalises and that, as a result, there are no innate abilities can use the distinction between dispositions to succeed and abilities to explain how we might be misled to believe that there are innate abilities. What does exist are innate dispositions to succeed. Those who think that there are innate abilities mistake innate dispositions to succeed for innate abilities. 9 The distinction between conditions of shape and situation can also be found in Sosa’s work and Greco countenances a similar distinction. However, both Sosa and Greco introduce the distinction by paradigmatic examples rather than offering a more explicit characterisation like the one above. 10 An even clearer example may be the following: suppose you have a grounded way W of producing layups in basketball that qualifies as an ability to score relative to some C. Currently, you are standing at the midcourt line and have two seconds to score a basket to win the game. Suppose you produce a shot via W, which, of course, doesn’t even get close to the basket. By the relevant instance of SGP Exercise, you exercise an ability to score. However, your shot is not competent. The shot you are taking is not within the range of the ability you exercise. 11 Alternatively, we could take the designated property to be knowledge. I will explore the prospects of this alternative and the corresponding knowledge first version of VR in Chapter 4. 12 I’d like to add two quick points. First, it is easy to see that the grounding condition will also deal with the cases of the alien cogniser and of Norbert, which caused trouble for variations of PR in the previous chapter. While I will leave it to the reader to work out the details, the key here is that the alien cogniser’s process is plausibly grounded and so may qualify as an epistemic ability, whereas Norbert’s process isn’t grounded. Second, note that a knowledge first version of VR will have an even better explanation of clairvoyant cases. Even if one is tempted to say that, in the above case, you do have an ability to form true beliefs about distant events, it is hard to deny that you don’t have an ability to know things about distant events. After all, while your way of belief formation does dispose you to acquire true beliefs here, it does not dispose you to acquire knowledge. Without a disposition to acquire knowledge, however, you simply cannot have an ability to know. Given that, on a knowledge first version of VR, a belief is justified only if it is produced by an ability to know, clairvoyant beliefs will not be justified. In this way, adopting a knowledge first version of VR may serve to strengthen the argument here.

3

Virtue Reliabilism Knowledge

The Because Relation With the virtue reliabilist account of justified belief in play, let’s move on to the account of knowledge. We already know that beliefs are taken to be performances with an aim. As a result, the general framework for performance normativity can be applied to the case of belief. Beliefs can be assessed as successful, competent and apt. Given that beliefs aim at truth, they are successful if and only if true. In addition, the previous chapter has provided a detailed account of a type of competent performance as well as of competent belief in particular. I also followed virtue reliabilists in identifying justified belief with competent belief, thus offering a detailed account of justified belief. This leaves aptness. A belief is apt if and only if it is successful because competent. Virtue reliabilists identify apt belief with knowledge. In other words, VR-K. One knows that p if and only if one aptly believes that p. Gettier cases, that is, cases of justified true beliefs that fall short of knowledge, constitute one of the most pressing problems for any account of knowledge such as VR-K. As a first observation, note that it is possible for a belief to be successful and competent without being apt. Given VR’s identification of true belief with successful belief, justified belief with competent belief and knowledge with apt belief, this means that it is possible for a belief to be true and justified without qualifying as knowledge. VR thus leaves room for Gettier cases. Of course, to provide a solution to the Gettier problem will require champions of VR to show not only that there is room for Gettier cases but also that Gettier cases are indeed a subset of cases of successful and competent but inapt belief. Moreover, a fully-fledged such solution will also be borne out by detailed accounts of the key notions of successful, competent, and apt belief. Since the last chapter has provided detailed accounts of successful and competent belief, in order to provide the kind of solution to the Gettier problem we are after, what remains to be done is to unpack the because relation at issue in apt belief. There are two

38 Virtue Reliabilism leading proposals in the literature. One, most prominently championed Greco, analyses the because relation in terms of explanatory salience, the other, most prominently associated with Sosa, in terms of the manifestation of an ability. In what follows, I will look at them in turn and outline how they venture to deal with the Gettier problem, starting with Greco’s account. Explanatory Salience VR Greco’s core idea is that achievements are successes that are explained by the exercise of ability. For instance, your hit in ARCH is apt if and only if the exercise of your arching ability explains why your shot hit the target. In order to develop the core idea in more detail, Greco appeals to an account of the pragmatics of causal explanations. Greco points out that, typically at least, in causal explanations we only give part of the full causal explanation of a given fact. For instance, we may explain the fact that you hit the target in terms of your ability. However, a complete causal explanation will include further facts—such as facts about the quality of the bow and arrow, the winds, etc. According to Greco, causal explanations cite “important or salient parts of the causal story” (Greco 2010: 74). Whether part of a causal explanation is important or salient depends at least in part on the interests and purposes of those who offer the causal explanation. In fact, what is salient may vary with interests and purposes of those giving the causal explanation. For instance, while in the context of a discussion among fans, your arching abilities will typically be salient, in the context of a discussion among equipment developers, the salience may lie with the superior quality of your equipment. Most importantly for present purposes, whether part of a causal explanation is important or salient depends also, again at least in part, on what is normal. More specifically, when something abnormal happens in the causal chain leading from exercise of ability to success, it tends to become important and salient. Abnormalities are salience magnets, as it were. To see how Greco’s account deals with Gettier cases, consider first a non-epistemic analogue of a Gettier case. You take a shot in ARCH, which is first brought off its trajectory by a gust of wind and then brought back on target through my helpful intervention with a wind machine. Crucially, something abnormal happens in the causal chain from exercise of arching ability to successful shot. An abnormal gust of wind interferes and success is only obtained because of another abnormal event: my helpful intervention. This abnormality functions as a salience magnet, with the result that your success is to be explained in terms of the gust of wind rather than your arching ability. Your shot, even if successful and competent, turns out not to be apt. The story for Gettier cases is in essence the same. You attain a true belief and do so via the exercise of an ability to form true beliefs such

Virtue Reliabilism 39 that the target belief is in the range of the ability. In Stopped Clock, for instance, you acquire a true belief that it is 8:22 and do so through the exercise of an ability to form true beliefs about the time. Crucially, however, something abnormal happens in the causal chain from exercise of ability to true belief. In the case under consideration, the abnormal element is the fact that the clock that you are taking a reading from is stopped and that you are lucky enough to end up with a true belief because it had stopped exactly 12 hours earlier. Accordingly, in Gettier cases, too, the abnormal element functions as a salience magnet. The stopped clock is what’s salient in the explanation of why you arrive at a true belief, not your ability to form true beliefs about the time. As a result, your belief, even if successful and competent, is not apt. In this way, Greco’s account can explain the absence of knowledge in Gettier cases like Stopped Clock. The Creditworthiness Dilemma One prominent problem for VR is the so-called creditworthiness dilemma, which has been developed in some detail by both Jennifer Lackey (2007, 2009) and Duncan Pritchard (2010, 2012a). The core idea here is that VR will manage to successfully predict the absence of knowledge in Gettier cases only on pain of predicting the absence of knowledge in cases where it shouldn’t, most prominently in cases of testimonial knowledge. And on the other hand, if the view is unpacked in such a way as to get cases of testimonial knowledge right, it will no longer solve the Gettier problem. Either way, then, VR is bound to get at least one case wrong. The view thus faces a dilemma. Let’s look at this problem in a bit more detail. Consider first the following case: Landmark. You arrive at the train station in an unfamiliar city and ask the first passer-by how to get to a famous landmark. Your interlocutor is a knowledgeable resident of the city who provides you with impeccable directions and you form the corresponding true beliefs. Note that, in this case, the testimonial beliefs you acquire qualify as knowledge. Can champions of Greco’s explanatory salience account accommodate this datum? There is reason to think that the answer to this question is no. After all, what is salient in the explanation of why you arrive at true beliefs about the location of the landmark are not your abilities. If anything, it is the abilities of your informant that enjoy salience in the explanation of why you get it right about the directions. When we want to explain why you get it right about the location of the landmark, we will appeal to the fact that you talked to a very knowledgeable resident of the city, not to your own abilities. But if your abilities are not salient

40 Virtue Reliabilism in the explanation of epistemic success here, by Greco’s lights, your belief is not apt and thus falls short of knowledge. These considerations suggest that if the explanatory salience account can successfully account for the absence of knowledge in Gettier cases, it threatens to make incorrect predictions elsewhere. In particular, it incorrectly predicts the absence of knowledge in standard cases of testimonial belief. Why? The plausible answer is that while abnormalities constitute a kind of salience magnet, other things can constitute such salience magnets as well. In particular, the contribution of the abilities of other agents to a given success can do the job. The problem is that other kinds of salience magnets are entirely compatible with the agent’s acquiring knowledge. If so, any account of knowledge that requires the absence of any salience magnet will fail. After all, it is bound to predict absence of knowledge in cases in which a ‘knowledge-friendly’ salience magnet is present. But couldn’t champions of the explanatory salience account avoid this kind of problem by modifying their account? Recall, for instance, that Greco maintained that a success will be because of ability when ability plays an “important or salient part” in the causal story. So, couldn’t salience in causal explanation just be dropped in favour of importance? To see that this type of response carries some promise, notice that in Landmark, you do exercise some abilities in acquiring your testimonial belief. For instance, you approach a person with your request, not a nonhuman animal or an inanimate object. In addition, you do not ask someone whom you have excellent reason to believe to be unreliable on the matter, such as a small child or a person with severe mental disabilities. You are also sensitive to a range of defeaters. If the testimony had been obviously false or if the testifier had displayed obvious signs of unreliability, you would not have formed the belief (Pritchard 2010: 41). It looks as though these abilities are of some importance in the causal explanation of why your testimonial belief is true. After all, had you not exercised these abilities, it is much less likely that you would have ended up with a true belief. To see this, compare yourself to an agent, me say, who does not discriminate at all between the quality of informants. While I happen to approach an adult human, I would have approached a small child and would have accepted its testimony at face value just the same. The probability of ending up with a true belief is significantly higher for you than it is for me. By the same token, then, the abilities you exercised in the landmark case make some contribution to your arriving at a true testimonial belief. Given that you exercise some epistemic abilities in the reception of testimony and given that the exercise of these abilities contributes to increasing the probability of success, it might be thought that the explanatory salience account can be rescued by replacing the requirement of salience by a requirement of importance in causal explanation. In order to count as knowing, then, one’s abilities must make an important enough

Virtue Reliabilism 41 contribution to the causal explanation of why one believes truly. In cases of testimony this is the case, even though one’s abilities are not salient in the explanation of epistemic success. Of course, the move from salience to importance will work only if the resulting account continues to predict the absence of knowledge in Gettier cases. The problem now is that it is far from clear that it will do so. After all, in Gettier cases, too, the agent exercises relevant epistemic abilities, including abilities parallel to those exercised in cases of testimonial knowledge. In Stopped Clock, for instance, you take a reading from a clock, not from a dog or a rock. In addition, you don’t take a reading from an obviously unreliable time-measuring device, such as a sun clock in the dark. Moreover, you are sensitive to a range of defeaters. If the clock had displayed a time that is obviously false (e.g. ‘88:88’) or if it had been labelled as being out of order, you would not have formed a belief about the time. What’s more, these abilities contribute to the causal explanation of why you get it right in much the same way as in the landmark case. Here, too, the probability of arriving at a true belief is increased by the exercise of ability. The problem that the explanatory salience account faces, then, is the following: the level of epistemic ability you exercise in the stopped clock case parallels the level of epistemic ability you exercise in the landmark case. Furthermore, the contribution to the causal explanation of why you succeed in the stopped clock case also parallels the contribution in the landmark case. As a result, if, in the landmark case, the contribution is important enough for your epistemic success to count as being because of ability, it is also important enough for your success to count as being because of ability in the stopped clock case. The explanatory salience account, then, secures the right verdicts in the former, but makes the wrong predictions in the latter. On the other hand, if, in the stopped clock case, the contribution is not important enough for your epistemic success to count as being because of ability, it is also not important enough for your success to count as being because of ability in the landmark case. In that case, the explanatory salience account secures the right verdicts in the former, but makes the wrong predictions in the latter. While the explanatory salience account can handle either case, it cannot get both cases right. It thus succumbs to the creditworthiness dilemma.

A General Problem? A number of critics of VR take the creditworthiness dilemma to constitute a general problem for the view. As I am about to argue, this is a mistake. Not only is there perfectly general reason to think that VR can escape the dilemma, but there is also a version of the view that does successfully do so.

42 Virtue Reliabilism In order to see why VR may be expected to avoid the dilemma, consider first the following widely accepted diagnosis of Gettier cases. The reason why agents in Gettier cases fall short of knowledge is that it is a matter of luck that their beliefs are true (G1). Moreover, as Linda Zagzebski (1994) has pointed out, Gettier cases follow a pattern of two strokes of luck that cancel out each other’s effects (G2). In Stopped Clock, for instance, when you acquire a true belief that it is 8:22 by looking at a clock that stopped exactly 12 hours earlier, it is a matter of luck that your belief is true (G1). Moreover, your taking a reading from a clock that is in fact stopped constitutes a stroke of bad luck. In contrast, the clock’s displaying the correct time anyway (since it had stopped exactly 12 hours earlier) constitutes a stroke of good luck (G2). Crucially, while these properties are characteristic of Gettier cases, they are not instantiated by cases of testimonial knowledge. For instance, in Landmark, it is not a matter of luck that you end up with true beliefs about the way to the landmark. Nor does the case follow a pattern of two strokes of luck that cancel out each other’s effects. In our toy case, it is not as if you ask someone who is out to deceive you or mistakes the landmark you are inquiring about for a different one and then gives you mistaken directions that lead you straight to your desired destination. Suppose that VR can identify a property of beliefs that are successful and competent but inapt that serves to predict G1 and G2 but is not also present in cases of testimonial knowledge. In that case, there is reason for optimism that VR will be able to avoid the creditworthiness dilemma. After all, VR can identify a structural difference between these cases. If so, there is reason to be hopeful that a proper unpacking of the because relation will bear out this difference with the result that the beliefs of agents in cases of testimonial knowledge come out as apt and the ones of agents in Gettier cases as inapt. In what circumstances can a belief be successful and competent but still fall short of aptness? Before answering this question, let’s ask the more general question in what circumstances a performance can be successful and competent without being apt. For instance, in what circumstances will your shot in ARCH be successful and yet be inapt? One plausible answer is that something must happen that tarnishes the performance. In the archery case, your shot may be deflected, the target may be moved and so on. Performances that are successful and competent but not apt, then, are tarnished performances. But, of course, when the agent’s performance is tarnished, the agent suffers a stroke of bad luck. When your shot is deflected by a gust of wind, you are suffering from bad luck. And, naturally, if the agent has bad luck only, the performance will end up unsuccessful. To ensure that the performance turns out to be successful after all, something else must happen that counteracts the bad luck the agent suffered. In this sense, the agent must also enjoy some good luck. In the archery case, your shot must be brought back on target, the target must

Virtue Reliabilism 43 be moved once more and so on. Cases of performances that are successful and competent without being apt, then, are cases in which the agent is afflicted by two strokes of luck: one is bad and tarnishes the performance, the other one is good and ensures that the performance is successful after all. Moreover, these considerations suggest that successful and competent but inapt performances are performances that are successful despite being tarnished. Since to succeed despite tarnished performance is to succeed by luck, in cases of successful and competent but inapt performance, the agent will be lucky that the performance ends up being successful. Applied to the epistemic case, cases of successful and competent but inapt performances are cases of beliefs that are successful (true) and competent (justified) without being apt (knowledge). Here, too, the agent’s performance will be tarnished. In Stopped Clock, your belief is tarnished by the fact that it was acquired by a reading taken from a stopped clock. When this happens, you suffer a stroke of bad luck. Moreover, to ensure that your belief ends up true after all, the effect of the stroke of bad luck must be cancelled out by a stroke of good luck. In the stopped clock case, the clock’s reading is still accurate since the clock had stopped exactly 12 hours earlier. This gives us G2. What’s more, given that to succeed despite tarnished performance is to succeed by luck, to end up with a belief that is true despite tarnished is to end up with a belief that is luckily true. In the stopped clock case, you are lucky to arrive at a true belief about the time despite taking a reading from a stopped clock. This gives us G1. In this way, VR can predict the two properties that are widely believed to be characteristic of Gettier cases. More importantly yet, there is also reason to believe that VR will not thereby get into trouble with standard testimony cases. The reason for this is that the agents’ beliefs in these cases are not tarnished. For instance, in Landmark, there is no tarnishing of your belief about the directions to the landmark. True, you rely heavily on your interlocutor for getting it right, but that’s not the same as having a tarnished belief. In the case of forming beliefs about the time by taking readings from clocks, you rely heavily on the clock. But that is not to say that your beliefs are ipso facto tarnished. For tarnishing to occur, something else must happen. For instance, taking a reading from a stopped clock will tarnish your belief. The same goes for cases of testimonial cases. The mere fact that you rely heavily on your interlocutor is not enough for tarnishing. Something else must happen, say that your interlocutor lies to you or that he mistakes the landmark you are inquiring after. Crucially, however, nothing of the sort happens in standard cases of testimonial knowledge. By the same token, there is reason to believe that the agents’ beliefs in these cases are not tarnished. What comes to light is that there is a property of successful and competent but inapt beliefs that is present in Gettier cases but not present in standard cases of testimonial knowledge. Moreover, this property serves

44 Virtue Reliabilism to predict two features that are widely believed to be essential to Gettier cases, viz. G1 and G2. Given that this is so, the present argument provides reason to think that VR’s approach to Gettier cases carries promise independently of any particular account of the because relation. That, of course, is good news. In addition, there is some hope that VR will be able to escape the creditworthiness dilemma. What needs to be done, it would seem, is to offer an account of the because relation that captures the relevant property of successful and competent but inapt beliefs. Fortunately, there exists such an account. It is the ability manifestation account. Ability Manifestation According to Sosa, apt performances are successes that manifest the agent’s ability. Here is Sosa: “A performance is apt,” Sosa tells us, “if and only if its success manifests a complete [ability].”1 ‘Complete ability’ is a technical term here. To see what it means, note first that Sosa distinguishes among the three components of abilities: seat (SE), shape (SH) and situation (SI). The seat is a set of properties of the agent in which the ability resides. Shape and situation correspond roughly to the homonymous conditions in the account of abilities from the previous chapter. Now, Sosa also distinguishes among three corresponding levels of competence: innermost, inner and complete. The innermost ability consists of the seat component of the agent’s disposition (SE). The inner ability combines the seat component with the shape component (SE and SH). Finally, the complete ability includes the situational component, besides the seat and the shape components (SE, SH and SI) (Sosa 2015: 95–6). Since aptness requires the manifestation of a complete ability, a performance will be apt only if the ability’s SE, SH and SI are in place. Consider once more your ability to hit the target in ARCH. You have this ability in virtue of having a way of shooting such that when awake, sober, etc. and there is enough light, winds are normal, etc., using this way of shooting would make it likely that you to hit the target. In that case, any shot you fire using the way of shooting will be apt only if you are in fact awake, sober, etc. and there is in fact enough light, winds are normal, etc. When you fire a shot in this way that is deflected by the pull of a magnet, the ability’s SI are not in place. As a result, you do not rise to the level of complete ability. Your shot will not qualify as apt, not even if it is successful after all, say because I, the helper with the wind machine, bring it back on target.2 And the same holds, mutatis mutandis, in the epistemic case. Here, too, a belief will be apt only if you rise to the level of complete epistemic ability, that is, only if the relevant ability’s SE, SH and SI are in place. To return to our toy case, consider your ability to form true beliefs about the time. You have this ability in virtue of having a way of belief formation such that when awake, sufficiently sober, etc. and there

Virtue Reliabilism 45 is enough light, the instrument you are using is functioning normally, etc., using this way of belief formation would make it likely that you to form a true belief about the time. Any belief you will form will be apt only if you are in fact awake, sufficiently sober, etc. and there is in fact enough light, the clock you are taking a reading from is in fact functioning normally, etc. When the SE or some SH/SI is not satisfied, you do not rise to the level of complete ability. What’s more, your belief will not be apt even if it turns out to be true after all. But how do we determine whether one’s rises to the level of complete ability in a particular case? I want to suggest that the answer is: by investigation trigger-manifestation conditionals. If the relevant triggermanifestation conditional is true of the agent in the SE, SH and SI that obtain in a particular case, the agent rises to the level of complete ability in that case. When in a particular case, your shot is deflected by a magnet, for instance, the trigger-manifestation conditional for your relevant arching ability is not satisfied. Hence, you do not rise to the level of complete ability in this case. Again, the story is just the same in the epistemic case. In order to determine whether an agent rises to the level of complete epistemic ability in a particular case, we investigate the relevant trigger-manifestation conditional. If the relevant triggermanifestation conditional is true of the agent in the SE, SH and SI that obtain in a particular case, the agent rises to the level of complete ability in that case. Now how does this apply to Gettier cases and standard cases of testimonial knowledge? Let’s first look at Gettier cases and, in particular, let’s return to Stopped Clock, our toy case. When you are taking a reading from a stopped clock, the trigger-manifestation conditional for your ability to form true beliefs about the time is not satisfied. After all, it is not the case that were you to form a belief about the time by taking a reading from a clock in this case, you would likely end up with a true belief. As a result, you do not rise to the level of complete ability here. The ability manifestation account thus delivers the right result for Gettier cases like Stopped Clock. What about Landmark, then? Here the relevant trigger-manifestation conditional is true of you in the SE, SH and SI that obtain in this case. That is to say, in those SE, SH and SI, if you were to form a belief about the location of the landmark, your belief would likely be true. After all, you are in a normal case of testimony. You are not drunk, confused, etc. and your interlocutor is not out to deceive you, etc. As a result, your true belief plausibly rises to the level of complete ability and so may well qualify as knowledge.3 Note also that the ability manifestation account tallies nicely with the above argument that VR can avoid the creditworthiness dilemma. Recall that we saw that in Gettier cases but not in standard cases of testimonial knowledge the agent’s performance is tarnished. At the same time,

46 Virtue Reliabilism according to the ability manifestation account, the SI of the relevant ability are satisfied in standard cases of testimonial knowledge but not in Gettier cases with the result that only in the former cases do the agents rise to the level of complete ability. What’s going on in Gettier cases, then, is that unfavourable SI compromise the agent’s ability. No surprise, then, that the performance produced turns out tarnished. It comes to light that the ability manifestation account of the because relation does not succumb to the creditworthiness dilemma. By the same token, this problem for VR can be solved. Ability Manifestation Precisified I believe that the ability manifestation account of the because relation carries promise. Unsurprisingly, then, I’d like to incorporate a version of it into the framework for SGPs from the previous chapter. Unfortunately, there is at least one obstacle that needs to be removed first. Recall that, according to Sosa, “A performance is apt if and only if its success manifests a complete [ability].” Moreover, a complete ability requires the presence of the ability’s SE, SH and SI conditions. Now, consider a case in which an agent attains a success and rises to the level of complete ability. Is his performance apt? Answering this question is actually a bit tricky. The reason for this is that it is not clear that cases in which performances are both successful and manifest a complete competence are by the same token cases of performances in which the success manifests a complete competence. The question remains, then, whether more is needed for apt performance than successful performance via complete ability. To answer it, notice that it is quite hard to see how cases of successful performance that manifest a complete ability could still fail to be cases in which the performance is apt. Think, for instance, of an archery case once more. Suppose you are taking a shot in perfect conditions (peak concentration, no winds, etc.) that hits the target. What reason could there possibly be to deny that your shot is apt? I, for one, can’t think of any. For that reason, I will take it that cases of performances that are successful and manifest complete abilities are also cases of apt performances. More specifically, my suggestion is to add the following precise account of aptness, and apt belief, to the normative framework for SGPs I developed in the previous chapter: SGP Achievement. A move in a given SGP is apt if and only if it is (i) successful, (ii) competent and (iii) the SI of the ability exercised are satisfied. Apt Belief. One aptly believes that p if and only if one’s belief that p is true, competent and the SI of the epistemic ability exercised are satisfied.

Virtue Reliabilism 47 It is easy to see that the remainder of the above solution to the creditworthiness dilemma can be adopted by champions of this account of aptness. According to Apt Belief, then, your belief in Landmark is apt because the SI of the epistemic ability exercised are satisfied. In line with the account of SGP abilities from the previous chapter, I want to say that this is because you retain the relevant disposition to form true beliefs in the SI that obtain in this case. And, in line with the above discussion of Sosa, this, in turn, is evidenced by the fact that the trigger-manifestation conditional comes out true in those SI. In contrast, in Stopped Clock, your belief is not apt because the SI are not in place. Again, this is due to the fact that you do not retain the relevant disposition, which, again, is evidenced by the fact that relevant trigger-manifestation conditional doesn’t hold in your SI in which you are taking a reading from a stopped clock.

