The Ethics of Wilfrid Sellars 1138708747, 9781138708747

Citation preview

The Ethics of Wilfrid Sellars

Wilfrid Sellars’s ethical theory was rich and deeply innovative. On Sellars’s view, moral judgments express a special kind of shared intention. Thus, we should see Sellars as an early advocate of an expressivism of plans and intentions, and an early theorist of collective intentionality. He supplemented this theory with a sophisticated logic of intentions, a robust theory of the categorical validity of normative expressions, a subtle way of reconciling the cognitive and motivating aspects of moral judgment, and much more­—all within a strict nominalism that preserves Sellars’s commitment to naturalism. The Ethics of Wilfrid Sellars offers the first systematic treatment of this sadly neglected aspect of Sellars’s work and demonstrates that his ethical theory­—just like his more widely discussed epistemology­—has much to contribute to current debates. Jeremy Randel Koons is Associate Professor of Philosophy at the Georgetown University School of Foreign Service in Qatar. He has published widely in epistemology, metaethics, philosophy of religion, and other areas. His most recent book, The Normative and the Natural (co­authored with Michael P. Wolf), appeared in 2016.

Routledge Studies in American Philosophy Edited by Willem deVries, University of New Hampshire, USA Henry Jackman, York University, Canada

Toward a Pragmatist Metaethics Diana B. Heney Sellars and Contemporary Philosophy Edited by David Pereplyotchik and Deborah R. Barnbaum Pragmatism and Objectivity Essays Sparked by the Work of Nicolas Rescher Edited by Sami Pihlström The Quantum of Explanation Whitehead’s Radical Empiricism Randall E. Auxier and Gary L. Herstein Peirce on Perception and Reasoning From Icons to Logic Edited by Kathleen A. Hull and Richard Kenneth Atkins Peirce’s Speculative Grammar Logic as Semiotics Francesco Bellucci Pragmatism, Pluralism, and the Nature of Philosophy Scott F. Aikin and Robert B. Talisse Pragmatism and the European Traditions Encounters with Analytic Philosophy and Phenomenology Before the Great Divide Edited by Maria Baghramian and Sarin Marchetti Community and Loyalty in American Philosophy Royce, Sellars, and Rorty Steven A. Miller Sellars and the History of Modern Philosophy Edited by Luca Corti and Antonio M. Nunziante The Ethics of Wilfrid Sellars Jeremy Randel Koons

For more information about this series, please visit: www.routledge.com /Routledge-­Studies-­in-­American-­Philosophy/book-­series/RSAP.

The Ethics of Wilfrid Sellars

Jeremy Randel Koons

First published 2019 by Routledge 52 Vanderbilt 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 © 2019 Jeremy Randel Koons The right of Jeremy Randel Koons 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 A catalog record for this title has been requested ISBN: 978-­1-­138-­70874-­7 (hbk) ISBN: 978-­1-­315-­20116-­0 (ebk) Typeset in Sabon by Apex CoVantage, LLC

For Lucy and Abby

Contents



Acknowledgments

ix

Introduction: Situating Sellars’s Ethical Theory in the Contemporary Landscape

1

  1 Sellars’s Synoptic Vision

12

  2 A New Naturalism

34

  3 Moral Judgments as Shared Intentions

54

  4 What Are Sellarsian We-­Intentions?

79

  5 Practical Reasoning and the Logic of Intentions

104

  6 Material Practical Inference

141

  7 Cooperative Rationality and We-­Intentions

158

  8 Defeasible Rules and the Particularist Challenge

180

  9 Rules, Pattern-­Governed Behavior, and Collective Attitudes

194

10 Moral Motivation 1­—Against the Humean Account

216

11 Moral Motivation 2­—Sellars’s Kantian Account

234

12 Against Moral Foundationalism

250

13 Categorical Validity and the Necessity of Community

271

14 Sellars’s Mistaken Formalism

309

15 Sellars’s Ethical Naturalism

324

Works Cited Index

345 355

Acknowledgments

In the first place, this book would not exist were it not for Bill deVries. In early 2015 I gave a paper on Sellars’s ethical theory at a conference (“Sellars in a New Generation”) at Kent State University. Bill encouraged me to expand these ideas into a book project­—an idea that had not previously occurred to me­—and offered substantial encouragement along the way. Writing about Sellars is a particularly daunting task, largely because he is such a systematic philosopher­—meaning that one often feels as though one must be conversant with every philosophical subfield to do justice to Sellars’s integrative philosophical vision. As this is impossible for mere mortals such as myself, it was essential during the writing process to get feedback from colleagues who could point out lacunae in my knowledge, literature I needed to address, or (just as often) simple argumentative lapses. I am very grateful to Bill deVries, Stefanie Dach, and Michael P. Wolf for reading a draft of the manuscript and providing very helpful feedback to me. Very intensive feedback was provided during a one-­day manuscript workshop sponsored by the Center for International and Regional Studies (CIRS) at Georgetown University’s Qatar campus. The manuscript was circulated in advance to a number of scholars, and we met on 5 November 2017 for an all-­day session to discuss the work. I received enormously helpful feedback from Bana Bashour, Arudra Burra, Dionysis Christias, Anjana Jacob, Bhaskarjit Neog, Amir Saemi, Lucas Thorpe, Jack Woods, and Bill Wringe­—feedback that led to substantial revisions (and a few chapters being overhauled from the ground up). I also presented some of the material from this book to audiences at Bilkent University and Boğaziçi University and am grateful to audiences there for stimulating discussion and for pushing me on several weak spots in my argument. Also, I am grateful to Luz Christopher Seiberth and members of the Sellars reading group he organizes (Bill deVries, Marc Joseph, Jim O’Shea, and Lionel Shapiro) for reading and discussing with me portions of Chapter 13. Bill deVries, Jim O’Shea, Kenneth Westphal, Matthew Chrisman, Robert Sugden, and Jack Woods were also patient

x Acknowledgments about responding to e-­mail inquiries and allowing me to bounce ideas off them. Finally, and always, my deepest gratitude goes to my wife, Lucy, for her endless patience and support. From reading and offering feedback on my prospectus, to helping me practice my conference presentations (and then holding down the fort while I traveled), to offering me every kind of practical and emotional support when I was overwhelmed by the immensity of the project I had undertaken: The list of ways, big and small, in which this project could not have succeeded without her could fill its own chapter. Lucy has never wavered in her love and commitment, and I will forever be in her debt.

Introduction Situating Sellars’s Ethical Theory in the Contemporary Landscape

In his early overview of Sellars’s ethical theory, David Solomon situates Sellars’s theory within what Solomon calls the tradition of “classical metaethics” (CM). According to Solomon, the battle lines in CM are drawn roughly as follows: Cognitivists accepted that there is a logical relation between statements of fact (e.g., evidence) and moral judgments but “denied, however, that any strong logical tie holds between” moral judgment and action (Solomon 1977, p. 153). Non-­cognitivists, on the other hand, hold that there is a strong connection between moral judgment and action but also hold that “a strong logical relation between [moral judgments and action would] preclude such a relation between [statements of fact and moral judgments]” (Solomon 1977, p. 153). Sellars is grappling with these fundamental (and intransigent) issues in his earliest writings on ethical theory. In IILO, Sellars expresses the motivating element of moral judgment in the following terms: “To know that there are certain things that one ought to do is to have a sense of duty” (IILO, p. 160), a view which he claims “smacks of emotivism” (IILO, p. 160). Thus, Sellars attributes to Ayer the view that “the so-­ called thought that one ought to do A here and now is not, strictly speaking, a thought at all, but rather a specific way of being moved to do A” (IILO, p. 161). This, of course, captures what Sellars takes to be the chief weakness of emotivism: While it captures the connection between moral judgment and moral motivation, it fails to capture the cognitive element of moral judgment, holding instead “that ethical concepts are ‘pseudo-­concepts’ and the logic of moral discourse is a ‘pseudo-­ logic’” (IILO, p. 162). As Ayer himself happily concedes, if emotivism is true, then opposing ethical judgments cannot, strictly speaking, be incompatible with each other: “It is plain that the conclusion that it is impossible to dispute about questions of value follows from our theory. . . . For as we hold that ‘Thrift is a virtue’ and ‘Thrift is a vice’ do not express propositions at all, we clearly cannot hold that they express incompatible propositions” (Ayer 1946, p. 110).1 Sellars expresses the cognitive element of moral judgment by writing that “obligation, by its very nature, is intersubjective” (IILO, p. 160). And, of course­—the Achilles heel of

2 Introduction emotivism­—Sellars is committed to the validity and soundness of certain inferences essentially involving moral premises.2 This intersubjective element of moral judgment, according to Sellars, “smacks . . . of intuitionism” (IILO, p. 160). However, Sellars is also unhappy with intuitionism, as the dominant forms rescue the cognitive element of moral judgment at the expense of explaining the connection between moral judgment and moral motivation: It was the signal merit of intuitionism . . . to have insisted on . . . the truly propositional character of prescriptive statements. . . . But the epistemological and metaphysical commitments of ethical intuitionism, which precluded it from understanding the logical connection between “thinking that one ought” and “being moved to do,” thus forced it to make a mystery both of the conduct-­guiding role of moral discourse, and of the uniqueness of prescriptive discourse which it had so happily emphasized. (IILO, p. 162) For example, Prichard (who merits critical mention in the early pages of IILO), in rejecting the Kantian view that the knowledge that an action is right can influence us to act, writes, “We are, I think . . . bound to abandon this view. For one reason, . . . if we face the purely general question, ‘Can we really do anything whatever unless in some respect or other we desire to do it?’ we have to answer ‘No’” (Prichard 2002, p. 38). The task for Sellars, then, is to reconcile these two seemingly incompatible elements of moral judgment. Thus, he must do justice to what the cognitivists recognize as important about moral judgments: they are intersubjective; they are in the space of reasons, and admit of (and require!) evidential support; they can stand in logical relations with other propositions; and so on. But he must also do justice to what the non-­ cognitivists want to preserve about moral judgments: Moral judgments are fundamentally connected to action. Ceteris paribus, one who judges that ϕ is obligatory is motivated (to some degree) to do ϕ. Sellars’s solution to this problem is powerful and ingenious. In the first instance, he argues that moral judgments express intentions and are thus conceptually connected to action. In arguing that moral judgments express a certain kind of intention, Sellars’s theory is a very early model of the expressivism of plans and intentions perhaps most famously associated with Gibbard­—although few philosophers working in this tradition know of Sellars’s early contributions to this theory. However, ordinary expressions of intention cannot, strictly speaking, contradict each other; nor can people share expressions of intention in ordinary cases. (If I intend to order vanilla ice cream, and you intend to order chocolate, we are not disagreeing; nor is there any conflict or contradiction between our intentions.) To solve this problem­—to explain

Introduction  3 how moral judgments are genuinely intersubjective­—Sellars develops his account of we-­intentions. Although the notion of we-­intentions has been developed in sophisticated ways by Bratman, Gilbert, Searle, Tuomela, and others, Sellars’s is (as far as I can determine) the first strictly philosophical account of the notion.3 So much for the connection to action. How does Sellars account for the cognitive aspect of moral judgment? Anticipating Bratman’s work, Sellars demonstrates that we can and do reason among intentions, and he develops a sophisticated logic for his expressivism that prefigures a similar logic developed­—50 years later­—by Mark Schroeder. A moral theory must accomplish much more than this, of course, and Sellars is careful to elaborate his moral theory in a way that satisfies a number of constraints. Because Sellars is such a systematic philosopher, he is able to offer an ethical theory that is integrated with and motivated by his larger philosophical project. His larger philosophical goal is to accommodate—via a thoroughgoing nominalism­—the framework of normativity (and of persons more generally) within a strongly scientific worldview. Thus, Sellars’s moral theory is constructed within a strictly naturalist philosophical framework. The systematic nature of Sellars’s work, combined with his deep familiarity with the perennial metaethical issues and problems that confront us just as they confronted Sellars’s contemporaries (such as Stevens and Hare), enables him to provide a theory that offers powerful and convincing solutions to a number of outstanding philosophical problems. As already noted, his view of moral judgments as expressions of we-­ intentions (combined with his sophisticated logic of intentions) allows him to explain how moral judgments can be both cognitive and motivating. He offers a subtle account of reasoning­—centered around material ­inference­—that offers us insights on practical reasoning, and also on how judgment is connected to action. He offers a version of ­constitutivism­— the theory that moral requirements stem from the nature of practical agency itself­—that has a Hegelian/social dimension, allowing him to explain the necessity of the moral community and the categorical validity of moral requirements. He offers a moral theory that reconciles consequentialist and deontological theories, filling in some of the problematic gaps in Kantian moral theory while managing to preserve key Kantian insights (such as the idea that the moral realm is the realm of autonomy as opposed to heteronomy). And this barely scratches the surface of the richness of Sellars’s account and what it has to offer. Further, all of this is developed from within a nominalism stricter even than Quine’s, one that allows Sellars to articulate a vision of the moral community consistent with the strictest naturalism. This isn’t to say, of course, that Sellars’s own ethical writings need no updating or supplementing. There are tensions between Sellars’s writings in ethics and some of his other writings. Also, there are elements

4 Introduction of Sellars’s theory that can be enriched by contemporary philosophical work­—work by later Sellarsian philosophers, by philosophers working on group intentions and cooperative rationality, and more. In resolving these tensions and updating Sellars’s theory with more recent work in the field, what emerges is a strikingly original and comprehensive theory that has much to contribute to contemporary debates. The goal of this book is to explain Sellars’s theory and to develop it into one that can­—and should­—be taken seriously by contemporary theorists. A note on methodology: There is a certain inevitable hubris inherent in trying to improve upon the work of one’s philosophical betters, and so I have attempted to be as conservative as possible in the modifications I make to Sellars’s own theory. The modifications I make fall into three classes, in ascending order of radicalness. First, there are simple additions I make in light of more recent work on the subject matter that do not, in my view, contradict anything Sellars himself says. For example, although Sellars offers (as I noted above) the first philosophical analysis of we-­intentions, a robust literature on the subject has developed since Sellars’s death, and I supplement Sellars’s account by drawing on more recent accounts. Second, there are amendments I make in light of what I take to be inconsistencies in Sellars’s own view. In such cases, I make a judgment (which I hope the reader shares) as to which element is more central to Sellars’s overall project. For example, a deep and central commitment of Sellars’s (and of later Sellarsians like Brandom) is a rejection of formalism about reason. That is, Sellars argues that we have to understand materially valid inference (e.g., “Fido is a dog; therefore, Fido is a mammal”) as prior in the order of explanation to deductively valid formal inferences. However, I will argue in Chapter 14 that Sellars embraces this formalism in certain aspects of his ethical theory. Given the absolute centrality of anti-­formalism to the rest of his philosophical project, the formalism in his ethical theory must give way­—but in a way that preserves the essential features of his ethical theory because the formalism served a purpose there that cannot simply be abandoned. Thus, I attempt to make maximally conservative revisions. Finally­—and most seriously­—there are elements of Sellars’s theory that I revise because they seem unsustainable in light of external criticisms or more recent developments in the literature. Pursuant to my conservatism, I try whenever possible to avoid these kinds of revisions, but there are places where they cannot be avoided. For example, I argue in Chapter 5 that certain elements of Sellars’s logic of intentions must be revised because his logic, as he presents it, cannot construct some core deontic notions, such as permissibility. In every case, though, I first explain Sellars’s original view before going on to supplement or modify it. Thus, although this book is not merely exegetical, I always begin at the source before going beyond it. The reader

Introduction  5 may not agree with all of my decisions, but fruitful philosophical discussion usually begins in disagreement. No work on a topic such as this can hope to be the last word on the subject or be in any way the definitive work on the matter; and it would be unforgivable hubris to have this as one’s goal in writing about Wilfrid Sellars’s ethical theory. Not every reader will agree across the board with my reading of Sellars. Nor will every reader agree with my decisions regarding how to reconcile what I take to be inconsistent elements in Sellars’s theory or how to supplement and extend Sellars’s theory using the philosophical tools that were developed after his death. Further, because Sellars’s ethical theory is so deep and rich, there are a number of topics that would welcome more in-­depth treatment. Indeed, I’m sure that some of the chapters of this work could admit of book-­length treatments or at least of more detailed discussion than I am able to offer here. There is much more to be said and much more work to be done. But if I succeed in nothing else, I hope to convince the reader of at least this: Sellars’s ethical theory is subtle and profound and is worth serious consideration­—not merely as an object of historical interest but as a flourishing research project that can be brought into fruitful dialogue with contemporary accounts, enriching them and being enriched by them. In light of this, the recent surge of interest in Sellars’s ethical writings can only be seen as a good thing.

Chapter Summaries On the assumption that most philosophers interested in ethics are not conversant with Sellars’s larger philosophical project, Chapter 1 (“Sellars’s Synoptic Vision”) gives a very brief overview of some of his chief aims and methods. Sellars’s chief philosophical goal is to accommodate the framework of normativity (and of persons more generally) within a hard-­ nosed scientific realism. His method for doing this is to understand distinctly philosophical expressions­—moral, epistemic, intentional, modal, semantic, and so on­—not as attempting to describe the world, but rather as nondescriptive expressions in the metalanguage. Readers who are already conversant with Sellars’s overall project may wish to skim this chapter or skip ahead to Chapter 2. In Chapter 2, I turn to Sellars’s claim that the normative is causally reducible but logically irreducible to the natural. This means that while (for example) the behavioral or neurological information conveyed by a belief ascription can in principle be wholly captured by an (ideal) scientific theory, such a theory wouldn’t say what intentional language said (in that it wouldn’t ascribe beliefs per se). What normative expressions say can only be said using normative language even if what they convey can be reduced to empirical psychology or neurology or some other branch of science. Serious objections have been raised concerning the tenability

6 Introduction of Sellars’s thesis of the “causal reducibility, but logical irreducibility” of the normative. Just as critically, Sellars’s powerful argument for scientific realism can be turned against his argument for the causal reducibility of the normative, so that the normative turns out to be both logically and causally irreducible. I will demonstrate that this conclusion leads to a more complicated naturalism, but one that is still compatible with Sellars’s austere nominalism. The thesis of causal irreducibility will turn out to be helpful in explaining, among other things, moral motivation (as we will see in Chapter 10). In Chapter 3, I turn to a core element of Sellars’s ethical theory, the key to his reconciliation of the cognitive and motivating aspects of moral judgment. Sellars argues that moral judgments are expressions of intention. This not only aids in this reconciliation­—intentions are conceptually tied to action and are also involved in reasoning­—but also is independently plausible. I try to motivate Sellars’s position by considering the role of intentions in practical reasoning as well as considering illuminating parallels with theoretical reasoning. Central to Sellars’s account of moral judgment is a special kind of intention­—a we-­intention. I show how Sellars’s account has the resources to make sense of phenomena like restricted-­scope or loyalty-­based obligations (i.e., special obligations I have qua member of some subset of rational humanity, such as qua member of a union or a family). Sellars does too little to explain what, precisely, a we-­intention is. In Chapter 4, I draw on recent accounts (like those of Gilbert and Tuomela) to enrich the Sellarsian account of we-­intentions. The resulting account rejects singularism­—it denies that group intentions are reducible to personal intentions­—but argues that this move is consistent with Sellars’s strict naturalism. Various elements of these more recent accounts are adapted to comport with various other features of Sellars’s project, such as his pragmatism and his view that an individual can express a we-­ intention not shared by the larger group. I conclude with some remarks on how Sellars’s account, while lending itself easily to a social practice account of rules and norms, offers a defense against relativism. Sellars’s account of moral judgment is broadly expressivist; and such accounts face well-­known problems such as the embedding problem. Sellars develops a very sophisticated logic of intentions, which foreshadows similar attempts by authors like Mark Schroeder by half a century. In Chapter 5, I explain the key features of Sellars’s logic of intentions and demonstrate how it provides us the tools to solve not only classical problems for expressivism (such as the embedding problem and the problem of negation) but also contemporary challenges to expressivist accounts (such as Schroeder’s challenge that such accounts cannot handle tense and modal operators). I argue that while the basic insights of Sellars’s account are sound, various modifications and extensions are required to his account so that we can have a fully satisfying intentional logic,

Introduction  7 one that is able to reconstruct all of the required deontic notions and relations. Sellars presents practical reasoning as based on nomologicals of the form: “Doing Ai if [in] Cj is causally necessary to the realization of Ek” (SM 7.VIII.§51/p. 195). This, however, misrepresents practical reasoning, as there is often more than one way of achieving an end; hence, the performance of no one means is in general a necessary condition on the achievement of one’s end. In Chapter 6, I introduce the notion of a material inference, which is central to Sellars’s philosophical project, and argue that the notion of a material practical inference gives us a more flexible model of practical reasoning. This, combined with Bratman’s notion of intentions as elements of partial plans, allows us to develop an account of practical reasoning that (a) more accurately represents practical reasoning (and the role of intention in practical reasoning) than does Sellars’s original, extremely rigid nomological-­based model and (b) is consistent with other elements of Sellars’s philosophy (such as his views on theoretical or causal-­explanatory material inference, and on the role and status of desires). Many discussions of rationality start with the assumption that all rationality must be conceived of as individual: Individual agents are acting on intentions with contents that might at most be parallel (in Sellars’s terminology) but cannot be shared. Drawing on the work of Gilbert, Sugden, Tuomela, and others, I argue in Chapter 7 that there is a notion of cooperative rationality that is not reducible to individual rationality, and that the moral point of view can be justified only from the standpoint of cooperative rationality­—not individual rationality. Further, Sellars’s account of we-­intentions goes hand in hand with an account of cooperative rationality to present a cohesive picture of the moral point of view. I conclude by discussing what rationality requires when individual and team rationality conflict, and I make some brief comments on relativism and supererogation. Rules play a central role in Sellars’s philosophy. As Sellars writes in LRB, “When God created Adam, he whispered in his ear, ‘In all contexts of action you will recognize rules, if only the rule to grope for rules to recognize. When you cease to recognize rules, you will walk on four feet” (p. 298). However, radical particularists in ethics argue (often on the basis of holism about reasons) that rules play no role in ethical judgment or moral justification. I argue in Chapter 8 that while radical particularism is not very plausible, it is incumbent upon the Sellarsian to develop a notion of a rule that has genuine force, but admits of exceptions, that can serve the role of Sellars’s important material inferential rules. Fortunately, recent years have seen the emergence of well-­ developed accounts of defeasible rules of inference­—both less formal accounts (like Lance and Little’s) and formal ones (like Horty’s). I show how these accounts can be pressed into service for Sellars’s philosophical project.