Fake Barn Cases and the Safety Dilemma Recall that, according to a widely held diagnosis, Gettier cases are cases in which the agent acquires a belief that is luckily true. Moreover, Gettier cases follow a pattern of two strokes of luck cancelling out each other’s effects. One reason why VR’s account of knowledge looks so promising is that it predicts these properties of Gettier cases. Let’s also briefly recap how VR manages to predict these properties of Gettier cases. Moves that are successful and competent but inapt are tarnished moves. Since the tarnishing constitutes a stroke of bad luck, for the move to end up successful a stroke of good luck is required that counteracts the effects of the stroke of bad luck. That’s how VR predicts one characteristic feature of Gettier cases, to wit, the pattern of two strokes of luck cancelling out each other’s effects. Moreover, since moves that are successful despite being tarnished are moves that are luckily successful, and since VR takes Gettier cases to be the epistemic incarnation of such moves, VR can predict the other characteristic feature of Gettier cases, to wit, that gettiered beliefs are luckily true. Fake Barn Cases Moves that are successful and competent but inapt are tarnished moves. If Gettier cases are the epistemic incarnation of such moves, then Gettier cases are cases of tarnished moves in inquiry. But is that correct? In particular, are all Gettier cases of tarnished moves in inquiry? To answer this question, consider first the following cases: Fake Barns. You are driving through the countryside, look out of the window of your car, see a barn in the field and form a belief that you are looking at a barn. Your belief is true as the structure you are looking at actually is a barn. Unbeknownst to you, however, it is one

48 Virtue Reliabilism of the few real barns in an area otherwise peppered with mere façades that are so cleverly constructed as to be indistinguishable from real barns from your position on the road (Goldman 1976). Fake Diamonds. You are a diamond expert and are currently handed a bag of fake diamonds that, thanks to a new technology, are so carefully crafted that they cannot be distinguished from real diamonds except by elaborate laboratory procedures. By some accident, one real diamond found its way into the bunch. You pick a stone at random, which happens to be the real thing, and check it for authenticity using a method that allows you to decisively discriminate real diamonds from all kinds of fakes except the ones produced by the new technology. Since the inventors of the new technology have been careful to keep its existence a secret, you do not know about it. You form a belief that you are looking at a real diamond. Visiting Twin. You have a new colleague who has started working with you only last week. Unbeknownst to you, your colleague has an identical twin brother who still lives in the faraway country the two brothers grew up in. Today the twin brother has come for the first time to visit his brother in his new town. You are passing by a café, see your colleague bending over the table writing something down and form the belief that your colleague is sitting at that table. What you don’t know is that your colleague’s twin is also at the café and the only reason why you are seeing your colleague and not his twin is that the twin just went to the bathroom. Had the twin been not been to the bathroom, your colleague would not have bent over to write something down. Rather, he would have reclined in the chair and as a result would have been hidden from view by a curtain. At the same time, the twin would have been in plain view in the window instead. Most epistemologists agree—and, accordingly I take it to be agreed— that, in all of these cases, the beliefs you form do not qualify as knowledge.4 Moreover, the cases also display the two characteristic features of Gettier cases: you suffer from two strokes of luck, one bad: there are many fakes barns/diamonds/an identical twin around, the other good: you are looking at one of the few real barns/the only real diamond/your actual colleague. It is pure luck that you arrive at a true belief here. At the same time, there is an important structural difference between these cases (which, in recognition of the original case, I will henceforth refer to as fake barn cases) and standard Gettier cases like Stopped Clock: in the fake barns case your move in inquiry is not tarnished.5 Unlike in Stopped Clock, where you take a reading from a stopped clock, in fake barn cases, you do not, for instance, mistake a fake barn for a real barn. Of course, you might so easily have looked at a fake instead in

Virtue Reliabilism 49 which case your move in inquiry would have been tarnished in much the same way as your belief in Stopped Clock. However, as a matter of fact, you do not look at a fake. What is going on in fake barn cases, then, is that your move in inquiry is untarnished but might so easily have been tarnished. This suggests that Gettier cases need not be cases of success despite a tarnished move. Rather, cases in which the agent’s successful move is untarnished but might easily have been tarnished will do as well. Unfortunately, this means trouble for VR. VR takes knowledge to be the epistemic incarnation of apt move. As a result, VR will be able to accommodate the intuition of ignorance across the full range of Gettier cases only if all Gettier cases are cases of successful and competent but inapt epistemic move. Now I argued that successful and competent but inapt moves are tarnished moves. If so, VR will be able to accommodate the intuition of ignorance across the full range of Gettier cases only if all Gettier cases are cases of tarnished moves. However, it turns out that this is not the case. There are Gettier cases, notably fake barn cases, in which the agent’s move is not tarnished. There is thus reason to believe that VR will not be able to accommodate the intuition of ignorance in all Gettier cases. In fact, things are worse than that. There is excellent evidence that agents who succeed via competent moves that are not tarnished attain aptness, even when their moves might easily have been tarnished. Just think of an archery case in which you produce a successful shot at a normal target when shooting at the only unsabotaged target at a shooting range otherwise full of sabotaged targets, think of a case in which you successfully prepare a tasty omelette but happened to take the only salt shaker in which the salt wasn’t replaced by sugar, or think of a case in which you successfully produce a beautiful monochrome after having taken the only can in which the colour wasn’t replaced by acid. In all of these cases, you produce successful and competent moves that are untarnished but might easily have been tarnished. There can be little doubt that, in all of these cases, your moves rise to the level of achievement and so qualify as apt. Given that, in general, agents who succeed via competent moves that are not tarnished attain aptness, we may expect the same to hold for the particular case of moves in inquiry. Given that fake barn cases fit the bill, we may expect that the agents’ beliefs in these cases are apt and so, by VR, qualify as knowledge. As a result, there is not only reason to believe that VR will be unable to accommodate the intuition of ignorance in all Gettier cases, but also that it will make the wrong predictions in fake barn cases. In other words, there is reason to believe that VR succumbs to Gettier-style counterexample after all. Unsurprisingly, champions of VR are aware of this problem. In fact, solving it has been one of their major occupations. An interesting fact about the proposals in the literature is that they tend to venture to accommodate the intuition of ignorance by adding, incorporating or otherwise

50 Virtue Reliabilism exploiting a safety or safety-like condition on knowledge. The core idea of safety is that in order to know one must be safe from error, and that is to say that one must avoid error not only at the actual world but also across a subset of nearby possible worlds. As I am about to argue, there is reason to believe that this kind of response remains ultimately unsatisfactory. My general strategy is to show that solutions to the problem of fake barn cases that exploit a safety-like condition are bound run into trouble elsewhere. The reason for this is that safety-like conditions that will do the trick in fake barn cases are arguably too strong to be necessary for knowledge. There are a number of cases that have been claimed to drive this point home, including the following: Halloween Party. “There is a Halloween party at Andy’s house, and I am invited. Andy’s house is very difficult to find, so he hires Judy to stand at a crossroads and direct people towards the house (Judy’s job is to tell people that the party is at the house down the left road). Unbeknownst to me, Andy doesn’t want Michael to go to the party, so he also tells Judy that if she sees Michael she should tell him the same thing she tells else (that the party is at the house down the left road), but she should immediately phone Andy so that the party can be moved to Adam’s house, which is down the right road. I seriously consider disguising myself as Michael, but at the last moment I don’t. When I get to the crossroads, I ask Judy where the party is, and she tells me that it is down the left road.” (Comesaña 2005: 397) Lucky Drink. “I am drinking a glass of water which I have just poured from the bottle. Standing next to me is a happy person who has just won the lottery. Had this person lost the lottery, she would have maliciously polluted my water with a tasteless, odorless, colorless toxin. But since she won the lottery, she does no such thing. Nonetheless, she almost lost the lottery. Now, I drink the pure, unadulterated water and judge, truly and knowingly, that I am drinking pure, unadulterated water. But the toxin would not have flavored the water, and so had the toxin gone in, I would still have believed falsely that I was drinking pure, unadulterated water” (Neta & Rohrbaugh 2004: 399–400). Atomic Clock. “[T]he world’s most accurate clock hangs in Smith’s office at a cereal factory, and Smith knows this. The clock’s accuracy is due to a clever radiation sensor, which keeps time by detecting the transition between two energy levels in cesium-133 atoms. This radiation sensor is very sensitive, however, and could easily malfunction if a radioactive isotope were to decay in the vicinity (a very unlikely event, given that Smith works in a cereal factory).

Virtue Reliabilism 51 This morning, against the odds, someone did in fact leave a small amount of a radioactive isotope near the world’s most accurate clock in Smith’s office. This alien isotope has a relatively short half-life, but—quite improbably—it has not yet decayed at all. It is 8:20 am. The alien isotope will decay at any moment, but it is indeterminate when exactly it will decay. Whenever it does, it will disrupt the clock’s sensor, and freeze the clock on the reading ‘8:22.’ (Don’t ask why; it’s complicated.) Therefore, though it is currently functioning properly, the clock’s sensor is not safe. The clock is in danger of stopping at any moment, even while it currently continues to be the world’s most accurate clock” (Bogardus 2013: 300). In all of these cases, the agents are said to know, whilst their beliefs are not safe. As a result, any of them will do to run the argument I am about to offer.6 That said, I will here focus on my own preferred counterexample to safety, which can aptly be described as an epistemic Frankfurt case. This is in recognition of the fact that the inspiration for this kind of case is due to Frankfurt’s famous (1969) counterexamples to the principle of alternative possibilities. The reason I am mentioning the other cases is just to highlight that the success of my argument does not hinge on the particular case that will guide my subsequent discussion. To repeat, any of the above cases would serve just as well. Here goes: Frankfurt Clock. Your arch-nemesis, a powerful demon, has an interest that you form a belief that it’s 8:22 when you come down the stairs this morning. In order to achieve this, your arch-nemesis has resolved to set the grandfather clock in your hallway (from which you habitually take a reading every morning) to 8:22 unless you come down the stairs at 8:22 of your own accord. Suppose, as it so happens, you do come down the stairs at 8:22. Your arch-nemesis remains inactive. You form a belief that it’s 8:22. It is 8:22. The grandfather clock is working reliably as always. In this case, you know that it is 8:22. After all, you have the ability to read the clock and form a belief via an exercise of this ability. Moreover, the clock is actually functioning properly and the reading is accurate. At the same time, your belief is not safe from error. After all, you might very easily have come down the stairs a minute earlier or later in which case your arch-nemesis would have intervened with the result that you would have taken a reading from an inaccurate clock and ended up with a false belief that it is 8:22. Since, in this case, your belief is at high risk of error, accounts of knowledge that feature a safety or safety-like condition will incorrectly predict the absence of knowledge here.7 The upshot of my argument is that champions of VR encounter another dilemma: this one has fake barn cases on one horn and Frankfurt cases

52 Virtue Reliabilism on the other. And while champions of VR can avoid the creditworthiness dilemma, this one is a harder nut to crack. Or so I contend. A Worry Before moving on, I’d briefly like to consider the following worry: Aren’t Frankfurt cases and fake barn cases structurally identical? And if so, isn’t that reason to think that the at least one of the intuitions is mistaken and hence ought to be rejected? My response is that I do not think that the two kinds of case are structurally identical. Here is one difference: in fake barn cases the relevant error possibilities are realised in the environment in a way in which they aren’t in Frankfurt cases. For instance, in Fake Barns the error possibilities take the shape of the actual presence of fake barns in the environment. In contrast, in Frankfurt Clock, no similar relevant error possibilities are realised in the environment. For instance, there are no stopped clocks around. As a result, the relevant error possibilities in fake barn cases are live in a way in the error possibilities in Frankfurt cases aren’t. By way of further evidence for this, note that, in fake barn cases, other agents actually fall prey to this error possibility, or so we may suppose. The same is not true of Frankfurt cases. It is not as if we can just assume that other agents in the same environment actually fall prey to the error possibility of forming a false belief by taking a reading from a stopped clock. After all, stopped clocks are simply nowhere to be found in the environment. Even if it is the case that if the two kinds of case are structurally identical, they should be treated alike, there is reason to think that they are not structurally identical.8 It may also be worth noting that while there is the structural difference between these cases, this difference cannot be detected by safety or safety-like conditions. Here is why. Even though the error possibility is not realised in the environment of a Frankfurt case, it might ever so easily have been realised. Moreover, Frankfurt cases are set up such that the closest worlds at which the error possibility is realised in the environment is also one at which the agent is fooled by it. As a result, in a Frankfurt case, the agent might ever so easily have ended up with a false belief. In contrast, in fake barn cases, the error possibility is realised at the actual world, but the closest world at which the agent is fooled by it is not the world at which it is realised (i.e. the actual world). Even so, the agent might ever so easily have been fooled by it. As a result, in a fake barns case, too, the agent might ever so easily have ended up with a false belief. Now, safety and safety-like conditions require absence of error at nearby (or perhaps very close nearby) possible worlds. Since the two cases do not differ on this front, the structural difference between the two cases is not one that can be detected by safety conditions. That’s why accounts of the absence of knowledge in fake barn cases in terms of a safety or

Virtue Reliabilism 53 safety-like condition encounter the above dilemma. And that’s why they remain ultimately unsatisfactory.

Lottery Cases Let’s start by considering the following case: Lottery Loser. I have recently bought a ticket in a fair lottery, l, with one million tickets and exactly one winner. In view of the unfavourable odds, you are pessimistic about my prospects of becoming rich soon. In particular, you believe that the ticket I bought—say it’s ticket 1— won’t win l. And you turn out to be right. My ticket does indeed not win. Although the probabilistic evidence supporting your belief that my ticket won’t win (call this proposition ) is excellent and the belief is true, you do not know . Can VR accommodate this datum? Unfortunately, there is reason to think that it can’t. To see this, let’s ask whether your belief is apt. It is not hard to see that the answer to this question is yes. After all, your belief is very plausibly produced an ability to form true beliefs about lottery outcomes. Your way of belief formation, which takes the probabilistic evidence as input and outputs belief about lottery outcomes, disposes you to form true beliefs about lottery outcomes across a wide range of conditions. Across a wide range of SH and SI, the trigger-manifestation conditional—were you use this way of belief formation to form a belief about the lottery outcome, you would likely enough form a true belief— holds. Moreover, we may assume that your way of belief formation has been shaped through interaction with the environment or else that you know that it is a highly reliable way of forming true beliefs about lottery outcomes. As a result it satisfies the grounding condition on abilities and so qualifies as an ability to form true beliefs about lottery outcomes. If so, provided you are also in suitable SH, your belief will qualify as being produced by an epistemic ability. Moreover, the target propositions, is a proposition about lottery outcomes and so will count as being in the range of the ability. Your belief will be competent. Notice, furthermore, that your belief is also true. After all, is ex hypothesi true. If, finally, you are in suitable SI, your belief will qualify as apt as well. The questions that remains to be answered is whether the SI you find yourself in are suitable. Recall that to investigate this question, we needed to ask whether in your SI the relevant trigger-manifestation conditional is true. And, again, the answer here is clearly yes. In your current SI (fair lottery, etc.), it is undeniably the case that were you to use the relevant way of belief formation to form a belief in , your belief in would likely be true. As a result, there is reason to believe that your belief

54 Virtue Reliabilism in is apt. In consequence, VR-K* rules that it qualifies as knowledge. There is thus reason to believe that my favourite traditionalist version of VR does not succeed in explaining the absence of knowledge in lottery cases. But perhaps this is just a difficulty my particular account faces. Perhaps other accounts do better. Before looking into this, let’s briefly survey the theoretical options for explaining the absence of knowledge in cases like Lottery Loser. Recall that your belief in this case is ex hypothesi true. This means that there are three options of dealing with such cases available to champions of VR. They can either say that your belief falls short of knowledge because (i) it is not competent, (ii) it is (true and) competent but not apt, or (iii) they can simply deny that we do not have knowledge of lottery propositions and offer an error theory for those who think otherwise. In what follows, I will look at each of these three options, starting with the last one. Knowledge Accepting that we can know lottery propositions is perhaps the initially least promising line to pursue. As I am about to argue there is good reason for resisting it. To see this, consider first the following principle: Minimal Coherence (MCH). If one knows that one is not in a position to know the conclusion of an argument one knows to be valid, then there is some premise of the argument that one is not in a position to know either. This principle is eminently plausible. Suppose you were to believe the premises of an argument you know to be valid, whilst knowing that you are not in a position to know its conclusion. Belief in the premises rationally commits you to the truth of the propositions you believe. Knowing the argument to be valid rationally commits you to the truth of its conclusion. If you simultaneously know that you are not in a position to know its conclusion, you are in the unfortunate situation of being rationally committed to the truth of a proposition that you know you are not in a position to know. Your doxastic state is incoherent in a manner reminiscent of a Moorean paradox. MCH captures the plausible thought that such incoherences must be avoided. With MCH in play, consider the following variation of Lottery Loser: No Winners. I have recently bought a ticket in a fair lottery, l0 , with one million tickets. While there may be a winner, there is no guarantee that there will be. In fact, there is a small chance that no ticket will win l0 . In view of the unfavourable odds, you are pessimistic about my prospects of becoming rich soon. In particular, you believe that the ticket I bought—say it’s ticket 1—won’t win l0 . And you turn out

Virtue Reliabilism 55 to be right. My ticket does indeed not win. As a matter of fact, on this particular occasion, the small chance of there being no winner materialises. All tickets lose. As a first observation, you know that there is a valid argument from {ticket 1 will lose l0 , . . . , ticket n will lose l0 } to {All tickets will lose l0 }. However, you know that you are not in a position to know that proposition. After all, the probability that it is true is vanishingly small. By MCH, it follows that there is some member of {ticket 1 will lose l0 , . . . , ticket n will lose l0 } such that you are not in a position to know it. Crucially, if there is some member of Γ = {ticket 1 will lose l0 , . . . , ticket n will lose l0 }, such that you are not in a position to know it, then you are not in a position to know any member of it. Here is why. First, in this particular case, all members of the set are true. Second, whether you are in a position to know a proposition supervenes on your epistemic position towards it and its truth value. Hence, you will be in a position to know some member, i ∈ Γ without being in a position to know some other member j ∈ Γ if and only if your epistemic position towards i is not the same as your epistemic position towards j. However, third, you are in exactly the same epistemic position towards each member of Γ. It follows that you are not in a position to know any member of Γ. But since one knows a proposition only if one is in a position to know it, you do not know any member of Γ. We will do best, then, to avoid trying to resist the received view which grants that lottery propositions cannot be known. Accordingly, we had better try and look for an account that explains why we don’t know lottery propositions. Inaptness Once the idea that we know lottery propositions is abandoned, we are left with two alternative options. We can either argue that the agents’ epistemic performances in lottery cases are not even competent—and, by VR-JB, not justified—or that they are successful and competent—true and justified—but not apt. Given that there is such strong reason to think that lottery propositions such as are true, at first glance it would seem as though the latter approach is more promising. As I am about to argue, however, there is reason to think that appearances are misleading here. To see why, consider first Kyburg’s (1961, 1970) famous lottery paradox: The Sufficiency Thesis (ST). If the probability of p on one’s evidence is very high, then p is justifiably believable for one. Conjunction Closure (CC). If p is justifiably believable for one and q is justifiably believable for one, then their conjunction, p and q, is justifiably believable for one.

56 Virtue Reliabilism No Contradictions (NC). No proposition one knows to be a contradiction is justifiably believable for one. While individually plausible, it turns out that ST, CC and NC are jointly inconsistent. To see this, notice that no matter how high we set the standards for satisfaction of the predicate ‘very likely’, there will be some fair lottery with exactly one winner such that it is very likely on my evidence that each ticket will lose. So suppose that a ticket will very likely lose if the chance that it will lose is at least (n − 1)/n and let l be a fair lottery I know to have n tickets and exactly one winner. By ST, for each ticket i ∈ l, it is justifiably believable for me that i will lose. By CC, it is justifiably believable for me that all tickets in l will lose. Since I also know that l has exactly one winner, by a further application of CC, it is justifiably believable for me that all tickets in l will lose and that exactly one ticket in l will win. However, I know that this is a contradiction and so, by NC, it is not justifiably believable for me. One of ST, CC and NC has to go. But which one? It is widely agreed that the least promising candidate is NC. After all, it is reasonable to think that NC is a basic principle of justified believability. In fact, there is independent reason to think that NC is not at the fault in the lottery paradox since the paradox can be generated without invoking NC. To see this, consider the following plausible thesis:

that ‘φ ’ but (henceforth No Moore Paradoxes (NMP). Propositions of the form I don’t also ‘Moorea know that φ ’ (henceforth also ‘Moorean propositions’) are not justifiably believable for one.9

Now consider No Winner once more. In this case, there is no guarantee that the lottery in question will have a winner. Accordingly, I do not know that there will be a winner. Instead, all I know is that there might be a winner, and hence that I do not know that all tickets will lose. It is not hard to see that the lottery paradox can be generated with ST, CC and NMP only. By ST and CC, it is justifiably believable for me that all tickets will lose. Since I know that I do not know that all tickets will lose, it is justifiably believable for me that I do not know that all tickets will lose. By another application of CC, it is justifiably believable for me that all tickets will lose but I do not know that all tickets will lose. By NMP this is not justifiably believable for me. We thus have a version of the lottery paradox that does not rely on NC.10 Denying CC is perhaps initially the most attractive option. After all, one might think that even if we have to give up CC, it might be possible to rescue at least a restricted version of CC. Thus, consider: CC0 . If p is justifiably believable for one and q is justifiably believable for one, then their conjunction, p and q, is justifiably believable for one, unless it is a contradiction.