8 Introduction I turn in Chapter 9 to addressing a second concern about rules—­ specifically, Wittgensteinian concerns that we cannot understand intelligent behavior in terms of following rules. The argument of this chapter attempts to establish two points. First, Sellars’s account of pattern-­governed behavior (behavior that is not rule-­following but that must be understood in relation to rules) responds to Wittgensteinian worries while preserving an essential role for rules, a role to be explored at length in this chapter. Second, I demonstrate that the social practices at the bottom of Sellars’s account, and the rules that govern them, must be understood in terms of collective attitudes­ —both we-­ intentions and collective beliefs­ —and that, therefore, individual rational agency strongly presupposes collective attitudes. This latter point is particularly important because it allows us to build toward the ultimate Sellarsian goal of proving the reality­—and, indeed, the necessity­—of the community of rational beings, which will bring us (in Chapter 13) to the final goal of establishing the categorical reasonableness of the formal end of morality. As I note at the beginning of Chapter 3, a persistent problem in metaethics is the question of how to reconcile the cognitive and motivational elements of moral judgment. The lynchpin to the “moral problem” is the Humean philosophy of mind, which holds that belief and desire are “distinct existences.” With the help of McDowell, I try in Chapter 10 to undermine allegiance to the Humean model, arguing that moral judgment doesn’t involve sterile cognition of a realm of facts but is a type of judgment that already embodies our concerns, attitudes, and normative commitments. Again, though, this doesn’t mean we are cognizing “queer” facts, because moral judgment isn’t in the first instance (for Sellars) a type of belief, and normative utterances aren’t descriptive­—even though they are very much in the space of reasons. Building on the argument from the previous chapter, and on the work of Sellars, Brandom, and McDowell, I present in Chapter 11 a Kantian account of moral motivation. I distinguish two cases: In the first case, we are motivated by explicitly moral judgments (e.g., “Kindness is good”) because such judgments are expressions of intention and, hence, conceptually connected to action. In the second case, we do not make an explicitly moral judgment, but judge that some feature of the world is salient in some way (e.g., “My friend is in trouble and open to being comforted”). Extending the argument from the previous chapter, I argue that to see certain features as salient is already to make a judgment embodying various concerns, attitudes, and commitments; hence, it is to make a judgment that is also essentially practical in nature (i.e., tied to intention and action). In Chapter 12, we begin our approach to categorical validity. Sellars rejects the idea that the moral ought can reside in the consequent of a hypothetical imperative, e.g., “If you intend E, then you ought to do A.” Following Jean Hampton, Joseph Heath, Christine Korsgaard, and others, I will argue that all reasoning cannot be reduced to instrumental

Introduction  9 reasoning; Sellars is correct that we need a substantive account of what ends are intrinsically reasonable. However, we can see that many attempts to “ground” morality are motivated by moral foundationalism­—by the —particularly when these “Given” transposed to the practical sphere­ attempts demand that morality be grounded in nonmoral reasons and given a “noncircular” justification. Our anti-­foundationalist, anti-­Given argument will demonstrate that the latter demand simply cannot be met in any justificatory enterprise­—be it epistemology, morality, or what have you­—and we should not hold practical reason up to this impossible standard. Just as Sellars rejects the Given in epistemology, so should the Sellarsian reject the Given in practical reasoning. Sellars was ambivalent as to the prospects of deriving the “­reality . . . of an ethical community consisting of all rational beings” (SM  7.XX.§144/p. 225) and, hence, the intrinsic validity of the formal end of morality. I think the prospects are rosier than he anticipated and set out to prove in Chapter 13 the reality of this community and, thus, the categorical reasonableness of the formal end of morality. Our dialectical opponent is the rational egoist; this is the point of view that Sellars saw as competing against the moral point of view for our allegiance. I argue that rational egoism (RE) cannot be established just by considering the nature of practical reason. Alternately, to establish RE on teleological grounds would require proving that an agent’s welfare is the only good (for her). I will demonstrate that this position cannot be consistently held, either. In Chapter 9, I will have already argued that the very norms that are expressed in an agent’s pattern-­governed behavior embody a range of we-­mode attitudes, and that pattern-­governed behavior must be understood in the we-­mode. Thus, rational agency itself cannot be understood apart from the community­—nor can the agent’s welfare, autonomy, or other essential goods or capabilities, which are, indeed, various essential elements of her identity. Thus, there is no meaningful way to argue that an agent’s well-­being matters but the welfare of the community­—which (partially) constitutes her agency, her welfare, and so on­—does not matter. Further, and more radically, to be a rational agent in the first place is already to occupy the perspective of the “we,” i.e., the community of rational (and, hence, as I will argue, moral) agents. In Chapter 14, I try to reconcile an inconsistency between Sellars’s ethical theory and his larger philosophical project. Sellars rejects formalism in theoretical reasoning, arguing that This dry, well-­made match is struck Therefore, the match will light is a good inference and doesn’t depend for its validity on the insertion of a major premise. In ethics, though, Sellars seems to embrace this formalism, arguing that

10 Introduction I am in Ci Therefore, Shallwe [I do Ai] is good qua instantiation of the intermediate moral principle Shallwe [I do Ai, if in Ci] which in turn is good only if derived from the supreme principle of morality Shallwe be [Our welfare is maximized] This formalism is inconsistent with Sellars’s other commitments. We should recognize that intermediate moral principles have original (but not foundational) authority; they are “auxiliary moves,” as described in SRLG. But there is still a role for the supreme principle of morality: It is a theoretical postulate that serves a dialectical role to resolve disputes among mid-­level intermediate moral principles and to bring consistency into our moral practice. Thus, we can preserve the formal structure of Sellars’s ethical theory while preserving his opposition to formalism per se. I begin this book with a discussion of Sellars’s strict naturalism­—his scientia mensura principle­—and the nominalism he developed to accommodate normative (and, more generally, philosophical) concepts within this principle. I conclude in Chapter 15 by tying together various elements in the book to demonstrate how Sellars’s ethical theory comports with his stringent naturalism. The key move for Sellars, of course, is his understanding of normative claims as nondescriptive. This obviates the need to “place” ontologically problematic normative facts into a scientific worldview, it avoids “queerness” worries as to how such facts can be intrinsically motivating, and it avoids is-­ought gap concerns about how natural facts could generate normativity. By drawing on social practice theorists like Joseph Rouse, I will demonstrate that while Sellars’s theory is nondescriptive, it paints a picture of our moral practice as ineliminably tied to the natural world through essentially world-­involving practices. We are thus left with a picture of our moral practice as robustly objective and involving binding normativity, yet fully consistent with Sellars’s strict nominalism and his scientia mensura.

Notes 1 Compare the emotivism of C.L. Stevenson, a contemporary of Ayer’s: “We must distinguish between ‘disagreement in belief’ (typical in the sciences) and ‘disagreement in interest.’ . . . Let me give an example of disagreement in interest. A. ‘Let’s go to a cinema tonight.’ B. ‘I don’t want to do that. Let’s go to the symphony.’ A continues to insist on the cinema, B on the symphony. This is a

Introduction  11 disagreement in a perfectly conventional sense. . . . It is disagreement in interest that takes place in ethics. When C says ‘This is good,’ and D says ‘No, it’s bad,’ we have a case of suggestion and counter-­suggestion. Each man is trying to redirect the other’s interest” (Stevenson 1937, p. 27). 2 Fundamentally, as Steven A. Miller notes, “Ayer ‘failed to distinguish between emotions and attitudes’ [SE, pp. 226–7] insofar as he needlessly ran together kneejerk reactions and carefully considered, thoughtfully habituated positions about the world” (Miller 2018, p. 20). 3 Although his account clearly did not appear in a vacuum, as he was influenced by sociologists like Durkheim and Bouglé (see Olen and Turner 2015, 2016) and, I strongly suspect, by philosophers like George Herbert Mead.

1

Sellars’s Synoptic Vision

Most philosophers who study ethics are not conversant with Sellars’s philosophical project, and most philosophers who have read Sellars have generally only read EPM. One of the obstacles to gaining familiarity with the philosophy of Wilfrid Sellars is that he is a systematic philosopher: He is one of the few 20th century thinkers with a view of how the various elements of his picture (language, ontology, ethics, epistemology, etc.) all hang together. Thus, one often feels it is difficult to understand any one element of Sellars’s project without understanding all of it­—or at least a large part of it. On the assumption that many readers may not be familiar with the details of Sellars’s overall project, I wish (briefly) in this chapter to give an overview of what I take to be Sellars’s chief philosophical goal: to accommodate—via a thoroughgoing nominalism­ —the framework of normativity (and of persons more generally) within a strongly scientific worldview. The goal of this chapter is not to present an exhaustive account of Sellars’s philosophical project, nor to defend his commitments in great detail. But I hope to lay out his view in broad outline and to motivate it­—­particularly those elements that are central to his project of situating the normative within a scientific worldview.1 Readers who are already conversant with Sellars’s overall project may wish to skim this chapter or skip ahead to Chapter 2. Sellars has a number of fundamental commitments that are, on their face, difficult to reconcile. First, Sellars is committed to the primacy of science as the final arbiter of what the world contains and what it does not contain. He describes his scientia mensura principle as follows: “In the dimension of describing and explaining the world, science is the measure of all things, of what is that it is, and of what is not that it is not” (EPM IX.§41/p. 83). Sellars’s view is quite radical. Sellars famously distinguishes between the manifest and scientific images of the world. The manifest image is “the framework in terms of which man came to be aware of himself as man-­in-­the-­world” (PSIM, p. 6). The manifest image contains not just persons and values but also qualia, mental states, ordinary objects (like tables and chairs), and the like. The manifest image is not unscientific­—Sellars notes that the manifest image has become

Sellars’s Synoptic Vision  13 scientifically sophisticated over the millennia, enriched by “the canons of inductive inference defined by John Stuart Mill, supplemented by canons of statistical inference,” and so on (PSIM, p. 7)­—but the manifest image does not contain the results of postulational science. Thus, it does not contain forces or atoms, quarks or electrons, and the like. As I said, Sellars’s view of the relation between the manifest image and the scientific image­—and the primacy of the latter­—is quite radical: On one classical interpretation, correspondence rules would appear in the material mode as statements to the effect that the same objects which have observational properties also have theoretical ­properties. . . . According to the view I am proposing, correspondence rules would appear in the material mode as statements to the effect that the objects of the observational framework do not really exist­—there really are no such things. (LT, p. 126) Thus, in PSIM, Sellars refers to Arthur Eddington’s famous “two tables” example (PSIM, pp. 35–6). Our manifest image table is hard, extended, solid, colored, and so on. But, as Eddington writes, My scientific table is mostly emptiness. Sparsely scattered in that emptiness are numerous electric charges rushing about with great speed; but their combined bulk amounts to less than a billionth of the bulk of the table itself. . . . There is nothing substantial about my second table. It is nearly all empty space­—space pervaded, it is true, by fields of force, but these are assigned to the category of “­influences,” not of “things.” (Eddington 1931, p. 72; quoted in O’Shea 2007, p. 14) Sellars would argue that the second, scientific description of the table is the real one, and that the table of the manifest image doesn’t really exist. Sellars is thus left with a very austere ontology. Related to the first commitment is a second: Sellars is a naturalist. In his introduction to NAO, Sellars describes his rejection of Platonism and Idealism­—but also his rejection of “wishy-­washy” forms of naturalism according to which “one could believe almost anything about the world and even some things about God, and yet be a Naturalist” (NAO, pp. 9–10). We have already noted Sellars’s commitment to the scientia mensura principle. What this means, for Sellars, is that we can posit no explanans that is not part of the causal-­explanatory order. As deVries puts the point, “For Sellars, naturalism includes the thesis that everything that exists is an element in the spatiotemporal causal nexus. He rejects, therefore, any kind of purported causal or metaphysical dependence on something outside space and time, such as God, souls, forms or other pure intelligibles” (2005, p. 16).

14  Sellars’s Synoptic Vision A concrete way in which Sellars realizes his commitment to naturalism and the primacy of science is through his embrace of nominalism. Indeed, as deVries points out in discussing Sellars’s nominalism, Sellars is even more ruthless than Quine in exorcizing abstract objects from his ontology (deVries 2005, p. 18). Sellars’s objection to abstract entities in general, and Platonism specifically, is that there is no good explanation for how such entities could be integrated into the causal order. Sellars explains the discontinuity between the theoretical posits of science and the abstract entities of the Platonist in NAO: Suppose a platonist with respect to attributes and/or classes to be of-­ factual relations between asked: “Must there not be matter-­ abstract entities and human minds by virtue of which abstract singular terms acquire a hook-­up with the world?” . . . The theory-­whole has specific things to say about the causal relations which connect micro-­physical objects with the sensory stimulations which bombard the sensory surfaces of experimentors looking at bubble chambers and photographic plates. The theory explains how we are in touch with micro-­physical objects. . . . This is not the case with such terms as “number,” “class,” “attribute,” and “proposition.” This fact . . . has important consequences for the problem of abstract entities. (NAO, 1.§§32–34/p. 13) Although Sellars rejects Platonism and embraces naturalism and nominalism, he cautions against an overcorrection from Platonism into psychologism, which he defines as “the mistake of identifying philosophical categories with those of psychology” (PPE, p. 5). Sellars is clear, from his very earliest work, that psychologism is unable to distinguish between empirical regularities and norm-­governed behavior: The psychologistic blunder as defined above is based on a still more fundamental error, that, namely, of confusing between (1) language as a descriptive category for which symbols are empirical classes to which events belong . . . by virtue of performing an empirical function, with (2) language as an epistemological category for which the relation of type to token is not that of empirical class to member. . . . For the moment it will help clarify the relation of symbol-­types to symbol-­tokens if we think of the former as norms or standards and of the latter as events which satisfy them. We can therefore, for the moment at least, contrast the above two senses of “language” as the descriptive and the normative respectively. (RNWW, p. 60) Thus, we can state Sellars’s next commitment: While Sellars is a naturalist, he thinks that normative discourse cannot be eliminated, nor

Sellars’s Synoptic Vision  15 can it be replaced by any empirical categories or descriptions. Thus, in EPM, Sellars writes, “Now the idea that epistemic facts can be analyzed without remainder­—even ‘in principle’­—into non-­epistemic facts, whether phenomenal or behavioural, public or private, with no matter how lavish a sprinkling of subjunctives and hypotheticals is, I believe, a radical mistake­—a mistake of a piece with the so-­called ‘naturalistic fallacy’ in ethics” (EPM, I.§5/p. 19). When Sellars says at the end of PSIM that “the conceptual framework of persons is not something that needs to be reconciled with the scientific image, but rather something to be joined to it” (PSIM, p. 40), he has set for himself a formidable task: making room for normativity, and persons, in a strict naturalist worldview.

1.1 The Necessity of Normativity As noted already, Sellars, from his earliest writings, opposes psychologism. In principle, there are certain things that we cannot do using a purely descriptive language, and our linguistic resources would be woefully impoverished were we to attempt to eliminate nondescriptive usages from our language. There are a number of reasons why specifically normative language is ineliminable from our language. For example, Sellars asks us to imagine giving a descriptive or behavioristic reduction of intentional terminology along the following lines: S has mental state M, and M means P ≡ S is in bodily/brain state B Sellars argues that while the left-­hand side of the equivalence may convey what the right-­hand side asserts­—namely, it might convey certain behavioral regularities on the part of S, or that S is in a particular state­—it asserts something that in principle cannot be asserted in purely descriptive language—namely, that M means P. Thus, as Sellars argues, even if the normative is causally reducible to the natural, in that what a normative sentence conveys (roughly, its extension) can be captured in purely extensional language, normative sentences say something that in principle cannot be said in a purely extensional, descriptive language and are therefore logically irreducible. The same would go not just for intentional equivalences but for semantic ones (P, uttered by S, means it is raining º S has certain habits and dispositions regarding the use of P), moral ones (A is morally right º A maximizes utility), and so on. Second, although Sellars has (as noted above) argued for the primacy of science in the dimension of “describing and explaining,” it has often been noted that the thesis of the objective authority of science is not a thesis that can be formulated from within science. Jean Hampton’s argument on this point is typical. She notes that any argument from a scientific standpoint to the conclusion of eliminativism about the normative

16  Sellars’s Synoptic Vision can explain why any person takes her methodology for understanding the world to be “right,” but it cannot establish that one of them is right. And if this is so, science is only authoritative for those who accept it, and (as Feyerabend says) the enterprise of science as we know it cannot really undercut ethics, because it has no objective authority to undercut anything. (Hampton 1998, p. 210) Thus, Sellars’s argument for the primacy of science is, and must be, made from a normative standpoint. Any other understanding of an argument for the primacy of science is inherently self-­undermining.2 In general, Sellars argues, the “framework of persons” is not something that can be reduced, without remainder, to the scientific image: To think of a featherless biped as a person is to think of it as a being with which one is bound up in a network of rights and duties. From this point of view, the irreducibility of the personal is the irreducibility of the “ought” to the “is” . . . the conceptual framework of persons is the framework in which we think of one another as sharing the community intentions which provide the ambience of principles and standards (above all, those which make meaningful discourse and rationality itself possible) within which we live our own individual lives. A person can almost be defined as a being that has ­intentions. Thus the conceptual framework of persons is not something that needs to be reconciled with the scientific image, but rather something to be joined to it. (PSIM, pp. 39–40) To understand a fellow human as an agent is to understand him or her as someone who, e.g., believes that the train runs to Union Station, fears the wages of sin and hopes for salvation in Christ, wishes she had gotten a pay raise last year, and so on. Someone who did not respond to reasons as reasons­—for example, someone who did not see something’s being toxic as a reason not to eat it, or someone who didn’t see something’s being fun as a reason to do it­—would seem less like an agent and more like a wanton. (Imagine, for example, a feral child who responds to causes and stimuli but doesn’t conceptualize them as reasons in the full sense and lacks agency and autonomy in the full, personal sense.) The framework of agency is, for Sellars, inextricably tied up with reasons and the normative concepts surrounding intentions and other related notions. Finally, as Willem deVries has noted, the world of artifacts in which we live is inherently bound up in prescriptivity. For example, to use deVries’s example (deVries 2005, p. 275), a hammer is not just something that happens to be used to drive nails­—we could use a rock or a hunk of metal for that. Rather, a hammer is something that is to be used to drive nails.

Sellars’s Synoptic Vision  17 Its very identity as the thing that it is is bound up with prescriptivity­—it is “fraught with ‘ought,’” as it were.3 Indeed, there are some artifacts whose identity is wholly constituted by a set of rules; their physical constitution is all but irrelevant. Sellars’s favorite example of this is chess pieces: What makes something a pawn in chess is that it is subject to certain rules within the game of chess. In principle, virtually anything­— a piece of wood, some computer code, or even a car4­—could serve as a pawn as long as it is made subject to these constitutive rules. Tuomela argues that institutions and joint attitudes (“we-­attitudes,” such as collective intentions and beliefs) are “formed and expressed in artifacts,” writing, For example, a red traffic light can be taken to presuppose the following kind of we-­attitude: I believe that I should stop and I also believe that others in this situation believe so (some of those others similarly face a red light, while some others­—coming from a different direction­—face a green light), and I also believe that it is a mutual belief that I should stop (while those coming from another direction and facing a green light should drive on). Notice that the mutual belief part is needed. . . . (Tuomela 2002, p. 98) Someone who didn’t understand our world of artifacts­—who didn’t understand the purpose of doors and windows, cars and streets and sidewalks, books and pens, tables and chairs, or the myriad of other objects whose identity is largely determined by their appropriate use­—would fail to understand the world of persons. It is all well and good, one might say, to argue that normative notions are essential to our language, and indeed to our self-­understanding. But Sellars has also embraced a very austere naturalism­—more austere, we have noted, than even Quine, who was at least willing to countenance classes! So has Sellars placed himself between a rock and a hard place? He has argued for the ineliminability of the normative while at the same time embracing a radical naturalism that makes it (seemingly) impossible to accommodate the normative. As we will see in the next section, Sellars effects a reconciliation through his innovative and thoroughgoing nominalism.

1.2 Nominalism and Sellars’s Theory of Meaning What are we doing when we say something like “S has mental state M, and M means P”? Specifically, what are we doing when we say that M means P? If science gives us the full measure of what is and what isn’t, then it seems that in doing so, we must be giving some kind of description of M­—maybe describing its causal relation with some worldly referent,

18  Sellars’s Synoptic Vision functional role. But of course, Sellars would or describing its causal-­ regard this as a serious error, one that failed to respect the logical irreducibility of the normative, intentional, and other philosophical categories. Sellars’s solution is that in saying “M means P,” we are not describing M at all. Indeed, this “descriptivist” philosophy goes hand in hand with a more insidious view of what philosophy is about: One might be tempted to say that the philosophical and non-­ philosophical questions . . . are those which deal with different aspects of the same subject-­matter. Thus, it has been said that psychology deals with thought as occurrent fact, whereas logic and epistemology deal with thought as true or false, valid or invalid, confirmed or disconfirmed. This formulation is not so much mistaken as dangerous. . . . It suggests that truth, validity, confirmation, and their opposites are characteristics belonging to thoughts as shapes belong to tables. Why, then, should the psychologists neglect them? (Sellars, “Psychologism,” quoted in Olen 2016, p. 168) If philosophical “properties” are aspects of, for example, mental states, then they should be suitable for empirical inquiry; and this road leads inexorably to descriptivism. The alternative is to say that these “properties” are aspects of mental states and the like but not aspects that are subject to empirical inquiry. This raises the specter of non-­naturalism, which is not at all congenial to Sellars’s worldview. Again, Sellars’s diagnosis is that we have fundamentally misunderstood the function of typically philosophical terms and predicates. An instructive place to start exploring his nominalism is with Sellars’s discussion of meaning and his treatment of universals and abstract entities, long a difficult problem for serious naturalists. According to Sellars, disputants at both ends of the spectrum­—both Platonists and empiricists­—are “­factualists,” in that they treat statements like “‘Rouge’ (in French) means red” as asserting a relationship between a word and some object, class, etc. The challenge becomes, then, identifying what this object is and then fitting it into a scientific worldview, as well as fitting into such a worldview the mechanism by which we can become acquainted with this object (whether it is a Platonic form, a class, or what have you). As deVries notes, Sellars takes Wittgenstein as an inspiration for much of his philosophy of language; and Wittgenstein sees the philosopher as facing a similar dilemma, created by a similar tendency to look for some object to “animate” signs and symbols and in virtue of which they might have meaning. As Genia Schönbaumsfeld writes, philosophers tend toward “what one might call ‘substantivism’­—the thought that we must be looking for entities corresponding to our signs­—as well as the scientistic tendency of wanting to answer philosophical questions as if they were scientific ones” (Schönbaumsfeld 2017, p. 175). Schönbaumsfeld

Sellars’s Synoptic Vision  19 continues, “Both tendencies seem intimately connected, and appear to spring from the notion that signs, by themselves, are ‘dead’ (BB 4) and need, as it were, to be ‘animated’ by thoughts, intentions or the presence of abstract objects. . . . Once such a metaphysically inflated picture of what meaning must be is in place, we are naturally going to be enticed either into some form of Platonism, or a hard-­nosed rejection of such a conception. . . . Consequently, the search for something that ‘corresponds to our substantive’ will tempt us to oscillate between two extreme positions: a ‘super-­naturalist’ Platonistic conception on the one hand, and a ‘sceptical,’ or ‘naturalist,’ rejection of such a view on the other” (Schönbaumsfeld 2017, p. 176). Although Sellars’s diagnosis is similar to Wittgenstein’s, his prescription is quite different. While Wittgenstein prescribes therapy, Sellars is not averse to offering substantive philosophical theory to explain (or even revise) ordinary usage. Sellars understands that his account must satisfy a number of important criteria of adequacy. Among these, it must (1) Preserve the normative character of meaning. (2) Accomplish (1) while respecting Sellars’s austere naturalism. (3) Accomplish (2) while giving us a satisfactory account of our apparent reference to abstract objects (e.g., triangularity) and universals (e.g., red). (4) Explain meaning and reference claims involving fictional or abstract objects (e.g., “‘Einhorn’ [in German] means unicorn”). This is a tall order indeed! And yet Sellars’s ingenious, non-­relational theory of meaning arguably accomplishes all of these. Let us begin our exploration with translational meaning statements because, as O’Shea notes, Sellars’s insights are most readily seen with such statements. Consider, again, the meaning claim C. “Rouge” (in French) means red From his earliest writings, Sellars distinguished between language as a physical, natural phenomenon and language as a formal system defined by a set of norms. Thus, in his first published essay, Sellars draws “a three-­fold distinction between (1) language as norm or type, (2) language as behavioral fact, and (3) items in the second class which token, and as tokening, items in the first class” (PPE, p. 9). In the above meaning claim C, Sellars interprets the subject-­term as what he calls a “distributive singular term,” or DST.5 A distributive singular term is a term that applies to or is true of all tokens or individuals in a class. For example, “The lion is tawny,” has the surface grammar of a sentence about a single lion, but the subject term (“the lion”) is a DST; hence, the statement is really making a claim about all lions. In claim C above, “rouge” functions like “the