Virtue Reliabilism 57 Another noteworthy consequence of the Moore paradoxical version of the lottery paradox is that this way of restricting CC won’t do the trick either. The reason for this is that the Moore paradoxical version of the lottery paradox shows that no contradiction is needed to generate the paradox. After all, the proposition that all tickets will lose but I do not know that all tickets will lose is not contradictory. Replacing CC by CC0 won’t do the trick. Of course, one could add to the list of riders. Thus, consider: CC00 . If p is justifiably believable for one and q is justifiably believable for one, then their conjunction, p and q, is justifiably believable for one, unless it is either a contradiction or a Moorean proposition. Now, CC00 smacks of ad hocness. What would be needed to remove the air of ad hocness is some unifying and independently plausible account of why CC00 should feature these riders. I cannot help but suspect that the only plausible candidate appeals to the notion of justified believability. The reason why CC00 makes exceptions for contradictions and Moorean propositions is that they are not justifiably believable. But that, of course, won’t do at all since holding that the conjunction of two justifiably believable propositions is justifiably believable unless it isn’t renders the principle trivial. All we would have done is replace ad hocness by triviality. Fortunately, the most common reason for rejecting CC is different. Here is the rough idea: We are fallible cognitive agents in the sense that, for a wide range of propositions, there will always be a small risk that, when we come to believe them, we make a mistake. At the same time, we are capable of justified belief in the sense that justified belief is attainable for a wide range of propositions, including a subset of propositions in the aforementioned range. Given that this is so, justified believability is compatible with a small risk of error. The aggregation of justifiably believable propositions involves an aggregation of small risks of error. Small risks of error accumulate to larger risks of error. Crucially, while justified believability is compatible with a small risk of error, it is not compatible with too large a risk of error. The problem with CC is that aggregation of justifiably believable propositions may lead to a risk of error that is simply too large. In that case, even though each member of the aggregate is justifiably believable, the aggregate as a whole is not. CC is bound to fail (e.g. Foley 1979; Kyburg 1997 an ancestor of the idea can be found as early as Ramsey 1990/1929). These considerations would seem to provide a more promising diagnosis of the lottery paradox. After all, for any fair lottery with n tickets and exactly one winner, and any subset of m tickets, the risk that all its members won’t lose is m/n. When n is sufficiently large and m sufficiently small, this risk will be very small, small enough, in fact, to allow that the

58 Virtue Reliabilism proposition that all members of the subset will lose is justifiably believable for one. At the same time, when m is large enough, the risk of error will be too large to allow the proposition that all members of the subset will lose to be justifiably believable for one. If this is right, CC is bound to fail in the lottery case. This diagnosis predicts that one cannot justifiably believe of all tickets in a fair lottery that they will lose. When the lottery is large enough, one cannot even justifiably believe of nearly all or even most tickets that they will lose. One would run too great a risk of error in so doing. This looks like a welcome result. After all, it appears independently plausible. In a lottery with one million tickets and one winner, one cannot justifiably believe that all one million tickets will lose. One also cannot justifiably believe that the first 999,567 or the first 500,000 tickets will lose either. One is too likely to be in error here. These results add to the attractiveness of the present proposal. I do not want to deny that this diagnosis is attractive at first glance. However, I will argue that, despite its initial appeal, in the final analysis it remains unsatisfactory. The reason for this is that the paradox can once again be generated in a way that bypasses the relevant principle, in this case CC. The principle is a variation of a principle we already encountered in the previous subsection, to wit, MCH. Here goes: Minimal Coherence, Justification (MCH-J). If one knows that the conclusion of an argument one knows to be valid is not justifiably believable for one, then some premise of the argument is not justifiably believable for one either. MCH-J can be motivated in much the same way as MCH earlier. Suppose you were to believe the premises of an argument you know to be valid, whilst knowing that you cannot justifiably believe its conclusion. Belief in the premises rationally commits you to the truth of the propositions you believe. Knowing the argument to be valid rationally commits you to the truth of its conclusion. If you simultaneously know that you cannot justifiably believe its conclusion, you are in the unfortunate situation of being rationally committed to the truth of a proposition that you know you cannot justifiably believe. Your doxastic state is incoherent in a manner reminiscent of a Moorean paradox. MCH-J captures the plausible thought that such incoherences must be avoided. MCH-J is inconsistent with ST. There are several ways of arguing this. Here is one. Note that, by NC, one class of propositions that are not justifiably believable for one are propositions one knows to be contradictory. As a result, MCH-J serves to motivate the following principle: Minimal Consistency (MCN). If one knows that the conclusion of an argument one knows to be valid is a contradiction, then some premise of the argument is not justifiably believable for one.11

Virtue Reliabilism 59 To see that MCN is inconsistent with ST, let, S1 be an agent who knows that there is a valid argument from the premise set {ticket 1 will lose l, . . . , ticket n will lose l, l has n tickets, l has one winner} to the conclusion ⊥. By MCN, some member of the premise set is not justifiably believable for S1 . At the same time, each member of the premise set is very likely to be true on S1 ’s evidence. By ST, each member of the premise set is justifiably believable for S1 . While I take this argument to provide strong reason to believe that ST is false, it does not serve to show that the above mentioned diagnosis in terms of accumulation of small risks is bound to remain unsatisfactory. Fortunately, there is another way of arguing that MCH-J is inconsistent with ST, which does the trick. Recall that, by NMP, no proposition of that ‘φ ’but (henceforth also ‘Moorean that φ ’ is (henceforth propositions’) also ‘Moorean are notfor justipropositions’) are not I don’t know justifiably believable one. Let the form S2 be an agent who knows: P1. that there is a valid argument from the premise set {ticket i will lose l, I don’t know that ticket i will lose l} to the conclusion ticket i will lose l but I don’t know that ticket i will lose l, for each i ∈ l, P2. that any such conclusion is Moore paradoxical and thus is not justifiably believable for him, P3. that he does not know that ticket i will lose, for each i ∈ l. Since S2 knows P1 and P2, by MCH-J, he cannot justifiably believe at least one of the argument’s premises. That means that either ticket i will lose l or I don’t know that ticket i will lose l is not justifiably believable for S2 . Since, by P3, S2 knows the latter premise to be true for each i ∈ l, it follows that ticket i will lose l is not justifiably believable for S2 , for each i ∈ l. At the same time, ticket i will lose l is highly probably on his evidence and so, by ST, justifiably believable for S2 . Crucially, the fact that small risks of error accumulate to larger risks of error does little to block the present argument. After all, we are aggregating justified believability for only two propositions. And while the risk of error associated with each proposition may be small enough to be compatible with justified believability and yet too large to be so compatible when aggregated, it need not be. Given that the probability of I don’t know that ticket i will lose l surpasses the threshold at issue in ST, we can set up l such that the odds against winning are high enough that the probability of ticket i will lose l but I don’t know that ticket i will lose l is still above the threshold at issue in ST. It comes to light that, on ST, Moore paradoxical propositions come out as justifiably believable. Since this cannot be the case, ST must be false. Of course, if the culprit in the lottery paradox is ST, then, by the same token, there is reason to believe that we cannot even justifiably believe lottery propositions. Since we have now seen that there is independent

60 Virtue Reliabilism theoretical reason to believe that ST is false, this means that there is independent theoretical reason to believe that we cannot justifiably believe lottery propositions. But given that this is so, we can now see why the present strategy of accounting for why we lack knowledge of lottery propositions is bound to fail. Just recall that this account aims to show that beliefs in lottery propositions are not true because competent. In other words, it grants that such beliefs are successful and competent, and ventures to establish that they fall short of aptness. Even if they can get all the cases right, given that the lesson the lottery paradox teaches is that lottery propositions aren’t even justifiably believable, this means that their account is bound to remain unsatisfactory. After all, knowledge entails justified believability, and so the absence of knowledge is already explained in terms of the absence of justified believability. Their account will be doomed to redundancy. Incompetence There is thus reason to believe that lottery propositions are not justifiably believable. This means that the right approach to lottery cases for virtue reliabilists is the remaining one, according to which beliefs in lottery propositions fall short of being competent. The question is how champions of VR can secure this result. One way to achieve this is to argue that we don’t have the epistemic abilities required for competent belief. Unfortunately, as we have already seen, the prospects for this strategy are rather dim as it is very plausible that beliefs in lottery propositions are frequently produced by abilities to form true beliefs about lottery outcomes.12 Another approach may be to exploit the fact that justification can be defeated. If so, even if the lottery paradox leads us to reject ST, we may still be able to hold on to the following qualified version of ST: ST0 . If the probability of p on one’s evidence is very high, then p is justifiably believable for one unless one’s justification for believing p is defeated. If we adopt this strategy, the key task will consist in countenancing a type of defeater such that the justification for all and only lottery propositions comes out as defeated. The good news is that, if this strategy turns out to be successful, it will be available to champions of VR as well, at least given that their theory of justified belief is compatible with the relevant account of defeat. Unfortunately, this strategy has suffered a significant setback. As Douven and Williamson (2006) have recently argued, the prospects of countenancing a defeater of the required type are dim. For an important class of proposals for such a defeater the strategy is bound to fail. More specifically, they have shown that no structural property can make for a defeater

Virtue Reliabilism 61 of the kind sought after. Roughly, a property of propositions is structural if supervenes only on its truth-functional and probabilistic properties or, in other words, if it can be defined in strictly logical and mathematical terms only. Roughly, the reason why structural properties don’t make for the right kind of defeater is that, for any proposition p with probability < 1 and structural property, P, we can run a lottery on p, as it were, by making it a member of a set, ∆, of equiprobable propositions which are jointly inconsistent. Given that CC and NC continue to hold, we can still generate the paradox. If so, the assumption that p has P does precious little to block the lottery paradox. Thus, structural accounts of the defeat property at issue in ST0 are bound to fail.13 What comes to light, then, is that the prospects of getting virtue reliabilism to provide the right kind account, one that explains why lottery propositions aren’t even justifiably believable, are dim. This, of course, is not exactly good news for VR.14

Conclusion In this chapter, I have looked into a number of potential problems for VR. While we saw that the version of the view I proposed in the previous chapter can steer clear of perhaps the most prominent problem for VR in the literature—the creditworthiness dilemma—it does not do equally well with respect to two other problems. In particular, the view runs into what I labelled the safety dilemma. While there is nothing wrong with VR’s treatment of standard Gettier cases, there is another kind of Gettier case that means trouble for VR. This kind of case is most famously exemplified by Fake Barns. In fact, champions of VR themselves acknowledge that their solution to standard Gettier cases does not carry over to fake barns cases. In order to deal with the problem, they incorporate a safety or safety-like condition into their views. Crucially, even though this does indeed enable them to handle fake barn cases, it only gets them into trouble elsewhere. In particular, I have argued that VR will now be too strong. It will mistakenly predict absence of knowledge in a range of cases, including in Frankfurt cases. In this way, champions of VR merely trade one problem for another. What’s more, VR also struggles to offer a satisfactory account of the absence of knowledge in lottery cases. After all, on the face of it, on VR beliefs in lottery propositions come out as apt. While one might try to embrace this result, it turned out that this approach remains unconvincing. Maintaining that beliefs in lottery propositions are successful and competent but not apt fared no better. After all, the lottery paradox calls for a solution according to which lottery propositions are not even justifiably believable for us. If so, any such account is at best redundant. The right thing to say about lottery propositions is that we cannot even justifiably believe them. The difficulty for VR is to secure this result. After all,

62 Virtue Reliabilism it is even harder to make a convincing case that beliefs in lottery propositions aren’t even competent (harder, that is, than to argue that they are not apt). Moreover, the obvious alternative of exploiting the anti-defeat condition on justification has suffered a significant setback. Champions of VR are thus facing a tough uphill struggle here.

Notes 1 Sosa (2010: 470). While this quote is from an earlier paper, there is every reason to believe that Sosa still subscribes to this view. To see this, note that Sosa still takes aptness to be “success that manifests [ability]” (Sosa 2015: 19). Moreover, he still distinguishes between three levels of ability: innermost, inner and complete (see below). Finally, it is clear from his discussion of cases that the level of ability required for aptness is complete ability. By way of evidence, consider how he analyses the case of the helper (see below): your shot “does not really manifest [ability]. And the reason for this, I suggest, is that the archer does not shoot when in appropriate shape, in an appropriate situation” (Sosa 2015: 103). 2 Sosa (2015: 102–3). The ability manifestation account of the because relation is also accepted by Turri (2011, 2016) and myself (2011a, 2013). 3 It may be worth noting that this solution to the creditworthiness dilemma is not the one Sosa himself has offered. 4 Recent experimental philosophy suggests that folk do not have the intuition of knowledge in Fake Barns (Colaço et al. 2014, see also Turri et al. 2015 which tests different cases with supposedly the same structure (though see en.34 below) and obtains similar results). This is thought to put pressure on philosopher’s judgements that the agents in these cases do not know. Even if these studies are methodologically relevant (pace e.g. Cappelen 2012; Deutsch 2015), I am not convinced that they will serve to rescue traditionalist virtue epistemology. Here are some of the reasons. First, all that is needed to refute traditional virtue epistemology is one counterexample. For all the studies have shown, Fake Diamonds, Visiting Twin or some other case may do the trick even if Fake Barns does not. Second, even if we ran enough empirical studies on such cases to make a good inductive case that, in all of these cases, folk are willing to attribute knowledge, I am unsure that this should move us philosophers to revise our verdicts on these cases. The reason for this is that it is not clear to me why we should prioritise laypeople’s judgements on these cases over expert judgements. On the contrary, I am attracted to side with (Devitt 2006, 2011, Hales 2006, Ludwig 2007, Williamson 2007, Grundmann 2010, Hofmann 2010, Horvath 2010, Kipper 2010, Wright 2010, Jackson 2011, Williamson 2011) that expert judgements should take precedence over laypeople’s. After all, this is standard practice in other domains as well. For instance, we’d accord much more weight to the judgements of expert physicists such as Einstein or Feynman on issues physical than to those of laypeople such as my neighbour, the local baker or even Feynman’s undergraduate physics students (Hales 2006: 172). In fact, we have little inclination to accord much weight at all to laypeople’s judgements in other domains. It is just not clear to me that philosophy should make an exception to this rule. Suppose I am mistaken about all this. In particular, suppose epistemologists have been mistaken about fake barn cases: agents in those cases do have knowledge. It may be worth noting that even if this turned out to be the case, this is at least not bad news for the kind of knowledge first epistemological

Virtue Reliabilism 63

5 6 7 8

9

version of virtue epistemology I will develop in the next chapter. After all, that view is entirely compatible with there being knowledge in fake barn cases. In fact, nothing about the structure of the view will have to be changed in order to accommodate the existence of knowledge in fake barn cases. In contrast, it is not clear the same goes for traditional versions of virtue epistemology that predict absence of knowledge in fake barn cases. Given that these views do feature a safety or safety-like condition and the beliefs of agents in fake barn cases do not satisfy this condition (see below), it would seem that they need to make structural revisions to their view (i.e. drop the safety or safety-like condition) to accommodate the presence of knowledge in these cases. Finally, suppose traditionalists can revise their views such that they accommodate the knowledge verdict in fake barn cases. Even so, my view does not stand without motivation. After all, the problem of lottery cases for traditionalist virtue epistemology (see below) and the direct reasons to favour the knowledge first version of virtue epistemology over its traditionalist rivals (Chapter 4) remain. For alternative accounts of this difference see Hetherington (1999) and Pritchard (2008). Or, to be more precise, any case will serve provided that it is successful as a counterexample to safety, a question I will not take up here. Appendix ‘The Safety Dilemma’ takes a more detailed look at how this dilemma affects the views of perhaps the most prominent champions of VR in the literature, to wit, Sosa, Greco and Pritchard. Recall that in en. 30, I mentioned a study by John Turri et al. which tests further cases with supposedly the same structure as Fake Barns in which folk also have the intuition that the relevant agents know. One issue with this study is that the case tested actually shares the same structure with Frankfurt Clock in that the relevant error possibility is not realised in the environment. This means that there is a structural difference between the case tested in this study and the original fake barns case. As a result, it is not clear whether the study serves to put pressure on the hypothesis that agents in fake barn cases lack knowledge. It is well-known that NMP is popular with knowledge first epistemologists. Crucially, one doesn’t need to be a knowledge firster in order to subscribe to NMP. In fact, NMP has received a perfectly compelling independent defence in Douven (2009). Here is a snapshot version of the argument. (Note that Douven himself states the argument in terms of rational belief. However, since the crucial assumptions, (i)–(iv) below, are no less plausible for justified belief than for rational belief, so my adaptation should be unexceptional.) Douven first offers and motivates a small number of very plausible assumptions about justified belief. Here they are (‘JBS (p)’ stands for ‘S justifiably believes p’, ‘KS (p)’ for ‘S knows p’, ‘→’ denotes the entailment relation, and CrS is S’s degree-of-belief function): (i) JBS (p&q) → JBS (p) (ii) JBS (p) → CrS (p) > CrS (¬p) (iii) JBS (p) → CrS (KS (p)) ≥ CrS (¬KS (p)) (iv) “A person [justifiably believes p] only if it does not readily follow strictly on the basis of the assumption of her [justifiably believing p] plus principles (i)–(iii) that her degrees of belief are not probabilities.” (2009: 369)

64 Virtue Reliabilism Douven then offers the following argument to show that, given these assumptions, Moorean propositions cannot be justifiably believed: (1) JBS (p&¬Ks (p)) [Assumption for RAA] (2) JBS (¬KS (p)) [(1), (i)] (3) CrS (¬KS (p)) > 0.5 [(2), (ii)] (4) JBS (p) [(1), (i)] (5) CrS (KS (p)) ≥ 0.5 [(4), (iii)] It is easy to see, however, that (3) and (5) jointly entail that S’s degrees of belief are not probabilities. It thus follows strictly on the basis of (1) and (i)– (iii) that S’s degrees of belief are not probabilities, thus contradicting (iv). (1) is thus inconsistent with (i)–(iv). Since (i)–(iv) are very plausible, (1) must be false. Hence, Moorean propositions cannot be justifiably believed. 10 In fairness to deniers of NC, it should be noted that this argument might turn out dialectically ineffective against those who reject NC. After all, if you are willing to reject NC, you will likely not be too attached to NMP either. At the same time, for the many who do not want to give up NC, it does provide an additional reason for not so doing: they would have to give up a further plausible principle governing justified believability. Thus, even if these considerations do not move foes of NC, I take them to provide some reason to think that NC is not the culprit of the paradox. Of course, if NC is not the culprit, it’ll have to be either CC or ST. 11 Notice that, besides receiving support by the plausible MCH-J and NC, MCN is independently plausible. After all, it captures the epistemic force of reductio arguments (Kaplan 1981). 12 To recap, consider once more your belief in Lottery Loser. Your way of belief formation, which takes the probabilistic evidence as input and outputs belief about lottery outcomes, disposes you to form true beliefs about lottery outcomes across a wide range of conditions. Across a wide range of SH and SI, the trigger-manifestation conditional—were you use this way of belief formation to form a belief about the lottery outcome, you would likely enough form a true belief—holds. Moreover, we may assume that your way of belief formation has been shaped through interaction with the environment or else that you know that it is a highly reliable way of forming true beliefs about lottery outcomes. As a result it satisfies the grounding condition on abilities and so qualifies as an ability to form true beliefs about lottery outcomes. It is hard to see how virtue reliabilists could plausibly hope to establish that we do not have the epistemic abilities required for competent belief. 13 Douven and Williamson also argue that their result applies to the most prominent solutions to the lottery paradox that venture to replace ST by ST0 , including the ones in Pollock (1995) and Ryan (1996) and Douven’s own earlier proposal in Douven (2002). 14 See Appendix ‘Lottery Cases’ for a more detailed discussion of Sosa’s, Greco’s and Pritchard’s approaches to lottery cases.

4

Knowledge First Virtue Reliabilism

In Chapter 2 I developed a normative framework for simple goal directed practices (SGPs), that is practices featuring a kind of target and kind of move and a designated property. A success in such an SGP is a move that has the designated property. A move is competent if and only if it is produced by an exercise of an ability to attain success and the SGP at issue is in the range of the ability exercised. It is apt if and only if it is successful, competent and the relevant ability’s SI are in place. I also proposed to view inquiry as a simple goal directed practice such that true answers to whether questions are the relevant kind of target, beliefs the relevant kind of moves, and the designated property is truth. A success in inquiry, then, is a true belief. A competent belief is a belief produced by an exercise of an ability to form true beliefs such that the target proposition is in the range of the ability. An apt belief is a belief that is true, competent, and the relevant ability’s SI are in place. The standard virtue epistemological identifications of justified belief with competent belief and of knowledge with apt belief deliver substantive accounts of justified belief and knowledge. Importantly, both accounts fall squarely within the purview of traditional epistemology. Both can be viewed as offering reductive analyses in the sense that they provide sets of non-circular necessary and sufficient conditions for, respectively, justified belief and knowledge. While this account was attractive in that it was able to improve on both process reliabilism and extant versions of virtue reliabilism on a number of counts, Chapter 3 argued that the view still encounters a couple of significant problems. It not only faces the safety dilemma but also runs into difficulties with lottery cases. The main aim of this chapter is to develop a knowledge first alternative and to argue that it compares favourably with its traditional cousin in that it can avoid these problems. Before getting down to business, however, I’d first like to say a few words on just how I take my account to be knowledge first epistemological.

66 Knowledge First Virtue Reliabilism

Knowledge First Epistemology What is knowledge first epistemology? Its founder, Timothy Williamson, characterises the view in the following passage: “Knowledge first” is a slogan for epistemology that takes the distinction between knowledge and ignorance as the starting point from which to explain other cognitive matters [W1]. It reverses the direction dominant in much twentieth-century epistemology, which treated belief as explanatorily prior to knowledge, attempting to analyse knowledge as belief that meets further conditions, such as truth and justification [W2]. By contrast, a knowledge first epistemologist might treat believing something as treating it as if one knew it [W3]. (Williamson 2010: 208) This characterisation of knowledge first epistemology contains three claims, which I have labelled W1, W2 and W3, respectively. Williamson’s characterisation already indicates just how ambitious his project is. In fact, it exceeds the boundaries of epistemology, at least as it is traditionally understood. By way of evidence, notice that W1 takes knowledge to be the starting point from which to explain other cognitive matters. Moreover, W3 anticipates one of Williamson’s famous theses, to wit, that knowledge is a mental state in its own right and that belief is to be analysed in terms of knowledge. Since accounts of mental states are traditionally taken to fall within the domain of the philosophy of mind and perhaps philosophical psychology, Williamson anticipates that he will go beyond the boundaries of epistemology as it is traditionally understood. I’d like to flag that I will not follow Williamson in his pursuit of this ambitious project. One reason for this is that I am interested developing a viable virtue reliabilism, one that is better than its traditionalist cousins. Since, for this purpose, the question as to whether knowledge is a state of mind in its own right is by-the-by, Williamson’s thesis concerning the philosophy of mind can safely be put to one side.1 At this stage, one might naturally ask the following question: in what sense exactly is my account going to be knowledge first epistemological? The answer is that I do adopt the core epistemological claim of Williamson’s characterisation of knowledge first epistemology, that is, the following version of W2: Knowledge First Epistemology (KF). Knowledge first epistemology reverses the traditional direction of analysis in epistemology: rather than analysing knowledge in terms of justified belief, justified belief is analysed in terms of knowledge. In other words, rather than treating justified belief as explanatorily prior to knowledge, knowledge is taken to have explanatory priority over justified belief.

Knowledge First Virtue Reliabilism 67 I submit that KF retains enough of the spirit of Williamson’s project to qualify as a knowledge first view. Since it reverses the direction of analysis, it is incompatible at least with the traditional attempt to analyse knowledge in terms of justified belief.2 Since it takes knowledge to enjoy explanatory priority over justified belief, there is an important respect in which knowledge comes first. For that reason, any epistemology that satisfies KF can plausibly be regarded as a knowledge first epistemology.

Knowledge First Virtue Reliabilism What exactly made the virtue reliabilism of Chapter 3 into a traditionalist version of the view? The answer, I submit, is its account of the designated property proposed for inquiry. Given that the property is truth, a success in inquiry is a true belief. Abilities to attain success are abilities to form true beliefs. Given the virtue reliabilist identification of justified belief with competent belief and knowledge with apt belief, both justified belief and knowledge are thus analysed in terms of abilities to form true beliefs. No surprise, then, that we end up with a traditionalist version of virtue reliabilism. The first step in my endeavour to develop a knowledge first version of virtue reliabilism is to offer an alternative proposal for the designated property in inquiry. It will come as no great surprise that I want to suggest that the property is knowledge, rather than truth. What we get, then, is the following account of successful belief: Successful Belief. A belief is successful if and only if it qualifies as knowledge. What’s more, we can apply the accounts of SGP abilities, their exercises, competent belief and apt belief from Chapter 2 to the epistemic case. This gives us first the following accounts of abilities to know and their exercises: Ability to Know. One has an ability to know propositions in a range, R, and relative to conditions, C, if and only if one has a grounded way of belief formation, W, such that, for any p ∈ R, using W in C disposes one to form beliefs that p that qualify as knowledge (henceforth ‘knowledgeable beliefs that p’ for short). Exercises of Abilities to Know. One exercises an ability, A, to know propositions in range R and relative to conditions C if and only if one has A and forms a belief via the way of belief formation at issue in A. Competent Belief. One competently believes that p if and only if one’s belief that p is formed by an exercise of an ability to know propositions in range R and relative to conditions C such that p ∈ R.