20  Sellars’s Synoptic Vision lion”­—as a DST­—and hence is making a claim about tokens of the word “rouge” considered as natural linguistic objects, as Sellars calls them, meaning instances of the sign-­design of the word “rouge.”6 So far, so good. What about the predicate term “red”? Sellars denies that this denotes an object to which tokens of “rouge” are related. Rather, this indicates roughly the function played by “rouge” in the French language. In other words, “red” serves as what Sellars calls an “illustrating sortal” (MFC, section IV). This means that “red” serves to indicate that “rouge” functions (in the French language) in approximately the same way that “red” does in English. As deVries notes in his commentary on Sellars’s philosophy of language, we should not, therefore, understand C as asserting a relation between “rouge” and a linguistic function. Rather, “red” is a metalinguistic sortal, and “a metalinguistic sortal attributes membership in some linguistic kind, for example, adverb or noun. An illustrating sortal is formed from an example of the kind in question. Thus in our sample sentence, ‘red’ serves as an example of the kind to which [‘rouge’] is being assigned” (deVries 2005, p. 30). As Sellars puts the point, “To say what a person says, or, more generally, to say what a kind of utterance says, is to give a functional classification of the utterance. This functional classification involves a special [illustrating] use of expressions with which the addressee is presumed to be familiar; i.e., which are, so to speak, in his background language” (MFC, p. 421). To indicate when a word is serving as an illustrating sortal, Sellars introduced the technical device of the dot quote. To indicate that “red” is serving as an illustrating sortal in C, we place it in dot quotes: •red•. Thus, C doesn’t assert a relationship between tokens of “rouge” and some second object; rather, it classifies “rouge”s as •red•s. C, then, is more perspicuously rendered as follows: C′. “Rouge”s are •red•s Again, all this means is that tokens of “rouge” function (in French) roughly as do tokens of “red” (in English). Understood this way, we can see that meaning claims are not relational at all. They do not have the form aRb Rather, they convey that a class of tokens functions in a certain way; they classify this class of linguistic tokens. Thus, as Sellars writes, “According to this analysis, meaning is not a relation for the very simple reason that ‘means’ is a specialized form of the copula” (MFC, p. 431). What about appeals to abstract entities that seem to reference them in noun-­like ways, such as references to redness or triangularity? These are given the same treatment as above.7 Thus, triangularity is the

Sellars’s Synoptic Vision  21 •triangular•, and to say that triangularity is a universal is merely to claim that it is a (one-­place) predicate. And so the problem of universals is given a nominalistic answer: “abstract entities . . .­—qualities, relations, classes, propositions and the like­—are linguistic entities. They are linguistic expressions. . . . Redness, as a first approximation, is the word •red• construed as a linguistic kind or sort which is capable of realization or embodiment in different linguistic materials, e.g. . . . the English word ‘red,’ the German word ‘rot,’ and the French word ‘rouge’” (AE I/¶1). Reference is given a similarly linguistic treatment. On a standard account, there is a reference relation between a word and an object. But Sellars denies that reference is a relation at all.8 Consider the myriad of different true referring statements we could make: “Grass” refers to grass “The smallest prime number greater than 200” refers to 211 “The Lone Ranger’s horse” refers to Silver “The official currency of the Eurozone” refers to the Euro “The last work composed by J.S. Bach” refers to The Art of the Fugue “The British House of Commons” refers to the British House of Commons It seems unlikely in the extreme that there is a single relation­—and, indeed, the same relation­—between the designating expression and its designatum in all cases given the wide variety of designata (fictional beings, mathematical beings, legal institutions, biological types, etc.). Any claim that there is such a universal reference relation will make a naturalist like Sellars uneasy, for the relation begins to seem somewhat magical in nature. Again, though, Sellars bypasses these worries by construing reference in purely linguistic (and non-­relational) terms. However, since reference is extensional, whereas meaning is intensional, we cannot give the same account of reference as we did of meaning. Thus, we cannot say, “‘The last works composed by J.S. Bach’ are •The Art of the Fugue•s,” for while these expressions might co-­refer, they do not have the same meaning. Thus, Sellars introduces the machinery of quantification and material equivalence to deal with reference. Consider the following reference claims: “The teacher of Aristotle” refers to Plato “Rational animal” refers to featherless biped Sellars analyzes reference claims in terms of material equivalence of senses: We also introduce the form Si is materially equivalent to Sj

22  Sellars’s Synoptic Vision examples of which would be •Rational animal• is materially equivalent to •featherless biped• which is true if and only if (x) x is a rational animal ≡ x is a featherless biped and •Plato• is materially equivalent to •the teacher of Aristotle• which is true if and only if (f) f(Plato) ≡ f(the teacher of Aristotle) (SM 3.VIII.§63/p. 84) Thus, we have only a material equivalence of senses rather than an identity; and this material equivalence rests on the underlying universal quantification. Senses, it will be remembered, are for Sellars “not . . . abstract objects . . . [but instead] ‘common nouns created by dot quoting’” (deVries 2005, p. 36). The practical upshot of a reference claim is not to claim some relation between two abstract entities (senses), but rather to permit (in extensional contexts) for the substitution of various co-­referring expressions. Thus, reference claims function for Sellars as limited substitution licenses and are therefore (again) metalinguistic. Now this move may seem to land the nominalist in the fire from the frying pan since Sellars’s account of reference seems to require us to quantify over whatever we are referring to. But how can we quantify over mathematical objects or fictional objects? As it turns out, Sellars’s account of quantification is more austerely naturalistic than other, dominant accounts; and (crucially) one can quantify over a domain without making an ontological commitment to the existence of any objects in that domain. On the Quinean “objectual” account, to quantify over a domain is to commit, ontologically, to these objects. On objectual interpretations, “(∀x)φx” is true only in case every object in a domain is in the extension of φ. Successfully quantifying over Euros, unicorns, or Bach compositions ontologically commits us to the existence of these items in the world. But Sellars rejects the objectual interpretation of quantification, arguing that it leaves the relation between the variable and the array of objects over which we are quantifying mysterious. As Sellars writes, The first and rather forlorn sub-­strategy, (A-­1), is to equate Something is a lion with Leo is a lion or Nixon is a lion or Gibraltar is a lion or (perhaps) the number 3 is a lion, or etc. This sub-­ strategy, though not without its temptations, runs into familiar road-­blocks. The “etc.” is doing a lot of unexplained work.

Sellars’s Synoptic Vision  23 As Ms. Anscombe has pointed out, the “etc.” (or the dots which may replace it) is not the “etc.” of laziness. (NAO 1.§13/p. 7) —particularly if comAn objectual interpretation of quantification­ bined with a relational interpretation of semantic terms­—makes the ability of terms to connect with all of these items in the universe seem truly magical. Sellars proposes an account (of both quantification and semantic notions) that is more naturalist-­friendly: He proposes an alternate, “substitutional” account of quantification according to which “‘something is a lion’ is true º some statement which makes a determinate reference to an object and classifies it as a lion is true” (NAO 1.§14/p. 7). Thus, for example, on a substitutional account, universal quantifiers are treated as universal conjunctions: “Universal . . . quantifiers are logical locutions that have the expressive function of making propositionally explicit conjunctive . . . substitutional commitments. Attributing commitment to a claim of the form (x)Px is attributing commitment to all claims of the form Pa” (Brandom 1994, p. 434). The problem­—and this problem was recognized by Sellars­—is that this interpretation of quantification only works if we have a word in our language for every item over which we quantify. As Sellars writes, “Statements are made in a language, and the resources of any natural language are always limited­—certainly with respect to determinate referring expressions . . . [A proponent of this strategy] must rather rely on the fact that a language not only consists of more than the grammatical strings which are actually deployed at any one time (which is obvious), but also of more than the grammatical strings which are available for deployment. It also includes, in a sense difficult to define, the resources by which the language could be enriched through being extended in specific ways” (NAO 1.§15/pp. 7–8). In a related vein, Mark Lance writes, “We not only do not, but, in the sense of possibility which I would urge to be relevant to this aspect of semantic content, cannot refer to specific hairs long since decomposed, nor to specific neutrinos in distant galaxies, individual grains of sand in Death Valley, or thoughts of beings with whom we will never communicate. We can, however, quite easily quantify over each of these” (Lance 1996, p. 489). Lance introduces the distinction between referential and arbitrary uses of singular terms to get around this difficulty. We use a singular term referentially when we mean it to refer to a particular entity, as when we say, “John has a fever.” We use a singular term arbitrarily when we don’t intend for it to refer to any particular item. For example: “We know there are transcendental numbers larger than any previously generated prime. Let α be such a number. Then α is not a ratio of integers” (Lance 1996, p. 489). As Lance notes, the use of α is not fictional, as there really is such a number, and it really is not a ratio of integers. But “equally clear, however, is that the use of ‘α’ is not referential in the sense noted above for it would be out of place to ask which number α is. The

24  Sellars’s Synoptic Vision whole point of this use is that it does not commit one to α’s being any particular number” (Lance 1996, p. 490). A substitutional interpretation of quantification requires that we permit the substitution not only of any term in the language but also of any arbitrary term. As Lance writes, The substitutional rendering was indeed onto something in attempting to assimilate universal quantifications to infinitary conjunctions. The difficulty arose by neglecting the existence of arbitrary terms and, hence, in failing to recognize that these offer a sort of expressive resource unavailable to a language with only referential terms. The solution to the difficulties faced by substitutional interpretations of ordinary language nominal quantifiers lies in seeing that commitment to (∀x)φx carries with it commitment not only to the referential substituends of this sentence, but to the arbitrary ones as well. (Lance 1996, p. 489) The substitutional interpretation of quantification, by giving a linguistic reading of quantification, has a number of advantages over objectual accounts. First, as Sellars noted, we don’t have to explain how, when we quantify, we can somehow referentially connect to everything that exists or ever has existed. Such “magical” referential connections should make naturalists uneasy, to say the least. Second, on Sellars’s suggestion, quantification licenses certain substitutions or inferences and does not by itself ontologically commit us. Thus, the statement “All unicorns have a horn” [i.e., (x) Ux → Hx] merely licenses a certain substitution [and the corresponding inference: i.e., “Rarity is a unicorn; therefore, Rarity has a horn.”]. Nevertheless, this is merely a question of whether we can substitute into the quantifier in a truth-­preserving way; we are not thereby committed to the existence of unicorns. This gives us a satisfactory account not only of quantification over fictional items but also of quantification over numbers, etc. Thus, while there is a lot of complicated machinery lurking in the background, Sellars’s account of reference is relatively straightforward. “A refers to B” does not assert a relation between a word, A, and an object, B, but instead conveys a material equivalence between the senses expressed by the two expressions, A and B. Later, purely intralinguistic accounts of reference (such as Brandom’s anaphoric account,9 which he explicitly compares to Sellars’s) can be seen as sophisticated extensions of the strategy proposed by Sellars. Does Sellars’s account satisfy the criteria of adequacy we set out above? Let us go through them one by one. First, Sellars’s theory needed to (1) Preserve the normative character of meaning. It is no doubt tempting for philosophers of a naturalistic inclina—say, tion to interpret Sellarsian linguistic functions naturalistically­

Sellars’s Synoptic Vision  25 as causal-­functional roles. But it is inaccurate to read Sellars this way; linguistic functions are normative roles. Thus, a meaning-­claim doesn’t relate a word to a causal-­functional role; rather, it classifies it according to the constellation of rules and norms constituting its identity (much as we classify a chess piece by identifying it as a pawn). Of course, for Sellars, normative assessments have descriptive/behavioral consequences.10 This is consistent with the relevant linguistic functions­—picked out by the illustrating sortals­—being normatively defined linguistic roles. Normativity is ultimately concerned with behavior, and while we don’t always behave in accord with the norms we espouse, Sellars thinks that it belongs to the concept of a norm or principle that in general, and subject to qualification, people will act on the norms and principles they espouse. Sellars expresses this commitment in what O’Shea calls the “norm/nature meta-­principle”: “Espousal of principles is reflected in uniformities of performance” (TC, p. 216; quoted in O’Shea 2007, p. 50). For example, if you know that two people are playing chess, you can predict the kinds of moves they will make based on the rules of chess (e.g., since it is a rule of chess that the bishop moves along the diagonal, you can predict that the players will move their bishops along the diagonal). This dual behavioral/normative feature of Sellars’s philosophy is his “naturalism with a normative turn” (again, O’Shea’s phrase) and demonstrates that Sellars’s use of linguistic functions is consistent with preserving the normativity of meaning. While meaning per se is not the topic of the present volume, normativity lies very close to the heart of our concern, and we will return to this crucial issue a number of times. Other kinds of normative ascriptions have corresponding behavioral consequences. As will be elucidated below (and at length later in the book), the content of linguistic items is constituted by the norms governing their use. Norms related to use ultimately concern people’s behavior with regard to an item; thus, competence with a term implies certain behavioral consequences. A person competent in the use of the term “red” will be disposed not to move from “X is red all over” to “X is green all over” or to “X is not colored”; and if a person were so behaviorally disposed, we would simply deny that she knew the meaning of “red.”11 Again, these behavioral regularities in accordance with norms functional roles (they are explained not because meanings are causal-­ aren’t) but because people generally do what they ought to do (thanks to proper training, about which more will be discussed in Chapter 9). (2) Accomplish (1) while respecting Sellars’s austere naturalism. Sellars’s understanding of “means”-­locutions should not offend even the strictest naturalist. DSTs are a common linguistic locution, and in “means”-­claims, they simply quantify over linguistic tokens, which are natural items or events.12 (And Sellars’s substitutional account of quantification is arguably more naturalist-­friendly than standard accounts.)

26  Sellars’s Synoptic Vision Illustrating sortals might give the naturalist more pause, but they need not. As deVries notes, “The fact that expressions have meanings is no more ontologically troublesome for Sellars than the fact that garden tools have uses” (deVries 2005, p. 30). As noted above, many artifacts are defined by various prescriptions regarding their use; and this, as it turns out, is also true of linguistic tokens. There is nothing mysterious or non-­ natural about this. Again, though, naturalism is central to discussions of ethical theory, and the degree to which Sellars’s project­—particularly his ethical theory­—is naturalistic will be a theme that weaves throughout the present work. (3) Accomplish (2) while giving us a satisfactory account of our apparent reference to abstract objects (e.g., triangularity) and universals (e.g., red). Again, as Sellars wrote, “Abstract entities . . . are linguistic entities. . . . Redness, as a first approximation, is the word •red•” (AE I/¶1). Thus, a purely nominalist account is given of universals and abstract entities. (4) Explain meaning and reference claims involving fictional or abstract objects (e.g., “‘Einhorn’ [in German] means unicorn”). Sellars’s account does very well here, as on his account the predicate term does not (attempt to) refer to an object in the world in the first place. A meaning claim like the one in (4) is rendered as “Einhorn”s (in German) are •unicorn•s Sellars’s theory also easily handles logical connectives and other non-­ referring expressions whose meaning is clearly constituted by their grammar. Thus, to use Sellars’s example, “Niet” (in Russian) means not is rendered as “Niet”s (in Russian) are •not•s thereby obviating the need for us to find a reference for “not.” Indeed, the fact that something like Sellars’s account almost has to be true for logical connectives and the like­—and probably for abstract objects like numbers­—combined with the fact that his account gives us a naturalist-­ friendly account of universals and abstract entities, offers us a powerful reason to accept it as a general account of the function of meaning claims.

Sellars’s Synoptic Vision  27 In general, Sellars’s nominalist strategy for semantic, logical, moral, modal, and other sorts of “philosophical” vocabulary is to go metalinguistic. For example, his strategy for understanding causal laws, such as one asserting a lawful causal connection between A and B, is to try to navigate a middle course between the empiricist, who claims that “the relation consists in a constant conjunction of A with B” (CDCM, p. 267/§56), and the rationalist, who (arguing for real connections and our epistemic access to these connections) claims that “the relation is a logical relation of entailment. . . . Actually, it might better be called ‘natural’ or ‘physical’ entailment” (CDCM, pp. 267–8/§56). Sellars argues that such laws are best understood as covertly metalinguistic claims asserting an inference ticket, which licenses assertion of B upon assertion of A (at least upon assertions of A to which one is entitled). Now, these claims are “covertly metalinguistic,” in that they often occur in the “material mode”­—that is, we often make claims about linguistic items by talking about the corresponding objects. This is why Sellars often calls modal and other related discourse a “transposed language of norms”­—transposed because discourse that is properly understood as metalinguistic is “transposed” into the material mode and takes place in what is apparently the object language.13 But, as Sellars argues, trying to understand the language of norms as actually being a part of the object language­—for example, understanding modal claims as part of the descriptive object language of science­—is “to court serious philosophical perplexity” (CDCM, p. 302/§102). For not only do we then have to find something in the world that we are describing with such talk of norms (a difficult task for the naturalist), but we are also faced with the difficulty of giving a descriptivist account of the subjunctive and counterfactual force of modal statements, prescriptives, and the like (a project whose prospects Sellars regards pessimistically and to which we will return). Again, though, if we understand these claims metalinguistically, then we don’t have to understand anything more mysterious than linguistic tokens and their uses/roles. This, for Sellars, is the most promising path for the naturalist to take. The key, for Sellars, is to break out of the mindset (one that comes so easily to the naturalist) that in stating a causal law or a moral claim or a meaning claim we are describing anything. As Sellars forcefully puts the point, “Once the tautology ‘The world is described by descriptive concepts’ is freed from the idea that the business of all non-­logical concepts is to describe, the way is clear to an ungrudging recognition that many expressions which empiricists have relegated to second-­class citizenship in discourse, are not inferior, just different” (CDCM, p. 282/§79); and famously, on the subject of knowledge: “In characterizing an episode or state as that of knowing, we are not giving an empirical description of that episode or state; we are placing it in the logical space of reasons, of justifying and being able to justify what one says” (EPM, VIII.§36/p. 169). Thus, we see from another perspective

28  Sellars’s Synoptic Vision how essentially and ineliminably our use of normative discourse is woven into our engagement with the world: Our deployment of logical vocabulary, our statement of causal laws, and so on are in the language of norms. It would of course be foolish to declare victory yet. Much more needs to be done; and the particular route Sellars takes in defending nominalism might seem to trade one set of problems for another. One can easily imagine the various objections that might be raised against Sellars’s account at this point: By giving a purely intralinguistic account of meaning and reference, haven’t you actually abandoned naturalism by cutting language off from the world? By giving a metalinguistic account of norms (including moral and scientific norms), haven’t you reduced norms to sets of socially approved roles and practices, thereby opening the door to a toxic relativism? If norms are metalinguistic, since there are many languages, does this raise the problem of incommensurability­—and the specter of there being as many normative systems as there are languages? Some of these problems will take the entire book to address. Some we can address right away. As to the last worry, Sellars understands metalinguistic roles as not tied to any particular language. For example, •red• is not a specifically English metalinguistic term. Rather, it names a concept that can be realized in multiple different languages, and is tokened in these different languages by a variety of natural linguistic objects (“red,” “rot,” “rouge,” etc.). Similarly, causal laws represent inference licenses, connections between senses that can be realized in multiple languages. Thus, There is lightning Therefore, there will be thunder can be rendered as Es gibt Blitz Deshalb wird es Donner geben because •thunder•s are •Donner•s and •lightning•s are •Blitz•s. We can also address the worry that Sellars’s approach to semantic vocabulary cuts language off from the world. Sellars doesn’t deny that language is connected, in a rich variety of ways, to the world. He just denies that there is a single relation that we can isolate as the reference relation or the meaning relation (and indeed thinks that we only tie ourselves in philosophical knots when we construe meaning, reference, and other such notions as relational in the first place). To understand how language is connected to the world, one must understand the kinds of moves one can make using the language.14 First, there are language-­language transitions. These are transitions from a position in the language game to another position in the language game. These are characteristically

Sellars’s Synoptic Vision  29 inferences, as when one moves from “S is red” to “S is colored.” For Sellars, these moves are most central to meaning (and arguably constitutive of meaning).15 But the language game involves other kinds of moves; two kinds of moves chiefly concern us here. First are language-­ entry moves, when one moves from an item that is not a position in the language game to a position in the language game. For example, one can make an observation and form the judgment “The light is red.” Finally, there are language-­departure transitions, where one goes from a position in the language game to something that is not a position in the language game (namely, an action). For example, one might form the intention to raise one’s hand and then do so. It is these language-­entry and language-­ departure transitions that provide the interface between language and the world. Suppose, for example, that “a” is a token linguistic expression, one that expresses the concept •a•. Suppose that it is a singular referring expression. Now, Sellars doesn’t understand reference in terms of word-­ world relations, but he does think that for •a• to do its job, tokens of “a” must be involved (qua natural linguistic objects) in the causal nexus in the right way, a way that is easy to state generally but somewhat vaguely (as Sellars does): In explaining the job of referring expressions in the base language, it is unilluminating to say that their job is to refer to certain objects. We must look instead to the semantical rules and uniformities in which they are involved. Thus: (1) Non-­ demonstrative referring expressions must themselves belong to the “natural” order and be connected with objects in a way which involves language entry transitions, intralinguistic moves (consequence uniformities) and language departure transitions (willings-­out-­loud). (2) There must be a relatively stable, if skeletal, framework of propositions (involving these referring expressions) which “describe the spatiotemporal location of these objects with respect to each other.” (3) A proper part of this skeletal framework must “specify location of the language user in his environment.” (4) Rehearsings of this skeletal framework must gear in with the use of demonstratives to “specify the location with respect to here-­now of the objects with which the referring expressions are correlated.” (SM 5.IV.§30/pp. 125–6) Thus, for a referring expression (like •apple•) to do its job, its tokens (“apple” or “Apfel,” or whatever) must be involved in a complicated

30  Sellars’s Synoptic Vision world-­language-­norm nexus; and no doubt this will involve language-­ entry transitions involving apples and language-­ departure transitions involving apples. But while this causal involvement is necessary for the expression to do its job, there is no one relation here we can single out as the referring relation. (And, as argued above, even if we were to isolate such a relation, would we find that same relation when we turned to other referring expressions in other domains, such as mathematics16 or law or what have you?) Thus, we should not fear that Sellars’s semantic theory has cut off language from the world; rather, we should appreciate that he has kept language intimately tied to the world while not attempting to appeal to mysterious (and ultimately difficult to naturalize) word-­ world relations such as reference. There is at least one other way in which language is connected to the world for Sellars. This is through Sellars’s (notoriously obscure) relation of picturing. Perhaps Sellars’s most lucid explanation of the picturing relation is in BBK, where Sellars distinguishes between signifying and picturing. Signifying is a relation between intentionally characterized items. Thus, if I say “Rot” signifies red I am not, as we saw above, asserting a relation between language and the world. Rather, I am asserting “Rot”s are •red•s As we saw, such claims classify “rot”s; they do not relate them to meanings, senses, or what have you. (Sellars talks in BBK of signifying as a “relation” belonging to the intentional order, but this strikes me as careless talk that encourages reification of linguistic roles and other related moves that are not congenial to his nominalism. As noted above, he elsewhere consistently denies that meaning claims have the form “ArB,” so it is perhaps best to stick with this non-­relational understanding of meaning claims.) However, language can only signify if it also pictures. Picturing, however, is a world-­world, not a word-­world, relation. That is, it takes overt linguistic performances (speech tokens) or mental events considered as natural linguistic objects­—not as linguistically characterized events­—and maps them onto the world. Consider, for example, a robot with a camera and internal memory that can explore an area and build an internal “map” of its environment. Sellars would say that there is an isomorphism between the internal representations of the robot and its external environment, even if we wouldn’t be tempted to characterize these internal states linguistically. Thus, in a sense, the internal states of the robot “picture” its environment. Similarly, our neurological and phonetic behavior

Sellars’s Synoptic Vision  31 “maps” our environment. But this is, again, a world-­world relation. For example, our sensations of whiteness “map onto” things in the world that are white in more or less systematic ways (e.g., there is an isomorphism), but it would be incorrect to say that sensation belongs to the conceptual order. A sensation is of whiteness in an analogical way, derivative upon our linguistic concept of whiteness; but the sensation is itself nonconceptual. It is in the “real order,” not the “logical order,” as Sellars puts the point. But the dependency goes both ways: If we cannot characterize our sensation except parasitically with reference to linguistically characterized concepts, these concepts cannot signify if our linguistic and neurological systems don’t picture the world. I am not going to explore Sellars’s notion of picturing in any detail. It is an interesting and fruitful topic, and in my view represents for Sellars another crucial piece linking language and the world. But there is also widespread disagreement as to the precise details of Sellars’s account of picturing and its centrality to his overall philosophical project. Further, this element of his philosophy is of very limited relevance to Sellars’s ethical project, and so a detailed exploration would take us too far afield. In any case, I think we can put to rest the worry that Sellars has, with his metalinguistic treatment of meaning and reference, cut language off from the world. But some problems related to Sellars’s naturalism will require lengthier treatment. The question of relativism will have to be postponed until Chapter 4. There is one problem we can (and must) tackle right away because it lies at the heart of Sellars’s philosophical enterprise and has serious consequences for how the current project is carried forward. Sellars has tried to establish the naturalist bona fides of his theory by making the strong claim that the normative is causally reducible (though logically irreducible) to the natural. There is reason to believe that this claim is implausibly strong, and indeed that Sellars himself has provided an argument undercutting his own view. In reconciling various arguments Sellars makes, I will maintain that we have to abandon his thesis of causal reducibility. Does this mean abandoning Sellars’s naturalism? Fortunately, I think the answer to this question is no, but why we need to abandon the thesis of causal reducibility­—and why this is consistent with a fairly strict naturalism­—will require careful discussion. It is to this issue that we turn in the next chapter.

Notes   1 For those seeking a more detailed presentation of Sellars’s philosophy, two good resources are deVries (2005) and O’Shea (2007).   2 For more on this argument, see Wolf and Koons (2016, chapter 2, esp. section 2.2).  3 Some (Stefanie Dach, private correspondence) have suggested that artifacts like this are defined teleologically rather than prescriptively. Certainly,

32  Sellars’s Synoptic Vision hammers have an intended purpose, and one doesn’t necessarily do anything wrong or incorrect if one uses a hammer for a different purpose. Nevertheless, the world of artifacts is the world not just of purposes but of norms: While hammers, doors, windows, and plates can all be repurposed without error, there is a shared understanding of the normal use of things, and this shared understanding sets the identity of these things. To take a trivial example, the (supposed) humor of Ariel’s attempt to identify and use human artifacts in the Disney film The Little Mermaid is that she is using them wrong (for example, using a fork as a comb). (And in any case, are socially instituted purposes and norms so easily separated?) Finally, the next example in the text­—artifacts whose identity is almost wholly shaped by rules of correct use­—is a much clearer example of prescriptively defined artifacts.   4 Sellars has us imagine (SRLG, §59/p. 344) a chess game where the pieces are different makes of cars and the squares different counties of Texas.   5 See, for example, AAE; SM (chapter 3).   6 For Sellars, “rouge,” “ROUGE,” and “rouge” are all sign-­designs that mean “red.”   7 I am eliding some complexities here; Sellars argues that there are contexts in which terms like “triangularity” function as DSTs, not as predicates. But again, he gives a nominalist account of this too. The details of his account are spelled out in AE.   8 As deVries notes, “Sellars treats ‘reference’ strictu sensu as a term in the semantic metalanguage. Despite his talk in the passages from Naturalism and Ontology about a ‘word-­world connection,’ he cannot maintain that reference is some particular object-­level relation between words and objects. Unfortunately, he occasionally writes in Naturalism and Ontology as if this is what he meant, but such passages must be read as shorthand for a longer story” (2005, p. 71).   9 Brandom (1984). 10 See, for example, Sellars AAE (§30/p. 188). 11 Of course, norms­—particularly moral norms­—involve ideals more than mere uniformities. That is part of what makes them norms rather than descriptions: We are able to deviate from them. So the implication of behavioral regularities is highly defeasible. Further, normative terms will not simply decompose into a normative component and a set of behavioral regularities (although Sellars sometimes talks as though they will). We will discuss in the next chapter how, although use of normative terms implies certain behaviors, there is no sense in which any normative term can be reduced to or associated with a naturally characterizable set of causal or behavioral regularities. 12 We will later have reason to revisit this claim­—that linguistic items can be regarded purely as naturalistically characterizable items or events­ —and regard it as something of an oversimplification, but let us for now rest with Sellars’s characterization. 13 Stefanie Dach has questioned (private correspondence) how moral utterances can be metalinguistic. “Killing is wrong” is not obviously metalinguistic; and, further, moral utterances are expressions of intentions, and it is not clear how such expressions would be in the metalanguage. I have not yet introduced the theoretical apparatus to answer this question; I will return to it in Chapter 15 when I give an overview of how Sellars’s moral theory is consonant with his strict naturalism. 14 See, for example, SRLG §22. 15 This is a point of disagreement among Sellars scholars. Some hold that language-­ entry and language-­ exit moves, which I take up next, are also

Sellars’s Synoptic Vision  33 indispensable to meaning. But all agree that at least language-­language moves are essential to meaning for Sellars. 16 Some points of clarification: When Sellars argues that referring expressions must be involved in the above complex causal relations to the world (even though reference doesn’t consist in these relations, and indeed isn’t even relational at all), he may be confining his discussion to factual expressions involved in the picturing relation outlined below­—thus, expressions involved in causal-­explanatory discourse. So we can make reference claims regarding expressions not necessarily involved in such relations, such as, say, mathematical expressions. Second, arguably mathematical claims often are involved in this causal nexus: We observe that there are, say, three cups on the table. Puzzles only arise when we try to figure out what the number 3 refers to and try to understand reference as a relation between a word and some entity. Ergo, trying to understand reference as a relation between language and some extralinguistic reality creates only metaphysical puzzles and is, for Sellars, a project best abandoned.

2

A New Naturalism

As I have noted, the normative is for Sellars causally reducible but logically irreducible to the natural. I have so far covered this issue only lightly, but it is time to give it a more detailed examination because this position turns out to be central to Sellars’s philosophical project. Serious objections have been raised concerning the tenability of Sellars’s thesis of the “causal reducibility, but logical irreducibility” of the normative, and we need to address these objections before moving on. Just as critically, Sellars’s thesis turns out to be inconsistent with some of his other central commitments, and resolving these inconsistencies is imperative if we are going to resolve several outstanding issues later in this book­—issues concerning the relation between the normative and the natural, concerning moral epistemology, concerning moral motivation, and more.

2.1 The Problem(s) Let us begin by explaining Sellars’s “causal reducibility, but logical irreducibility” thesis and then noting the problems it raises for his larger philosophical project. Sellars’s view of the relation between the scientific and manifest images is complex (and perhaps not always consistent). The core of his view is the scientia mensura principle­—that insofar as we are concerned with describing and explaining a world of physical objects, the manifest image is the world of mere appearances, and that the scientific image is our guide to what is actually real: “Speaking as a philosopher,” Sellars writes, “I am quite prepared to say that the common sense world of physical objects in Space and Time is unreal­—that is, that there are no such things. Or, to put it less paradoxically, that in the dimension of describing and explaining the world, science is the measure of all things, of what is that it is, and of what is not that it is not” (EPM IX.§42/p. 173). As we have seen, though, Sellars doesn’t think that the point of normative vocabulary (be it modal or semantic, epistemological or moral) is to describe the physical world, and this insight is the key to his attempted reconciliation of the manifest-­image

A New Naturalism  35 notion of persons (an inherently normative category) with the scientific image (whose business is wholly descriptive and explanatory). Despite the fact that normative utterances are not in the first instance descriptive, Sellars thinks that there are important connections between such utterances and descriptive regularities, and that such regularities will ultimately find their place in the scientific image. He writes, “It is indeed important to see that (in principle) the world, including human behavior, could be described and predicted without using semantical discourse. But the proper way to interpret this fact is not by propounding an ‘extensionalist thesis’ according to which everything can be said without using semantical discourse, but rather offering a careful account of the interrelationships which would obtain between semantical discourse and an ideal behaviorese” (MMB, §4.2/p. 94). However, even if “an ideal behaviorese” could fully describe human behavior, this doesn’t imply that normative vocabulary is eliminable. For there are things that we can and must do with normative vocabulary that we cannot do with merely descriptive vocabulary: “Whatever users of normative discourse may be conveying about themselves and their community when they use normative discourse, what they are saying cannot be said without using normative discourse. The task of the philosopher cannot be to show how, in principle, what is said by normative discourse could be said without normative discourse, for the simple reason that this cannot be done” (SSMB IX/p. 256).1 And as we saw in the previous chapter, the normative is ineliminable, for Sellars, as any kind of modal statement (whether one of causal or logical relations or one of moral or epistemic necessity) is couched in the language of norms. The best way, I have said, to understand Sellars’s thesis is as follows: Equivalences between normative and natural claims, such as “X is right ≡ X maximizes utility,” should be understood (in contemporary terms) as conveying an equivalence in extension between these two claims but not as asserting an equivalence of meaning. As noted above, Sellars asks us to imagine that we have a purely extensional language (PM-­ese), and he envisions an equivalence between, say, intentional claims and claims in this purely extensional language.2 Sellars’s claim is very strong: It is that every intentional claim (such as “Otto believes it is raining”) is correlated with an extension that can be specified in purely descriptive (perhaps behavioral or neurological) terms. This is the sense in which the normative is causally reducible. However, the normative cannot be eliminated because while it is causally reducible, it is not logically reducible. The normative doesn’t even aim at description in the first place, and giving a physical description of the person’s behavioral dispositions (or neurological state) in principle is not saying, for example, that Otto believes it is raining.3 To give an analogy, contemporary moral naturalists often say that moral and natural terms can have the same extension but not the same

36  A New Naturalism meaning, for asserting a synonymy (rather than mere co-­extensiveness) between the moral and the natural would commit the naturalistic fallacy. In a similar vein, Sellars is saying that normative expressions in the manifest image always retain a logical or conceptual surplus over their scientific image counterparts and therefore resist complete elimination in favor of the latter. As Sellars puts the point in “Phenomenalism,” “the idea that persons ‘really are’ such multiplicities [of microparticles] does not require that concepts pertaining to persons be analysable into concepts pertaining to sets of logical subjects. Persons may ‘really be’ bundles, but the concept of a person is not the concept of a bundle” (PH VIII/p. 101). Crucially, the scientific image will give us a descriptive analysis of the normative concepts as such, and so we will be able to associate a naturalistic extension with normative claims. Thus, as O’Shea writes (discussing the closing paragraphs of the preface to SM),4 “It is clear that the adequate scientific or naturalistic explication of human beings’ rational recognition of norms is to be an account of such norms qua the norms that they are, and not merely an account of a separate scientific subject matter” (2009, p. 200).5 There are at least two problems with Sellars’s thesis of causal reducibility but logical irreducibility. One objection, raised by Olen, is that it generates what he calls an “explanatory gap.” The second objection is that it is inconsistent with Sellars’s own argument for scientific realism. Addressing these objections­—and making Sellars’s own arguments consistent­—will ultimately require abandoning the thesis of causal reducibility.6 But let’s not get ahead of ourselves. Sellars’s thesis is, as noted above, a strong one: He claims that “the rule-­governed uniformities (not all of which, of course, are purely syntactical) which constitute a language (including our own) can, in principle, be exhaustively described without the use of meaning statements” (NAO 4.VII.§65/p. 78). Olen asks, if empirical psychology can give a complete descriptive account of human behavior, “why [are] philosophical accounts of behavior ... needed” (Olen 2016, p. 117)? Granted, as Olen notes, Sellars argues that “normative discourse­—especially as it concerns rules­—cannot be reduced to purely extensional terms (in what would be a purely descriptive behavioral language)” and “normative terms must be seen as logically irreducible because their linguistic role cannot be accounted for by descriptive terms” (Olen 2016, p. 119). But Olen is left wondering “why we need explanatory resources over and above those found in behavioral psychology” (Olen 2016, pp. 119–20). Olen suggests that Sellars’s later accounts of normativity face the “explanatory gap”: If we can in principle explain all of human behavior in terms of descriptive psychology (or other empirical sciences), then what is left to explain by a set of purely normative concepts?

A New Naturalism  37 The problems Olen sees fall into roughly two categories: First, if a purely descriptive account fails to capture some aspect of agency­—some aspect that only normative language can capture­— what is the difference between capturing (accounting for, characterizing, etc.) and explaining? Second, by meshing normative and behavioral concepts, hasn’t Sellars sharpened the problem? If rules are “generalization[s] written in flesh and blood,” then aren’t normative concepts doing explanatory work? And can’t this work be done by counterpart scientific or psychological concepts? And in that case, what need is there for a specifically normative set of concepts? These are both difficult challenges for Sellars. As to the first point, I think Sellars can evade the “explanatory gap” because, as a nominalist, he resists the “factualizing” move that tries to understand, say, meaning as a relation or thing to be explained descriptively. Rather, as a pragmatist, what Sellars is trying to characterize or account for is what we are doing when we make a moral (or semantic or epistemic . . .) claim. For as I have already noted, although Sellars (in)famously embraces the scientia mensura principle, he is also insistent that the point of modal and normative vocabulary isn’t to describe or explain: “Once the tautology ‘The world is described by descriptive concepts’ is freed from the idea that the business of all non-­logical concepts is to describe, the way is clear to an ungrudging recognition that many expressions which empiricists have relegated to second-­class citizenship in discourse, are not inferior, just different” (CDCM, §79/p. 282). Normativity turns out not to be a thing or a relation or a property. It is a grammatical mode, and no amount of talking in a descriptive mode will ever amount to making a normative claim. Thus, you can exhaustively describe the behavior of people acting under a set of rules and normative constraints, but to do so would involve normativity in a “shallow” sense (one would be as an anthropologist, describing a culture without undertaking any of the described normative commitments). Nor would your description justify any of the behavior in question. To undertake a commitment, we must leave the descriptive mode and enter a different mode­—and hence do something that is not captured by (descriptive) explanation. Sellars would argue that the “explanatory gap” argument proceeds from the thought that once you have described human behavior, you have done all the engagement with human behavior that you can. But again, as Sellars stresses at the end of PSIM, if we have only the resources of descriptive language, we lose sight of persons because we can no longer talk about norms, responsibility, reasons, and so on. Thus, the “explanatory gap” argument is based on a false presupposition.

38  A New Naturalism Sellars’s answer to the second point is tied to his answer to the first. To say that normative claims are not descriptive is not to say that they don’t convey descriptive content. Thus, to know that Smith uses “red” as a •red• conveys certain behavioral information about him­—e.g., that he will, ceteris paribus, describe this apple as “red,” that he will deny that this green object is red, and so on. However, as Sellars has noted, while normative claims convey descriptive information along these lines, they say something that can only be said using normative terms. Thus, like explanatory language, normative language is used to convey descriptive information; but the primary purpose of normative language isn’t explanatory, and hence it cannot be replaced by such language. A helpful way of understanding Sellars’s position is by distinguishing, as Mark Lance and Heath White do, between two different approaches to agency and personhood: metaphysical approaches and stance approaches (Lance and White 2007). A metaphysical approach “begins by asking what a person, or agent, or subject is, perhaps by attempting to supply necessary and sufficient conditions” (Lance and White 2007, p. 2). Lance and White classify Sellars as a stance theorist who begins “not with a straightforward account of persons or freedom or subjects, but of what it is to take something to be a person/free agent/subject” (2007, p. 2). Now obviously, one must give conditions for when it is appropriate to take something to be a person; and one might think that one could collect these conditions into a metaphysical account of personhood. In this way, the stance approach would collapse into the metaphysical approach. Thus, let us designate P the set of conditions under which it is appropriate to attribute personhood to x. Can’t we then just give a metaphysical account of personhood as follows? x is a person iff P(x) What is left for the stance approach to do? (I take it that this is a way of reformulating Olen’s worry.) Lance and White’s reply to this line of thought is that such an approach would “miss whatever non-­cognitive aspects there are to, say, personhood” (2007, p. 3). Thus, while it is true that x is a person iff P(x), it is just as true that there is a conceptual connection between the personhood of x and holding x responsible, taking x to have commitments and entitlements, and so on. These are the noncognitive elements of personhood and agency that are not captured by the metaphysical account. (“Noncognitive” is probably the wrong term to use on the present account. For Sellars, states like “holding someone responsible” or “taking someone to have commitments and entitlements” would be states with cognitive content. But the crucial point for Sellars is that no specification of descriptive or factual circumstances will ever capture what you are doing in making a normative claim. For as I have said, normativity

A New Naturalism  39 is not a thing, it is a grammatical mode; and no amount of talking in a descriptive mode will ever amount to making a normative claim. Thus, there is no way of eliminating the normative in favor of the descriptive because describing and prescribing are different things. This is a claim that is central to Sellars’s view on normativity and one I will return to repeatedly throughout this book.) Is this a satisfying response to the kind of worries raised by Olen? Of course it will satisfy some and not others. The key vulnerability in Sellars’s position lies in his claim that normative claims are causally reducible. In Lance and White’s language, this would be the claim that there is some way of filling out the claim “x is a person iff P(x)” where the predicate P is spelled out in purely naturalistic terms. If this is possible, then this would create the concern that nothing is left to do for specifically normative language. As it turns out, the thesis of the causal reducibility of the normative is inconsistent with Sellars’s argument for scientific realism. Thus, in rejecting the causal reducibility of the normative, we can bring consistency to Sellars’s account and eliminate the source of Olen’s worry (i.e., rejecting the possibility of a metaphysical approach to personhood)­—while setting the stage for a better form of naturalism that will help us address other issues later in the book (such as the connection between judgment and moral motivation). But while we must abandon the thesis of causal reducibility, we do want to hold on to some version of the claim attributed to Sellars above: While normative claims are not descriptive, they do convey descriptive content. Part of what the metaphysical approach is right about is that the stance approach owes us a story about what licenses our taking a particular stance; and this story has to be consistent with a robust naturalism. As Lance and White say, “The danger facing stance approaches is that they either implicitly advocate an irrealist answer to the metaphysician’s question [What is a person?], when this at least needs arguing, or they simply fail to address the question of when a certain non-­cognitive attitude is appropriate” (2007, p. 3). Thus, I hope to answer Olen’s challenge by denying causal reducibility. But the advantage of causal reducibility was that it gave a concrete answer to the metaphysician’s question and allowed the normative to fit snugly within Sellars’s scientific naturalism. So we must demonstrate how a “new naturalism” can still fit within Sellars’s austere scientific naturalism. Let us begin by looking at Sellars’s argument for scientific realism.

2.2 Sellars’s Argument for Scientific Realism To understand the inconsistency in Sellars’s austere scientism, we must first understand his powerful argument for scientific realism­—and how this same argument will establish a more complicated naturalism about the normative. As Marc Lange notes, in arguing for scientific realism,

40  A New Naturalism Sellars is responding to what Hempel calls the “Theoretician’s Dilemma”: Suppose you have a theoretical law, couched in terms of unobservable entities, of the form “All T1s are T2s.” But suppose you have bridge laws tying these unobservables to observables: “All T1s are O1s” and “All T2s are O2s.” Thus, it would seem that the theoretical law can simply be reformulated in terms of observable entities­—“All O1s are O2s”­—and talk of unobservable entities is eliminable. Now, one can challenge this conclusion about the eliminability of theoretical concepts (e.g., T1) in two different, but related, ways: (1) One could argue that there are, in fact, no bridge laws linking theoretical concepts like T1 and observation concepts like O1. This would in essence be arguing that the theoretical kind is “shapeless” at the observational level: that is, that the set of observations that instantiate the theoretical concept do not form a kind that is recognizable without use of the theoretical concept. For example, let us assume for the moment (for I know this is controversial) that species are natural kinds. Arguably, there simply are no facts about speciation that are recognizable at the level of, say, chemistry or quantum physics. Call this the no bridge laws strategy. (2) One can argue that observation-level generalizations (e.g., “All O1s are O2s”) are not independently law-­like and/or do not have any independent justificatory force­—independent, that is, from the theoretical laws on which they are parasitic (e.g., “All T1s are T2s”). Call this the theory-­dependent inference relations strategy. Because Sellars’s argument can be difficult to follow, let us first illustrate with an easy and familiar case: the relation of folk-­psychological kinds to physical kinds. While there aren’t folk-­psychological laws per se, there are folk-­psychological generalizations, suitably qualified: for example, “If someone believes that p, and believes that if p then q, then (ceteris paribus) she believes that q” and “If someone desires that p, believes that if q then p, and is able to bring it about that q, then (ceteris paribus) she brings it about that q” (Lange 2000, p. 231). The question is: Can these generalizations be reformulated as generalizations at the purely physical level, in terms that eliminate all reference to folk-­psychological kinds? That is, can you replace talk about folk-­psychological regularities (F1 → F2) with talk about regularities formulated purely in physicalist language (P1 → P2)? We can answer no and defend our answer along the two lines suggested above: First, we can deploy the no bridge laws strategy, arguing that there are no bridge laws linking folk-­psychological kinds (e.g., F1) to physical kinds (P1). This is due to the familiar fact of multiple realizability and the consequence that within reasonable parameters, an indefinite number of physical systems can count as realizing various folk-­psychological states.

A New Naturalism  41 Thus, the “rules” linking the folk-­psychological states and their physical instantiations will not be law-­like but will link these states with a broad and disparate disjunction of physical realizations. The physical correlates of folk-­psychological kinds will not themselves be recognizably physical kinds. Second, we could employ the theory-­dependent inference relations strategy, arguing that the laws or generalizations that hold true at the level of folk-­psychological kinds (F1 → F2) will not hold true when transposed to the physical level (P1 → P2). Marc Lange and Ruth Millikan focus in particular on the role of ceteris paribus clauses in such generalizations. Consider, again, a generalization such as “If someone desires that p, believes that if q then p, and is able to bring it about that q, then (ceteris paribus) she brings it about that q” (Lange 2000, p. 231). Now there are various states that could interrupt the transition from desiring p and believing that if q then p to bringing it about that q: cognitive impairment, distraction, depression, a belief “that r would be a better means than q for bringing it about that p” (Lange 2000, p. 231). However, when we transpose this generalization to the level of physical kinds, the generalization ceases to be even remotely law-­like, largely because folk-­psychological terms like “cognitive impairment” will have wildly different physical instantiations in different systems. As Millikan notes, Characteristically, the same function could, at least in principle, be performed by many differently constituted items. But . . . then the supporting conditions required for them to effect this function must differ as well. Brain cells performing the division algorithm require oxygen whereas computer chips require electric currents, and so forth. . . . The result is that there are no ceteris paribus laws governing all items having a certain function. (Millikan 1993, pp. 226ff; quoted in Lange 2000, p. 232) Lange comments, “The physical cashing-­out of ‘so long as there is no cognitive impairment’ would have to be very long; it would have to list each physically possible kind of mental hardware and what it takes for that hardware to be working properly. But this clause would then, as Teller says, be too ‘fat’ (1984, pp. 58ff.) to appear in a statement of natural law; in an important sense, the various entries on the list have nothing in common” (Lange 2000, p. 232). Thus, even if there are folk-­ psychological laws, there will be no corresponding laws at the purely physical level. Or, if there are, the force of these laws will be wholly parasitic on the corresponding folk-­psychological laws. The upshot of these arguments is that folk-­psychological concepts (F1) and generalizations (F1 → F2) have theoretical autonomy and cannot be eliminated in favor of physicalist concepts (P1) or generalizations (P1 → P2).7

42  A New Naturalism I use this particular illustration because it is familiar and (I hope) easy to understand. The above example­—an example of an argument to prove the autonomy and independence of folk psychology vis-­à-­vis a more scientific theory of the mind­—would not be the kind of use Sellars would have in mind for this argumentative strategy. Typically, Sellars is arguing for the autonomy of the theoretical with respect to the observational; or, as in “Phenomenalism,” he uses these arguments to demonstrate the necessity of material-­object talk and why we cannot dispense with it or reconstruct it purely in terms of phenomenal concepts. That essay is one place these two strategies work hand in hand for Sellars. Consider a material-­object generalization (that is, an instantiation of “All M1s are M2s”) such as “All lightning is followed by thunder.” We might think that we can identify phenomenal counterparts (P1, P2) of lightning and thunder (M1, M2). However, in “Phenomenalism,” Sellars argues that there simply are no bridge laws linking, for example, the material-­object concept lightning to a set of phenomenal concepts; the former cannot be identified with any set of the latter. This, in turn, undermines the possibility of there being any law-­like generalizations stateable at the purely phenomenal level. As a consequence, whether we are justified in inferring P2 from P1 depends on whether P1 is in fact a case of M1­—something we cannot determine without deploying the relevant material-­object concepts and inferential relations. In other words, what gets included in the antecedent of the phenomenal generalization (the instantiation of “All P1s are P2s”) will appear gerrymandered, and we will only be able to infer P2 from any instance of P1 if we are justified in including this phenomenal instance in P1 by virtue of the fact that it is an M1­—that is, by virtue of the fact that it is a genuine instance of lightning. Consider: What is the phenomenal difference between lightning versus a photograph of lightning versus a video of lightning (perhaps without audio or with the sound muted)? Surely, there is a way to distinguish between these such that the generalization “All P1s are P2s” is preserved­—but Sellars’s point is that in deciding which cases to include in the antecedent, and which cases to exclude, we will be making tacit appeal to our judgment as to which are actual cases of lightning (versus simulacra, or reproductions, etc.). In other words, in deciding which cases go into the antecedent, and hence which cases justify an inference to the consequent, we necessarily must appeal to the material-­object concept. Thus, we cannot determine what elements belong in the “set” of instances comprising P1 without appeal to the material-­object concept M1. The material-­object concept is crucial, in the context of justification, in deciding what properly belongs in the antecedent of a conditional that would justify us in inferring the consequent. There is also (not incidentally) the question of what gets left out of the antecedent. As Sellars notes, much of what is included in our phenomenal experience is highly contingent, as our phenomenal experience is necessarily autobiographical. Thus, as Sellars writes,

A New Naturalism  43 The uniformities I find are bound up with the fact that my environment has included wallpaper of such and such a pattern, a squeaky chair, this stone fireplace, etc., etc. My having had this pattern of sense contents has usually eventuated in my having had that sense content, because having this pattern of sense contents guarantees, for example, that I am awake, not drugged, wearing my glasses and looking at the fireplace. And a generalization which is an expression of the contingencies of my existence can scarcely be one of the generalizations which, in the intersubjective conceptual heritage of the race, support the phenomenally conditional sense contents postulated by the phenomenalist. Thus, even granting that there are inductively warranted generalizations which permit the definition of phenomenally conditional sense contents, the latter will be logically tied to the peculiarities of my environment in such a way that they cannot be transferred to other things in other places. (PH, III.§51/pp. 82–3) Such regularities are “essentially autobiographical,” as Sellars notes, and hence cannot support any law-­like generalizations. A law-­like generalization will have to extract those specific sense contents that are occasioned by lightning and make an inductive generalization based on these­—but obviously, to do so is to make an appeal to the material kind in question. Virtually all of Sellars’s arguments for scientific realism make use of the theory-­dependent inference relations strategy. As Sellars argues, “Furthermore, theories not only explain why observable things obey certain laws, they also explain why in certain respects their behaviour obeys no inductively confirmable generalization in the observation framework . . . [For example,] it is because a gas is­—in some sense of ‘is’­—a cloud of molecules which are behaving in certain theoretically defined ways, that it obeys the empirical Boyle-­Charles law” (LT III.§§42 and 41/p. 121). Thus, to use deVries’s elaboration of Sellars’s example, the Charles law relating the temperature, pressure, and volume of Boyle-­ gas holds linearly up to a certain pressure and then breaks down. Using purely observational predicates, we could create a mathematical function that would capture this breakdown, but the function would seem arbitrary, from the purely observational perspective; there would be no explanation for it. However, if you import a theoretical understanding of gases­—that they are molecules “bouncing” off each other­—then this breakdown makes sense. “As the pressure increases, a point will be reached at which the particles become too crowded, and the volume will shrink ever more slowly” (deVries 2005, p. 155). Thus, in the context of justification and explanation, the Theoretician’s Dilemma disappears: Inferences from O1 to O2 can be justified only by the prior justifiability of the inference from T1 to T2, and (what is more important, perhaps) the

44  A New Naturalism theoretical relationship behind T1 and T2 (and the underlying theoretical apparatus) explains why O1 and O2 are so related. The theoretical terms are ineliminable, and scientific realism is thereby established. As Sellars writes, “Thus, microtheories not only explain why observational constructs obey inductive generalizations, they explain what, as far as the observational framework is concerned, is a random component in their behavior, and, in the last analysis it is by doing the latter that microtheories establish their character as indispensable elements of ­scientific explanation and . . . as knowledge about what really exists” (LT III.§43/p. 122).