68 Knowledge First Virtue Reliabilism Apt Belief. A belief is apt if and only if it is (i) successful, (ii) competent and (iii) the SI of the ability to know exercised are satisfied. Recall that traditional virtue reliabilism identifies justified belief and knowledge with different normative standings of beliefs as performances. In particular, they identify justified belief with competent belief and knowledge with apt belief. I would like to follow traditional virtue epistemologists on this front. That is to say, I would like to offer the following accounts of justified belief and knowledge: KFVR-JB. One justifiably believes that p if and only if one competently believes that p. KFVR-K. One knows that p if and only if one aptly believes that p.3 Let me say a few words to explain these accounts of justified belief and knowledge. They are superficially indistinguishable from their traditionalist rivals. At the same time, once they are properly unpacked, there are substantial differences between the two. Starting with KFVR-JB, notice that, on my account, competent belief is analysed in terms of abilities to know. While the accounts of abilities to know and their exercises do not figure directly in KFVR-JB they are of crucial importance to it. After all, they contribute to making the core notions in terms of which justified belief is defined more precise. It is easy to see that the resulting version of virtue reliabilism will qualify as a knowledge first version of the view in the sense at issue in KF. After all, justified belief is analysed in terms of abilities to know. Thus, the direction of explanation is from knowledge to justified belief, which is all that KF requires for the view to qualify as a version of knowledge first epistemology. Let’s move on to KFVR-K, then. First, it is of crucial importance to keep in mind that, unlike its traditionalist cousin, KFVR-K does not (and actually could not) offer a reductive analysis of knowledge. The easiest way to appreciate this point is by taking note of the fact that apt belief requires successful belief and successful belief is knowledgeable belief. According to KFVR-K, then, one key condition a belief must satisfy in order to qualify as knowledge is that it qualifies as knowledge. KFVR-K is blatantly circular and thus unfit for the purposes of reductive analysis. Second, notice that it follows from Successful Belief and KFVR-K that a belief is successful if and only if it is apt. After all, by Successful Belief, a belief is successful if and only if it qualifies as knowledge and, by KFVR-K, a belief qualifies as knowledge if and only if it is apt. Now, it might be thought that this is a somewhat undesirable result. After all, recall that we noted earlier that aptness, competence and success can come apart in various ways. In particular, recall that a successful move in an SGP need not be competent. And moves that are both successful and competent can still fall short of aptness. Recall also that both of

Knowledge First Virtue Reliabilism 69 these points are nicely illustrated by the case of ARCH. When you take a shot in ARCH while completely drunk, your shot is not competent. Even so, it may be that, by an incredible stroke of luck, your shot hits the target anyway. Your shot is successful without being competent. Alternatively, when you take a shot at a shooting range that is sabotaged by a magnet, your shot even if competent will not be taken in hospitable SI and so will not be apt. This is so even when your shot is brought back on target by a helper with a wind machine and so ends up successful after all. Your move is both successful and competent and yet falls short of aptness. If it turns out that, in the case of inquiry, a move is successful if and only if it is apt, it may appear as though inquiry is a bit of an oddball among SGPs. And that, in turn, may make us suspicious that something is wrong with either Successful Belief or KFVR-K, which jointly deliver this result. Fortunately, on reflection, there is little cause for concern after all. While it is true that there are SGPs such that aptness, success and competence come apart in this way, things are different for other SGPs. Consider the practice of joining consenting adults in lawful marriage (henceforth LM). LM can be understood as an SGP. Targets here are couples of consenting adults, moves are tokens of ceremonies of a certain type, call it c, and the designated property is the property of effecting the lawful marriage of the couple. A success in LM is a token of c that effects the lawful marriage of the target couple. (The definitions of LM ability, exercise thereof, competent move and apt move are as expected.) Crucially, there is reason to believe that one joins partners in lawful marriage only if one competently conducts a token of c. To see this, suppose you, the master of ceremony, fail to competently conduct a token of c. This may be for two reasons. You may fail to conduct a token of c altogether, say because you are too drunk and so do not manage to say enough of the relevant words. Or else the SGP may not be in the range of c, say because one of the partners is under age and so a different kind of ceremony is needed. In either case, the couple will not end up lawfully married. Since joining the couple in lawful marriage is the success condition of LM, this means that the success of LM will not be attained. If so, there is reason to believe that a move in LM will be successful only if competent. What’s more, there is also reason to believe that one joins partners in lawful marriage only if the SI of the ability exercised are satisfied. To see this, suppose that the situational conditions are not satisfied. For instance, one key situational condition for the ability to join partners in lawful marriage is that any caveat lodged against the marriage has been discharged. So suppose that this SI is not in place. There remains an undischarged caveat against the marriage. If so, again, the couple will not end up lawfully married, even if you have competently conducted a token of c. Since joining the couple in lawful marriage is the success condition of LM, this means that the success of LM will not be attained. As a result,

70 Knowledge First Virtue Reliabilism there is also reason to believe that a move in LM will be successful only if the SI of the ability exercised are satisfied. These considerations motivate the following condition on success in LM: LM Success. A move in LM is successful only if it is competent and the SI of the ability exercised are satisfied.4 LM Success is an informative condition on success in LM. After all, it does not hold for successes in other SGPs such as ARCH and so captures relevant information about success in LM. At the same time, LM Success is unfit to constitute part of a reductive analysis of success in LM/joining consenting couples in lawful marriage. After all, it features the notion of competence, which, in turn, is unpacked in terms of an ability to attain success in LM/joining consenting couples in lawful marriage. Now, let ∆ be any SGP such that a move in ∆ is successful only if it is competent and the SI of the ability exercised are satisfied. Since, trivially, a move in ∆ will be successful only if it is successful, it follows that a move in ∆ is successful only if successful, competent and the SI of the ability exercised are satisfied. However, by the relevant instance of SGP Aptness, we get the result that a move in ∆ is successful if and only if apt. Since, by LM Success, LM is an SGP that features the relevant condition on success, we get: LM Aptness. A move in LM is successful if and only if it is apt. It comes to light, then, that while there are SGPs in which aptness and success come apart in the sense that one can attain success while falling short of aptness, in others aptness not only entails success, but success also entails aptness. What’s more, on reflection, it is independently plausible that, in the case of inquiry, moves are successful if and only if apt. In fact, an argument parallel to the one that made this point for LM will do the trick here too. Here goes. To begin with, there is reason to believe that knowledge requires competent belief. To see this, suppose you, an epistemic agent, form a belief that you are facing a barn, say, that falls short of being competent. This may be for two reasons. You may fail to exercise an ability to know, say because you form your belief on the basis of a coin toss. Or else the proposition may not be in the range of your ability, say because it has been produced by an ability to recognise colours. In either case, you end up not knowing that you are facing a barn. Since knowledge is success in inquiry, this means that success in inquiry will not be attained. If so, there is reason to believe that a move in inquiry will be successful only if competent. What’s more, there is also reason to believe that knowledge requires that the SI of the ability exercised are satisfied. To see this, suppose that

Knowledge First Virtue Reliabilism 71 the situational conditions are not satisfied: you are in fake barn county, are currently looking at a fake barn etc. If so, again, you end up not knowing that you are facing a barn. Since knowledge is the success condition of inquiry, this means that success in inquiry will not be attained. As a result, there is also reason to believe that a move in inquiry will be successful only if the SI of the ability exercised are satisfied. Just as in the case of LM before, these considerations motivate the following condition on success in inquiry: Inquiry Success. One’s move in inquiry is successful only if it is competent and the SI of the ability exercised are satisfied. But now recall that we just saw that, for any SGP such that this condition holds, it follows that a move in that SGP is successful if and only if apt. Since we have seen that there is independent reason to think that Inquiry Success is true, this means that it is independently plausible that a belief is successful if and only if it is apt. It may also be worth noting that, just as in the case of LM Success, Inquiry Success is an informative condition on success in inquiry and yet is not suited to constitute part of a reductive analysis of success in inquiry/knowledge. It is informative because the parallel condition does not hold for successes in other SGPs such as ARCH and so captures relevant information about success in inquiry. It is unfit to constitute part of a reductive analysis of success in inquiry/knowledge because it features the notion of competence, which, in turn, is unpacked in terms of an ability to attain success in inquiry/knowledge.

Knowledge as Success in Inquiry At this stage, there are two versions of virtue reliabilism on the table. One is traditionalist, in that it offers reductive analyses of knowledge and justified belief, the other is knowledge first epistemological, in the sense at issue in KF: it analyses justified belief in terms of knowledge. The crucial difference between the two proposals concerns their respective accounts of the designated property for inquiry. While traditionalists opt for truth, knowledge firsters go for knowledge. Of course, we are free to propose whatever property we like as the designated property in inquiry. This leaves open the question which of the two proposals we have before us is preferable. One way to answer it is by assessing which of the resulting versions of virtue reliabilism is preferable by putting them next to each other and comparing how they fare with respect to solving various relevant problems. This is a strategy I will pursue in the next section. There I will return to the problems for traditional virtue reliabilism that I outlined in the last chapter and argue that the above knowledge first version of virtue reliabilism escapes them.

72 Knowledge First Virtue Reliabilism However, one might wonder whether isn’t a more direct way of arguing in favour of one view over the other and if so, what this argument might look like. Interestingly enough, such a line of argument is available. In fact, in the remainder of this section, I will outline two ways of arguing that a knowledge first version of virtue reliabilism is preferable to its traditionalist competitor. Here goes. Recall that both versions of virtue reliabilism are instantiations of a more general normative framework for simple goal directed practices. One key thesis of the general framework is that a move in an SGP is successful if and only if it has the relevant designated property. Given that both versions of virtue reliabilism take the relevant SGP to be inquiry and agree that moves in inquiry are beliefs, this means that one way to investigate which thesis concerning the designated property is preferable by asking which thesis about success in inquiry is preferable. The two relevant theses here are of course: Knowledge is Success in Inquiry (KSI). One has attained success in inquiry into whether p at time t, if and only if, at t, one has a knowledgeable belief on whether p. True Belief is Success in Inquiry (TSI). One has attained success in inquiry into whether p at time t, if and only if, at t, one has a true belief on whether p. To argue that KSI is preferable to TSI, I will first need to motivate some further general theses about SGPs. To get there, note that practitioners of SGPs can be committed to attaining success in an SGP. There are a number of types of commitment one may have towards attaining success in an SGP. The nature of the commitment depends on how it arises. For instance, as a practitioner of ARCH, you may have an intention to hit the target out of a desire to do so. In this case, I will say that you have a practical commitment to attaining success in ARCH. You may also have promised someone to hit the target. In this case, you have a moral commitment to hitting the target. Or you may have been contracted to hit the target in which case your commitment will be contractual. Other forms of commitment are conceivable. What’s more, an agent may simultaneously have various different kinds of commitment towards attaining success in an SGP. Besides having originally formed an intention to hit the target in ARCH out of a desire to do, you may now have promised and signed a contract that you would do so. In this case, you are practically, morally and contractually committed to hitting the target in ARCH. One way in which an agent can be released from any commitment he has towards attaining success in an SGP is by accomplishing the feat, i.e. by attaining the relevant success. For instance, when you have hit the target in ARCH, you will be released from your practical, moral and contractual commitments to hit the target in ARCH. This gives us the relevant thesis:

Knowledge First Virtue Reliabilism 73 Commitment Release (CR). If one has attained a given success in a given SGP at t, then, at t, one is released from all commitments towards attaining that success in that SGP.5 Now consider the following case. Detective 1. You are a private detective. I have hired you for one month to find out (= attain success in the inquiry into) whether ( =) someone in my family is the murderer of my wife. The first suspect you investigate is my uncle who, let us suppose, has a particularly strong motive for the deed. Fortunately for you, my uncle credibly admits to having committed the crime upon questioning and even signs a confession in writing. On the basis of this evidence, you come to believe that someone in my family has indeed murdered my wife, thereby making your move in the relevant inquiry. With your move made on the first day, you pack your bags, including the confession, get on the next flight to the Caribbean where you intend to take a holiday for the rest of the month. Meanwhile, it becomes widely known that my uncle’s claim to have murdered my wife is false. In fact, he has a watertight alibi for the time of the deed. You are currently sipping cocktails in the sun and are entirely unaware of the news about my uncle’s statement. Notice that we may assume that is true: someone in my family did indeed murder my wife. It is just that it wasn’t my uncle. That means that you have a true belief that .6 According to TSI, you have attained success in your inquiry into whether . By CR, you are released from all commitments towards finding out whether . Now here is the 64,000 dollar question: Is this prediction correct? In particular, are you released from your contractual commitment to find out whether ? The answer to this question is very plausibly no. You may have an excellent excuse for not working on the case. At the same time, as far as your contract is concerned, what you ought to be doing is work on the case rather than sit on the beach in the Caribbean. If this isn’t immediately obvious, put yourself in my shoes (i.e. in the shoes of your employer) for a moment. Suppose I find out about your situation. While I might concede that you have an excuse for no longer working on the case, I could rightly insist that, as far as the contract is concerned, what you ought to be doing is work on the case rather than sip cocktails on some beach in the Caribbean. What’s more, I may rightly take steps to get you back to work, to fulfil your contract. Notice that when I do this, there is no need to negotiate a new contract with you. All I need to do is remind you of the old contract, which is still binding. None of this

74 Knowledge First Virtue Reliabilism would make any sense if you had been released from your contractual commitment. Thus, TSI’s prediction that you are released from all commitments towards finding out whether is mistaken. In contrast, KSI does not make a mistaken prediction here. After all, the case is a standard Gettier case and so your belief falls short of knowledge. According to KSI, then, you fall short of attaining success in the relevant inquiry and are therefore not released from your contractual commitment to finding out what you were employed to find out. Since TSI but not KSI makes mistaken predictions about this case, there is reason to think that KSI is favourable to TSI. As I pointed out above, this means that the thesis that knowledge is the designated property in inquiry is preferable to the thesis that it is truth. As a result, there is a first reason to favour the knowledge first construal of the designated property over its traditionalist rival.7 Here is a second argument. Again, let’s start with a general thesis about SGPs. There can be varying degrees of progress towards success in SGPs. Progress here is a function of times and distances from success. One has made progress on an SGP between ti and tj if and only if, at tj , one is closer to success in that SGP than at ti . Suppose you are engaging in a version of ARCH, ARCH100 , that requires you to hit the target 100 times. Now that you have hit the target 98 times already you have made progress on ARCH100 compared to when you had done so only 13 times. This is because now you are closer to success than you were before. It is not hard to see that this account of progress supports the following thesis: Progress. If, at tj , one has not attained success in a given SGP, S, and if between ti and tj one makes progress towards success in S, then one has not attained success in S at ti either. Suppose you are engaging in ARCH100 and have currently, at t2 , hit the target 96 times. At t2 , you have not attained success in ARCH100 . Suppose you have made progress between t1 and t2 , say because at t1 you had hit the target only 34 times. In that case, you haven’t attained success in ARCH100 at t1 either. With this general thesis about SGPs in play, let’s return to inquiry once more. In particular, consider the following case: Detective 2. You are a private detective. I have hired you for one month to find out (= attain success in the inquiry into) whether . The first suspect you investigate is my uncle who, let us suppose, has a particularly strong motive for the deed. Fortunately for you, my uncle credibly admits to having committed the crime upon questioning and even signs a confession in writing. On the basis of this evidence, at t1 , you come to believe that . At t2 , however, you discover that my uncle’s S’s confession was false. In fact, he has a watertight alibi for the time of the crime and was trying to protect the perpetrator. Accordingly, at t2 , you abandon your belief that .

Knowledge First Virtue Reliabilism 75 Again, we may assume that is true: someone in my family did indeed murder my wife. It is just that it wasn’t my uncle. Now here is what I take to be an overwhelmingly plausible claim about Detective 2: you make progress on your inquiry into whether between t1 and t2 (henceforth also ‘the crucial claim’). By way of support for the crucial claim notice, first, that, at t2 , you have discovered that a certain piece of evidence that, at t1 , appears to support is misleading. Second, at t2 , you can tick one person off the list of suspects you were not able to tick off at t1 . The second reason why KSI is preferable to TSI is that only KSI is compatible with the crucial claim. Here is why: According to the crucial claim, you made progress on the inquiry into whether between t1 and t2 . However, at t2 you don’t even have a belief that . Since success in inquiry into whether requires a belief that (recall that is true), you have not attained success in inquiry at t2 . By PROGRESS, it follows that at t1 you did not attain success in inquiry into whether . However, at t1 you have a true belief that . According to TSI, you did attain success in inquiry into whether at t1 . TSI is thus unable to accommodate the crucial claim. In contrast, KSI encounters no difficulties here. After all, at t1 , your belief that is gettiered and so does not qualify as knowledge. According to KSI, you do not attain success in inquiry into whether at t1 . As a result, KSI is entirely compatible with the crucial claim. Given the plausibility of this claim, this means that there is a second reason to favour the knowledge first construal of the designated property over its traditionalist rival.8

The Problems for Traditional Virtue Epistemology We have seen that there is reason to favour the knowledge first version of our virtue reliabilist accounts of knowledge and justified belief over its traditionalist competitor. The reason for this is that there is independent reason to think that the traditionalist thesis that true belief is success in inquiry is problematic. In contrast, the knowledge first competitor steers clear of these problems. Another way in which knowledge firsters may venture to make headway vis-à-vis their traditionalist competitors is by arguing that their view offers a better treatment of various epistemological problems that we want such an account to address. Recall that the last chapter outlined a couple of relevant problems. I argued that traditionalists encounter difficulties with lottery propositions. In particular, the lottery paradox posed a significant obstacle for them. Moreover, we saw that extant traditionalist versions of virtue reliabilism face the safety dilemma. In what follows, I will argue that the knowledge first version of the view can steer clear of these problems. As a result, we will have even further reason to prefer KFVR over its traditionalist rival.

76 Knowledge First Virtue Reliabilism The Lottery Paradox Recall that, according to Sufficiency Thesis (ST), if the probability of p on one’s evidence is very high, then p is justifiably believable for one. Conjunction Closure (CC) has it that if p is justifiably believable for one and q is justifiably believable for one, then their conjunction, p and q, is justifiably believable for one. Finally, No Contradictions (NC), states that no proposition one knows to be a contradiction is justifiably believable for one. The lottery paradox shows that ST, CC and NC are jointly inconsistent. In addition, the discussion in Chapter 3 indicated that the right solution to the lottery paradox is one that denies ST. However, the task of providing an adequate solution to the lottery paradox is made more difficult by Douven and Williamson’s generalisation of the lottery paradox which showed that a large class of prospective solutions are bound to fail. In what follows, I will argue that KFVR serves to offer a solution to the lottery paradox that denies ST without succumbing to Douven and Williamson’s generalisation of the paradox. In order to achieve this, we first need a principle connecting KFVR-JB with the notion of justified believability. To begin with, I take it that a proposition is justifiably believable for one only if: if one has done what one is in a position to do to decide whether it is true and thereupon has come to believe it, then one’s belief is justified (cf. Williamson 2000: 95). Given that justified believability is understood in this way, here is a natural way of connecting it with KFVR-JB: KFVR-JBY. p is justifiably believable for one only if one is in a position to believe p via an exercise of an ability to know propositions within range R and relative to C such that p ∈ R. To see how KFVR-JBY solves the lottery paradox, consider first arbitrary lottery proposition, say ‘ticket 3 won’t win the fair lottery I am about to hold’ (henceforth ‘’). By KFVR-JBY, is justifiably believable for you only if you are in a position to believe that via an exercise of an ability to know propositions in range R and relative to conditions C such that ∈ R. This in turn requires you to be in possession of a grounded way of belief formation, W, such that using W disposes you to form a knowledgeable belief that in at least some C. Crucially, however, when the only evidence bearing on you have at your disposal is the probabilistic evidence concerning the low odds of winning, no grounded way of belief formation in your possession disposes you to form a knowledgeable belief in , no matter what conditions you may find yourself in. In consequence, is not in the range of any of your abilities to know. Since is an arbitrary lottery proposition, the result generalises to all propositions of the form ‘ticket i won’t win fair lottery l’.

Knowledge First Virtue Reliabilism 77 It may also be worth noting that we cannot avoid this result by fiddling with your history and the learning process by means of which you acquired the relevant way of belief formation. The reason for this is that, when the only evidence you have at your disposal is probabilistic, it is not just a contingent matter of fact that your ways of belief formation do not dispose you to acquire knowledgeable beliefs about lottery propositions. Rather, it could not be otherwise: when the only relevant evidence you have is probabilistic, no way of belief formation could dispose you to acquire knowledgeable beliefs about lottery propositions. In this kind of evidential situation, no way of belief formation could qualify as an ability to know lottery propositions. In consequence, when the probabilistic evidence is the only evidence you have, you are not in a position to believe a given lottery proposition via an exercise an ability to know such that the lottery proposition is in the range of that ability. Hence, by KFVR-JBY, in such cases lottery propositions are not justifiably believable for you.9 On KFVR, ST fails for lottery propositions. The lottery paradox can be solved. Finally, let’s turn to the question as to whether this solution falls prey to Douven and Williamson’s generalisation of the lottery paradox. Recall that the upshot of Douven and Williamson’s argument is that solutions to the lottery paradox that venture to replace ST by the weaker principle ST 0 —if the probability of p on one’s evidence is very high, then p is justifiably believable for one, unless one’s justification for believing p is defeated—are bound to fail for any account of defeat according to which it is a structural property. Here’s why KFVR’s solution to the lottery paradox escapes Douven and Williamson’s generalisation. The reason why, according to KFVR we don’t have justification for believing lottery propositions is not that a defeater present. By the same token, KFVR is not committed to endorsing even ST 0 . On the contrary, by the lights of KFVR’s solution, lottery propositions are never in the ballpark for justification to begin with. Champions of KFVR will thus want to reject ST 0 . Since Douven and Williamson’s generalisation targets solutions that hold on to ST 0 and KFVR’s solution escapes the threat here. With these points in play, it is easy to see that KFVR will also be able to offer a satisfactory account of why we do not know lottery propositions. Recall that in the last chapter, we also saw that since the lottery paradox requires a solution according to which lottery propositions are not justifiably believable for us, the correct explanation of why we don’t know them to be true is in terms of the lack of justified believability also. Since, on KFVR’s solution to the lottery paradox, lottery propositions turn out not to be justifiably believable, at least not when the only evidence available is probabilistic, it offers the right kind of explanation of why lottery propositions are not known. In this way, KFVR can avoid yet another problem its traditionalist rival encountered.10

78 Knowledge First Virtue Reliabilism The Safety Dilemma In essence, the safety dilemma is the problem of offering an account of knowledge that predicts absence of knowledge in fake barn cases and presence of knowledge in Frankfurt cases. The former kind of case has driven many contributors in the literature, including the most prominent defenders of virtue reliabilism, to adopt a kind of safety condition on knowledge. The latter kind of case suggests that a safety condition that can successfully deal with fake barn cases will be too strong. KFVR can easily avoid the safety dilemma. As a knowledge first epistemology, KFVR can allow that the difference between agents in fake barn cases and agents in Frankfurt cases is fundamentally epistemic and cannot be understood without making use of the concept of knowledge. That said, KFVR can offer a slightly more detailed account of what is going on in these cases. Recall that, according to Inquiry Success, there is an informative competence condition on knowledge. More specifically knowledge requires (i) competent belief and (ii) that the SI of the relevant ability to know are satisfied. The reason why, in Frankfurt cases, you know, while, in fake barn cases, you don’t is that you satisfy (ii) only in Frankfurt cases. Crucially, however, the range of the SI of an ability to know that must be satisfied cannot be understood without invoking the concept of knowledge. After all, whether your SI are within the range of an ability to know depends on whether the relevant way of belief formation continues to dispose one to acquire knowledgeable beliefs in those SI. We thus need the concept of knowledge to understand the concept of range of ability to know, which, in turn, serves to explain the presence/absence of knowledge in Frankfurt/fake barn cases. While KFVR can explain presence and absence of knowledge in these cases, the explanation is going to be circular. Since KFVR abandons the aspiration of offering a reductive analysis of knowledge, there is nothing amiss with the circularity in explanation here. In fact, we can find the same kind of circularity in analogous non-epistemic cases. Consider: in the archery case your shot missed the target because it was taken in conditions outside the range of your ability (say the winds were too strong). This appears to be a perfectly fine explanation of why the shot was unsuccessful. But here, too, we must not hope to offer an account of the notion of range of the ability without invoking the notion of a hit. After all, the range of your ability to hit the target is delineated by whether the relevant way of shooting continues to dispose you to produce hits. In much the same way as we needed the concept of knowledge to understand the concept of range of ability to know, we need the concept of hit to understand the concept of range of ability to hit the target. Compatibly with that, there is nothing amiss with explaining a particular miss in terms of the agent’s being in conditions outside the range of his ability to hit the

Knowledge First Virtue Reliabilism 79 target, nor with explaining why a particular belief falls short of knowledge in terms of the agent’s being in conditions outside the range of his ability to know. It comes to light that KFVR can offer a promising solution to the safety dilemma. Crucially, unlike its traditional virtue reliabilist competitors, KFVR does not try to explain the presence and absence of knowledge in Frankfurt and fake barn cases in purely non-epistemic terms. Instead, it takes the difference between agents in the two types of cases to be a fundamentally epistemic one.

Conclusion In this chapter, I have developed a knowledge first version of virtue reliabilism. Like its traditionalist cousin, this view identifies justified belief with competent belief and knowledge with apt belief. However, the two views differ in how the key notions of competent and (hence of apt) belief are unpacked. Ultimately, the differences between the two views are rooted in different construals of the designated property for inquiry. Whereas, on the traditionalist view, the designated property is truth, the knowledge first version of virtue reliabilism takes it to be knowledge. As a result, the two views yield different views of what a success in inquiry amounts to—true belief vs. knowledge—and what it means to have an ability to attain such a success—an ability to form true beliefs vs. an ability to know. Since competent and apt belief are analysed in terms of abilities to attain success in inquiry, it follows that we end up with different accounts of competent and apt belief. The traditionalist view, which opts for an account in terms of abilities to form true beliefs, is very much in line with the ambition of an offering reductive analysis of knowledge in terms of justified belief. In contrast, there is little hope that this can be made to work on the knowledge first alternative. The reason for this is of course that on the knowledge first version of virtue reliabilism, justified belief is analysed in terms of abilities to know. As a result, a reductive analysis of knowledge in terms of justified belief will no longer be an option. On the contrary, it is easy to see that this kind of virtue reliabilism reverses the traditional direction of analysis and so qualifies as a knowledge first version of the view. In addition, I have offered two lines of argument supporting knowledge first virtue reliabilism over its traditionalist competitor. The first one was direct. Here I provided reason to believe that the knowledge first account of success in inquiry is preferable to the traditionalist alternative since only the knowledge first account can accommodate a number of plausible theses about simple goal directed practices, to wit, Commitment Release and Progress. In contrast, the second line was more indirect. I have made a case that KFVR compares favourably with its rival because

80 Knowledge First Virtue Reliabilism KFVR can steer clear of the problems that I had earlier argued beset traditionalist versions of virtue reliabilism. In particular, KFVR can avoid the safety dilemma and can offer a satisfactory solution to the lottery paradox according to which lottery propositions turn out not to be justifiably believable. As a result, KFVR can also give the right kind of explanation of why we don’t know lottery propositions to be true. Finally, KFVR can of course also hold on to the motivations for its process reliabilist and traditional virtue epistemological ancestors. Just like its ancestors, KFVR can also allow for a connection between justified belief and truth without jeopardising the non-factivity of justified belief. It can accommodate paradigm cases of justified belief, including beliefs based on perception, memory and competent deduction. After all, the relevant processes qualify as abilities to know. It can accommodate paradigm cases of unjustified belief such as belief based on wishful thinking and guesswork. The relevant processes here do not qualify as abilities to know. Finally, champions of KFVR are free adopt process reliabilism’s solutions to the regress problem and the problem of scepticism as well as traditionalist virtue reliabilism’s solutions to the problems of world bound reliability, the new evil demon problem, the problem of clairvoyant cases and the generality problem.