2.3 The Manifest Image, Revisited The crucial lesson to be gleaned from the above is that discourse type A object discourse or theoretical discourse) cannot be (e.g., material-­ replaced by discourse type B (e.g., phenomenal discourse, observational discourse) if we simply cannot do things in the context of justification (or explanation or what have you) purely in discourse type B without recourse to discourse type A. However, it would seem that this relation of irreducibility holds between all philosophical discourse types and the purely physical. Sellars’s thesis of causal reducibility is false since such reducibility presupposes that philosophical concepts (i.e., intentional, epistemological, moral, etc.) will find extensionally equivalent concepts in the scientific image­—a presupposition that we now have the tools to show is wrong. By way of illustration, we have already discussed above how mental terms do not seem to pick out recognizable kinds at the physical level. Thus, the same absence of bridge laws that Sellars noted between theory and observation in science also seems to exist between the intentional and the physical, primarily due to multiple realizability. Thus, if we hoped to generate Sellarsian equivalences of the form S believes that P ≡ S is in physical state B then we would appear to be out of luck, as whatever we put on the right-­ hand side of the equivalence will be hopelessly disjunctive and gerrymandered. What they have in common­—what unites them­—is that they are all cases of believing that P. Thus, we cannot understand the “lower-­ level” extension without appealing to the “higher-­level” concept. Sellars’s hope that we could associate intentional notions with descriptive extensions that form recognizable kinds and can do explanatory work, when viewed purely from a descriptive, non-­intentional standpoint, is in vain. This extension only makes sense when seen as the extension of an inherently intentional concept. As deVries puts the point, intentional concepts “pick out a level of uniformity in the world that would not be salient

A New Naturalism  45 without the introduction of a new level of concept not strictly definable in physicalistic . . . terms” (deVries 2005, p. 200). One could, of course, try to fill in the right-­hand side of the equivalence with “behaviorese” instead of attempting to cash it out in terms of physical states. And of course Sellars (especially in his earlier works) talks as though this is how a science of the mind would proceed. This would allow for multiple realizability, all while allowing us to avoid intentional terms on the right-­hand side of the equivalence. But such a solution cannot be acceptable to the mature Sellars. Such a behaviorist analysis would essentially be dispositional: To want A is to be disposed to take various actions to acquire A. But these are dispositional and defeasible: If I want to go to the movie this afternoon, I will be inclined to look up movie times, go to the box office and buy tickets, and so on­—unless, of course, I have a looming deadline, in which case I will probably not do so. Or I may do so but lie about it if you ask. The problem is not (merely) that these dispositions are complicated and holistically interrelated. This is true, and it implies that it will probably be impossible to spell out, behavioristically, the extension of any particular intentional state. This was, of course, a standard critique of early theories of philosophical behaviorism. But we can safely ignore this criticism, for a deeper concern lurks. We may suppose (although this may be a fantasy) that some ideal behavioristics could map out all of these complex interrelations and connections. The real problem is that behavioristic dispositions, for Sellars, aren’t the real stuff of science, for dispositions are always dispositions of some underlying substance, and the real scientific explanation must be in terms of the properties of this underlying substance. As Sellars writes in “The Structure of Knowledge,” “I introduced [Verbal Behaviorism] as a simple model, and while I have been polishing and defending it, it was introduced in order to be transcended. It correctly represents a basic stratum in our conception of what thinking is, but it is only a part of a larger picture, to which I now turn. In the case of the dispositions and propensities of material things, we distinguish between the propensities and dispositions themselves, which are iffy states definable in terms of test conditions and empirically ascertainable results, on the one hand, and the explanation of these dispositions and propensities, which theoretical physics has made available, on the other” (SK IV.§§49–50). The first lesson, then, is that Sellars’s argument for scientific realism can, it seems, be deployed to demonstrate that normative terminology is not causally reducible to the scientific image. We have seen this for beliefs; we will see, in Chapter 10, how McDowell extends this argument to moral terms. (And Brandom [2015, Chapter 6] has argued more generally that there are no “cross-­sortal identities” of the sort that would be required to reduce manifest-­image objects to or identify them with objects in the scientific image.) The real payoff, however, is that if

46  A New Naturalism normative terminology is not causally reducible, then Olen’s explanatory gap argument disappears. We cannot exhaustively account for our normative practices purely in descriptive terms, and so the normative is (to that extent, at least) not otiose and hence ineliminable.

2.4 Further Consequences of the Argument Indeed, our recent progress forces us to revisit some of Sellars’s earlier claims that I reported more or less uncritically in the previous chapter. As I noted, Sellars argues that language is integrated into the causal order via linguistic tokens. For example, “red” and “rot” are tokens of •red• and as such are natural linguistic objects, i.e., empirical items or events that can be treated as objects of scientific study. Sellars often treats such tokens as purely naturalistically characterizable events. Indeed, Sellars sometimes even gives a purely phonetic characterization of observation sentences (e.g., “Jonz-­iz-­ceeing-­redd” [PPE II/p. 12]), as though such tokens could be identified as the tokens they were purely by their auditory qualities. Again, though, this cannot be right: Which sounds count as instances of language in the first place depends on a prior deployment of a variety of normative and intentional notions. Thus, “That is green,” as spoken by Smith, is a •That is green•. But if there is a species of bird whose distinctive call is “That-­iz-­green! That-­iz-­green!” it is not thereby speaking English. So we see the same dependence between the normative and the natural here as we saw between theory and observation in Sellars’s argument for scientific realism: Which tokens ought to be included as “natural linguistic objects” will depend on a judgment that must itself rely on inherently intentional notions, such as who is a language user and other interpretive questions. Thus, we find ourselves in the same position with respect to laws formulated at the lower level. Can we expect an utterer of “That-­iz-­green” to be disposed to utter “That-­iz-­ kuhlurd”? We can certainly expect a language user to make the inference from “That is green” to “That is colored”; but whether we can make the lower-­level inference depends on whether we decide the “That-­iz-­green” in question is an instance of a “That is green” (meaning, more generally, •That is green•). [The reader might at this point wonder whether I have punted on the naturalism question or introduced circularity to the account; I will return to these questions at the end of the chapter.] Thus, we see the applicability of the theory-­dependent inference relations strategy­—we can only make the inference at the lower level if we decide that the lower level is, in fact, an instantiation of the higher level; lower-­ level inferability is parasitic on our characterization of this lower-­level item as an instance of a •That is green•­—with the accompanying inferential licenses. To be sure, once you have identified a natural linguistic object as such, it can be studied as a physical event. But the point is that the class of such events that will count as natural linguistic objects can

A New Naturalism  47 only be picked out using normative and semantic concepts. They will not form a kind that is recognizable as such without reference to such normative and semantic concepts. What goes for intentional concepts (such as belief, desire, and so on) will go for other normative concepts­—including moral concepts. As I noted above, John McDowell has extended Sellars’s argument to demonstrate that moral concepts have no extension at the natural level that is recognizable as such when viewed purely from the descriptive standpoint. Only by deploying the moral concept in question can one see why these particular items all belong together; without the aid of the concept, the group seems gerrymandered and wildly disjunctive. Thus, McDowell is employing something like the no bridge laws strategy to argue that the lower-­level instantiation is “shapeless” if not viewed from the perspective of the higher-­level concept. We will examine this argument­—and its important consequences for the question of moral motivation­—in Chapter 10. So while I agree with Sellars that the normative is logically irreducible to the descriptive, I think he was wrong to claim that the normative is causally reducible to the descriptive or natural. This, I think, alleviates Olen’s worry, which I discussed above. His worry was that if the normative is causally reducible to the descriptive, why not just eliminate the normative? What specific explanatory work is left for the normative to do if it is so reducible? Sellars was arguing for the causal reducibility of the normative­—that, as O’Shea says, in the scientific image we should give “an account of such norms qua the norms that they are, and not merely an account of a separate scientific subject matter” (O’Shea 2009, p. 200). But, again, there is not going to be an independent scientific account of these norms because they form a gerrymandered kind when viewed from the natural/scientific level. We can see certain events as courageous or kind or as instances of knowledge or as speech acts only when employing the tools of the manifest image. Ergo, the normative is not causally reducible. If the normative is not reducible in this way, then this worry at least disappears. But perhaps this move only eliminates one problem by creating another: If we deny that the normative is causally reducible to the descriptive, does this mean that we have given up on Sellars’s naturalism before we have even begun? It does not. The important concept here is what John McDowell and Jim O’Shea call our “second nature.” Our second nature is our battery of linguistic and conceptual resources that enables us to see the world as this way or that way.8 Looking around, I see the things around me as books, as a desk, as a window, a street, etc. But these judgments all deploy acquired concepts that I nevertheless deploy non-­inferentially in understanding the world around me. Seen from this perspective, there is nothing more mysterious in seeing an act as cruel than there is in a scientist’s seeing a trail as an

48  A New Naturalism electron­—or in my seeing a particular organized combination of paper, ink, and glue as a book. And Sellars will still insist on his scientia mensura principle: Science tells us what really exists. Although we use the tools of the manifest image to identify an event as intentional, it can then be treated as a natural linguistic object, and the causal powers it has, it has purely in virtue of its descriptive properties. (By analogy, a hammer has the causal powers it has in virtue of its physical properties, not in virtue of the web of normative proprieties that constitutes its identity as a hammer.) Even if our acquaintance with such objects and events is necessarily mediated by manifest-­image concepts, the manifest image is not the ultimate arbiter of our ontology for Sellars; he repeatedly counsels us that what is prior in the order of knowing need not be prior in the order of being. When we tie together the elements from chapters 1 and 2, I do not see, so far, any reason to think that Sellars’s philosophy is incompatible with his stringent naturalism. The elements of his philosophy are all things that we can make sense of as physical creatures interacting in physical ways with a purely physical environment: • Normative concepts, universals, abstract entities, and so on are understood metalinguistically. Thus, for example, •red• doesn’t designate a Platonic object, but rather picks out a role played by the relevant linguistic tokens (the “red”s and “rot”s and “rouge”s, etc.). • That “red” has the particular meaning it does because it is a •red• is no more mysterious than the fact that this piece of metal and rubber is the particular tool it is because it is a hammer. • That we can cognize particular actions as cruel or kind (in virtue of our second nature) is no more mysterious than the fact that we can cognize certain items as books or electrons or what have you. Let me close by stating that if Sellars is correct (more accurately, if I am correct about Sellars), the aspirations of the naturalist must be modest along at least one dimension. We can be naturalists. We can even say that science has a privileged position as an account of what there is. We can even say that something like Harman’s explanatory requirement is a good test of ontological commitment, and so something must have real existence in the space-­time continuum to be real. But it doesn’t follow that the only things that we can say are scientific or descriptive or causal-­ explanatory, for there are many discourse types that serve a different purpose. And here is where naturalism must embrace a bit of modesty: We must never lose sight of the fact that even if we succeed in giving a very exhaustive and detailed account of the descriptive circumstances that are satisfied by a meaning claim (say, “‘Red’ means red”), we should not be puzzled when, in looking at this list of descriptive circumstances, we somehow feel that we have failed to give an account of meaning­—for

A New Naturalism  49 of course we would have failed. A central lesson of Sellars’s naturalism is that the normative (i.e., semantic, moral, etc.) aspect of a claim can never be captured by any set of descriptions. And so while a description might possibly display the conditions of meaningfulness, it will not explain meaning because meaning is an inherently normative concept, and that­— just about by definition­—cannot be captured by a set of descriptions. (Recall the discussion of Lance and White [2007] from the beginning of the chapter­—even if we give a metaphysical account of personhood, such that x is a person iff P[x], where P is a set of necessary and sufficient conditions­—we will have failed to capture what they call the “noncognitive aspects” of personhood. For example, there is a conceptual connection between x’s personhood and holding x responsible or attributing commitments and entitlements to x. And none of this will be captured by the metaphysical account of x in terms of P.) This is, fundamentally, the lesson of the closing lines of PSIM: Even if the scientific image ultimately tells us that moral values, semantic contents, and so on aren’t really part of the ultimate ontological constituents of the world, for us to talk only in scientific, descriptive language would be to lose sight of personhood­— it would be to lose sight of the dimension of holding each other responsible and placing each other in the space of reasons. Therefore, as agents, we must retain the nondescriptive language of norms. Thus, any account we give of the normative is going to be “modest,” to use the word in McDowell’s sense (a point we will return to in Chapter 9). That is, we will not be able to display or justify the normativity to someone wholly outside of the system who doesn’t already have a foothold in it. You can’t explain meaning to someone who doesn’t already have a language; you can’t justify morality to someone who is fully outside the moral point of view. We can display some features of moral justification­—but only to people who are already, as it were, playing the moral game. There is no bootstrapping in matters normative. This doesn’t mean, again, that we have given up on naturalism; the act I classify as cruel is a spatiotemporal event. It just means that there is no lever of Archimedes to justify or explain the normative standpoint to someone outside of this standpoint. (Stated baldly, the point should be obvious; but the demand for “noncircular” justifications, or naturalistic reductions of normative concepts, seems like demands for just such an Archimedean point­—a point from which we could rationally convince someone to adopt the moral point of view without using any moral premises at all.9 This point we will return to in Chapter 12.) I think this modesty is implicit in Sellars’s theory of the function of the “means” locution. Sellars isn’t even attempting to give what McDowell would call a “full-­blooded” theory of meaning. His theory of meaning is not attempting to reconstruct, from outside of meaning, what it is for a word to have semantic content. The theory itself presupposes a different function for meaning-­talk­—that the pragmatic point of such talk is

50  A New Naturalism that it takes place between people who already have access to meaningful language. Thus, if I tell you, “‘Rouge’ (in French) means red” (i.e., “‘Rouge’s [in French] are •red•s”), I am using •red• as an illustrating sortal, which presupposes you know the meaning of “red.” Means-­talk can’t serve its purpose if I don’t use an illustrating sortal whose meaning you already understand. The purpose of meaning-­talk, for Sellars, isn’t to show exhaustively the non-­semantic facts that constitute meaning, but rather to unite linguistic communities, convey that different words play a similar functional role, and in general to facilitate mutual understanding among people who are already language users. This allows me to circle back and address a point from earlier that might have been troubling the reader. As I noted above, I pushed back against Sellars’s claim that we can treat natural linguistic objects from a purely scientific standpoint. For whether a particular utterance of, say, “That-­iz-­green” should be treated as a linguistic object in the first place depends on a host of normative considerations; it is primarily a question of whether the utterer is an agent and whether the utterance belongs in the space of reasons. To the reader, this might have raised dual concerns: First, does this mean that our account of meaning is circular since we are admitting that we cannot identify natural linguistic objects independently of normative considerations? Second, have we given up on naturalism since normative considerations are creeping in to the naturalistic basis of meaning? To answer the first question, what our brief discussion of modesty has (hopefully) made clear is that the only way to avoid all appearance of circularity is to give a reductive account, and that is simply not possible. As Tuomela writes (in discussing Marras’s charge that Sellars’s account of conceptuality is circular, in that it relies on rules of language), “though conceptuality can be illustrated and elucidated in terms of rules, no reductive analysis is possible . . . the concept of language and the concept of conceptual activity are mutually intertwined” (Tuomela 2002, p. 51). In any case, the demand for reduction misunderstands the nature of normativity: It thinks of normativity as something that must be placed in the world and, hence, something that must ultimately be identified with something more respectable and scientific. But as I noted earlier in the chapter, Sellarsian normativity is not a thing, it is a grammatical mode, and no amount of talking in a descriptive mode will ever amount to making a normative claim.10 Thus, there is no way of eliminating the normative in favor of the descriptive because describing and prescribing are different things. There is a sense, then, in which circularity is unavoidable in matters normative. We will return to this question in Chapter 12. Again, in making a normative utterance, one conveys certain descriptive information. Thus, if I believe “Smith means red by ‘red,’” then I can predict certain behaviors on his part­—that, for example, he will be disposed to respond to certain apples with the word “red,” that he

A New Naturalism  51 would deny that something completely red is completely green, and so on. This list is open-­ended, indefinite, and not the same for everyone. (People’s behavioral dispositions regarding words depend a lot on their other collateral commitments, and so even a reductive account in terms of behavioral dispositions is hopeless.) So normative concepts are tied to descriptive contents­—but saying things in the normative mode is not describing, and we shouldn’t expect that we can, in the end, ground prescribing in describing. As for the second concern, I am not denying that we can treat certain utterances (e.g., tokens of “red”) as natural linguistic objects suitable for scientific study. But to do so, we must first identify them as natural linguistic objects; and in doing so, we are importing considerations from the manifest image. Again, though, this is no more puzzling than saying that a hammer is something that is defined by a web of norms and appropriatenesses. It doesn’t follow that a token hammer is not a physical thing with physical properties that can be examined, measured, and so on. It just means that in identifying it as a hammer we are not talking about it descriptively; we are classifying it in a particular way, a way that involves conveying the normative proprieties that identify the item as the item that it is. In the same manner, in saying the meaning of “red,” we engage in a similar classification of the token rather than mere description of it. Thus, there is nothing here to offend against naturalism. There is much more to say about the naturalistic bona fides of Sellars’s philosophical project. But this book is concerned specifically with Sellars’s ethical theory, and it is to this I will now turn. Sellars’s naturalism will be, however, a consistent theme throughout this work. In the final chapter, I will return to this important topic and will gather up various threads discussed in these introductory chapters, and later in the book, and demonstrate how Sellars’s ethical theory is thoroughly naturalistic­—how Sellars manages to defend a robust account of ethical objectivity within the strict limits laid out by his scientia mensura principle.

Notes   1 Sellars leaves the distinction between saying and conveying mostly intuitive. O’Shea explains the distinction as concerning “what the statement says, what it describes or mentions explicitly­—as opposed to further information that is pragmatically conveyed, or implied, or presupposed by the statement” (2009, p. 202).   2 As I have argued elsewhere (Koons 2006), the Sellarsian claim is that the normative claim (on the left side of the equivalence) has a meaning surplus over and above the descriptive claim (on the right side of the equivalence), and this stands in the way of complete reduction (or elimination) of the normative to the descriptive.   3 Sellars seems to reject the idea (most clearly in LT IV.52/p. 126) that physical objects in the manifest image can be correlated with objects, or sets of objects, in the scientific image. But as the quote below from NAO (4.VII.§65/p. 78)­—

52  A New Naturalism which originated as the John Dewey lectures from 1973–74­ —indicates, he held throughout his career the view that norm-­governed uniformities could be described in the scientific image.   4 The quote O’Shea is discussing runs as follows: “. . . unless and until the ‘scientific realist’ can give an adequate explication of concepts pertaining to the recognition of norms and standards by rational beings his philosophy of mind must remain radically unfinished business” (SM, p. x).   5 It actually isn’t clear to me that by “causal reducibility” Sellars means that we will find, in the scientific image, concepts that are coextensive with normative terms (or that we will otherwise be able to give a naturalistic account of the extension of normative terms as such). When introducing the notion of causal reducibility, Sellars writes only, “For our present purposes, a concept will be said to be causally reducible to descriptive concepts if (roughly) it is either definable in descriptive terms (the trivial case), or occurs in the antecedent of a properly constructed causal explanation only as a subordinate element in a descriptive mentalistic context (e.g., as ‘entails’ occurs in ‘Jones believes that responsibility entails determinism’). . . . If we use ‘ethical assertion’ in such a way that ‘Jones ought to pay his debt’ is an ethical assertion, but ‘Jones feels that he ought to pay his debt’ is not, then we can say that to claim that Ought is causally reducible to Is is to claim that one can give a causal explanation of the history of moral agents without making ethical assertions” (SSMB, ¶6). Setting aside “the trivial case,” it seems clear that the second element in the disjunct does not commit Sellars to any kind of strong, co-­extensionalist reductionist thesis. It is true that throughout SSMB Sellars uses an example of identifying the extension of mental predicates with Ideal Behaviorese sentences, but this may be seen as a tool to illustrate his point rather than as an indication of Sellars’s own commitments as to what is possible (“Even if you accept the possibility of such equivalences, they still fail fully to reduce the mental to the descriptive.”). This reading is particularly plausible in the light of other things Sellars claims, such as in LT, where he denies that manifest-­image items will find coextensive equivalents in the scientific image: “On one classical interpretation, correspondence rules would appear in the material mode as statements to the effect that the same objects which have observational properties also have theoretical properties, thus identifying the denotation, but not the sense, of certain observational and theoretical expressions. . . . According to the view I am proposing, correspondence rules would appear in the material mode as statements to the effect that the objects of the observational framework do not really exist­—there really are no such things. They envisage the abandonment of a sense and its denotation” (LT §52/p. 76). However, I fear I am swimming against the current here, as most commenters on Sellars agree with O’Shea’s interpretation of “causal reducibility.” Further, it must be admitted that in some later writings (which I quote in this chapter), Sellars does seem to move toward the more radical version of causal reducibility attributed to him (although, as noted, LT provides contrary evidence). Given these facts, I will go ahead and address this common reading of “causal reducibility.”   6 Christias (2015) offers a subtle and sympathetic reconstruction of Sellars’s theory, arguing that it evades the “explanatory gap” argument. But Christias’s reconstruction is premised on the truth of the causal reducibility of the normative, which (as I will argue) is untenable.  7 For an interesting critical assessment of Lange’s argument, see Christias (2016).   8 Of course, our second nature needs to be seen as connected to and arising in connection with our “first nature”­—our non-­normative, physical ability to

A New Naturalism  53 respond differentially to elements in our environment, an ability that has an evolutionary rather than a normative explanation. The melding of our first and second natures, and the corresponding integration of the normative agent into its physical environment, is crucial for Sellars and is an issue to which we will return in chapters 9 and 15.   9 Obviously, you can train a nonmoral child to become a moral being; and you can teach a nonlinguistic child to speak and understand language. But these are matters of training (and training agents who already have a particular “first nature”­—certain emotional and cognitive capacities that make them capable of moral agency, certain neurological structures that make them capable of linguistic competence). Modesty doesn’t mean training is impossible; it only implies the impossibility of a view from nowhere, of the ability to justify a normative system from somewhere outside that system. It is a rejection of the Given: a rejection of the idea that you could make moves with, say, moral or semantic normative import from entirely outside a system of moral or semantic norms. 10 It is this fundamental insight that leads many so-­called “left-­Sellarsians” to conclude that Sellars was fundamentally misguided to try to identify some kind of descriptive, behavioral regularity with which to associate intentional and other normative concepts. To be sure, normative utterances must have behavioral consequences, but this need not entail causal reducibility.