Notes 1 At the same time, knowledge first virtue epistemology is compatible with Williamson’s thesis that knowledge is a state of mind (henceforth also ‘KSM’) and the thesis that knowledge has explanatory priority of belief (henceforth ‘KPB’). That said, my account neither entails KSM, or KPB for that matter, nor does either thesis play a key role in it. I take this to be a benefit of my account. After all, both theses remain a highly controversial in the philosophy of mind. For that reason, it will be of some interest if it is possible to reap at least some of the epistemological benefits of knowledge first epistemology, without taking on the heavyweight commitments in the philosophy of mind that KSM and KPB bring with it. 2 Note that KF is not strictly speaking incompatible with the traditional analytical project. After all, it does not rule out the possibility of an analysis of knowledge in terms other than justified belief. For instance, the view is entirely compatible with an analysis of knowledge in terms of safe and/or sensitive belief, where justification is not itself analysed along those lines. For what it’s worth, I think that the safety dilemma indicates that prospects of an analysis in such terms are dim. Since this is very plausibly the most promising strategy for a reductive analysis of knowledge in terms that do not involve justification, I take the prospects for an analysis of knowledge in terms other than justified belief to be no brighter. Crucially, however, for present purposes, I do not need to take a stance on the issue. For that reason, rather than foreclosing the possibility of a successful reductive analysis of knowledge, I will rest content with highlighting my reservations about it.

Knowledge First Virtue Reliabilism 81 3 On the present proposal, knowledge is both apt belief and successful belief. At the same time, success is necessary but not sufficient for aptness. While this may seem terribly confused at first glance, I’d like to ask you to bear with me. I’ll explain how this can be possible momentarily. 4 Other plausible candidates for types of performances such that success entails a competence-in-SI condition are certain types of intentional action, such as winking, waving, applauding and reading as well as Zagzebski’s (1996) acts of virtue. 5 Attaining success in an SGP is not the only way of being released from a commitment to doing so. In the case of ARCH, you may lose interest and as a result drop your desire and intention to hit the target. This will release you from your practical commitment. Of course, even when you have lost interest, you may still be bound by your promise or the contract you entered. In order to be released from the corresponding moral and contractual commitments, you may need the cooperation of, here, the person you made the promise to and the other party to the contract. In addition, one may have a moral commitment to fulfil other commitments, such as contractual ones, that one has taken on. In that case, one won’t be released from one’s moral commitment unless one has been released from one’s contractual commitment. The same may hold for other types of commitments. 6 In addition, your belief is formed on the basis of excellent evidence, evidence that is strong enough to ensure that your belief is also justified. 7 Since your belief in Detective 1 is not only true but also justified (en. 46), the present argument serves to show that KSI is preferable not only to TSI but also to the theses (i) that justified belief is success in inquiry and (ii) that justified true belief is success in inquiry. 8 Again, in Detective 2, your belief at t1 is not only true but also justified (for the same reasons as your belief in Detective 1, see en. 46). As a result, this argument, too, serves to show that KSI is preferable not only to TSI, but also to the alternative theses that identify success in inquiry, respectively, with justified belief and justified true belief. 9 There may be cases in which you have other ways of forming beliefs about lottery propositions. For instance, you may have been told by a reliable informant that a certain ticket will lose the lottery because the lottery is rigged against it. Since, at least in certain conditions, believing lottery propositions on the basis of the informant’s say-so disposes you to acquire knowledgeable beliefs about lottery propositions, in this situation you are in a position to believe the lottery proposition via an exercise of an ability to know. By KFVR-JBY, the lottery proposition is justifiably believable for you. I take this to be the right result. 10 It may be worth noting that KFVR also successfully handles variations of lottery cases in which the lottery is rigged. After all, so long as the agent’s evidence is probabilistic, the lottery proposition still cannot be justifiably believed by her.

5

The Competition

The last chapter developed a novel version of virtue reliabilism. It contrasts with traditionalist versions of the view in that it takes knowledge not to be something that is built up from the relevant form of epistemic success and further conditions. Rather, knowledge itself is a form of epistemic success, success in inquiry. In combination with the virtue reliabilist framework from Chapter 2, this yielded accounts of competent and apt belief that are knowledge first epistemological in that these notions are now defined in terms of knowledge. The standard virtue reliabilist identification of justified belief with competent belief and knowledge with apt belief delivered my specific knowledge first version of the view. That said, my view is not the only knowledge first version of virtue reliabilism on the market. Two notable alternatives have been defended in recent literature. One is due to Alan Millar, and the other is due to Lisa Miracchi. In this chapter, I will discuss their views and compare them to my own, i.e. KFVR. Unsurprisingly, my aim is to show that KFVR comes out on top once again.

Millar The View Let’s start with Millar’s version of knowledge first virtue reliabilism. One of Millar’s main reasons for adopting this kind of view is that it best reflects the way in which the concept of knowledge is applied in our everyday practices. We do not seem to apply the complex conditions that have been proposed in analyses of knowledge when we judge that S knows p. What we do is different: We happily count people as knowing that something is an F when they see an F, and they may be presumed to have what it takes to tell of something they see that it is an F from the way it looks. [ . . . ] The conceptual level at which we encounter the perceptual knowledge that we have, or that others have, is that of knowing that p through seeing or otherwise perceiving that p, by means of an ability to tell

The Competition 83 that such a thing is so from the look or other appearance of what is perceived. (Millar 2010: 133–4)

According to Millar, we should not try to give a reductive account of knowledge, but rather attempt to elucidate it by means of an investigation of the specific abilities exercises of which allow us to gain it, as it is precisely these abilities that we seem to latch on to when we ordinarily ascribe knowledge to someone. For instance, in the case of perception, which I will focus on in what follows, the relevant abilities are perceptual-recognitional abilities, which are ways of telling that things are so from their appearances, where appearances are just the way things look, sound, smell, etc. For subjects to have these abilities, it is important that the presented appearances are distinctive of the recognised objects in the sense that “[w]hen an appearance of something is distinctive of Fs, not easily could something have this appearance and not be an F” (Millar 2010: 125). Perceptual-recognitional abilities are thus environment-dependent. This accords with the intuition that you cannot know that there is a barn in front of you when you are looking at one of the few real barns in fake barn county. In such a scenario, there are too many fake look-a-likes around that make the appearance of the real barn no longer distinctive of a real barn. Millar even goes as far as to claim that you, when in fake barn county, do not just fail to exercise the requisite perceptual-recognitional ability, but even lack this ability altogether (Millar 2010: 126). In contrast, in a situation in which the environment is in fact favourable but one nevertheless does not recognise something for what it is, one fails to exercise a perceptual-recognitional ability that one does possess. This might happen when one is careless in ones judgement, or is just unlucky enough to encounter the only fake look-a-like in the entire environment. In any case, the important point is that, as Millar construes it, the notion of exercise of an ability is a success notion (Millar 2010: 125): one cannot that (henceforth that ’also (henceforth ‘Moorean propositions’) ‘Moorean propositions’) are not justi- are not one φ s. Applied to also the case of perceptualexercise an ability to φ ’unless recognitional abilities, this means that one cannot exercise a perceptualrecognitional ability without knowing that such-and-so is the case. One thing that is interesting about Millar’s view is that while he opts for a knowledge first virtue reliabilist account of knowledge, he does not venture to extend this view to justified belief as well. Rather, Millar opts for an accessibilist account of justified belief, albeit a knowledge first version of it. To get a first handle on what Millar is up to consider the following passage: [E]pisodes in which I come to know that an animal I am looking at is a zebra are, barring rare, dire confusion, or rare and bizarre deception, episodes in which the fact that I see that the animal is a zebra is available to me as a reason to believe that it is a zebra and

84 The Competition to continue to believe that it was thereafter. The intimate connection between perceptual knowledge and justified belief is accommodated by acknowledging that the fact that I see that the animal is a zebra can constitute a reason I have to take it to be one. But instead of explaining the knowledge as, so to speak, built up from justified belief, we treat the knowledge as what enables one to be justified in believing. (Millar 2010: 139) On Millar’s account, justification has to do with being in possession of clinching reasons for belief, i.e. reasons that settle it that the belief is true. Possession of such reasons in turn implies that one “[stands] in some relation to a (distinct) consideration in view of which one is justified” (Millar 2010: 112). Thus, even though Millar does not think that knowledge must be built up from justified beliefs, he does take justification to consist in having accessible, clinching reasons for belief. In paradigmatic cases of visual perceptual knowledge, these reasons are constituted by the fact that S sees that p. Gettier Strikes Back With Millar’s view on the table, let’s ask how plausible it is. In particular, I’d like to focus on whether Millar did well in combining a virtue epistemological account of knowledge with an accessibilist account of justified belief. I aim to show that the answer to this question is no. In particular, there is reason to think that Millar’s accessibilism runs into trouble with Gettier cases. One might think that this is crazy. Once we abandon the traditional analytical project in favour of a knowledge first approach to epistemology, it is hard to see how Gettier cases could even begin to cause trouble for us. Of course, this is true in the sense that we will no longer need to search for a condition on knowledge that deals with Gettier cases, or, at the very least, not one that can be specified without invoking the concept of knowledge. But now note that, besides the familiar point that agents in Gettier cases lack knowledge, the beliefs of agents in Gettier cases are justified. For instance, while, in Fake Barns, you do not know that there is a barn before you, your belief that there is is justified. This should not come as much of a surprise. After all, Gettier cases were intended as counterexamples to the justified true belief account of knowledge. If it weren’t plausible that the relevant agents’ beliefs are justified, Gettier cases wouldn’t have made for convincing counterexamples to this view in the first place. And unlike the absence of knowledge, the presence of justified belief in Gettier cases is of interest at least to knowledge first epistemological accounts of justified belief. It is a datum such accounts need to accommodate. Accordingly, we have before us what I will henceforth refer to as ‘the new Gettier problem’, to wit, the problem of accommodating the presence of justified belief in Gettier cases.

The Competition 85 To assess how Millar fares on this score, note first that, on Millar’s view, justified belief entails knowledge. The reason Millar is committed to this is that, according to him, possession of the kinds of factive reasons required for justified belief are sufficient for knowing. For instance, in the case of the visual perceptual belief that p, justification requires that one sees that p. At the same time, seeing that p is said to be a way of knowing that p. In consequence, one will satisfy Millar’s conditions for justified belief that p here only if one knows that p. Justified perceptual beliefs that fall short of knowledge turn out to be impossible. But, of course, this means that Millar’s view does encounter the new Gettier problem.1 Couldn’t Millar avoid the problem simply by weakening his account of factive reasons such that, whilst factive, they are not sufficient for knowledge? No. To see why not, note first that even if Millar can resist the entailment from justified belief to knowledge, he will still be committed to the idea that justified belief is factive. And, arguably, that’s enough to cause trouble. Just consider the non-lucky counterparts of Gettiered agents who end up with false beliefs. (In what follows, I will also refer to these cases as ‘Counterpart cases’.) For instance, consider a variation of Fake Barns in which you end up looking at a fake barn. When, in this case, you form the belief that you are looking at a barn, your belief will be false but still justified. Millar is of course aware that his account runs into this problem. He ventures to address it in the following passage: [T]he notion of justified belief that figures in traditional analysis and in descriptions of Gettier cases is [ . . . ] very weak. It has everything to do with a kind of reasonableness that renders one blameless in thinking that something is so, but little to do with the kind of well groundedness that settles that something is so and on that account entitles one to take it to be so. (Millar 2010: 102) I believe that this response remains unsatisfactory for two reasons. First, it collapses a normative difference worth marking in epistemology. Second, the intuition of justification is not the only reason to think that agents in Gettier and Counterpart cases have justified beliefs. To see the first point, consider the following two cases: Insanity. You have gone insane. As a result, you form your beliefs in all sorts of crazy manners. When hearing the wind blow you think your long lost love is speaking to you, when the sky is red at sunset, you think that doom is impending, and so on. Benighted Isolation. You are part of an isolated and benighted community the members of which share a common belief that thunderstorms indicate that their 20-eared deity is about to scratch its largest left ear.

86 The Competition Just now you are witnessing a thunderstorm and come to believe that the deity is about to scratch an ear.

In both of these cases you form a blameless belief. This is confirmed by the following two widely accepted theses concerning blamelessness: (B1) ’ (henceforth ‘Moorean propositions’) ’ (henceforth are not also ‘Moor one is blamelessthat for φ -ing if it is outalso of one’s control thatthat one φ s; (B2) onejusti’ (henceforth (henceforth ‘Moorean also propositions’) ‘Moorean not justiif that one φ s’ also in the light of good reason propositions’) to are believe that are not justiis blamelessthat for φ -ing that φ -ing ’ (henceforth also ‘Moorean propositions’) not1998). justi-2 Insanity is an is permissible (e.g. Zimmermann 1997;are Haji instance of (B1). Here it is out of your control that you believe that doom is impending. You have gone insane. Benighted Isolation is instance of (B2). In this case, your belief about impending ear-scratching is formed in the light of good reason to believe that it is permissible. After all, you reasonably believe that thunderstorms indicate ear-scratching and that a thunderstorm has occurred. If so you have good reason to believe that it is permissible for you believe as you do. Moreover, since you base your belief on reasonable beliefs that provide you with good reason to believe that it is permissible, you believe in the light of this reason. Crucially, there is a difference between agents in Gettier and Counterpart cases on the one hand, and agents in cases like Insanity and Benighted Isolation on the other. Agents in the Gettier and Counterpart cases form their beliefs in perfectly fine ways, ways that normally lead them toward epistemic goods such as true belief or knowledge. In contrast, agents in cases like Insanity and Benighted Isolation form their beliefs in highly problematic ways, ways that will normally not bring them on the path toward epistemic good. In this way, agents in the former cases are in a much stronger epistemic position than agents in the latter cases. Their beliefs have a connection with knowledge and truth that the beliefs of agents in cases like Insanity and Benighted Isolation lack altogether. To see this, compare, for example Fake Barns and Benighted Isolation. In Fake Barns, you are simply unlucky not to acquire knowledge on this occasion, whereas, in Benighted Isolation you fail to acquire knowledge because you are part of a community that is on the wrong epistemic track entirely. This difference is a difference worth marking in epistemology. One reason for this is that marking this difference enables us to select the right course of action when working toward improvements of the agents’ future epistemic performances. For agents in cases like Benighted Isolation and Insanity, we will need to get agents to change the ways in which they form their beliefs. For agents in Gettier and Counterpart cases, in contrast, this is not necessary. Accordingly, when aiming for future improvements, a different strategy would seem more appropriate. For instance, we might consider engineering a more hospitable epistemic environment. In fact, the difference is a difference worth marking as a distinctively normative difference in epistemology. It makes sense to evaluate

The Competition 87 the beliefs of agents in Gettier and Counterpart cases positively and the beliefs of agents in cases like Insanity and Benighted Isolation negatively. To see that it makes sense to evaluate the beliefs of agents in Gettier and Counterpart cases positively, note that so doing will reinforce their ways of proceeding as epistemic agents, which is a good thing because agents will start to reap epistemic goods again as soon as they are back in epistemically more hospitable territories (cf. Bird 2007: 106). In contrast, it makes sense to evaluate the beliefs of agents in cases like Insanity and Benighted Isolation negatively because so doing will discourage agents from continuing in their ways of proceeding as epistemic agents, which is also good thing given that no epistemic goods are to be gained by their way of proceeding. (Of course, it also makes sense grant these agents an excuse for proceeding in the way they do, but this point is by-the-by here.) With these points in play, it can be argued that Gettier and Counterpart cases pose a problem for Millar independently of the intuition that agents in these cases have justified beliefs. To see this notice first that it is hard to see how Millar’s account of justified belief can accommodate the aforementioned normative difference between agents in Gettier and Counterpart cases and agents in cases like Insanity and Benighted Isolation, at least on the present strategy. If the only thing that the beliefs of agents in Gettier and Counterpart cases have going for themselves is that they are blameless, then they are on par with the beliefs of agents in cases like Insanity and Benighted Isolation. The normative difference between agents in Gettier and Counterpart cases on the one hand and agents in cases like Insanity and Benighted Isolation is going to be lost. Since this difference is a difference worth marking, this means Millar’s account of justified belief collapses a difference worth marking.3 Let’s move on, then, to the question of whether there is reason to think that beliefs of agents in Gettier and Counterpart case are justified, other than the relevant intuition. I believe that the answer to this question is yes. The beliefs of agents in Gettier and Counterpart cases have a couple of properties that are widely considered to be hallmark properties of justified belief, to wit, the property of enjoying a strong connection to truth and knowledge (P1), and the property of having positive normative status (P2). That the beliefs of agents in Gettier and Counterpart have P1 was argued above. And while I haven’t strictly speaking shown that these beliefs also have P2, I believe that I have come pretty close. After all, I have effectively argued that it makes sense to accord positive normative status to these beliefs. But since it plausibly makes sense to accord positive normative status to these beliefs only if they do indeed have positive normative status, it is also plausible that the beliefs of agents in Gettier and Counterpart cases have P2. The fact that the beliefs of agents in Gettier and Counterpart cases have properties that are widely considered to be hallmark properties of justified belief confirms the hypothesis that these beliefs are indeed justified.4

88 The Competition The fact that the blamelessness response collapses a normative difference worth marking in epistemology suggests that this response remains unsatisfactory. This suggestion is reinforced by the fact that we have more reason to believe that the beliefs of agents in Gettier and Counterpart cases are justified than just the relevant intuition. In this way, there is not only evidence that the blamelessness response is unsuccessful, but also some positive reason to think that the beliefs of agents in Gettier cases are justified after all. If so, Millar’s account does succumb to the new Gettier problem.

Miracchi The View According to Miracchi, knowledge is the manifestation of an epistemic competence. Two points are important here: first, an epistemic competence is taken to be a competence to know and, second, the is here is the is of identity. Miracchi’s view contrasts with standard versions of virtue reliabilism on both counts. Traditionalists take knowledge to require the manifestation of an epistemic competence, alongside other conditions, rather than to be identical with it. And, of course, traditionalists do not construe the relevant epistemic competences as competences to know. Note also that this latter ingredient makes the view into a knowledge first version of virtue reliabilism. Given this view of knowledge, won’t Miracchi will also be committed to holding that justified belief is knowledge? After all, if she adopts the classical virtue epistemological view according to which justified belief is belief that manifests epistemic competence, and manifestations of epistemic competence are knowledge, doesn’t it follow that any belief that is justified will also qualify as knowledge? Finally, if so, doesn’t Miracchi also fall prey to the new Gettier problem in just the same way as Millar did before? Fortunately, the answer to all of these questions is no. Miracchi’s knowledge first virtue reliabilism does keep justified belief and knowledge apart. Her key move is to distinguish between manifestations and exercises of a competence to know such that one can exercise a competence to know without manifesting it and to identify justified belief with the exercise of a competence. Let’s look at how this works out in more detail, starting with Miracchi’s account of the manifestation of a competence to know. First, Miracchi takes any competence to know to be associated with a way of knowing. For instance, the competence to visually recognise chanterelles is associated with a way of coming to know about the presence of chanterelles by vision. And what Miracchi calls the manifestation conditions of a competence to know, then, are the conditions that constitute knowing in a

The Competition 89 particular way. The manifestation conditions of a competence to visually recognise chanterelles are the conditions that constitute visually knowing about the presence of chanterelles. Miracchi takes these manifestations conditions to break down into subpersonal mechanisms and external conditions. In the case of a competence to visually recognise chanterelles, the manifestation conditions break down into certain visual processes and external conditions such as sufficient light, no obstruction from view, etc. Finally, for a competence to know to be manifested is for its manifestation conditions to obtain. For the competence to visually recognise chanterelles to be manifested is for the relevant visual processes to be run in the right external conditions. Here is Miracchi’s account of the manifestation of a competence to know in her own words: Manifestation Condition. “The manifestation conditions of CK are whatever operations of subpersonal cognitive mechanisms and external conditions together (against a background of possession of CK ) constitute a particular case of knowing that p in the way characteristic of the competence WCK . A competence to know is manifested just in case its manifestation conditions obtain.” (Miracchi 2015a: 45–6) Manifestation Condition offers a precise account of what’s going on in cases in which knowledge is present. But now recall that Miracchi also wants to allow for cases of justified belief without knowledge. Moreover, recall that she wants to do so by distinguishing manifestations and exercises of competences and identifying justified belief with the exercise of competence. To see how she makes these rough ideas precise, let’s first take a closer look at one further condition on competences Miracchi countenances: Basis Condition. “The basis of CK is fully constituted by the subpersonal cognitive mechanisms of the subject S whose operations partially constitute S’s knowing that p in way WCK .” (Miracchi 2015a: 46) Basis Condition isolates the subpersonal cognitive mechanism at issue in Manifestation Condition. For instance, in our toy case, the basis of a competence to visually recognise chanterelles will be the relevant visual processes. To see Miracchi’s next step towards making room for justified beliefs that fall short of knowledge, consider her definitions of the notions of a degenerate exercise of a competence to know and of an exercise of a competence to know: Degenerate Exercise. “A competence to know is degenerately exercised just in case its basis is operative, but not all of the manifestation conditions obtain.” (Miracchi 2015a: 48)

90 The Competition Exercise. “A competence to know is exercised just in case it is either manifested or degenerately exercised.” (Miracchi 2015a: 48) In conjunction, Degenerate Exercise and Exercise allow for cases in which a competence to know is exercised but the agent does not come to know. Consider a version of our toy case in which you run the visual processes that constitute the basis of your competence to visually recognise chanterelles. At the same time, suppose that not all the manifestation conditions of your competence are in play, say because you are looking at a hologram of a chanterelle. In this case, by Degenerate Exercise, you degenerately exercise your competence to visually recognise chanterelles. But, of course, by Exercise this is enough for exercising this competence. The result is that you exercise your competence to visually recognise chanterelles even though you don’t come to know about the presence of a chanterelle by vision. In this way, for Miracchi manifestations of competences to know and their exercises come apart. The account of justified belief is straightforward. Miracchi takes a belief to be justified if and only if it is an exercise of some competence to know (2015a: 48). It is now easy to see that, as a result, Miracchi allows for cases of beliefs that are justified and fall short of knowledge. In fact, all cases of degenerate exercises of competences to know will qualify. Crucially for present purposes, this view allows Miracchi to accommodate the presence of justified belief in standard Gettier cases. To see this, consider once more the case in which you are looking at a hologram of a chanterelle. We have already seen that, in this case, you exercise your competence to visually recognise chanterelles. By Miracchi’s account of justified belief, you arrive at a justified belief that there is a chanterelle before you. All we need to do to turn the case into a Gettier case is add that your belief is also true, say because the hologram happens to be projected onto a chanterelle. In this way, Miracchi can accommodate the presence of justified belief in Gettier cases. While this much is good news, here is a wrinkle. Consider yet another case of a degenerate exercise of your competence to visually recognise chanterelles. But suppose in this case, the manifestation condition that is missing is slightly different: the light isn’t good enough and as a result you are prone to mistake a range of things aren’t chanterelles for chanterelles. The problem is that, unlike in the stopped clock case, here it is rather implausible that any belief about chanterelles you form will be justified (cf. “World-Bound Reliability” section). However, Miracchi’s view seems to predict that it is and so runs into trouble. Fortunately, this wrinkle can be ironed out. Miracchi countenances another condition on competences that will do the job. Here goes: Proficiency Condition. “The proficiency condition of CK requires that the objective probability of the manifestation conditions obtaining

The Competition 91 conditional on the basis of the competence being operative be sufficiently high. I.e., Pr(M|OB) ≥ n, for some sufficiently high n ∈ (0; 1].” (Miracchi 2015a: 46) Note that Proficiency Condition develops the sense in which competences must be reliable. For instance, for the competence to visually recognise chanterelles to qualify as a genuine competence to know, it must be the case that there is a high enough chance of coming to know about the presence of chanterelles by vision given that the visual processes that form its basis are operative. Crucially, given Proficiency Condition, not all manifestation conditions of a competence to know are created equal. In particular, the absence of certain manifestation conditions of a competence to know compromise the competence altogether. That’s what happens in the case of your competence to visually recognise chanterelles in too poor lighting conditions. Since the lighting conditions are sufficiently poor, the probability of coming to know about the presence of chanterelles given that the visual processes are run is insufficiently high. By Proficiency Condition, the result is that you don’t have the competence to visually recognise chanterelles. And since you cannot exercise a competence that you do not possess, you do not exercise this competence either. In this way, Miracchi can secure the right result that your belief about the presence of a chanterelle is unjustified. While your competence to visually recognise chanterelles is indeed compromised in the poor lighting case, the same does not hold for your competence visually recognise chanterelles when you are looking at a hologram of a chanterelle. After all, since there are many real chanterelles in your environment and you are just unlucky to look at the one hologram, the probability of coming to know about the presence of a chanterelle given that you are running the cognitive mechanisms underlying your competence to visually recognise chanterelles remains high. Proficiency Condition is satisfied. Your competence to visually recognise chanterelles is not compromised. You exercise it, albeit degenerately, and so arrive at a justified belief.5 Return of the Fake Barns While Miracchi can accommodate the presence of justification in standard Gettier cases, she takes a turn that’s at least initially surprising when it comes to fake barn cases. Here is Miracchi: My view is that such subjects [i.e. subjects in fake barn cases] fail even to be justified because hostile environmental conditions deprive them of the relevant epistemic competences. (Miracchi 2015a: 33, fn. 12)

92 The Competition Miracchi thus accepts that agents in fake barns cases don’t have justified beliefs. As a result, Miracchi encounters the new Gettier problem. Before moving on, I’d like to take a quick look at exactly why Miracchi claims that agents in fake barn cases aren’t justified. Couldn’t she say instead that they have the competence but that the manifestation conditions aren’t satisfied? If so, agents in fake barn cases do come out as having justified beliefs and the new Gettier problem will be avoided. And wouldn’t that be the better option? Perhaps. Crucially, however, Miracchi argues that the option isn’t available to her. To see why, note first that she agrees with Sosa that agents like the archer who is shooting at the only target that isn’t sabotaged at the range can produce apt performances. Moreover, she thinks that once we agree on this, the only plausible thing to say about agents in fake barn cases is that they lack the relevant competence to know. Here is Miracchi: [B]eing lucky to be in a good situation is not the kind of luck that precludes manifestation of competence. I think this view is plausible, and I do not see how one could retain it and still claim that the precariousness in [fake barn cases] prevents subjects from manifesting their competences. We must conclude that the only way such precariousness might affect whether a subject manifests a competence is by precluding possession of the competence in question. That is, the subjects not only might easily have gone wrong, but would have too easily have gone wrong for their performances to count as competent. (Miracchi 2015b: 367) It comes to light, then, that it is no accident that Miracchi holds that agents in fake barn cases don’t have justified beliefs. By the same token, the new Gettier problem threatens to affect her with unmitigated force. Miracchi does not explicitly address this problem. However, she does address a related issue and what she has to say here may serve as a basis for dealing with the new Gettier problem. Let’s look at the related issue first, then. Miracchi considers the objection that agents in fake barn cases plausibly do possess the relevant competence, contrary to what she claims. In response, she ventures to explain the plausibility of this claim away. The key idea is to distinguish between competences and abilities. While these two concepts are often run together in ordinary parlance, it is important to keep them apart. In particular, according to Miracchi, they differ in at least one important respect: exercises of competence must make success likely, whereas exercises of ability need only make success possible. Those who find it plausible that agents in fake barn cases have the relevant competence might not be distinguishing between abilities and competences. However, once we are clear that there is a difference between abilities and competences and just how they differ, the claim that agents in fake barn cases have the competence is less and less plausible (Miracchi 2015b: 367–8).

The Competition 93 It is not hard to see how we might try to develop this story into an error theory for those who judge that agents in fake barn cases believe justifiably. All that’s needed are the following claims. First, agents who make these judgements mistake a belief from ability for a belief from competence. Second, competent belief entails justified belief. Third, belief from ability does not entail justified belief. To see whether this error theory works, let’s consider first the following non-epistemic case. You are a good runner who is currently about to start a race against a bunch of primary school kids. It is pretty clear that we would judge that you have the competence to win the race here. By way of evidence, note that were you to win the race and nothing out of the ordinary were to happen, we’d also judge that you would have won the race from competence. But now suppose that just before the race is about to begin, the kids are replaced by professional runners. In this case, we would no longer judge that you possess the competence to win the race, especially once we are clear on difference between competences and abilities. On the contrary, there can be little doubt that we would now judge that you don’t have this competence, even if we grant that you have the ability. After all, now that you are up against a bunch of pros, it’s just too unlikely that you will actually win the race. And were you to win the race—say because due to some incredible series of coincidences all the professional runners drop out due to injury and you are the only one who makes it across the finish line—we would judge that it was not from competence that you did so. The key point that I’d like to bring into focus is that, in this case, we do not mistakenly attribute the competence to win the race to you. By the same token, even if you retain the ability to win the race here, and even we often mistake abilities for competences, we don’t do so in this particular case. Consider next the parallel epistemic case: You are a good barn spotter. You are about to come up to a barn on your right in a part of the world in which everything that looks like a barn is a barn. It is pretty clear that we would judge that you have the competence to recognise barns. By way of evidence, note that were you to form a belief about the presence of a barn and nothing out of the ordinary were to happen, we’d also judge that you would have formed this belief from competence. But now suppose that just before the barn you are about the reach comes into sight, all but one barn in the area are randomly replaced by indistinguishable fakes. In this case, we would no longer judge that you possess the competence to recognise barns, especially once we are clear on difference between competences and abilities. On the contrary, there can be little doubt that we would now judge that you don’t have this competence, even if we grant that you have the ability. After all, now that you are surrounded by fakes that are indistinguishable from real barns, it’s just too unlikely that you will actually recognise a barn. What’s more, were you to end up

94 The Competition looking at the one real barn in the area—say because that’s the one that was randomly selected not to be replaced—we would judge that it was not from competence that you got it right about the barn. Here’s the trouble. First, what we have here is just a version of the fake barn case (in which the fakes have been placed very recently). As such it is no less plausible that the relevant agents have justified beliefs than in the original version of the case. At the same time, the case is also structurally analogous to the above running case. As a result, it is no less plausible that we would not mistakenly attribute the relevant competence here than in the running case. Likewise, it is no less plausible here than in the running case that even if you retain the relevant ability, and even we often mistake abilities for competences, we don’t do so in this particular case. But of course this means that the error theory remains ultimately unsatisfactory. Of course, Miracchi might venture to look elsewhere for a workable error theory. That said, it may also be worth recalling what the discussion of Millar’s view brought to light, viz. (i) that the obvious candidate for an error theory, which appeals to a confusion between genuine justification and blamelessness, won’t work; and (ii) that there is independent theoretical reason for thinking that agents in Gettier cases have justified beliefs, which applies equally to fake barn cases and to standard Gettier cases. As a result, fake barn cases continue to pose a problem for Miracchi. Finally, even if Miracchi manages to deliver the needed goods, she’ll still be in an awkward spot. She wants to offer an account of justified belief that allows for justified belief in Gettier cases and may well succeed, at least as far as standard Gettier cases are concerned. At the same time, already this success comes at a cost. On her account, justified belief turns out to be a disjunctive kind, with knowledge and justified belief as the relevant disjuncts (Miracchi 2015a: 48). At the very least, this isn’t particularly elegant. What has come to light now is that, in addition, her account doesn’t accommodate the presence of justified belief in fake barn cases. This means that she will have to offer a different treatment for fake barn cases, in terms of some error theory. As a result, she will be at a further disadvantage vis-à-vis my view, which not only doesn’t take justified belief to be a disjunctive kind, but can also offer a uniform treatment of both kinds of Gettier cases (see “KFVR” section). What’s worse, we may also expect her to be at a disadvantage against other rival knowledge first accounts of justified belief. For instance, consider a variation of Miracchi’s view, call it the simple view, according to which a belief is justified if and only if it is a manifestation of a competence to know and devises some error theory to explain away the appearance of justified belief in Gettier cases. Since we may expect that whatever error theory Miracchi may come up with for fake barn cases will work equally well for standard Gettier cases, the simple view will be no worse off than Miracchi’s more complex alternative in terms of its account of the cases. At the same time, just as my own view, the simple view also

The Competition 95 has the advantages of securing the benefit of a uniform treatment of all Gettier cases and avoiding the costs of construing justified belief as a disjunctive kind. In fact, note that Miracchi’s view just is the simple view with some extra complexity (a distinction between a manifestation and an exercise of a competence, an account of degenerate exercises of competences, a disjunctive account of exercise of competence, an account of justified belief as a disjunctive kind, etc.), which is added to deal with cases like Gettier cases. Once it is clear that Miracchi’s view doesn’t even accommodate all of the relevant cases, one may well start to wonder whether it is really worth adding all of the proposed extra complexity in first place. My two cents is that it isn’t. Isn’t this too quick? After all, Miracchi does offer some considerations that may just serve to provide a theoretical reason in support of her view. Here is Miracchi once more: Exercises of a competence to know, then, whether cases of knowledge or cases of justified belief that fall short of knowledge, are qua exercises of competences to know likely to be cases of knowledge. Moreover, cases of knowledge and cases of merely justified belief that are exercises of the same competence to know will be equally likely to be cases of knowledge (qua exercises of that competence). Thus, insofar as reliability with respect to an epistemic good is itself an epistemic good, all justified beliefs share a common positive epistemic status, one that derives from its relation to knowledge—reliability with respect to knowledge. (Miracchi 2015a: 49) The key thought here is that, on her account, mere beliefs that are degenerate exercises of competences to know have positive epistemic status, which provides us with some reason for thinking that such beliefs are justified. But, of course, there are many properties that give beliefs positive epistemic status. True, Miracchi’s property of being identical with an exercise of a competence to know is one. But there are many others. Consider: (i) being identical with a manifestation of a competence to know (in Miracchi’s sense), (ii) being competent in the sense at issue in my view, (iii) being equally likely to constitute knowledge in some other sense of probability, (iv) being reliably produced in Goldman’s sense, (v) being defensible against challenges, and so on. The question remains as to why we should identify any one of these properties that give beliefs positive epistemic status with the property of being justified. To settle it, we will very plausibly have to look how the various proposals fare with respect to empirical adequacy and theoretical virtues such as simplicity, uniformity and elegance. And, as we have seen, there is reason to think that Miracchi’s proposal simply doesn’t fare as well as other views on these counts.

96 The Competition Finally, it may be objected that I have said nothing about Miracchi’s positive argument for her view. If this argument works against rival views, such as my own view or the simple view, it may still be that Miracchi’s view comes out on top in the end. After all, her argument may be powerful enough to make palatable the costs of both an inelegant account of justification and of having to offer separate treatments for standard Gettier cases and fake barn cases. Miracchi’s positive argument targets a certain kind of virtue epistemological account of knowledge. In essence, it is a development Zagzebski’s recipe for generating Gettier cases that is said to apply not only to traditional versions of virtue reliabilism but also to knowledge first cousins that grant that competences to know can be exercised without yielding knowledge. I will here focus on the part of the argument that targets knowledge first versions of virtue reliabilism, for obvious reasons. Here goes: This difficulty [posed by Zagzebski’s argument] is not resolved if we suppose the success condition to be knowing, and maintain that the exercises of competences to know can obtain either in cases of knowledge or in cases that fall short of it. If this were true, then what it is to know would be specifiable independently of the particular competence in question, and there would have to be a general relation that holds between cases of knowledge and exercises of competences to know. However, it should be clear that the [i.e. Zagzebski’s] procedure for generating systematic Gettier cases would be equally applicable to a proposal along these lines. (Miracchi 2015a: 41–2) As a first observation, note that Miracchi’s argument really is an argument against certain accounts of knowledge, including knowledge first virtue reliabilist ones. It is not an argument for her own account of justified belief. It will come as no surprise, then, that even if the argument is successful, her account of justified belief remains in trouble. After all, the simple view according to which a belief is justified if and only if it is a manifestation of a competence to know is entirely compatible with her account of knowledge, which, of course, she is firmly committed to taking to escape the argument. Second, the argument will work only against accounts of knowledge that venture to offer informative sufficient conditions for knowledge. The reason for this is that, for any account that doesn’t offer informative sufficient condition for knowledge, it simply won’t be possible to construct a Gettier-style counterexample in the first place. Consider a view that countenances informative necessary conditions on knowledge but not (jointly) sufficient ones. Even if you can construct a case in which all of the account’s informative conditions are met and the agent doesn’t have knowledge, given that the informative conditions are only necessary

The Competition 97 but not (jointly) sufficient, the case will simply not be counterexample material for the account in question. It may be worth adding that it is not only possible to hold this kind of view, but actually very much in the spirit of knowledge first epistemology (Williamson 2000). In fact, my own account of knowledge is but one case in point. Hence, my view does not fall prey to Miracchi’s argument. Miracchi’s positive argument does not tip the balance in favour of her view either.

KFVR Can my own view do better? To begin, here it is once more: KFVR-JB. One justifiably believes that p if and only if one competently believes that p (alternatively: if and only if one’s belief that p is formed by an exercise of an ability to know propositions in range R and relative to conditions C such that p ∈ R). What does KFVR-JB have to say about Gettier cases of both the fake barns and the standard variety? To answer this question, let’s first return to our toy version of a fake barns case, to wit Fake Barns. Recall that here you are in Fake Barn County, see what appears to be a barn and thereupon come to believe that you are facing a barn. In this case, you acquire your belief that you are looking at a barn in a way, W, that does not qualify as such an ability relative to your actual situational conditions, SIA . To see this, recall that whether or not a given set of conditions is within the range of a given SGP ability depends on whether using the way of move production at issue in it one continues to dispose its possessor to attain the SGP’s success in those conditions. Crucially, in SIA , using your way of belief formation does not dispose you to acquire knowledgeable beliefs about the presence of barns. SIA are not within the range of your ability. In other words, you do not have the ability relative to SIA . That said, in more favourable situational conditions (SIF ), using W does dispose you to form knowledgeable beliefs about the presence of barns. For instance, it does so in certain situational conditions such that everything that looks like a barn actually is a barn. In addition, since you are a normal human being, there is every reason to believe that your way of belief formation is grounded in the way required to qualify as an ability. After all, it is safe to assume that normal human beings acquire their ways of forming perceptual beliefs via exactly the kinds of learning processes that result in grounding. By Ability to Know, W qualifies as an ability to know propositions about the presence of barns relative to SIF . Since you form your belief via W, by Exercise of Ability to Know, you form your belief via the exercise of an ability to know propositions about the presence of barns relative to SIF . Moreover, the target proposition is within the range of this ability as it concerns the presence of a barn. Since you form your belief via the exercise of an ability to know such that the target proposition is

98 The Competition within the range of this ability, by Competent Belief, your belief is competent and so, by KFVE-JB, it is justified. It comes to light that KFVR-JB makes the right predictions in Fake Barns. It is easy to see that the story for standard Gettier cases is in essence the same. To this this, let’s take another look at our toy case, Stopped Clock. The way in which you form your belief qualifies as an ability to know propositions about the time, for the same reasons as your way of belief formation in Fake Barns qualifies as an ability to know propositions about the presence of barns. If so, by Exercise of Ability to Know, your belief that it is 8:22 is the product of an exercise of an ability to know propositions about the time. Since the belief you form is a belief about the time, the target proposition is within the range of the ability to know that produced your belief. By Competent Belief, your belief that you are facing a barn is competent and so, by KFVE-JB, it is justified.

Conclusion The last chapter introduced KFVR and argued that it compares favourably with its traditionalist virtue reliabilist competitors. This chapter has done the same for rivals in the knowledge first camp. More specifically, we have seen that knowledge first accounts of justified belief encounter the new Gettier problem. Even if they will not have to account for the absence of knowledge in these cases, they will have to account for the presence of justification. What this chapter has argued is that alternative knowledge first versions of virtue reliabilism in the literature, that is, Millar’s and Miracchi’s, both succumb to the new Gettier problem. On Millar’s view, justified belief entails knowledge. As a result, no Gettier case comes out as a case of justified belief. Miracchi’s account of justified belief, in contrast, is slightly weaker. Even if this allows her to explain the presence of justified belief in standard Gettier cases, her account still rules out justification in other cases. In particular, I have argued that fake barn cases continue to constitute a problem for the view. In contrast, I have argued that KFVR-JB can accommodate the presence of justified belief in Gettier cases of both the standard and the fake barns variety. One way of seeing why this works is that the external condition on knowledge is even weaker than Miracchi’s. Justified belief does not require an objectively high probability of knowledge, it only requires sufficiently high probability in favourable conditions. Since conditions in Gettier cases aren’t favourable, the fact that (in some of them at least) the objective probability of coming to know is very low constitutes no threat for the view. All in all, I take myself to have provided a reasonable amount of support for KFVR, enough, I submit, to rest my case for the time being.

The Competition 99

Notes

1 Note that Millar is not the only one who encounters this problem. There are other views in the knowledge first camp are equally in trouble, including all knowledge first accounts according to which justified belief entails knowledge such as Littlejohn’s (2015, Forthcoming), Sutton’s (2005, 2007) and Williamson’s (2000, 2010), alongside Millar’s. In this way, the problem is one that is not specific to Millar. What’s more, the solution that Millar offers (see below), according to which the relevant agents are not justified but only blameless, is the standard response to this problem. All of the above have ventured to deal with this problem via some version of this move or other. By the same token, the argument (developed below) that the distinction between justification and blameless won’t do the trick for Millar generalises also. The generalised version of this argument is developed in more detail in Kelp (2016) and Simion, Kelp & Ghijsen (2016). 2 See Kelp & Simion (Forthcoming) for a full account of blameless action that of Could good reason tocase believe incorporates both (B1) and (B2). it be the thatthat one φ s-ing in the light of of good goodreason reasonto to believe believe that that φ -ing is permissible whilst one should have known d reason to believe that that φ -ing is impermissible? Perhaps. However, for present purposes, this issue of goodassume reasonthat to believe that φ -ing is of little importance. Accordingly, I will simply one cannot in the light of good reason to believe that φ -ing is permissible when one should have known that it isn’t. 3 Littlejohn (Forthcoming) draws a distinction among blameless agents between agents who get an excuse and agents who get an exemption. In the case of exemptions the rational capacities of the blameless agent are absent or compromised. In contrast, agents who are blameless but have intact rational capacities get excuses (Littlejohn Forthcoming: 10). Could Millar venture to accommodate the difference between agents in Gettier and Counterpart cases on the one hand and agents in cases like Insanity and Benighted Isolation on the other by means of the distinction between exemptions and excuses? It is easy to see that the answer to this question is no. While agents in cases like Insanity come out as getting an exemption on Littlejohn’s view, agents in cases like Benighted Isolation will get an excuse. After all, their rational capacities are in perfect working order. Even with the proposed distinction in play, champions of Millar’s account of justified belief will be unable to accommodate the difference between agents in Gettier and Counterpart cases on the one hand and agents in cases like Benighted Isolation on the other. What’s more, it might be thought that even if the distinction did serve to get the cases right, it is still bound to be insufficient for understanding the normative status at issue in Gettier cases. After all, there is a clear normative difference between an excuse and a justification that this view threatens to collapse. 4 It is easy to see that the beliefs of agents in cases like Insanity and Benighted Isolation do not have all of these hallmark properties of justified belief. In particular, they do not have P1 and P2. If so, we have reason to think that the beliefs of agents in these cases are not justified. 5 It may be worth noting that this is not how Miracchi motivates Proficiency Condition. Rather, for her Proficiency Condition is a key element in her account of why she can avoid the Gettier problem (Miracchi 2015a: 47). That said, I did not mean to suggest that the above is the only way for Miracchi to motivate Proficiency Condition. After all, the two are certainly compatible. What’s more, I cannot see why Miracchi wouldn’t allow that cases like the above provide additional motivation for Proficiency Condition.

The Safety Dilemma

In Chapter 3 I argued that traditionalist virtue reliabilist accounts of knowledge aim to account for fake barn cases by introducing a safety or safety-like condition on knowledge. I also argued that, as a result, they are facing a dilemma with fake barn cases on one horn and cases of unsafe knowledge on the other. In this appendix, I will take a look at a few of the most prominent virtue reliabilist accounts of knowledge and show how the dilemma affects them.

Sosa In order to deal with fake barn cases, Sosa first introduces a distinction between animal knowledge, reflective knowledge, and knowledge full well. Animal knowledge is simple first-order apt belief. Reflective knowledge is reflectively apt belief. Knowledge full well is apt belief such that its aptness derives from its reflective aptness. Crucially, according to Sosa, reflectively apt performance requires safe success. To see this, note that a performance is reflectively apt if and only if one has aptly assessed the risks of inaptness. In particular, one must have formed an apt belief that the performance would be successful, if one were to produce it. Since apt belief is factive, one’s performance is reflectively apt only if it would be successful, were one to produce it; alternatively, given a standard possible worlds semantics for counterfactual conditionals, one’s performance is reflectively apt only if at close possible worlds at which one produces it, it is successful. Since fully apt performance entails reflectively apt performance, fully apt performance also requires safe success in this sense. Second, knowledge full well is fully apt belief and so entails belief that is safe in the sense that one knows that p full well only if were one to believe that p, p would be true; alternatively, one knows that p full well only if at close possible worlds at which one believes that p, p is true. Finally, third, Sosa understands “human knowledge as requiring knowledge full well” (Sosa 2015: 85). As a result, human knowledge features the aforementioned safety condition. With these points in play, it is easy to see how Sosa can handle fake barns cases. Sosa agrees that, in the fake barn case, your belief that you

The Safety Dilemma 101 are facing a barn qualifies as animal knowledge. You attain simple firstorder aptness, just as the agents in the analogous practical cases (e.g. archery) do. However, since you are in an area in which indistinguishable fakes predominate, your belief is not safe. At very close possible worlds at which you form a belief that you are facing a barn, you are looking at a fake barn. At those worlds, your belief is false. In consequence, you do not rise to the higher levels of reflective knowledge and knowledge full well. Since human knowledge requires knowledge full well, it follows that you fall short of human knowledge. In this way, the absence of knowledge is duly explained. It is not hard to see that if Sosa’s story handles fake barn cases in a satisfactory manner, it is bound to get him into trouble with Frankfurt cases and other counterexamples to safety. Let’s return to Frankfurt Clock. In this case, your belief that it is 8:22 is not safe. There are some very close worlds at which you come down the stairs a minute earlier or later. At those worlds, the demon manipulates the clock to read ‘8:22’ anyway. You form a false belief that it is 8:22. Since your belief is unsafe, it falls short of reflective knowledge and knowledge full well. Since human knowledge requires knowledge full well, Sosa’s account incorrectly predicts absence of (human) knowledge here. Now, at some point, Sosa suggests to accommodate conflicting intuitions about certain cases in terms of their sensitivity to different levels of knowledge (Sosa 2015: 81). Couldn’t he also use this strategy here? That is to say, couldn’t he maintain the intuition of the absence of knowledge in the fake barn case picks up on the absence of knowledge full well, while the intuition of the presence of knowledge in Frankfurt Clock picks up on the presence of animal knowledge? On reflection, there is reason to believe that the answer to these questions is no. As a first observation, Sosa claims that fully apt performance is “what really matters” and that “the fully desirable status for performances in general is full aptness” (Sosa 2015: 85). Applied to the epistemic case, this means that the state that’s of genuine epistemic value is knowledge full well. If so, it is unclear why we should ever be interested in mere animal knowledge and, consequently, why the presence of animal knowledge should ever trigger an intuition of knowledge when the state of genuine value, knowledge full well, could have been attained but wasn’t. Perhaps more importantly, Sosa’s explanation, if it works all, will do best in accounting for differences in intuitions across subjects. That is to say, it will work best in accounting for why some people are willing to grant that agents in fake barn cases know, while others aren’t. Those who are interested in animal knowledge have the intuition, whereas those interested in knowledge full well don’t. (It may also be worth noting that this is how Sosa envisages the explanation to work.)

102 The Safety Dilemma What is less plausible is that this kind of explanation will serve to explain divergent intuitions about cases in a single subject. It is less plausible that it will successfully explain why one and the same person has an intuition of presence of knowledge in some cases and an intuition of absence of knowledge in other cases. After all, for the explanation to work, one and the same person would have to pick up on the presence of animal knowledge in one case and the absence of knowledge full well in the other. Why should that be? Why should I be picking up on the presence of your animal knowledge that it is 8:22 in Frankfurt Clock, and the absence of your knowledge full well that you are facing a barn in Fake Barns? Even if we bracket the issue of why we should ever be interested in anything other than human knowledge, the obvious answer in terms of a difference in interest carries little promise. Why should I be interested in animal knowledge in the former and knowledge full well in the latter? Finally, to see just how difficult it will be to offer a satisfactory answer to these questions, note that the intuitions remain even when the cases are considered side-by-side and the judges are confident that what they are interested in and what they are assessing is the presence or absence of the state of genuine value, i.e. human knowledge. For instance, when considering Fake Barns and Frankfurt Clock side-by-side, I am inclined to say that you don’t know that you are facing a barn, while you do know that it is 8:22. I am also very confident that what I am interested in and what I am assessing is the presence or absence of human knowledge. That is to say, I am inclined to say that you don’t have human knowledge that you are facing a barn, while you have human knowledge that it is 8:22. For the envisaged explanation to do the trick here, it would have to be the case that, despite all this, I am picking up on animal knowledge in Frankfurt Clock and on knowledge full well in Fake Barns. Why should that be the case? I cannot see even a remotely plausible answer on the horizon.1 Without a satisfactory answer to this question, however, Sosa does not successfully escape the safety dilemma.