3

Moral Judgments as Shared Intentions

3.1 Why Intentions? 3.1.1 Moral Judgments, Evidence, and Motivation David Solomon situates Sellars’s ethical project within “a common view of the problematic of ethical theory which is largely shared by Anglo-­American moral philosophers from Moore to Hare” (1977, 151). The chief difficulty facing classical metaethics (CM, as Solomon calls it) is a problem that should prove instantly familiar to anyone working in contemporary metaethics. The problem is how to reconcile two different aspects of moral judgment. First, we present evidence for our moral judgments, which entails (a) that moral judgments are belief-­like or cognitive and (b) that moral judgments can be involved in reasoning and logical relations. As deVries points out, “There are valid inferences in which our moral or normative vocabulary occurs essentially, and this needs to be taken account of” (2005, p. 252), offering as an example the perfectly legitimate-­seeming inference, “Jones ought to do A and B. So, Jones ought to do A.” Thus, our account of moral judgment must account for how they answer to evidence and appear in valid forms of practical reasoning. Second, morality is concerned with acting rightly or well, and so moral judgment must be (it seems) connected to motivation and action. But CM has found it difficult to reconcile the cognitive and motivating aspects of moral judgment. Michael Smith, a contemporary metaethicist, has famously summed up the tension in a set of apparently inconsistent “platitudes” regarding moral judgment: (1) Moral judgments of the form “It is right that I ϕ” express a subject’s beliefs about an objective matter of fact, a fact about what it is right for her to do. (2) If someone judges that it is right that she ϕs then, ceteris paribus, she is motivated to ϕ. (3) An agent is motivated to act in a certain way just in case she has an appropriate desire and a means-­end belief, where belief and desire are, in Hume’s terms, distinct existences. (Smith 1994, p. 12)

Moral Judgments as Shared Intentions  55 Although I think Smith critically prejudges certain issues­—for example, his gloss on the cognitive nature of moral judgments is that they must be beliefs­—I think it is also clear that Smith’s summary of the “moral problem” shows a continuity of metaethical concern in Anglo-­American philosophy for at least a century. In this context, Sellars’s conclusion that moral judgments must express intentions is in fact ingenious. Intentions are by their nature connected to motivation and action; and in Chapter 11 I will argue, at much greater length, for the connection between moral judgment and moral motivation in. However, intentions are also cognitive, in that one can reason with intentions. Indeed, most of his major expositions of his ethical theory begin with a discussion of the logic of intentions (an issue we will explore at length later in Chapter 5), and throughout the course of his career, Sellars continued to develop and refine his account of practical reason. (It is worth noting, before we continue, that Sellars’s novel reconciliation of the cognitive and motivational aspects of moral judgment allows him to neatly bypass Smith’s third platitude, which is supposed to show the difficulty of reconciling these two aspects. We will return to this theme in Chapter 11, where we discuss at greater length Sellars’s Kantian­—rather than Humean­—theory of practical reason.) I will argue in this chapter that treating moral judgments as expressions of intentions has many theoretical virtues: It is independently plausible, it solves a number of difficult problems in metaethics, it predicts a number of features of normative judgments, and so on. I do not think that Sellars himself adequately “sells” his own account or makes clear its many virtues; let me attempt to do this for him. Further, because of the intimate connection between intentions and plans, Sellars’s view­—that intentions are the constitutive state expressed —anticipates by decades views often thought to by moral judgments­ originate with authors like Alan Gibbard, whose plan-­expressivism has been so influential. Sellars’s theory, it turns out, will fit squarely in the intention-­or plan-­oriented expressivist camp and will turn out to be a rich and sophisticated version of this theory. 3.1.2 Intentions in Practical Reasoning Sellars has an extensive botany of states in his metaethical theory, and it is crucial to understand the relations among them. The central concept is that of the expression of an intention. For Sellars, practical reasoning­—both prudential and moral­—begins and ends with intentions, and judgments about what one ought to do (both categorically ought and ought relative to a prudential end) are analyzed in terms of expression of an intention. Sellars reserves the term “shall” for the expression of an intention, as in I shall go to the park tomorrow

56  Moral Judgments as Shared Intentions To simplify the logic of intentions (which we will discuss in Chapter 5), Sellars proposes using “shall” as an operator that operates on the contents of intentions. Thus, the above intention can be represented formally as Shall [I go to the park tomorrow] I will alternate (as Sellars does) between the formal and informal methods of expressing intentions. Expressions of intention should not be conflated with ascriptions of intention (e.g., ORAV §§32–3/p. 84). For example, if I say Jones intends to go to the park tomorrow I ascribe an intention to Jones. But if I say Jones shall go to the park tomorrow I express my intention that he will go. To ascribe to oneself an intention, however, is also to express that intention, just as to ascribe a belief to oneself is to express a commitment to that belief (ORAV §33/p. 84). “Will,” in Sellars’s idiolect, merely makes predictions about what will happen. If I say, “I will go to the park tomorrow,” I don’t necessarily ascribe or express any intention; perhaps I think I will be kidnapped and taken there against my will. Ceteris paribus, though, “I will do X” follows from “I shall X” because expressions of intention are connected to volitions and hence to action. Volitions are essentially intentions whose time has come. If I form the intention “I shall leave the office in 10 minutes,” after 10 minutes the intention will (again, ceteris paribus) give rise to the volition “I shall leave the office now,” where “now” is used by Sellars to indicate the expression of a specific kind of intention­—namely, a volition. (One can see here what Michael Bratman cites as the essential connection between intentions and planning, about which we will have more to say anon.) A Sellarsian volition is what in present-­day terminology might be called an intention in action, whereas a Sellarsian intention (say, “I shall leave the office in 10 minutes”) is what we would call a prior intention. Intentions, via volitions, are essentially tied to action. In a section of AAE that is worth reading very carefully (§§22ff/pp. 185ff), Sellars discusses the difference between claiming that S said that-­p and our merely noting that S made the same noises that one would make if one said that-­p. (Think, for example, of a prelinguistic child, or an animal, putting together phonemes in the order of a word in a natural language­—what would have to be true so that we would claim the child or animal in

Moral Judgments as Shared Intentions  57 question said the word?) Sellars writes, “It would be absurd to say: Tom said (as contrasted with uttered the noises) ‘It is not raining’ but has no propensity to avoid saying ‘it is raining and it is not raining’” (AAE §29/p. 187). To ascribe a saying (rather than a mere noise-­making) to someone is to ascribe a certain functioning to them­—not merely a predictive functioning but a logical functioning, a commitment to a set of norms regarding inference, compossibility, and the like. Note that part of what Sellars is saying is that ascribing this kind of functioning to a person is to ascribe to them a commitment to engage in certain behaviors (and refrain from others): A belief that Fido is a dog will result in the behavior (in appropriate circumstances) of concluding that Fido is a mammal; believing that-­p involves refraining from concluding ~p; and so on. Sellars extends this account to concepts governing our practical lives.1 Concerning expressions of intention (and their connection to volition and action), Sellars writes, Consider the child who is learning to use sentences which, as we say, formulate intentions; sentences involving . . . “shall.” From the standpoint of non-­functional description, it is a matter of learning how to use sentences involving the sound “shall”: I shall now raise my hand. Clearly, the child has not learned how to use this sound unless he acquires the propensity to raise his hand, ceteris paribus, upon uttering (or being disposed to utter) the sound “I shall now raise my hand.” (AAE §44/pp. 192–3) To return to our botanizing, Sellars distinguishes in LTC between statements: ought-­ to-­ be’s and ought-­ to-­ do’s. different kinds of ought-­ Running in parallel to these two kinds of oughts are different expressions of intention. Thus, I might intend as follows: Shall be (my children have a good education) Just as ought-­to-­be’s imply ought-­to-­do’s (LTC, p. 508), Shall-­be’s imply do’s (ORAV §45/p. 85, §68/p. 88) via a principle Sellars calls Shall-­ “Shall-­be.” This is important for Sellars because Shall-­be’s, while expressing genuine intentions, often take as their objects states of affairs or other things that we cannot directly will. Thus, these states of affairs (or whatever is intended) must be brought about indirectly via other actions that we can intend as Shall-­do’s. Thus, from the above “Shall-­be” follows: Shall (I do that which contributes to bringing it about that my children get a good education)

58  Moral Judgments as Shared Intentions From which I might reason to further, more specific intentions (= means of fulfilling my prior intentions): Shall (I set up a college savings account for my children) Finally, and crucially for our project here, there is the distinction between personal intentions and we-­intentions. I can intend something as an individual, as when I intend that my children get a good education. But also, notes Sellars, I can intend something qua a member of a group. Suppose all of us, as members of the Whooping Crane Society, adopt the following intention: Shall be [each of us WCS members do what he can, ceteris paribus, to promote the survival of whooping cranes] (ORAV §175/p. 98) In acting as a member of the WCS, I am acting under the following description: Shall [I, qua member of the WCS, do what I can, ceteris paribus, to promote the survival of whooping cranes] Since I am acting as a member of a group, my intention is not a personal intention but a group intention­—a we-­intention­—and can be represented as follows: Shallwe [I do what I can, ceteris paribus, to promote the survival of whooping cranes] For Sellars, moral judgments are a special class of we-­intentions, where the scope of the “we” is “rational beings generally.” He writes, “Ought, as an expression of intention, is a special case of shallwe” (IILO 12/p. 204, emphasis removed).2 I will have much more to say about this as we go along. My present task is to motivate one aspect of Sellars’s account of practical reason­—to demonstrate why Sellars thought it was plausible that moral judgments consisted of expressions of intention. Tying practical reasoning to expression of intention is a very deliberate and carefully thought-­out move for Sellars­—and an ingenious one. It is extremely plausible and allows Sellars’s moral theory to satisfy a number of competing demands. However, Sellars doesn’t always explicitly spell out the motivation for adopting the view that he does, nor is he always very explicit about the advantages his view has over other views. Perhaps he assumes the reader will be familiar with disputes concerning the theories of his contemporaries and will immediately see how his own work fits into the contemporary scene. Sellars’s silence has led to an

Moral Judgments as Shared Intentions  59 underappreciation of the subtlety and power of his ethical theory, and so I want to attempt to make explicit some of these strengths. Sellars almost always begins his discussions of ethical theory by first talking about individual practical reasoning, and in particular prudential reasoning that takes the form of hypothetical imperatives. (See, for example, IILO, ORAV, SM Chapter 7.) The strategy behind his method is, I take it, that we can most easily see that practical reasoning issues in an intention to act if we first focus on means-­end reasoning by individuals. (I will have much more to say regarding the connection between practical judgment and action in chapters 10 and 11.) I suspect that there is a temptation among philosophers to assimilate practical reasoning into the more formal elements of theoretical reason. However, misleading analogies between the two can create misunderstandings of the sort Sellars aimed to correct. Thus, many philosophers might see the following as an unobjectionable instance of practical reasoning: (1) I need to pick up my daughter from school. (2) If I need to pick up my daughter from school, then I should leave now. C. Therefore, I should leave now While this may seem unobjectionable, it in fact completely misrepresents actual practical reasoning. If theoretical reasoning proceeded in parallel fashion, then we would reason as follows in theoretical cases: (3) I believe that Socrates is a man. (4) If I believe that Socrates is a man, then I ought to believe that Socrates is mortal. C. Therefore, I ought to believe that Socrates is mortal While this argument is sound, it misrepresents theoretical reasoning. The conclusion of a theoretical argument for p is not a metabelief to the effect that I ought to believe that-­p. The conclusion of a theoretical argument is p, and the conclusion of theoretical reasoning is (ceteris paribus) simply a belief that-­p. By the same token, we should reject the treatment of practical reasoning that parallels the above misleading representation of theoretical reasoning.3 As Richard Moran writes, “Practical reflection . . . does not conclude with a normative judgment about what would be best to do, but with the formation of an actual intention to do something” (2001, p. 59). Harman points out the same fact, claiming, “I understand practical reasoning to be the reasoned revision of intentions” (1986, p. 77). Of course, for Sellars, ought-­statements are to be analyzed in terms expressions. Thus, practical reasoning issuing in an of intention-­

60  Moral Judgments as Shared Intentions ought-­statement does, in the final analysis, issue in an expression of an intention, and hence (ceteris paribus) in an action. But the point of the present discussion is to explain why Sellars thinks this and why we should resist the (seemingly natural) conclusion that practical reasoning issues in a belief­—specifically, in a belief about what one ought to do. For example, if you feel thirst, your response to this is not a believing but a doing­—you will perhaps go to the sink and draw a glass of water or go to the refrigerator and pour yourself a glass of juice. If you are walking home and you see that the sidewalk is closed, you may reflect on your options (e.g., do I cross the street and use the sidewalk on the other side? or do I turn right down this side street and take alternate route X?), but the outcome of such deliberation is an action­—it is the execution of a new plan. One must be in the grip of a certain dogmatic view of how reason works to argue that the outcome of this deliberation must be the formation of a belief of the form “I ought to take route X,” or something of the sort; and that our task is to explain the connection between this belief and the ensuing action (namely, taking route X). It is natural to point out at this juncture that the outcome of deliberation frequently is belief unaccompanied by action. Thus, suppose I am sitting on the sofa watching TV and I feel thirst. While normally such a feeling would result in action, we can imagine that I am overcome by accidie and just cannot be bothered to walk all the way to the kitchen. Thus, I might have the belief “I really would like to get a drink”­—I might even think, “I really ought to get something to drink before I get a headache from not drinking enough water”­—but my feeling of thirst leads only to belief, not action. However, I think it would be a mistake to conclude from cases like this that practical reasoning in general issues in belief (rather than in intention or volition). To conclude this would be a manifestation —to generalize from of another unfortunate tendency in philosophy­ exceptional cases to all other cases rather than explaining exceptional cases in terms of their deviation from normal cases. Consider a parallel: Suppose I read all of the arguments for animal rights, and I find them very convincing. I don’t find any plausible reason why animals should not be afforded a very broad range of rights, and I don’t see any plausible rebuttals to the pro-­rights arguments. However, I find myself resistant to embracing this view­—perhaps because I am resistant to the practical consequences that such a belief might have for my life. (Philosophers who focus on weakness of the will in moral cases often fail to note that there can be parallel failures of rationality in epistemic cases.) Thus, while a theoretical argument of this sort would normally have the conclusion C2. Animals have rights X, Y, and Z

Moral Judgments as Shared Intentions  61 the argument failed to produce this belief in me; instead, it only produced the following metabelief in me: C1. I ought to believe that animals have rights X, Y, and Z Although such cases of “epistemic weakness of the will” occur, they are of course exceptional, and we should not generalize from them. In particular, they should not lead us to believe that theoretical reasoning in general takes the following form:   (5)   (6)   (7)   (8)

I believe that P. I believe that Q. I believe that R. If I believe that P, Q, and R, then I ought to believe that animals have rights X, Y, and Z.

C1. Therefore, I ought to believe that animals have rights X, Y, and Z C2. Therefore, animals have rights X, Y, and Z Rather, we should understand that theoretical reasoning proceeds as usual:  (9) P (10) Q (11) R (12) If P, Q, and R, then animals have rights X, Y, and Z. C2. Therefore, animals have rights X, Y, and Z And we should understand my failure to embrace C2 not as a failure to move from C1 to C2 but as a situation-­specific rational failure on my part­—a failure to move from 9–12 to C2­—induced by my reluctance to face the consequences of embracing a potentially life-­altering theoretical conclusion. The practical case is precisely parallel to the theoretical case. Feeling thirsty would normally produce in me a volition that would find overt expression as follows: I shall now get water from the kitchen However, accidie prevented the formation of this volition, and as a result my thirst produced (instead of a volition) a metabelief­—a belief about what volition or intention I should form or what action I should perform: I should go to the kitchen and get something to drink

62  Moral Judgments as Shared Intentions Just as in the theoretical case, though, it would be a mistake to conclude from this that practical reasoning in normal cases results in a metabelief about what one ought to do. Rather, it makes more sense (just as in the theoretical case) to explain that the normal, fluid movement from motive (feeling of thirst) to volition was interrupted by some intervening state (such as accidie, tiredness, or laziness). This is a simpler explanation, and it avoids over-­intellectualizing the process of practical reasoning. And indeed, when one considers the countless times in a day one acts­—sometimes without deliberation, sometimes after brief deliberation­—it becomes clear that in the overwhelming majority of cases, a decision issues directly in action. Thus, for Sellars, it is entirely plausible that when I engage in means-­end reasoning, the issue of such reasoning is an intention to act. This explains why in normal cases such reasoning results not in a metabelief about what one ought to do but instead in a doing. Alternately, one could say that in such cases one really does express a genuine ought-­judgment­—i.e., one really does express an intention­—but the intention is robbed of its power by a countervening desire or some other intervening psychological state. It is consistent with Sellars’s view that the overt expression of what one ought to do is often characteristic of cases of weakness of the will. In normal cases, one forms an intention and acts in a fluid sequence; one doesn’t overtly or consciously express the intention. Only when something (like a countervening desire) blocks the intention from being efficacious does the intention become “present-­to-­hand,” in Heidegger’s terminology­—just like, say, a malfunctioning computer mouse becomes present-­to-­hand as it defies fluid and skillful manipulation. But the mouse and the intention retain their identity even as their function is hampered by malign causal influences. I will return to this issue in the chapters on moral motivation. The focus on intentions also captures another important part of our practical lives. (Michael Bratman, for one, has written about this aspect extensively and insightfully; see Bratman 1999.) Our lives involve planning. We plan, generally, by forming an intention and then treating this intention as a basis from which to reason about means of fulfilling this intention. This reasoning takes the form of reasoning from the original intention to subsidiary intentions, which are means or conditions of fulfilling the original intention. Thus, to use Bratman’s example, suppose in January I form a plan­—that is, I develop an intention­—to go to Boston in April. I do­—indeed, if I am rational, I must­—treat this intention as a basis from which to infer further intentions concerning means of achieving this prior intention. As Bratman writes, “I treat my intention to go to Boston as directly relevant to the rationality of the further intention concerning means that I reach in such reasoning. I see it, for example, as directly relevant to the rationality of my decision to take a certain United flight” (1999, p. 22).

Moral Judgments as Shared Intentions  63 Thus, an important feature of our practical lives involves reasoning from settled intentions (“I shall write a book on Wilfrid Sellars”) to subsidiary intentions seen as means of achieving this end (“I shall acquire this book of essays on the philosophy of Wilfrid Sellars”). Thus, Sellars treats inference among intentions as central to the practical lives of agents. In crucial respects, the moral case is parallel to the case of individual means-­end reasoning. In normal cases, reaching a moral decision doesn’t generate an output of the form “I ought to do X.” Instead, it generates a doing of X. If I find a lost wallet at work and begin walking to the security office to hand it over to lost and found, it is because my discovery has resulted in an intention to turn the wallet in, not in a belief that I ought to do so. There can, of course, be moral weakness of the will­—just as there can be prudential and epistemic akrasia­—but I again urge the reader to generalize from the typical case rather than the exception. Not everyone finds the analogy between practical reasoning and moral reasoning convincing. There is still a tendency in philosophy to regard moral reasoning as issuing not in an intention but in a normative belief about what one ought to do, a move that is unfortunate chiefly because it renders mysterious the connection between moral reasoning/judgment and action. Of course, this view of moral reasoning makes it a species of theoretical reasoning, not practical reasoning. Anscombe was, of course, a fierce critic of this line of thought: If all that distinguishes moral reasoning is its subject matter (and not the fact that it issues in action), then there ought to be as many types of reasoning as there are subjects. As she wryly remarks, “Contemplating the accounts given by modern commentators, one might easily wonder why no one has ever pointed out the mince pie syllogism” (Anscombe 1963, §33/p. 58). Again, I will have much more to say about the connection between moral judgment and motivation in chapters 10 and 11. However, let me preview one of the arguments there, as it gives a good reason for rejecting this view of “moral” reasoning as a kind of theoretical reasoning. Michael Smith (1994) has convincingly argued that an explicit theoretical belief that such and such is morally obligatory does not ordinarily play a role in the motivating reason of the virtuous person. Thus, Smith (who embraces the Humean theory of motivating reasons) argues that someone who is motivated by the following belief-­desire pair displays not virtue but “moral fetishism”: Belief that I ought to give to Oxfam Desire to do what I ought to do According to Smith, “Good people care non-­derivatively about honesty, the weal and woe of their children and friends, the well-­being of their fellows, people getting what they deserve, justice, equality, and the like, not just one thing: doing what they believe to be right, where this is read de dicto and not de re. Indeed, commonsense tells us that being so

64  Moral Judgments as Shared Intentions motivated is a fetish or moral vice, not the one and only moral virtue” (Smith 1994, p. 75). As I note in Chapter 10, this line of thought is similar to Bernard Williams’s “one thought too many” criticism of utilitarianism. If I see someone in need of help, for example, the thought “. . . and helping her is the right thing to do” is not the thought that ought to motivate me. I should be motivated directly by concern for her welfare and not merely derivatively by way of concern for the moral law. Thus, representing moral reasoning as reasoning that issues in a theoretical reason about what I ought to do (rather than an action) does not accurately represent the virtuous person. Again, I will argue for this at greater length in chapters 10 and 11. Indeed, if there is no conceptual connection between moral reasoning and action, then moral discourse loses its point. As James Lenman points out, the amoralist might be conceivable­—but only against a background of people who are not amoralists. We simply cannot conceive of an entire community of people who engage in moral discourse with no propensity to act in accordance with their moral judgments. As Lenman puts the point, If you and I agree about what to do, how to live, how to deliberate, what to pursue, what to shun, whom to reward, whom to punish, whom to love, whom to hate, but disagree completely about the proper application of words like “good,” “right” “ought,” while taking these words to have no bearing on what to do, it may seem that these words, in our mouths, have become something close to empty noise.4 (Lenman 2014, pp. 17–18) But if moral reasoning issues in beliefs (or if motivational externalism is otherwise true), then the situation described by Lenman is perfectly coherent. Thus, Sellars’s decision to construe the moral ought as a species of intention-­expression­—thereby securing a conceptual connection to action­—is, as I said, ingenious. For now, though, I hope part of Sellars’s motivation for focusing on intentions has been made clear. Practical reasoning­—both prudential and moral­—is just that: It is practical. It is a misrepresentation of it to understand it as issuing in metabeliefs about what one ought (morally or prudentially) to do. In normal cases, such reasoning issues not in beliefs but in actions. Thus, it is more accurate to represent the outcome of practical deliberation in terms of intentions than beliefs. 3.1.3 Personal Intentions Versus We-­Intentions Although individual means-­end reasoning and moral reasoning both issue in intentions, there are important differences between the two kinds of