Greco To see how Greco ventures to deal with fake barn cases, note that he maintains that attributions of abilities afford a contextualist semantics: context determines which ability to attain success, S, in conditions C is picked out by ‘ability to S’. In the case of attributions of the epistemic abilities at issue in VR-K, the virtue epistemological account of knowledge, the story is even more complex: context first determines a practical reasoning context, which may be the agent’s, the attributors’ or some third party’s. The relevant epistemic ability picked out by ‘epistemic ability’ is then fixed in accordance with what Greco takes to be a core function of the concept of knowledge, to wit, flagging actionable information and

The Safety Dilemma 103 sources of information, where what is actionable is determined by the practical reasoning context fixed at the first step (Greco 2010: 78–9). Greco’s account of fake barn cases exploits the context sensitivity of attributions of ability. More specifically, Greco aims to show that in the context that obtained when we considered Fake Barns and in which we found it intuitive that you do not know that you are looking at a barn, the semantic value of ‘ability to form true beliefs about the presence of barns’ is something like ability to form true beliefs about the presence of barns while moving through an area featuring mostly indistinguishable fakes. The idea here is that the epistemic ability’s C include the wider area in which you find yourself and in which fake barns predominate. Moreover, the ability’s C are not restricted to circumstances in which you happen to look out of the window just when you pass a real barn, but also include circumstances in which you look out of the window a little bit earlier or later and end up looking at a fake barn. But now notice that, given that the C are thus understood, the majority of nearby worlds at which you are in these C and form a belief about the presence of a barn are worlds at which you end up looking at a fake barn. Since, at those worlds, you end up with a false belief that you are looking at a barn, it is not the case that you attain true beliefs about the presence of barns at a high rate across nearby worlds in these C. On Greco’s account of abilities, then, you do not have the ability to form true beliefs about the presence of barns relative to these C. In consequence, in our context, the attribution of the relevant epistemic ability to you is false and the denial of knowledge is true. But now notice that, on Greco’s account, sentences of the form ‘S knows that p’ and ‘S does not know that p’ do not express propositions unless both of the following have been determined: (i) a practical reasoning context and (ii) what counts as actionable information in the context at issue. While Greco effectively endorses this consequence of his view (Greco 2010: 79–80), it does appear to generate some trouble for him.2 To see this, notice that the description of Fake Barns appears to leave both of these issues wide open. But given that this is so, the sentence ‘You do not know that you are looking at a barn’ doesn’t express a proposition in the context that obtained when you first read Fake Barns. At the same time, you agreed (if only implicitly), with my assertion, that I took it to be agreed that you do not know that you are looking at a barn. Now, this is rather puzzling. If Greco is right and ‘You do not know that you are looking at a barn’ does not express a proposition in that context, we’d expect competent speakers of English like yourself to disagree with this assertion. After all, the object of agreement is a proposition. If ‘You do not know that you are looking at a barn’ doesn’t express a proposition, there simply is no suitable object for agreement. We’d expect competent speakers of English to notice this and to refuse to accept my assertion. If this isn’t immediately obvious, compare: You are standing in a room

104 The Safety Dilemma with fifty paintings by a variety of artists. I walk in and assert, ‘I take it to be agreed that this is a Picasso’ offering no indication of what ‘this’ is supposed to refer to. Given that you are a competent speaker of English, we would expect you to disagree with my assertion. If you decided to respond, we would not expect you to say, ‘Yes, you are right.’ Rather, we’d expect you, for instance, to point out that before you can agree you will have to know which painting I have in mind. In other words, we’d expect you to spot the problem and refuse to accept the assertion. For that reason, the fact that you agreed with my assertion about you suggests that ‘You don’t know that you are looking at a barn’ did express a proposition when I asserted ‘I take it to be agreed that you do not know that you are looking at a barn’ in the context that obtained when you first read Fake Barns. Greco thus faces a difficulty. I assume that Greco’s response to this difficulty is that there is a default practical reasoning context and a default standard for actionable information, which are in play in contexts in which no explicit information about the practical reasoning context and what counts as actionable information is provided. If so, then it is the default practical reasoning context and standard for actionable information that are in play in the context that obtained when you agreed that you do not know that you are facing a barn upon reading Fake Barns. For Greco’s account of the absence of knowledge in Fake Barns to work, the C at issue in the default practical reasoning context and given default standards for actionable information must be wide enough to include the wider area you find yourself in as well as circumstances in which you look out of the window a bit earlier or later and end up looking at a fake barn. With these points in play, let’s return to Frankfurt Clock. Notice that, here too, we have not been given explicit information about the practical reasoning context and the standards for actionable information. Yet, again we have no difficulties in agreeing that you know. We may thus assume that the default practical reasoning context and standard for actionable information obtain. Now, crucially, if, in Fake Barns, the C at issue in the default practical reasoning context and given default standards for actionable information are wide enough to include circumstances in which you look out of the window a bit earlier or later and end up looking at a fake barn, then, in Frankfurt Clock, the C that are in play by default will be wide enough to include circumstances in which you look at the clock a bit earlier or later. Of course, in those cases, the demon shows his hand and you end up with a false belief about the time. The difficulty for Greco is that the majority of nearby worlds are worlds at which one of these circumstances obtains. It is thus not the case that you attain true beliefs about the time at a high rate across nearby worlds in the relevant C. On Greco’s account of abilities, then, you do not have the ability to form true beliefs about the time relative to these C. The

The Safety Dilemma 105 attribution of the relevant epistemic ability to you is false in the context that obtained when you read Frankfurt Clock and, in consequence, the denial of knowledge is true in that context. Greco’s account of the fake barn case, if successful, will lead him to make the wrong verdict in Frankfurt Clock. Greco thus does face the dilemma.

Pritchard Finally, consider Pritchard’s anti-luck virtue epistemology: Anti-Luck Virtue Epistemology (ALVE). S knows that p if and only if S’s safe true belief that p is the product of her relevant cognitive abilities (such that her safe cognitive success is to a significant degree creditable to her cognitive agency). (Pritchard 2012a: 273) Here is how Pritchard unpacks safety: Safe Belief. S’s belief is safe if and only if in most near-by possible worlds in which S continues to form her belief about the target proposition in the same way as in the actual world, and in all very close nearby possible worlds in which S continues to form her belief about the target proposition in the same way as the actual world, her belief continues to be true.3 It is now easy to see that your belief in Fake Barns is unsafe. Given the predominance of fake barns in your environment, at the majority of nearby possible worlds at which you continue to form a belief about the presence of a barn in the same way as in the actual world, you will be looking at a fake barn and end up with a false belief. Since safety doesn’t tolerate false belief at most nearby worlds, your belief turns out unsafe. It would seem, then, that ALVE can explain the absence of knowledge in fake barn cases. What about counterexamples to safety such as Frankfurt Clock, then? It is easy to see that the beliefs of agents in cases of unsafe knowledge are just that: unsafe. For instance, in Frankfurt Clock, at the majority of nearby possible worlds, some of which are very close, you come down the stairs a minute or so earlier or later. At those worlds, the demon shows his hand with the result that you wind up with a false belief about whether it is 8:22. As a result, ALVE predicts, now incorrectly, that agents in these cases lack knowledge. ALVE also succumbs to the safety dilemma. In a number of recent pieces, Pritchard has taken on the challenge of addressing this problem of cases of unsafe knowledge. I will here focus on his response to my Frankfurt case.4 Pritchard’s general strategy is to mount a case that, in Frankfurt Clock, you don’t know that it is 8:22. More specifically, he offers a couple of arguments in support of this claim

106 The Safety Dilemma and an account of how those who have the intuition could have become entrapped in error. Here is his first argument: [You are] effectively finding out the time by looking at what is (for [you] anyway) a stopped clock, since whatever time [you come] downstairs the clock will say ‘8:22 am’. But one cannot gain knowledge about the time by consulting a stopped clock, even when one happens to form a true belief! (Pritchard 2012b: 187) I must confess that I don’t find this argument particularly convincing. To begin with, notice that it is a crucial part of the case that the clock is functioning properly and is thus not stopped. The issue of whether the clock is stopped is no more agent-relative than the issue of whether the earth revolves around the sun. It is no more plausible that, in Frankfurt Clock, the clock is stopped for you, than it is plausible that the sun revolves around the earth for Ptolemy. Of course, Pritchard is free to define an agent-relative concept of stopped, but then he evidently cannot appeal to the plausible idea that one cannot acquire knowledge from a stopped clock in order to argue that an agent who satisfies this agent-relative concept does not know. There is further reason to believe that Pritchard’s argument will not do the trick for him. To see this, consider the following variation of the case: Frankfurt Clock’. Your demon nemesis wants you to believe either that it is 8:22 or else that it is past 8:22. He has resolved that if you come down before 8:22 he will set the clock to 8:22 and if you come down at 8:22 or later, he will do nothing. You come down at 8:22, the demon remains inactive, and you acquire a true belief by taking a competent reading from a perfectly functioning clock. Just as in Frankfurt Clock, you also know that it is 8:22 here. Moreover, your belief remains unsafe. After all, you might so easily have come down the stairs a minute earlier, in which case the demon would have manipulated the clock with the result that the reading would have been inaccurate. Finally, however, in the present version of the case, the clock is not stopped for you in the sense envisaged by Pritchard. To see this, note that it is not the case that whatever time you come downstairs, the clock will read ‘8:22’. You might very easily have come down a minute later, in which case the demon would also have remained inactive and the clock would have read ‘8:23’. If so, Pritchard’s response is bound to fail anyway. Let’s move on to Pritchard’s second argument that you do not know. Here the idea is that since Frankfurt Clock is structurally analogous to

The Safety Dilemma 107 Fake Barns and since, in Fake Barns, you lack knowledge, we should accept that the same is true in Frankfurt Clock (Pritchard 2009: 40). There are a number of problems with this argument. First and foremost, as I argued in Chapter 3, there is reason to believe that the two cases are actually not structurally analogous. But let’s bracket this point for a moment. Even if the two cases are relevantly analogous, it is far from clear that the lesson to be learned is that, in Frankfurt Clock, you lack knowledge rather than that, in Fake Barns, you have knowledge. At the very least, this point affords argument. Unfortunately, Pritchard fails to deliver on this front. That said, still assuming that the two cases are indeed analogous, Pritchard might be able to discharge at least part of the burden that he now finds himself settled with by offering a plausible error theory for the intuition that you know in Frankfurt Clock. In order to achieve this, Pritchard distinguishes between knowledge on the one hand and cognitive achievement on the other. A cognitive achievement is, roughly, a cognitive success that is primarily creditable to cognitive ability. Knowledge, in contrast, features both a weaker ability condition and the safety condition (see above). The idea then is that, while knowledge and cognitive achievement often go hand-in-hand, they can sometimes come apart. Cases like Frankfurt Clock constitute one type of case in which they do. Those who have the intuition that you have knowledge in this case mistake cognitive achievement for knowledge (Pritchard 2009: 40, 2012b: 183, 2013: 160). There are two problems with Pritchard’s error theory. To see the first, notice that what Pritchard owes us, at this stage, is an explanation of why our intuition leads us astray in Frankfurt Clock but not in Fake Barns. Pritchard’s story does not achieve this. After all, Fake Barns, too, is a case of cognitive achievement without knowledge. If, in such cases, we are prone to mistake cognitive achievement for knowledge, the question remains why we don’t do so in Fake Barns also. The issue that affords diagnosis—why intuition leads us astray in Frankfurt Clock but not in Fake Barns—thus remains unresolved. Pritchard’s diagnosis fails on yet another and more significant count: there are variations of Frankfurt Clock in which you do not secure a cognitive achievement, at least not by Pritchard’s own lights. To see this, notice that, besides cases of cognitive achievement without knowledge, Pritchard also grants that there are cases of knowledge without cognitive achievement. Most importantly, cases of testimonial knowledge in which the greatest part of the cognitive work is done by the testifier and not the receiver of testimony fall under this heading (Pritchard 2008: 446 and Pritchard et al. 2010: 40–3). That means that if we can offer a (suitable) variation of Frankfurt Clock involving testimonial knowledge in which the bulk of the cognitive work is done by the testifier, Pritchard is committed to treating the case as one in which the agent does not

108 The Safety Dilemma secure a cognitive achievement. If so, his diagnosis of Frankfurt Clock in terms of mistaking cognitive achievement for knowledge is bound to fail. This leaves us with the task of providing a suitable case. Here goes: Frankfurt Clock”. As you come down the stairs, your little cousin, who cannot yet read the clock, asks you what time it is. You look at the clock, see that it reads 8:22 and tell your cousin that it is 8:22. On that basis, your cousin acquires a true belief that it is 8:22. Unbeknownst to both of you, your cousin’s demon nemesis was lurking in the background, prepared to set the clock to 8:22 had you not happened to look at it at just the right time, i.e. at precisely 8:22. Your cousin’s belief qualifies as knowledge. At the same time, this belief is not safe. At many nearby possible worlds, including very close ones, at which your cousin acquires the belief via testimony from you, you come down the stairs a minute earlier or later. Since, in that case, the demon shows his hand, at those worlds your cousin ends up with a false belief. Crucially, the bulk of the cognitive work is done by the testifier (here, you). After all, you are the one who has the crucial cognitive ability in this case. If so, by Pritchard’s lights, the belief about the time your cousin forms does not qualify as a cognitive achievement. We thus have a case of unsafe knowledge without cognitive achievement. Pritchard’s diagnosis of the intuition of knowledge in terms of mistaking cognitive achievement for knowledge fails on yet another count. Despite Pritchard’s best efforts, ALVE does not manage to escape the safety dilemma.

Notes 1 I do not mean to say that it is impossible to imagine some such explanation. A contextualist semantics for ‘human knowledge’ according to which, when considering Frankfurt Clock, ‘human knowledge’ means animal knowledge and, when considering Fake Barns ‘human knowledge’ means knowledge full well, in conjunction with a good dose of semantic blindness ensuring that I am blissfully unaware of the context shift as I move from one case to the other might do the trick. However, I take this move to be so implausible that it can safely be set aside. 2 In fact, there is reason to believe that it is a vital feature of our concept of knowledge that it is applicable without information about the specificities of some practical reasoning context (Craig 1990). 3 Pritchard (2007: 292, 2009: 34). It may be worth noting that, in more recent work Pritchard has moved to a full-blown sliding scale account of the tolerance for error at nearby worlds. The idea is that at nearby worlds closest to actuality, safety tolerates no false beliefs and the further we move away from

The Safety Dilemma 109 actuality the more false beliefs safety is compatible with. However, Pritchard has not given a precise statement of safety in recent work. Note also that, in recent references to “versions of the safety principle” in the literature, Pritchard (2012b: 170, fn.3, 2013: n.10) lists Safe Belief as the last version in his own work. 4 Note that, for present purposes, this should be perfectly fine. After all, if this response remains unsatisfactory, the safety dilemma remains unresolved.

Lottery Cases

In Chapter 3 I sketched three ways of dealing with the apparent absence of knowledge in lottery cases such as Lottery Loser and No Winner. I provided theoretical reason for thinking that, according to the right approach to these cases, we do not even have justification for believing lottery propositions and that that’s why we don’t know them to be true either. In what follows, I will once more look at a number of the most prominent virtue reliabilist treatments of lottery cases.

Sosa Perhaps somewhat surprisingly, Sosa accepts that we do know in lottery cases. In order to make this response palatable—if only to those of us who, like me, find it entirely implausible—Sosa does two things. First, he produces some evidence that the likes of lottery propositions can be commonsensically asserted. Second, he offers an error theory for the intuition that we don’t know lottery propositions. Let’s first look at the relevant linguistic evidence, which can be found in the following passage: I have claimed that lottery propositions are knowable, which may provoke incredulous outrage. And yet, about someone’s infidelity after twenty-five years of marriage, I might say: “You see, you never know!” This I might say commonsensically even if it would be wrong to add that no one ever knows that their spouse is faithful [E1]. And to someone in tears because their hopes were dashed when their ticket lost in a zillion-ticket lottery, we can say: “Why are you surprised? You should have known that you would lose. You knew the odds!” [E2] And to someone who trusted their shaman’s prediction that their ticket would win, and is surprised when it loses, we can say: “Aw, come on, you should have known better.” [E3] Finally, consider someone anxiously holding their ticket as they look expectantly at the screen where the winner is about to be announced. It seems perfectly appropriate to tell them, seriously and with no hint

Lottery Cases 111 of jocularity: “Don’t set yourself up for disappointment. Trust me. Your ticket has lost.” [E4] (Sosa 2015: 122) As a first observation, note that what matters is not what we might commonsensically say. To see this, note that we might commonsensically say lots of false things. What matters is not what one might commonsensically say, but what one can correctly say. If we want to use what we might say as evidence for how things are, we had better at least consult our wellconsidered judgements on the relevant assertions and check whether the assertion is appropriate given our well-considered judgements. With this point in play, let’s take a closer look at Sosa’s evidence. There are four pieces, which I have labelled [E1] – [E4]. I’d like to say straightaway that I just don’t agree that with Sosa on [E2] and [E4]. In particular, given a well-considered judgement, “You should have known that you would lose” and “Your ticket has lost” cannot be appropriately asserted. [E2] and [E4] don’t serve to support Sosa’s case. In the case of [E3], it is not even clear what the content of the assertion is. Of course, “You should have known better” might express the proposition that you should have known that your ticket will lose. However, it might also express the proposition that you should have known not to trust your shaman. If it is clear that the assertion expresses the former proposition, I am again unwilling to concede that it is a proposition that can be appropriately asserted, given a well-considered judgement. If it expresses the latter proposition, it is entirely appropriate, even after wellconsidered judgement. However, in that case it provides no support for Sosa’s case. Finally, [E1] does not do better either. At the very best, it serves to show that the mere fact that we say things like “You never know” in lottery cases does not provide evidence that people don’t know that tickets won’t lose. But that’s of course not evidence for Sosa’s view. In fact, on reflection, there is reason to think that the relevant linguistic evidence tells against Sosa. To see this, notice first that “You see you never know” has the form of a generic just like “Mosquitos transmit malaria.” Some such sentences (call them ‘mere generics’) can be true even if some (perhaps many) of its instances are false. “Mosquitos transmit malaria” is true, even though “This mosquito transmits malaria” is false. Similarly, “You see you never know” in Sosa’s case may be true even if “X doesn’t know that his spouse has been faithful to him” is false. However, other sentences that have the form of generics express genuine universal generalisations. “Ravens are black” is one example. If so, the reason why “You never know” is true in lottery cases could be that it expresses a mere generic or a genuine universal generalisation. There is excellent reason to believe that it is the latter. Just as “This raven is black” is true of all ravens we have ever encountered, “X knows

112 Lottery Cases that this ticket won’t win” is true of all agents we have ever encountered who only have the probabilistic evidence against winning at their disposal. Let’s move on, then, to the second part of Sosa’s story on lottery propositions, to wit, his error theory. The key idea here is that in taking lottery propositions to be unknowable, we are confusing the true safety condition on knowledge with the false sensitivity condition. Recall that, according to safety, one knows that p only if were one to believe that p, p would be true (Bp € p). Sensitivity is the contrapositive of the counterfactual at issue in safety. According to sensitivity, then, one knows that p only if were p false, one would not believe that p (¬p € ¬Bp). Crucially, while material conditionals contrapose, counterfactual conditionals don’t. Hence, safety does not entail sensitivity. As a result of missing this important fact about counterfactuals, we are “misled into accepting a sensitivity condition by confusing it with a safety condition.” (Sosa 2015: 120). More specifically, Sosa wants to allow that lottery propositions satisfy safety. For instance, he grants that, in Lottery Loser, your belief that my ticket won’t win is such that were you to believe it won’t win, it would be the case that it won’t win. However, beliefs in lottery propositions are not sensitive. To see this, consider, once more, your belief that my ticket won’t win. If my ticket were to win, you would still believe that it won’t. Hence, your belief is not sensitive. Even though you really know that my ticket won’t win, we are confusing the true safety with the false sensitivity condition and so are led to the mistaken intuition that you do not know that my ticket won’t win. There are a number of problems with Sosa’s account. First, as Frankfurt cases suggest, there is reason to believe that we don’t even accept a safety condition on knowledge. If so, there is no basis for being misled into accepting the sensitivity condition on knowledge in terms of which Sosa wants to explain our error. But even if we bracket this issue, Sosa’s error theory gets him into trouble elsewhere. To see this, consider the following case: [Take] a world where apple has grown so large that a million operators stand by constantly to provide answers to any simple apple question, and suppose technology randomizes in such a way that there are a million-minus-one operators no farther modally from me than the one I actually reach when I place my call. And suppose there is one bad apple in the lot, a liar who would give me an incorrect answer. But suppose further that apple’s service continues to be great, so that all the other million-minus-one operators are infallibly reliable in their answers to such questions. Is that enough to preclude me from learning the answer to my question from the operator I actually reach? (Sosa 2015: 119)

Lottery Cases 113 Crucially, as Sosa himself acknowledges, the answer to his last question is no. That is to say, the agent does have knowledge in this case. But now note that this case has exactly the same modal profile as the lottery case. Both cases are cases of safe but insensitive belief. If we are prone to confusing safety and sensitivity and if this leads us to mistakenly judge knowledge to be absent in the lottery case, we would expect to be led to a mistaken judgement that knowledge is absent in this case, too. However, this is not what we find. As I already said, we grant the agent knowledge in this case. If successful in lottery cases, Sosa’s error theory thus threatens to make mistaken predictions in other cases, including his own case of the bad apple. Of course, Sosa might want to say that in the latter case we simply do not fall prey to the confusion. But this simply raises the question why our tendency to confuse safety and sensitivity leads us to mistaken judgements in some cases (e.g. Lottery Loser) but not in others (e.g. the bad apple case). Until Sosa has provided a satisfactory answer to this question, his error theory for lottery cases remains unsatisfactory.

Greco Recall that, according to Greco, a belief is apt just in case epistemic ability saliently explains why one believes truly. Crucially, according to Greco, salient luck may trump the salience of epistemic abilities. This is exactly what, according to Greco, happens in lottery cases. More specifically, Greco claims that “the very idea of a lottery has the idea of chance built right into it” (2003b: 124). For instance, in Lottery Loser, in conversational contexts in which your belief in is under discussion, luck becomes salient. It trumps the salience of epistemic ability so that what is salient in the explanation of why you truly believe is not your epistemic ability but the fact that you were lucky enough to believe in the way that happened to correspond with the facts (e.g. 2003b: § IV). Hence, Greco’s view predicts that ‘you know that my ticket won’t win l0 is bound to express a falsehood, thus explaining the absence of knowledge in our toy case.1 Unfortunately, there is reason to think that Greco’s account of lottery cases remains unsatisfactory. To see this, consider the following variation of the case: Rigged Lottery. I have bought a ticket in a lottery with one million tickets and exactly one winner, l00 . In view of the low chances of winning, you believe that my ticket won’t win l00 (). Unbeknownst to me and you, l00 has been rigged by X and Y. They have made arrangements that no ticket will win l00 . Just as in the original version of the case, you don’t know , even though your belief is true and based on excellent evidence. At the same

114 Lottery Cases time, there is excellent reason to think that Greco will struggle to explain the absence of knowledge in this version of the case. To see this, notice that l00 is evidently not a chancy event. After all, the outcome has been fixed in advance. No ticket is going to win. Greco’s core claim that “the very idea of a lottery has the idea of chance built right into it” turns out to be incorrect.2 As a result, it should now be possible to construct conversational contexts in which luck is not salient. Here is one: X and Y are discussing whether anyone has got wind of their foul play. X knows that you believe and worries that you might believe this because you know that l00 is fixed. However, Y is quick to appease X’s worry: “They don’t know that Kelp’s ticket won’t win. They believe that Kelp’s ticket won’t win because of the low chances of winning. They have no inkling that the lottery is rigged.” What is salient in this context, if anything, is the non-luckiness of my ticket not winning the lottery. What Y effectively points out is that your belief falls short of knowledge despite the fact that it is not a matter of luck that my ticket will lose. Luck is simply not salient in this context. As a result, Greco’s explanation fails. Salient luck does not trump the salience of epistemic abilities here. In fact, his account predicts, mistakenly, that in the present context “you know that my ticket won’t win” expresses a truth and Y’s claim that you don’t know turns out false.

Pritchard To see why ALVE’s account of these cases remains unsatisfactory, recall first Sosa’s case of the bad apple. As Sosa (2015: 118–9) forcefully argues, this case forces ALVE into a dilemma. After all, we have seen that Sosa’s case has the same modal profile as a lottery case. At the same time, unlike in lottery cases, the agent in Sosa’s case knows what he is told. In consequence, ALVE faces the following dilemma: either the safety condition is construed as strong enough to ensure that, in Lottery Loser, you do not satisfy it. In that case, it can correctly explain the absence of knowledge. However, since the bad apple case has the same modal profile as Lottery Loser, the agent in the bad apple case is bound not to satisfy this strong safety condition either. That is to say, if ALVE construes safety as strong enough to correctly predict absence of knowledge in Lottery Loser, it is bound to make incorrect predictions in the bad apple case. Alternatively, Pritchard can construe the safety condition as weak enough to allow that the agent in the bad apple case satisfies it. However, since the lottery case has the same modal profile, it follows that the agent in the lottery case satisfies this weak safety condition also. That means that if ALVE construes safety as weak enough to allow for knowledge in Sosa’s case, it is bound to be too weak to explain the absence of knowledge in Lottery Loser. Either way, then, ALVE is in trouble.

Lottery Cases 115 What’s more, there is reason to believe that even on the strong construal of safety, safety is too weak to explain the absence of knowledge in lottery cases. To see this, consider again the case of a rigged lottery. In particular, note that a lottery, l000 may be safely rigged, perhaps only in the sense that it is ensured that a particular person’s ticket, mine say, would not win, no matter which ticket I might buy. If so, when you acquire a belief that my ticket won’t win l000 , you will acquire a safe belief.3 However, when the belief is based only on the high odds against winning, it falls short of knowledge. Or consider a case in which you believe, again based on the high odds against winning, that the ticket I bought won’t make me rich. Your belief falls short of knowledge. Compatibly with that, my ticket may have been destroyed since I bought it. So suppose it has been. If so, there will be no close possible worlds at which it will make me rich. Your belief once again ends up safe. It comes to light that the safety condition is too weak to explain the absence of knowledge in all lottery cases. ALVE thus does not manage to offer a satisfactory account of lottery cases.4

Notes 1 It is worth noting that Greco does not discuss beliefs about lottery losers in his 2010 book. However, he develops the account outlined above in a number of earlier papers (Greco 2003a,b, 2004). What’s more, this account is still endorsed in Greco & Turri (2013: §6). 2 In fact, it is not hard to see exactly how Greco went wrong. While the idea of a fair lottery is by its nature the idea of a chancy event, the idea of an unfair or rigged lottery clearly isn’t. 3 Note that Pritchard himself effectively acknowledges as much. By way of evidence, consider the following passage in which he describes a recipe of generating cases in which the agent has safe beliefs that fall short of knowledge: Imagine an agent forming her beliefs about a certain subject matter on an epistemically problematic basis, and then stipulate some feature of the environment—a helpful demon, for example—whose overriding concern is to ensure that what the subject believes is true, to the extent that the evil demon changes the facts to ensure that what the subject believes is the case. Here we have the right kind of modal profile for the subject’s true belief to be safe, in that this is a true belief that could not easily have been a false belief . . . (Pritchard 2014: 158) Your belief that my ticket won’t win l000 fits the description of Pritchard’s recipe like a glove. As a result, we may expect Pritchard to have no qualms with the safety of his belief in this case. 4 Couldn’t Pritchard venture to explain your lack of knowledge in this case in terms of the virtue condition on knowledge? No. To see why, notice that Pritchard endorses a weakened version of Greco’s virtue condition on knowledge. However, as we have seen above, even Greco’s version of this condition is too weak to do the trick in lottery cases. In fact, Pritchard himself acknowledges as much: First off, notice how strong virtue epistemology struggles with the lottery case. After all, [your epistemic] success does seem to be explained by the

116 Lottery Cases exercise of her relevant [epistemic] abilities, doesn’t it? Put another way, if it is not [your epistemic] ability that explains [your epistemic] success, then what does explain it? For note that the only plausible candidate here is the (epistemic) good fortune that [my] ticket is indeed a losing ticket. However, given the odds involved, it is hard to see how this eventuality could be considered a matter of fortune at all. (Pritchard 2012a: 266–7) As a result, the prospects of explaining the absence of knowledge in lottery cases in terms of the virtue condition are dim, to say the least. The fact that the safety condition is not up to the task either means that these cases do constitute a genuine obstacle for ALVE.

Bibliography

Aikin, S. 2005. Who is afraid of epistemology’s regress problem. Philosophical Studies, 126, 191–217. Aikin, S. 2008. Meta-epistemology and the varieties of epistemic infinitism. Synthese, 163, 175–85. Aikin, S. 2011. Epistemology and the Regress Problem. London: Routledge. Alston, W. 1988. Justification and knowledge. In: Epistemic Justification. Ithaca, NY: Cornell University Press. Alston, W. 1995. How to think about reliability. Philosophical Topics, 23, 1–29. Audi, R. 2001. An internalist theory of normative grounds. Philosophical Topics, 23, 31–45. Beddor, B. 2015. Process reliabilism’s troubles with defeat. Philosophical Quarterly, 65, 145–59. Beebe, J. 2004. The generality problem, statistical relevance and the tri-level hypothesis. Noûs, 38, 177–95. Bird, A. 2007. Justified judging. Philosophy and Phenomenological Research, 74, 81–110. Bishop, M. 2010. Why the generality problem is everybody’s problem. Philosophical Studies, 151, 285–98. Bogardus, T. 2013. Knowledge under threat. Philosophy and Phenomenological Research, 86, 289–313. BonJour, L. 1980. Externalist theories of empirical knowledge. Midwest Studies in Philosophy, 5, 53–73. BonJour, L. 1985. The Structure of Empirical Justification. Cambridge, MA: Harvard University Press. BonJour, L., & Sosa, E. 2003. Epistemic Justification: Internalism vs. Externalism, Foundations vs. Virtues. Malden, MA: Blackwell. Cappelen, H. 2012. Philosophy Without Intuitions. Oxford: Oxford University Press. Carter, A. 2016. Robust virtue epistemology as anti-luck epistemology: a new solution. Pacific Philosophical Quarterly, 97, 140–55. Carter, A., Jarvis, B., & Rubin, K. 2015. Varieties of cognitive achievement. Philosophical Studies, 172, 1603–23. Chisholm, R. 1977. Theory of Knowledge. 2nd edn. Englewood Cliffs: Prentice Hall. Cohen, S. 1984. Justification and truth. Philosophical Studies, 46, 279–96. Colaço, D., Buckwalter, W., Stich, S., & Machery, E. 2014. Epistemic intuitions in fake-barn thought experiments. Episteme, 11, 199–212.

118 Bibliography Comesaña, J. 2005. Unsafe knowledge. Synthese, 146, 395–404. Comesaña, J. 2006. A well-founded solution to the generality problem. Philosophical Studies, 129, 27–47. Conee, E., & Feldman, R. 1998. The generality problem for reliabilism. Philosophical Studies, 89, 1–29. Conee, E., & Feldman, R. 2001. Internalism defended. In: Kornblith, H. (ed), Epistemology: Internalism and Externalism. Malden, MA: Blackwell, pp. 231–60. Conee, E., & Feldman, R. 2004. Evidentialism: Essays in Epistemology. Oxford: Oxford University Press. Craig, E. 1990. Knowledge and the State of Nature. An Essay in Conceptual Synthesis. Oxford: Clarendon Press. Deutsch, M. 2015. The Myth of the Intuitive: Experimental Philosophy and Philosophical Method. Cambridge, MA: MIT Press. Devitt, M. 2006. Ignorance of Language. Oxford: Clarendon Press. Devitt, M. 2011. Experimental semantics. Philosophy and Phenomenological Research, 82, 418–35. Douven, I. 2002. A new solution to the paradoxes of rational acceptability. British Journal for the Philosophy of Science, 53, 391–410. Douven, I. 2009. Assertion, Moore and Bayes. Philosophical Studies, 144, 361– 75. Douven, I., & Williamson, T. 2006. Generalising the Lottery Paradox. British Journal for the Philosophy of Science, 57, 755–79. Foley, R. 1979. Justified inconsistent beliefs. American Philosophical Quarterly, 16, 247–57. Foley, R. 2012. When Is True Belief Knowledge? Princeton, NJ: Princeton University Press. Frankfurt, H. 1969. Alternate possibilities and moral responsibility. The Journal of Philosophy, 66, 829–39. Gettier, E. 1963. Is justified true belief knowledge? Analysis, 23, 121–3. Ghijsen, H. 2016. The Puzzle of Perceptual Justification: Conscious Experience, Higher-Order Beliefs, and Reliable Processes. Dordrecht: Springer. Ginet, C. 1975. Knowledge, Perception, and Memory. Dordrecht: Reidel. Goldman, A. 1976. Discrimination and perceptual knowledge. Journal of Philosophy, 73, 771–91. Goldman, A. 1979. What is justified belief? In: Pappas, G. (ed), Justification and Knowledge. Dordrecht: Reidel. Goldman, A. 1986. Epistemology and Cognition. Cambridge, MA: Harvard University Press. Goldman, A. 1988. Strong and weak justification. Philosophical Perspectives, 2, 51–69. Goldman, A. 1993. Epistemic folkways and scientific epistemology. Philosophical Issues, 3, 271–85. Goldman, A. 2008. Immediate justification and process reliabilism. In: Smith, Q. (ed), Epistemology: New Essays. Oxford: Oxford University Press. Goldman, A. 2012. Reliabilism and Contemporary Epistemology. New York: Oxford University Press. Graham, P. 2012. Epistemic entitlement. Noûs, 46, 449–82. Greco, J. 1999. Agent reliablism. Philosophical Perspectives, 33, 273–96.

Bibliography 119 Greco, J. 2000. Putting Skeptics in Their Place: The Nature of Skeptical Arguments and Their Role in Philosophical Inquiry. Cambridge: Cambridge University Press. Greco, J. 2003a. Further thoughts on agent reliabilism: replies to Cohen, Geivett, Kvanvig and Schmitt and Lahroodi. Philosophy and Phenomenological Research, 66, 466–80. Greco, J. 2003b. Knowledge as credit for true belief. In: DePaul, M., & Zagzebski, L. (eds), Intellectual Virtue: Perspectives from Ethics and Epistemology. Oxford: Oxford University Press. Greco, J. 2004. How to preserve your virtue while losing your perspective. In: Greco, J. (ed), Ernest Sosa and His Critics. Oxford: Blackwell. Greco, J. 2007. The purpose of knowledge and the nature of ability. Philosophical Issues, 17, 57–69. Greco, J. 2010. Achieving Knowledge. Cambridge: Cambridge University Press. Greco, J. 2012. A (different) virtue epistemology. Philosophy and Phenomenological Research, 85, 1–26. Greco, J., & Turri, J. 2013. Virtue epistemology. In: Zalta, E.N. (ed), The Stanford Encyclopedia of Philosophy, winter 2013 edn. Grundmann, T. 2010. Some hope for intuitions. A reply to Weinberg. Philosophical Psychology, 23, 481–509. Haji, I. 1998. Moral Appraisability. Oxford: Oxford University Press. Hales, S. 2006. Relativism and the Foundations of Philosophy. Cambridge, MA: MIT Press. Heller, M. 1995. The simple solution to the generality problem. Noûs, 29, 501–15. Hetherington, S. 1999. Knowing failably. The Journal of Philosophy, 96, 565–87. Hofmann, F. 2010. Intuitions, concepts, and imagination. Philosophical Psychology, 23, 529–46. Horvath, J. 2010. How (not) to react to experimental philosophy. Philosophical Psychology, 23, 447–80. Jackson, F. 2011. On Gettier holdouts. Mind and Language, 26, 468–81. Kaplan, M. 1981. A Bayesian theory of rational acceptance. The Journal of Philosophy, 78, 305–30. Kelp, C. 2011a. In defence of virtue epistemology. Synthese, 179, 409–33. Kelp, C. 2011b. What’s the point of ‘knowledge’ anyway? Episteme, 8, 53–66. Kelp, C. 2013. Knowledge: the safe-apt view. Australasian Journal of Philosophy, 91, 265–78. Kelp, C. 2016. Justified belief: knowledge first-style. Philosophy and Phenomenological Research, 93, 79–100. Kelp, C., & Simion, M. 2017. Criticism and blame in action and assertion. Journal of Philosophy, 114, 76–93. Kipper, J. 2010. Philosophers and grammarians. Philosophical Psychology, 23, 511–27. Klein, P. 1999. Human knowledge and the infinite regress of reasons. Philosophical Perspectives, 13, 297–325. Klein, P. 2012. Infinitism and the epistemic regress problem. In: Toldsdorf, S. (ed), Conceptions of Knowledge. Berlin: De Gruyter. Klein, P. 2014. Infinitism is the solution to the regress problem. In: Steup, M., Turri, J., & Sosa, E. (eds), Contemporary Debates in Epistemology, 2nd edn. Malden, MA: Wiley-Blackwell.

120 Bibliography Kornblith, H. 1980. Beyond foundationalism and the coherence theory. Journal of Philosophy, 72, 597–612. Kornblith, H. 2002. Knowledge and Its Place in Nature. Oxford: Clarendon Press. Kornblith, H. 2004. Does reliabilism make knowledge merely conditional? Philosophical Issues, 14, 185–200. Kornblith, H. 2009. A reliabilist solution to the problem of promiscuous bootstrapping. Analysis, 69, 263–7. Kvanvig, Jonathan. 2007. Two approaches to epistemic defeat. In: Baker, D. (ed), Alvin Plantinga. Cambridge: Cambridge University Press, pp. 107–24. Kyburg, H. 1961. Probability and the Logic of Rational Belief. Middletown, CT: Wesleyan University Press. Kyburg, H. 1970. Conjunctivitis. In: Swain, M. (ed), Induction, Acceptance and Rational Belief. Dordrecht: Reidel. Kyburg, H. 1997. Probabilistic acceptance. In: Proceedings of the Thirteenth conference on Uncertainty in artificial intelligence. UAI’97, 326–33. Lackey, J. 2007. Why we don’t deserve credit for everything we know. Synthese, 158, 345–61. Lackey, J. 2009. Knowledge and credit. Philosophical Studies, 142, 27–42. Lehrer, K. 1974. Knowledge. Oxford: Clarendon Press. Lehrer, K. 1990. Theory of Knowledge. London: Routledge. Lehrer, K., & Cohen, S. 1983. Justification, truth, and coherence. Synthese, 55, 191–207. Littlejohn, C. 2013. The Russellian retreat. Proceedings of the Aristotelian Society, 113, 293–320. Littlejohn, C. 2017. How and Why Knowledge is First. In: Carter, A., Gordon, E., & Jarvis, B. (eds), Knowledge First: Approaches in Epistemology and Mind. Oxford: Oxford University Press. Littlejohn, C. Forthcoming. A plea for epistemic excuses. In: Dorsch, F., & Dutant, J. (eds), The New Evil Demon: New Essays on Knowledge, Justification and Rationality. Oxford: Oxford University Press. Ludwig, K. 2007. The epistemology of thought experiments: first person versus third person approaches. Midwest Studies in Philosophy, 31, 128–59. Lyons, J. 2009. Perception and Basic Beliefs. Zombies, Modules, and the Problem of the External World. Oxford: Oxford University Press. Millar, A. 2010. Perceptual knowledge and recognitional abilities. In: Pritchard, D., Millar, A., & Haddock, A. (eds), The Nature and Value of Knowledge. Oxford: Oxford University Press. Millikan, R. 2000. On Clear and Confused Ideas. Cambridge: Cambridge University Press. Miracchi, L. 2015a. Competence to know. Philosophical Studies, 172, 29–56. Miracchi, L. 2015b. Knowledge is all you need. Philosophical Issues, 25, 353–78. Neta, R., & Rohrbaugh, G. 2004. Luminosity and the safety of knowledge. Pacific Philosophical Quarterly, 85, 396–406. Peijnenburg, J. 2007. Infinitism regained. Mind, 116, 597–602. Plantinga, A. 1993. Warrant: The Current Debate. Oxford: Oxford University Press. Pollock, J. 1995. Cognitive Carpentry. Cambridge, MA: MIT Press. Pritchard, D. 2005. Epistemic Luck. Oxford: Oxford University Press. Pritchard, D. 2007. Anti-luck epistemology. Synthese, 158, 277–97.

Bibliography 121 Pritchard, D. 2008. Greco on knowledge: Virtues, contexts, achievements. The Philosophical Quarterly, 58, 437–47. Pritchard, D. 2009. Safety-based epistemology: Whither now? Journal of Philosophical Research, 34, 33–45. Pritchard, D. 2010. Cognitive ability and the extended cognition thesis. Synthese, 175, 133–51. Pritchard, D. 2012a. Anti-luck virtue epistemology. The Journal of Philosophy, 109, 247–79. Pritchard, D. 2012b. In defence of modest anti-luck epistemology. In: Becker, K., & Black, T. (eds), The Sensitivity Principle in Epistemology. Cambridge: Cambridge University Press. Pritchard, D. 2013. Knowledge cannot be lucky. In: Steup, M., Turri, J., & Sosa, E. (eds), Contemporary Debates in Epistemology, 2nd edn. Oxford: Blackwell. Pritchard, D. 2014. Epistemic luck, safety and assertion. In: Littlejohn, C., & Turri, J. (eds), Epistemic Norms: New Essays on Action, Belief and Assertion. Oxford: Oxford University Press. Pritchard, D. 2017. Knowledge, luck, and virtue: resolving the Gettier problem. In: de Almeida, C., Klein, P., & Borges, R. (eds), Explaining Knowledge: New Essays on the Gettier Problem. Oxford: Oxford University Press. Pritchard, D., Millar, A., & Haddock, A. 2010. The Nature and Value of Knowledge. Oxford: Oxford University Press. Ramsey, F. 1990/1929. F.P. Ramsey: Philosophical Papers. Cambridge: Cambridge University Press. Riggs, W. 2002. Reliability and the value of knowledge. Philosophy and Phenomenological Research, 64, 79–96. Riggs, W. 2007. Why epistemologists are so down on their luck. Synthese, 158, 329–44. Riggs, W. 2009. Two problems of easy credit. Synthese, 169, 201–16. Russell, B. 1948. Human Knowledge: Its Scope and Limits. London: Allen & Unwin. Ryan, S. 1996. The epistemic virtues of consistency. Synthese, 109, 121–41. Sartwell, C. 1991. Knowledge is mere true belief. American Philosophical Quarterly, 28, 157–65. Simion, M., Kelp, C., & Ghijsen, H. 2016. Norms of belief. Philosophical Issues, 26, 375–92. Smithies, D. 2012. Moore’s paradox and the accessibility of justification. Philosophy and Phenomenological Research, 85, 273–300. Sosa, E. 1980. The raft and the pyramid. Midwest Studies in Philosophy, 5, 3–25. Sosa, E. 1991. Knowledge in Perspective. Selected Essays in Epistemology. Cambridge: Cambridge University Press. Sosa, E. 2007. A Virtue Epistemology. Apt Belief and Reflective Knowledge. Vol. 1. Oxford: Oxford University Press. Sosa, E. 2010. How competence matters in epistemology. Philosophical Perspectives, 24, 465–75. Sosa, E. 2011. Knowing Full-Well. Princeton, NJ: Princeton University Press. Sosa, E. 2015. Judgment and Agency. Oxford: Oxford University Press. Steup, M. 1999. A defense of internalism. In: Pojman, L. (ed), The Theory of Knowledge: Classical and Contemporary Readings. Belmont, CA: Wadsworth. Sutton, J. 2005. Stick to what you know. Noûs, 39, 359–96.

122 Bibliography Sutton, J. 2007. Without Justification. Cambridge, MA: MIT Press. Turri, J. 2011. The express knowledge account of assertion. Australasian Journal of Philosophy, 89, 37–45. Turri, J. 2016. Knowledge as achievement, more or less. In: Fernandez, M. (ed), Performance Epistemology. Oxford: Oxford University Press. Turri, J., Buckwalter, W., & Blouw, P. 2015. Knowledge and luck. Psychonomic Bulletin and Review, 22, 378–90. Wedgwood, R. 2002. The aim of belief. Philosophical Perspectives, 16, 267–97. Williamson, T. 2000. Knowledge and Its Limits. Oxford: Oxford University Press. Williamson, T. 2007. The Philosophy of Philosophy. Malden, MA: Blackwell. Williamson, T. 2010. Knowledge First Epistemology. In: Bernecker, S., & Pritchard, D. (eds), The Routledge Companion to Epistemology. New York: Routledge, pp. 208–18. Williamson, T. 2011. Philosophical expertise and the burden of proof. Metaphilosophy, 42, 215–29. Wright, C. 2010. On intuitional stability: the clear, the strong, and the paradigmatic. Cognition, 115, 491–503. Zagzebski, L. 1994. The inescapability of Gettier problems. The Philosophical Quarterly, 44, 65–73. Zagzebski, L. 1996. Virtues of the Mind: An Inquiry into the Nature of Virtue and the Ethical Foundations of Knowledge. Cambridge: Cambridge University Press. Zimmermann, M. 1997. Moral responsibility and ignorance. Ethics, 107, 410–26.

Index

abilities 20, 24–28; ability/competence distinction 92–94; attributions of 102–105; and dispositions: 20, 34; epistemic 30–31, 67; exercise of: 28–31, 38–41, 83, 91–92, 95–96; grounding condition on: 26–28, 31–32; levels of: 44–46; SGP abilities: 24–29; range of 28, 30, 67, 78; reliability condition on: 20; shape and situational conditions on: 29, 44; as ways of performing: 25–26, 28. achievement 17–18, 46, 49, 107–108 Alston, W. 13 because relation: explanatory salience account of: 38–39; ability manifestation account of: 44–46; precisified ability manifestation account of: 46–47; see also creditworthiness dilemma, safety: dilemma Beebe, J. 13 belief: as epistemic performance 18–19; as moves in inquiry, 30, 70–71; success condition on: 19, 67; see also inquiry, justified belief Bishop, M. 13, 32 blamelessness 85–88, 94 Bogardus, T. 51 Comesana, J. 13, 15, 32, 50 competences see abilities creditworthiness dilemma 39–46 dispositions 20–27

Douven, I. 76–77 examples: archery 17–18, 20–29, 33, 38, 42, 44–46, 49, 69–72, 74, 78, 92, 101; benighted cogniser 85–97; clairvoyant 6–7, 11–13, 22–23, 31–32; fake barn 47–52, 70–71, 78–79, 83–86, 91–94, 96–98, 100–105; Frankfurt Clock 51–52, 101–102, 104–108; Gettier cases 37–49, 74–75, 84–88, 90–92, 94–99, 110–116; insanity 85–87; landmark 39–47; lottery 50, 53–61, 75–77; radical deception/new evil demon problem 5–6, 9–11; stopped clock 39, 41–43, 45, 47–49, 98 externalism see internalism/ externalism debate Frankfurt, H. 51 generality problem 13–14, 22–23, 32–34 Gettier 3–5; see also examples: Gettier cases Goldman 1–4, 9–11, 14–16, 48 Greco 38–40, 102–105, 113–115 Heller, M. 13 inquiry 30, 67–79 internalism/externalism divide 4–5 justified belief: as competent belief 19, 31, 68; justified believability 55–61, 76–77; knowledge first accessibilist account of 83–84; knowledge first disjunctive account of 89–90;

124 Index process reliabilist account of 1–14; see also belief, internalism/ externalism divide, performance: competent Knowledge: as success in inquiry 71–75; as apt belief 19, 37, 46, 68, 100; levels of (animal, reflective, full well) 100–102; perceptual 82–84; testimonial 39–43, 45–46, 107–108 Knowledge first 66–68, 75, 78, 82–84, 88, 96–97 Kyburg, H. 55, 57 Lackey, J. 39 Lottery paradox 55–61; see also examples: lottery cases Lyons, J. 12–13 Millar, A. 82–85, 87–88 Millikan, R. 26–27 Moorean paradoxes/propositions 54, 56–58 Miracchi 88–92, 94–97

normativity of 17–19, 32; successful 17–19, 42–44, 46, 68; tarnished 42–43, 45–49; see also achievement, belief, justified belief, knowledge Pritchard, D. 5, 39–40, 105–109, 114–116 reductive analysis 68, 70–71, 78 risk of error 57–59 Rohrbaugh, G. 50 safety: condition on knowledge 50, 100, 112–115; counterexamples to 50–51; dilemma 50–53, 78–79, 100–108 sensitivity condition on knowledge 112–113 simple goal directed practices 24, 30, 69–71 scepticism 4–5 Sosa 20, 44, 46–47, 100–102, 110–114 virtues see abilities

Neta 50 Williamson 60, 66–67, 76–77 performances: apt 17–19; competent 17–19, 29, 32–33, 42–44, 46, 68;