Moral Judgments as Shared Intentions  65 reasoning; and Sellars argues that individual means-­end reasoning cannot capture the moral point of view. It is not necessarily (or not merely) the content of personal intentions that misses the moral point of view. One can personally intend an action that is morally right, or that is altruistic, for example. But one fails thereby to will the action as a moral action. Suppose Smith and Jones intend as follows (ORAV, p. 96): Jones: Shall be [the nations remain at peace] Smith: Shall be [the nations remain at peace] Since Jones and Smith are expressing personal intentions, although the content of each intention is (in a sense) the same, the expressions of intention are not identical­—they are at most parallel. “Traditional emotivists,” he writes, “would speak here of an ‘agreement in attitude’” (SM 7.VI.§35/p. 189). This raises all of the familiar problems about such accounts­—can there be genuine moral disagreement rather than mere disagreement in attitude?­—but Sellars has an additional worry in mind. The standpoint of personal intention misses the moral point of view because it cannot account for the fact that to intend something under a moral description is to intend something as a member of a moral community and not merely as an individual. Thus, to realize fully the moral point of view, we must move from an egocentric notion of intention to one that better captures the idea that we are part of a group. (We will return to the necessity of community and group membership in Chapter 13.) Thus, Sellars writes, All intentions are explicitly or implicitly action I-­referential. To put it in this way, however, obscures the fact that although “in the last analysis” all action is action by individuals, and although all volition has “in the last analysis” the form Shall [I do A here and now] there is an important sense in which even such intentions can transcend I-­centeredness. The key point lies in the fact that one can think of oneself as a member of a group. (ORAV §§142–4/p. 96) Crucially, in thinking of oneself as a member of a group, it can become possible to intend qua member of that group. Thus, although I might have a personal intention to go to Bob’s Wine ‘n’ Dine for dinner, I might intend qua member of the Whooping Crane Society to participate in a peaceful protest at the headquarters of Acme, LLC, to protest its destruction of whooping crane habitat. To express such a notion, Sellars of course introduces the notion of a we-­intention. We-­intentions are expressed formally as follows: Shallwe [I do A]

66  Moral Judgments as Shared Intentions The move to we-­intentions solves a number of problems for Sellars. First, Sellars claims that an account of morality must make sense of the “intersubjectivity and truth” of the moral ought (SM 7.I.§4/p. 176). But while moral ought-­statements are capable of truth and falsity, expressions of personal intentions are not (ORAV §57/p. 87); and as we saw above, expressions of personal intentions lack a key element of “intersubjectivity and truth”: Namely, different individuals can neither share the same personal intention nor have, strictly speaking, contradictory (as opposed to merely practically conflicting) personal intentions. If true moral oughts are to express intentions, Sellars must find a species of intention-­expression capable of the marks of intersubjectivity. We-­ intentions fit the bill: We can share we-­intentions; you and I can have we-­intentions that logically contradict each other; and (as we will see in Chapter 13) Sellars holds that we-­intentions can be intersubjectively and categorically reasonable­—that is, binding on all rational beings. As deVries writes, “Sellars does not think an ordinary expression of intention, say, ‘I’m going to have the raspberry tart for dessert’ is either true or false. . . . Moral judgments are expressions of intention, but, because of their universality and intersubjectivity, the vocabulary of ‘true’ and ‘false’ is appropriate to them” (deVries 2005, p. 311, n. 8).5 To continue: Although a we-­intention issues in individual action, one acts as a member of the group. Thus (ORAV, §§153ff/pp. 96–7), if we intend (α) Shall be [each of us have a new car] this intention implies the following intention, which results from replacing “each of us” with an instantiation of the same: (γ) Shall be [I, Dick, have a new car] As Sellars notes, (γ) follows from (α) because “Dick has come into the picture as ‘one of us’” (ORAV §156/p. 97), and so (γ) might be reformulated in the following, more illuminating way: Shall be [I, Dick, because I am one of us, have a new car] The thought behind this reformulation is crucial to understanding how an individual action is related to group intention. As Sellars writes, “It is important to realize that because is just as much a part of the content of the above intention as it is a part of the belief content expressed by, ‘The Pirates should win because they are the better team,’ and has the same rationale” (ORAV §156/p. 97). Of course, two things are crucial for a we-­intention to express the moral point of view. The first is the scope. A we-intention where the

Moral Judgments as Shared Intentions  67 scope of the “we” is only the Whooping Crane Society, for example, does not in itself express the moral point of view. Sellars argues that the scope of this “we” must be “rational beings generally” (ORAV §206/p. 101). The second thing necessary for a we-­intention to express the moral point of view deals with the intention’s content, for a shared intention to, say, open a lemonade stand or put on a community theater production of “South Pacific” or paint the spare bedroom chartreuse need not have any particular moral content at all. There must be a we-­intention that is categorically reasonable. For Sellars, this is the we-­intention that our general welfare be maximized. We will return to this important topic­— categorical validity­—in Chapter 13. At the end of this chapter (and again at length in Chapter 13), we will explore in more depth this idea that the scope of the “we” must be “rational beings generally.” But first, I should remark briefly on Sellars’s theory of action.

3.2 Sellars’s Causal Theory of Action As I will discuss at greater length in later chapters, intentions and actions fit into Sellars’s larger botany of linguistic transitions. For Sellars, there are language-­entry transitions, as when photons strike my retina and I form the belief that there is an orange pencil in front of me. These take one from a “situation which is not a position in the [­language] game [to] a situation which is a position in the game” (SRLG, §22/p. 329). Second, there are language-­language transitions­—chiefly, inferences­— as when I move from the belief that the pencil is orange to the belief that the pencil is colored. Such transitions take one from one position in the language game to another position in the language game. Finally, there are language-­departure transitions, as when I say to myself, “I shall raise my hand now,” and then raise my hand. These transitions take one from a position in the language game to a behavior that is not a position in the game. For Sellars, all of these transitions are in the first instance conditioned transitions­—that is, stimulus-response transitions­—and so entrenched in the causal order. (We will say more in Chapter 9 about what makes these causal transitions also part of the rational order.) It seems clear, then, that Sellars’s theory would fall under the heading of causal theories of action. Rosalind Hursthouse defines such theories as follows: “Intentional actions are actions (or movements) caused, perhaps in a certain way, by certain mental states or events, whose occurrence explains the occurrence of the action or movement” (Hursthouse 2000, p. 84; quoted in Wiseman 2016, p. 109, n. 1). Causal theories are perhaps the dominant type of theory at present, but they are of course not the only show in town. Anscombe was famously critical of such theories, arguing (among other things) that if the intention were separate from the action­—that is, if we understand the intention as an inner mental event and the action as an external event caused by the intention­—it would be

68  Moral Judgments as Shared Intentions impossible to explain how an action itself could be good or bad, virtuous or vicious. As Rachael Wiseman notes in her commentary on Anscombe’s Intention, there are a couple of ways to try to combat the “interiorization” of intention that is part of the causal theory of action. “First,” she suggests, “one might say: ‘intention’ refers to something whose existence is not purely in the sphere of the mind” (Wiseman 2016, p. 159). In this view, intention extends beyond the mind into the realm of action. Thus, the assumption that intentional action involves a “language-­exit transition” is misleading, in that it envisions action as itself outside the sphere of the conceptual, linguistic, or rational. I take it that this is McDowell’s move when he criticizes Sellars for “mak[ing] unavailable to himself the idea that bodily behavior can itself be informed by practical thought, not just a transition from it” (McDowell 2011, p. 122). McDowell writes, If we follow Sellars’s recommendation about how to conceive non-­ overt conceptual goings-­on, and apply it to his account of acting on expressed intentions, according to which acting is an exit from the sphere of the linguistic, the result should be a conception of actions as exits from the sphere of the conceptual. And that seems hard to square with the idea that the sort of practical thought that intention is can be present in intelligent behavior, not just a separable item to which intelligent behavior responds. It should be possible to see conceptual capacities as actualized in acting itself, not just in something that is the starting-­point of a transition one makes in acting. (McDowell 2011, p. 121) This view can be seen as a sympathetic extension of Sellars’s view. Anscombe, however, would reject even this, as it regards intention as a specific mental state, albeit one that is carried over into action. As Wiseman writes, Anscombe rejects any psychological analysis of intention (and with it the idea that intention is specifically a mental state that causes action). Rather, the function of intention-­talk in the language is different. When we say that an action was done intentionally or with the intention that-­x, we are “assign[ing] it to a calculative order” (Wiseman 2016, p. 161). That is, in Anscombe’s famous account, we are saying that it is the kind of action to which a certain kind of why question is appropriate and to which a certain kind of reasoning­—namely, practical reasoning­—applies. Thus, to use Anscombe’s example, I might describe a person’s action as follows: He is moving the pump handle up and down In describing this action as intentional, I am placing it within a calculative order: I am saying that it is appropriate to ask why­—to what end or

Moral Judgments as Shared Intentions  69 purpose­—he is moving the pump handle up and down. And we can nest this description within further descriptions that not only describe the action but also give the further purposes he intends in performing the action: He is moving the pump handle up and down Why? In order to pump water to the house Why? In order to poison the inhabitants of the house Why? In order to prevent the inhabitants from carrying out their evil plan Etc. Eventually, the nesting comes to an end because we arrive at an “answer [that] does not describe something he is doing in moving his arm but only something that he is moving his arm in order to do” (Wiseman 2016, p. 126). So we see that an intention is a form of description of an action rather than a psychological state posited to explain an action: While all of these are descriptions under which the man’s action might fall (“Pumping water to the house,” “Poisoning the inhabitants of the house,” etc.), we are not committed to these describing the actual mental state of the person performing the action. Rather, we are putting the person’s action in the rational, calculative order; we are presenting conditionals to which the person is committed (and, presumably, would in ordinary circumstances assent if asked). I suspect that Sellars’s broader ethical theory could accommodate an Anscombean theory of intention. The key lies in what Anscombe says about (a) practical reasoning and (b) intentions as belonging to the calculative order. Anscombe’s theory of practical reason­—like her broader account of intentions­—gives a formal/logical rather than psychological account of the subject. In this, she would find an ally in Sellars. Because her account is formal, practical reasoning does not assert a relation between (for example) a desire and an end. To do so is to confuse a (potential) cause of action with the object of a desire. Anscombe illustrates this distinction with the story of a child frightened by a piece of red fluff on the stairs who subsequently reports that his nurse told him it was a piece of Satan. (No doubt, Anscombe writes, the nurse told the child it was satin.) The object of his fright­—“what he was frightened of”­—was the bit of red fluff. On the other hand, “the cause of his fright was his nurse’s remark” (Anscombe 1963, §10/p. 16). Practical reasoning represents actions as rational to pursue in light of some object that is itself good. But even so, the aim or major premise of practical reasoning is not the object as such­—it is rather the object under a certain description that makes its pursuit intelligible. Thus, Anscombe writes, “We must always remember that an object is not what what is aimed at is; the description under which it is aimed at is that under which it is called the object” (Anscombe 1963, §35/p. 66). Let us see how this latter consideration fits with Anscombe’s (and Sellars’s) broader view of practical reasoning. Consider the conditional:

70  Moral Judgments as Shared Intentions (P) There are certain substances in the soil implies (Q) There will be spectacular plant growth For Anscombe, theoretical and practical reasoning are underlain by the same conditionals; the difference is the use to which the conditional is put. In theoretical inference, we will reason from P to Q­—that is, from a causal antecedent to a consequence. In an explanatory inference, we will reason from Q to P­—that is, from an effect to a cause we hypothesize to explain the effect. And finally, in practical inference, we also reason from Q to P­—that is, from an effect we wish to produce to a means to bring about the effect. This meshes well with Sellars’s account of practical inference, where we would reason from Shall be [My plants grow spectacularly] to Shall [I put those substances in the soil] That is, I reason from Q to P. But in Sellars’s account of practical reasoning, my intention expresses an end­—an object­—that is (in some way or other) rational or desirable to pursue. Thus, it presents an object or end as the starting point of practical inference, as in Anscombe’s account. Consider how this account of practical inference fits into Anscombe’s account of what intentions are in the first place. Remember that for Anscombe, an action is intentional if a certain sort of why question is appropriately asked of it­—a why question showing why the action was reasonable or justifiable, i.e., what end or object it promoted. Thus, for example, we can ask why the man is moving the pump handle up and down but not why he is casting a particular shadow on the rock as he does so.6 (Or if we ask, he will say, “I didn’t know I was doing so,” thereby carving out this action as non-­intentional, in Anscombe’s account.) Thus, we can attribute to the man the intention Shall (I move the pump handle up and down) But we cannot (in this particular case) attribute to the man the intention Shall (I cast a certain shadow on this rock) Having labeled the first action as intentional, and thereby placing it in the calculative order, we can thereby ask why he is doing it­—that is, to what end or purpose. And we can attribute to the man wider intentions corresponding to the broader intentional descriptions under which the action falls.

Moral Judgments as Shared Intentions  71 Shall (I pump water to the house) Shall (I poison the inhabitants of the house) Shall (I prevent the inhabitants from carrying out their evil plan) Etc. Practical inference, in a Sellars-­cum-­Anscombe account, could be seen as reasoning from wider descriptions of actions to narrower descriptions that would concretely realize these descriptions. Thus, Shall (I prevent the inhabitants from carrying out their evil plan) (I prevent the inhabitants from carrying out their evil plan) implies (I poison the inhabitants of the house) Therefore, Shall (I poison the inhabitants of the house) Etc. In general, reasoning from a wider description to a narrower description will involve reasoning from an end to a means of fulfilling this end. (I don’t know if this is always the case. If not, then there will be two patterns into which practical reasoning will fall. I don’t think anything in particular hangs on this.) Further, I think that Anscombe’s view of the point of intention-­talk is congenial to Sellars’s view of the role of intentional (i.e., mental and linguistic) vocabulary more broadly. As Sellars famously writes, “In characterizing an episode or a state as that of knowing, we are not giving an empirical description of that episode or state; we are placing it in the logical space of reasons, of justifying and being able to justify what one says” (EPM VIII.§36/p. 169). For Anscombe, talk of intentions does not ascribe a mental state to someone. Rather, it serves to place someone’s actions in the rational or calculative order­—that is, to place it in a particular space of reasons, the space of reasons where certain kinds of why questions are appropriate. Moreover, I think that Sellars’s regimentation of “positions” into those that are in the language game and those that are not is driven by his particular construal of what naturalism entails. For Sellars, as we saw in chapters 1 and 2, an ideal science of the person would be able to give a reductionist account of the extension of linguistic and mental terms. Thus, I think for Sellars it is important that we give, from the start, naturalistic characterizations of these “events,” such as actions and behaviors­—and making them intentional or somehow conceptually laden complicates this project. However, I don’t think that in principle there is any more difficulty in characterizing a physical action (like pumping water) intentionally than there is in characterizing spoken words or any other physical processes (like brain events) intentionally. This isn’t to say that this isn’t hard to do­—it’s just to say that it doesn’t present any new problems for the naturalist, and so Sellars shouldn’t have any principled opposition to doing so. (And as I

72  Moral Judgments as Shared Intentions noted above, McDowell urges the Sellarsian to extend the intentional to include actions.) Indeed, Anscombe’s approach is consistent with Sellars’s approach to normativity, which avoids the placement problem by insisting that normativity is not something that needs to be located within the world; rather, intentional and normative vocabulary is a way of characterizing episodes and agents and placing them within the space of reasons. Finally, I should note that while still in its infancy, there is work being done to develop an Anscombian account of collective agency.7 In a way, a strict naturalist might be more comfortable with such an account, as it doesn’t require positing a group attitude or group intention to explain group action. I argued above that such notions are innocuous, but not all naturalists are quite so sanguine. At any rate, as intentions are not states for Anscombe, an explanation of intentional action doesn’t begin with the positing of an intention. For example, one might begin explaining group action by arguing that “what constitutes joint activities is a particular experience­—Searle’s ‘sense of doing something together’” (Seemann 2009, p. 648). Thus, “we think of collective events that can be individuated in response to particular ‘Why’-­questions as joint intentional actions” (Seemann 2009, p. 650). Thus, in the same way as individual intentional action, collective action can be responsive to why questions in a way that places it in the calculative order: We are marching on the Capitol Why? To protest racism and misogyny Etc. I will not develop these Anscombean themes any further here. The point is, Sellars’s model of moral judgments as expressions of intentions and his account of practical reasoning as reasoning among intentions, collective action, etc., seem as though they could be accommodated if we moved to a more Anscombean model of intention. Ultimately, what I am trying to present here is an account of Sellars’s ethical theory, and although I think his theory is flexible enough to accommodate various emendations, it is beyond the scope of the project to pursue every possibility.

3.3 Restricted-­Scope We-­Intentions In the Kantian spirit of his project, Sellars repeatedly emphasizes the universality of moral judgment: The scope of the “we” is, he argues, “rational beings generally.” However, it seems difficult to reconcile some intuitive features of morality with this conception of the scope of the “we.” For example, it seems uncontroversial that you have a special obligation to members of your immediate family, obligations you do not have toward

Moral Judgments as Shared Intentions  73 distant strangers. It is also plausible that there is a range of loyalty-­based virtues: As Andrew Oldenquist writes in his seminal article “Loyalties,” “it is the most common thing in the world for a person to decide that he should (or should not) do so-­and-­so on grounds of loyalty to his friends, family, organization, community, country, or species” (1982, p. 173). For example, suppose I am a member of a union voting to strike. Suppose I personally voted against striking, but out of loyalty to my union and its members, I choose not to cross the picket line. Cases like this seem, on their face, to have two features: (1) They seem clearly to be cases of action on moral principle. (2) They seem to be group action where the scope of the “we” is less than “rational beings generally.” Thus, I seem to be acting on a principle like “Shallwe (I do not cross the picket line),” where the scope of “we” is union workers who seem to have a special obligation not to cross the line in virtue of membership in the union and the obligations that accrue in virtue of this membership. Such cases would seem to be problematic for Sellars, who holds that the universal scope of the “we” is at least a regulative ideal for moral judgments. We can find Sellars’s answer to this puzzle by examining his discussion of hypothetical imperatives. In this discussion, Sellars notes that even hypothetical imperatives, which have the general form A. “I shall bring about Ek” implies “I shall do Ai if in Cj” are universal in scope (i.e., binding on all rational beings), and that “any restriction [in scope] belongs in the circumstance clause” (SM  7.VIII.§51/p. 195). Thus, if I wish to restrict the scope of the imperative to, say, WASPs, it would be wrong to render the imperative as B. “I shall bring about Ek” implies (for all WASPs) “I shall do Ai if in Cj” Instead, Sellars claims, the imperative’s general form should be rendered as C. “I shall bring about Ek” implies “I shall do Ai if in Cj◊” where “‘Cj◊’ differs from ‘Cj’ by including the additional characteristic of being a WASP. In this sense general hypothetical imperatives can be said to hold ‘for all rational beings’” (SM 7.VIII.§52/p. 195). Sellars’s move is not ad hoc. As he points out, B straightforwardly implies that being a WASP is causally implicated in doing Ai in Cj in order to bring about Ek. But that is not what is intended at all. Rather, what is

74  Moral Judgments as Shared Intentions intended is that being a WASP is a condition of the applicability or normative relevance of the hypothetical imperative rather than, say, a sufficient condition on one’s being causally able to bring about Ek by doing Ai in Cj.8 All of this suggests a parallel treatment for loyalty-­or group-­based obligations. Consider the derivation of particular moral principles in Sellars’s account. We start with the intrinsically reasonable categorical imperative D. It shallwe be the case that our welfare is maximized Then, using “the vast number of complicated instrumental nomologicals” of the form E. “x so acts as to maximize GW” implies “x does Ai if in Ci” we derive concrete categorical imperatives of the form F. It shallwe be the case that each of us does Ai in Ci Let us focus our attention on E, for that is the crucial step. One way of deriving loyalty-­based principles would be by instantiating E with nomologicals like the following and deriving categorical imperatives as follows: G. “x so acts as to maximize the general welfare” implies (for members of my union) “x does Ai if in Ci” Therefore, H. It shallwe◊ be the case that each of us◊ does Ai in Ci [where ◊ indicates the restriction of the scope of “we” and “us” to “members of my union”] The problem with this derivation lies in G. The reasoning is identical to the case of the general hypothetical imperative. G, as it stands, implies that being a member of my union is somehow causally responsible for one’s ability to maximize the general welfare by doing Ai in Ci. Again, though, that was never the point of the scope restriction, and it will often turn out to be false that my membership in a group makes me uniquely causally able to maximize the general welfare by doing Ai in Ci. Rather, the point of the scope restriction was to indicate that being a member of my union was a sufficient condition on my being obligated to do Ai in Ci, and that while others might have a similar obligation, they would not have this obligation in virtue of such membership. Their obligation would not be loyalty-­based. To illustrate, membership in the union does not make Jones uniquely able to maximize union welfare by refusing to cross the picket line. A

Moral Judgments as Shared Intentions  75 non-­union worker is equally able to maximize union welfare by refusing to cross the picket line; union membership plays no causal role here, which is why Sellars objects to the implication that group membership plays a causal role in the nomological. To be sure, a non-­union member might still have an obligation not to cross the picket line, although I do not take up that argument here. But what this specific restricted-­scope we-­intention tries to capture is that the nature of the union member’s obligation is different from the non-­union worker’s obligation. The union worker’s obligation is of a particular nature because of his membership in the union, and this is what is expressed by tying his obligation specifically to his membership in the union. Thus, instead of using G in our above derivation, we should (to parallel the hypothetical imperative case), use G'. “x so acts as to maximize the general welfare” implies “x does Ai if in Ci◊” [where Ci◊ differs from Ci by including the additional characteristic of “being a member of my union”] This would allow us to draw the following, correct conclusion: H'. It shallwe be the case that each of us does Ai in Ci◊ [where Ci◊ differs from Ci by including the additional characteristic of “being a member of my union”] Obviously, there often is a causal connection between being a member of a particular group and being able to maximize the general welfare by doing Ai in Ci. A frequently made point is precisely that, for example, parents are best able to contribute to the general welfare by attending to the needs of their own children, and members of my union are best situated to achieve their workplace goals by acting in solidarity with each other, which is something only they can do. Often, the point of such restricted-­scope obligations is precisely that they take into account who is best situated to bring about certain outcomes, and the causal story is part of that. But inserting the scope restriction into the antecedent of what is essentially a nomological statement (like G) is misleading, in that it implies that in all cases one’s membership in the relevant group confers some ability to bring about the relevant end by doing Ai in Ci. As demonstrated above, this is not always the case. Perhaps just as importantly, though, by putting the scope restriction directly into the scope of the “we,” it eliminates what is possibly the most intuitive way of distinguishing moral action from other sorts of coordinated group activity that is not action on a moral principle: namely, that moral actions express categorically valid we-­intentions binding on all rational creatures, not merely on a subset of such creatures. Some support for Sellars’s line of thinking can be found in the literature on agent-­relative duties. The distinction between agent-­neutral and

76  Moral Judgments as Shared Intentions relative duties can be traced back to Nagel’s The Possibility of agent-­ Altruism. Nagel asserts (similarly to Sellars) that reasons-­statements quantify universally over all persons. However, within a universally quantified reason statement, the conditions for the application of the reason can either be universal or restricted in scope. Nagel puts it in terms of whether the conditions for the application of the principle contain “a free occurrence of the variable p” (1970, p. 91), where p is the variable used to range over persons. For example, suppose Anne decides to take her therapy animal to the local nursing home on the grounds that it will make some of the residents happy. She might be acting on the following principle: I. (p, A) (If A will make someone happy then p has reason to promote A). (Ridge 2011, p. 3) (We needn’t, of course, suppose that this reason is decisive or overriding.) This principle is agent-­neutral because p doesn’t appear in the antecedent. Thus, the reason doesn’t make specific reference to any circumstances or relations of the agent in question; this reason could therefore be a reason for anyone and so is agent-­neutral. Suppose, by contrast, that Anne wishes to visit her friend Sue to cheer her up and wants to do so specifically because Sue is her friend. In this case, the fact that Sue is her friend is part of Anne’s reason to visit her. Thus, our principle I above does not capture Anne’s reasoning; her reasoning is better captured by something along the following lines: J. (p, A) (If A will make p’s friend happy then p has reason to promote A). (Ridge 2011, p. 4) As Michael Ridge writes, since “the sufficient condition for the application of the reason predicate (the condition given by the antecedent of the principle corresponding to the reason) includes . . . a free-­agent variable then the reason is agent-­relative” (2011, p. 4). A condition for the application of the reason is that it is p’s friend who is made happy, and this is a condition on p having a reason to act. Thus, the reason is agent-­ relative: It is a reason for Anne but not necessarily for anyone else. The importance of this discussion is two-­fold. First, loyalty-­based reasons like the ones I have been discussing are plausibly understood as agent-­relative reasons. Thus, I have a reason to promote my children’s welfare because they are my children. When I do so, I do not act on the following principle: K. (p, A) (If A will promote the welfare of children, then p has reason to promote A) Rather, I am acting on something much more like the following principle:

Moral Judgments as Shared Intentions  77 L. (p, A) (If A will promote the welfare of p’s children, then p has reason to promote A) Similarly, if I join in solidarity with my union and refuse to cross the picket line, I am not acting on the following principle: M. (p, A) (If A violates union solidarity, then p has reason not to do A) I may well endorse this principle­—and I probably do, if I am a devoted union member­—but in the case at hand, I am motivated by loyalty to my union, not to the principle of respect for unions generally. Thus, my reasoning is better represented by the following principle: N. (p, A) (If A violates solidarity with p’s union, then p has reason not to do A) So that is the first important conclusion of this discussion: Loyalty-­ based reasons are most plausibly thought of as agent-­relative reasons. The second important conclusion of this discussion is that the scope restriction of such reasons appears in the circumstances of application. On this reading of moral principles, such principles are still universal in scope, but the agent-­relativity appears in the conditions of application for the principle. This, of course, aligns neatly with Sellars’s contention that “any restriction [of scope] (say, it only applies to WASPs) belongs in the circumstance clause” [SM 7.VIII.§51/p. 195]. Thus, such principles are binding on all rational beings even if most will never find themselves in the conditions specified in the circumstances of application.

Notes 1 Although in point of fact, Sellars thinks that the practical/theoretical distinction is somewhat artificial: “Of course, if one gives to ‘practical’ the specific meaning ethical then a fairly sharp separation of these activities can be maintained. But if one means by ‘practical’ pertaining to norms, then so-­called theoretical reason is as larded with the practical as is practical reasoning itself” (LTC, p. 511). 2 Also, “In their primary use, ‘ought’ and ‘good’ are special cases of ‘shall’” (TA, I/p. 106); and “‘Ought’ statements express (presumptively) shared intentions” (SE §53). 3 And indeed, I think it is a quite deliberate decision on Sellars’s part to begin ORAV with a discussion not of practical reason but of theoretical reason, and to point out that “To ‘see’ that ‘P’ implies ‘Q’ is, ceteris paribus, to have the propensity to believe ‘Q’ if one believes ‘P.’” Thus, theoretical reasoning normally results directly in belief, not metabelief. Why would one think otherwise? The key, as I shall be arguing, is to think in a precisely parallel way about practical reason. 4 Lenman attributes this argument to Gibbard (2003, chapters 1 and 7). 5 Sellars’s views on moral truth are subtle, and he is not always explicit about them. As noted, Sellars claims that “expressions of intention are neither true

78  Moral Judgments as Shared Intentions nor false” (ORAV §57/p. 87). Just as valid theoretical reasoning preserves truth, valid practical reasoning preserves categorical reasonableness. However, Sellars is insistent that moral oughts are truth-­apt, arguing that “a theory of practical reasoning in morals which denies the in principle intersubjectivity and truth of the ought-­to-­be’s and ought-­to-­do’s of everyday life must face the challenge of the ought-­to-­be’s and ought-­to-­do’s of theoretical reason” (SM 7.I.4/p. 176). And, of course, the moral ought expresses a certain kind of we-­ intention. But how can the moral ought be truth-­apt if expressions of intention cannot? The best way of reconciling all of this, I think, is understanding Sellars as claiming that expressions of personal intention are neither true nor false­— for his claim always comes at a point in the dialectic when he is still talking about personal intentions and hasn’t yet introduced we-­intentions. Moral we-­ intentions are “universal in three dimensions” (SM 7.XIX.133/p. 222): They are universal in content (applying to anyone in the same circumstances); they are shareable by all rational beings; and they are objective, in that “there is, in principle, a decision procedure with respect to specific ethical statements” (SM 7.XIX.133/p. 222). Thus, moral we-­intentions have characteristics that allow us to apply the truth predicate to them­—especially if we have (as Sellars does) a deflationary view of truth. Ergo deVries’s claim, just quoted, about why Sellars thinks moral we-­intentions are (unlike other expressions of intention) truth-­apt. For a discussion of how Sellars fits moral truth within his larger theory of truth as semantic assertibility, see Klemick (forthcoming). 6 Of course, we could ask why the man is casting a certain shadow. But we would be asking for a causal explanation; the why would have a different sense. Again, Anscombe’s point is that an action is intentional if a certain sort of why question is appropriately asked of it. 7 See, for example, Seemann (2009); Laurence (2011). 8 “For if being a WASP is irrelevant to bringing about Ek by doing Ai in Cj, then there is no point in including it anywhere; while if it is relevant, the relevance is a causal one and belongs in the content of the implication and not as a limitation on those for whom it holds” (SM 7.VIII.§51/p. 195).

4

What Are Sellarsian We-­Intentions?

Michael Tomasello has famously argued that one of the fundamental divides between humans and other primates­—the feature of our psychology that has allowed humans to make enormous, cumulative social advances in the last thousands of years­—is our capacity for cooperation, which is built on an underlying capacity for shared intentionality and joint activity. Even the apparently coordinated hunting behavior of wolves, lions, and chimpanzees does not involve, on their part, “any kind of joint goals or plans” (Tomasello 2009, p. 63). When apes hunt, they “are engaged in group activity in I-­mode, not in We-­mode” (Tomasello 2009, p. 63). Humans, by contrast, display the capacity to form joint goals with partners from soon after their first birthday (Tomasello 2009, p. 63). This capacity for group intentionality creates the basis for all human social institutions, including the one most relevant for our purposes­—the institution of morality.

4.1 We-­Intentions and Group Attitudes The previous chapter’s discussion of we-­intentions leaves a number of questions unanswered. For example: What characterizes we-­intentions (as opposed to I-­intentions)? What is the relation between individually held we-­intentions and the we-­intentions attributable to the larger group? • What is the relation between we-­intentions and other sorts of we-­ attitudes (like beliefs)?

• •

Sellars, unfortunately, doesn’t answer all of these questions for us. Indeed, as Olen and Turner write, “Given its centrality to his philosophy, Sellars’s discussion of collective intentionality is astonishingly slim” (2015, p. 955). Fortunately, the contemporary literature offers rich accounts of collective attitudes (intentions, beliefs, and so on) that demonstrate how one can embrace a range of such attitudes in a manner consistent with

80  What Are Sellarsian We-­Intentions? Sellars’s strict naturalism.1 We can mine this literature to see which elements we can borrow to supplement Sellars’s own account. 4.1.1 Group Attitudes A distinction Sellars doesn’t address adequately is the distinction between a group attitude and an individual’s possessing an attitude qua member of a group. The former has been addressed in recent literature under the heading of joint commitment, collective action, collective belief, and so on. Arguably, possessing an attitude qua member of a group is parasitic on a prior notion of group attitude, a notion that cannot be constructed out of purely personal attitudes (like personal beliefs or personal intentions). The group attitude need not necessarily be the same attitude as your individual attitude, but your possession of an attitude qua group member will depend on our ability to ascribe some appropriate attitude to the group and on your being properly related to this group attitude. Or so authors like Gilbert and Tuomela argue. I think they are right, and that their account will help us illuminate both our present account of we-­ intentions and the Sellarsian account of social practices. Outside of ethics, it is almost entirely uncontroversial that we can predicate actions of groups as a whole, where no individual member of the group can be said to perform the action in question. Natural language contains many instances of non-­distributive plural predication, where we predicate something of a group of objects that cannot be predicated of any member of that group. As Thomas McKay writes, “It is often possible to interpret a plural claim, such as . . . The Fs are G . . . in a way that entails a universal: (∀x:Fx) (Gx) . . . But plural predication often involves properties that apply to a plurality without applying to any individual of that plurality” (1994, p. 559). Thus, for example, one can make the following claims: We surrounded the building The rocks rained down The toys were scattered across the floor All of these involve non-­distributive plural predication, in that something predicated of the group cannot be predicated of any particular member of the group. Other natural language examples seem to attribute actions to groups­—actions not obviously attributable to individual members of

What Are Sellarsian We-­Intentions?  81 the group (e.g., “The Senate passed the measure,” “The team won the game,” etc.). As, for example, Frederick Stoutland notes, it is characteristic of such groups that they can perform actions that individuals cannot perform: “The senate does things individual senators cannot do, like pass laws or issue a resolution. The university appoints a new president, which no member can do” (Stoutland 1997, p. 47). Given that we do attribute actions to groups, and given that these are often actions that, prima facie, we cannot attribute to any individual member of the group, it remains to understand what it is for a group to act together­—and what it is to act as a member of a group. For Gilbert, it is a commonplace that people often act together, and that this is not a matter of two people acting in parallel to achieve some goal. For example, when we work together to lift and move a table, it is not (merely) that you are lifting one end and I am (at the same time) lifting the other end. We are lifting the table together and together carrying it to the other side of the room. For Gilbert, the notion of acting together­—acting as a group­—is built around the idea of joint commitment. Joint commitment is a type of group attitude and contrasts with personal commitment, as when a person acting alone makes a personal decision. Gilbert writes, “Two or more people are acting together if they are jointly committed to espousing as a body a certain goal, and each one is acting in a way appropriate to the achievement of that goal, where each one is doing this in light of the fact that he or she is subject to a joint commitment to espouse the goal in question as a body” (Gilbert 2014, p. 34).2 Joint commitments imply individual commitments. However, these are not personal commitments; they are what Gilbert calls “dependent individual commitments,” and they differ from personal commitments, in that the former derive from joint commitments. A personal commitment is one that, for example, might arise from an individual decision­—say, Andrea’s decision to have lunch at the Nutmeg Restaurant today (Gilbert 2014, p. 31). A joint commitment, on the other hand, is a commitment of two or more people. Thus, if Heinrich and Andrea “jointly commit to espouse a certain goal as a body . . . [then their] joint commitment is not composed of a personal commitment of Heinrich’s and a personal commitment of Andrea’s. Rather, it is their commitment, the commitment of the two of them” (Gilbert 2014, p. 31). Tuomela concurs, writing that “the intention to perform one’s part has ‘holistic’ content as it involves joint action. This personal (but still ‘nonprivate’) intention is based on the joint intention in question” (Tuomela 2002, p. 20). When Tuomela describes this derivative intention as “personal but nonprivate,” this should be read as equivalent to Gilbert’s “individual but not personal.” For reasons of clarity, I will use Gilbert’s terminology. For both Gilbert and Tuomela, there is a fundamental difference between a we-­mode action and a purely personal action. For Tuomela,

82  What Are Sellarsian We-­Intentions? “functioning in the we-­mode is necessarily connected to a group reason” (Tuomela 2007, p. 83), something he connects (2002) to the possession of a we-­attitude. For Gilbert, group action involves acting on a joint commitment. Thus, one is only participating in a joint action to the extent that one is acting on a joint commitment­—or, to put it otherwise, to the extent that one is acting on a dependent individual commitment derived from a joint commitment rather than acting on a personal commitment. One finds echoes of Gilbert’s view in Sellars’s claim that “Shallwe [I do A]” amounts, essentially, to “Shall [I, because I am one of us, do A]”; and that the “because is just as much a part of the content of the above intention as it is a part of the belief content expressed by, ‘The Pirates should win because they are the better team,’ and has the same rationale” (ORAV §156/p. 97). Gilbert (2014, p. 263) offers an example that can help us understand the relation between collective attitudes, dependent individual attitudes, and personal attitudes: It is the view of the Joneses that Johnnie must be home by midnight. Now, this is not the personal view of either Mr. Jones (who doesn’t see why Johnnie shouldn’t come home whenever he pleases) or Mrs. Jones (who thinks Johnnie should be home by ten). But, as Gilbert writes, “The Joneses have discussed the matter and compromised. They have arrived at what they may properly characterize as ‘our view’­—not his, nor hers, but ‘ours,’ and in this case, ours only. Each understands that it is now incumbent upon each one to express ‘their’ view (that is, the compromise view) in front of Johnnie” (2014, p. 263). Ergo, this is the collective attitude of the Joneses. Further, Mr. and Mrs. Jones each have a dependent individual attitude (which is realized in their individual interactions with Johnnie)­—but this attitude is parasitic upon, and cannot be understood apart from, their collective attitude. Finally, their dependent individual attitudes cannot be understood as mere personal attitudes since (again) the collective view of Mr. and Mrs. Jones does not reflect the personal view of either. Thus, collective attitudes are not reducible to an agglomeration of personal attitudes. As Olen and Turner write, there is in Sellars a categorical difference between “the genuinely collective [versus] the merely enumerative fact of various individuals having parallel intentions or beliefs” (2015, p. 964, emphasis added). Gilbert’s example helps us understand Gilbert and Tuomela’s insistence that notions like joint commitment and collective action cannot be constructed out of personal commitments and attitudes: The existence of a group attitude (and hence a dependent individual attitude) does not entail the existence of a corresponding personal attitude; ergo, group attitudes are not simply agglomerations of personal attitudes. Gilbert argues against “singularist” accounts of social phenomena, according to which intentions always have to be intentions of an individual person even if they are not personal intentions. Tuomela argues that the we-­mode and the I-­mode are mutually irreducible. To begin, Gilbert argues that one

What Are Sellarsian We-­Intentions?  83 cannot construct an account of joint commitment out of personal commitments because the latter (even when aggregated) lack crucial properties of the former. The example of the Joneses illustrates one way in which this is true. Further, Gilbert points out that personal commitments are individually rescindable: If I make a personal commitment to eat an apple each day or go to the movies after work on Thursday, I can change my mind and rescind this commitment without having violated any obligation. However, joint commitments are not individually rescindable (Gilbert 2014, pp. 31–2).3 Not only does this mean that joint commitments aren’t simply an aggregate of personal decisions, but it also shows that personal decisions are poor tools for solving coordination problems and the like. As Gilbert writes, “A personal decision is rescindable at will by the one who made it. As I wonder whether or not to push off from the shores of reflection, it may occur to me that you just might change your mind before you act, knowing that, after all, I might change mine” (2014, p. 92). (Gilbert here is keen to point out that the individual rescindability of personal commitments makes them worse than joint commitments for solving coordination problems. But I think her main point is that joint commitments aren’t personal commitments because the two have different properties.) Further, as Gilbert notes, joint commitment gives one the standing to rebuke another for violation. If joint commitments are constituted by personal commitments, it is not clear where this standing would come from, as I don’t ordinarily have the standing to rebuke you for failure to follow through on a personal commitment. The normative force of values presupposes that they are social and not merely personal, and our account of we-­intentions should reflect this. Let us turn to characteristics of we-­intentions, or we-­attitudes more broadly. Tuomela has a helpful botanization of different kinds of intention. First, one may have a strictly personal (I-­mode) intention, as when I intend to go for a walk today. Second, one can act as a group member in a weak (I-­mode) sense. Central to the notion of any group is an ethos, which Tuomela defines as “the set of constitutive goals, values, beliefs, standards, norms, practices, and/or traditions that give the group motivating reasons for action” (2007, p. 16). Tuomela writes that acting as a group member in the weak sense “is based on the group members’ sharing the ethos but being only privately (and not collectively) committed to it” (2007, p. 29), and gives the following example: “Thus there may be a group of persons who grow flowers in the village commons and intend to make their small village look beautiful in this and perhaps other ways. Each of them is only privately committed to making the village beautiful, and they mutually know or believe that the others are similarly committed” (2007, p. 29). Finally, one can act as a group member in a strong (we-­mode) sense. One does this, roughly, when one acts because of a group reason. Tuomela gives the following general definition of a we-­attitude:

84  What Are Sellarsian We-­Intentions? A we-­attitude in its strongest, full-­blown sense is defined as follows for a person, say x: (a) x has ATT(p) and (b) he believes also that the others in the group, g, have ATT(p) and he also (c) believes (or at least is disposed to believe) that it is mutually believed . . . that the members have ATT(p). (2002, p. 23) Based on the above definition, Tuomela then constructs an account of acting for a group reason as follows: Acting for a we-­attitude related to ATT(p) entails acting in the right “ATT-­realizing” way for the reason content p (cf. clause (a)). It also involves acting for the social reason that the others in the group have ATT(p) and are satisfying it or at least are disposed to satisfy it (clause (b)), and also acting in part on the mutual belief in question (clause (c)). (Tuomela 2002, p. 23)4 This is a general account of a group having an attitude and an individual acting on this group reason (and therefore acting in the we-­mode in the strong sense). I will emphasize a couple of features of Tuomela’s account. First, essential to having a group attitude is the notion of mutual belief. (We will return to this crucial issue in Chapter 9.) Thus, for ATT(p) to be a truly collective attitude, it is not enough that each of us individually holds ATT(p); rather, each of us must believe that the others hold it too. For example, if five people independently walking through the park independently form the belief “The park needs to be cleaned,” this is not a group belief even though each person shares a belief with the same contents. (This mutual belief requirement will turn out to be too strong and in need of modification as we go along.) Second, group action requires acting for a group reason as defined above (including the crucial element of mutual belief). Thus, if our five hypothetical park walkers each start cleaning the park, they are not acting as a group even though they are all doing the same thing, at the same time, in the same place, based on the same belief. To illustrate group action, Tuomela offers the following example: For instance we can have p = “Our club house is beautifully decorated” and ATT = want. When some persons act for (because of) this we-­want to have the house beautifully decorated, each person has a composite full reason because of the collective end in question. This full reason “internalistically” described consists of his wanting to have the house beautifully decorated and his believing that also the

What Are Sellarsian We-­Intentions?  85 others want so and his also believing that it is mutually believed that the members want to have the house beautifully decorated. (Tuomela 2002, p. 23) Tuomela thinks that acting in full-­blown we-­mode conceptually requires acting for a group reason­—acting on a we-­attitude­—but argues that this group reason need not be a psychological or motivational reality. The social reason need not enter the agent’s psychology but can “function as a kind of presuppositional and unreflected ground serving to characterize the very activity in question” (2002, p. 88). Thus, for example, when discussing the social constitution of money, Tuomela notes that when using coins to pay you for something, I must recognize coins as money, and I must believe that you recognize coins as money, and I must believe that you believe that I recognize coins as money. But these are conceptual presuppositions of the social institution of money and need not be consciously entertained by me during the transaction.5 Gilbert’s account of collective action and collective belief is to my mind less “intellectualist” than Tuomela’s in ways that will emerge as the chapter progresses. Gilbert doesn’t rely on the notion of mutual belief, nor on the notion of acting on a group reason. As noted above, Gilbert defines group action such that it involves joint commitment. She doesn’t explicitly parse “commitment” in terms of belief, which allows for a more expansive view of what commitment entails. Nor does she quite say that joint action involves acting for a group reason (rather than merely acting on a joint commitment). What, precisely, is joint commitment? Gilbert writes, A given joint commitment can always be described in a sentence of the following form: the parties are jointly committed to X as a body. Acceptable substitutions for “X” are psychological verbs such as “believe,” “intend,” and so on. What is it to be jointly committed to X as a body? As I understand it, this joint commitment has two related aspects. First, the parties are jointly committed to bring it about, as far as is possible, that the parties emulate a single body that Xs. Second, it is understood that their doing so is to be a function of the joint commitment in question. (Gilbert 2014, p. 175) Using the tools offered to us by Tuomela and Gilbert, we can construct an argument that norms and social institutions rely on collective attitudes, and that acting on such norms involves acting in the we-­mode in the strong sense (in Tuomela’s terminology). As noted above, there are differences between Tuomela’s and Gilbert’s accounts, and so one might be more suitable than the other. Let us begin by looking at Tuomela’s attitudes. Tuomela argument that social institutions presuppose we-­

86  What Are Sellarsian We-­Intentions? illustrates this with reference to institutions, but the applicability of this argument is much wider. Recall that Tuomela’s definition of having a group attitude and acting on it involves the element of mutual belief and acting on this mutual belief. Social institutions constitutively depend on this mutual belief and on acting on this mutual belief­—if not psychologically, then conceptually/presuppositionally. To return to the example of money, Tuomela notes, “Suppose x lacks the belief that y believes that x believes that Coins are correct exchange money for the group in question. Here x thus believes that it is not (fully) rational for y to engage in the exchange, because y, in accordance to what has been assumed, might not believe that x has a belief to the effect that . . . Coins are correct exchange money” (Tuomela 2002, p. 177). But the same applies for norms generally and not just for institutions. If, for example, I want to speak to you using the word “bird” to mean bird, then not only must (a) I believe that “bird”s are •bird•s, but also (b) I must believe that “bird”s are •bird•s for you, and (c) I must believe that you believe that “bird”s are •bird•s for me.6 Again, none of this should be supposed to be a psychological reality; only in “breakdown” situations will we consciously entertain any of these beliefs. (In the famous words of Inigo Montoya from The Princess Bride, “You keep using that word. I do not think it means what you think it means.”) But these commitments are conceptual presuppositions of language use and indeed of the use of norms generally. Thus, institutions and norms presuppose we-­attitudes in Tuomela’s sense and are thus in the basic case instances of behavior in the we-­mode in the strong sense. (I won’t take up the question as to whether following norms always involves behavior in the we-­mode in the strong sense. Tuomela argues that a group can survive a certain amount of weak [I-­mode] group activity, but such activity must be seen as parasitic upon the prior existence of group attitudes and strong we-­mode attitudes and activities. As always, I think it is more fruitful to focus on the basic case and understand “defective” cases in terms of deviation from the basic case.) There are reasons to believe that Tuomela’s mutual belief requirement is too strong as it stands. For many cases, the mutual belief requirement is quite plausible­—particularly in the kinds of examples that populate the literature on group action (e.g., two people carrying a table together). But when we look at large social collectives (like users of a particular language) or small groups engaged in skillful activity (like the quarterback and his wide receiver), the mutual belief requirement seems too strong. Something like mutual belief is at play, but the wide receiver doesn’t have an occurrent belief that the quarterback believes that he will run a particular route­—any more than I have an occurrent belief that you believe that I use the word “bird” to mean bird when I speak. We shouldn’t wholly abandon Tuomela’s insight; something like mutual belief is at work here. I wouldn’t use the word “bird” in speaking

What Are Sellarsian We-­Intentions?  87 to you if I didn’t in some sense think that there was a shared understanding of what the word meant. The key is fleshing out this “in some sense.” With what should we replace the mutual belief requirement if it is too strong? I will return to this issue later (particularly in Chapter 9), but let me preview some themes in order to answer this specific question now. As I note in Chapter 6, Brandom argues that on a pragmatist methodology, one should “prefer an order of explanation that begins with what is implicit in practice (what people do) and proceeds to an account of what they explicitly believe or say, over one taking the opposite tack” (Brandom 1994, p. 101). Any social practice­—even one in which significant disagreement exists­—presupposes massive agreement. Indeed, the very possibility of disagreement presupposes massive agreement. The ability to use language presupposes an agreement in what Wittgenstein would call “forms of life.” These forms of life are not, in the first instance, beliefs or the like. Rather, they are habits of behavior (including linguistic behavior). Consider an analogy. Imagine someone walking a meandering path that threads its way through a series of ornamental ponds. The person may never entertain the belief that the path will support her weight or that the water in the pond will not or that if she attempts to walk across the pond she will get wet. But implicit in her behavior is a set of expectations about the way the world behaves. If the path instead meandered through a series of grassy patches, the person might cut across the grass, her behavior demonstrating her differing expectations about the behavior of the world. “Expectation” is a misleading word here­—it implies an occurrent mental state where there probably is none. Again, let us remind ourselves of Brandom’s comment that what we are doing here is making explicit commitments that are merely implicit in behavior rather than attributing to the agent occurrent states to explain the behavior in question. The case is similar with agents who share a form of life. When I utter an English sentence to a fellow English user, implicit in my behavior is the expectation that this will have a certain effect on the person in question. It might simply result in an updating of the other agent’s “scorecard.” Or it might result in a certain corresponding behavior on the part of the other agent (if, for example, I have asked her for the time). But implicit in our real-­time interactions with sharers of our form of life is the expectation that we do share this form of life­—that we share a language or a set of norms for greeting (e.g., shaking hands) or for conducting transactions in the checkout line at the grocery store or for rowing a coxed eight or for whatever the particular shared activity or set of norms in question is. Indeed, this element of mutual expectation­—that each of us understands our role, that we will fulfill our part, and so on­—is a fundamental and ineliminable part of the form of life. (What would be the point of bringing groceries to the checkout counter if I didn’t expect the cashier to sell

88  What Are Sellarsian We-­Intentions? them to me? What would be the point of my ringing up your purchase if I didn’t expect you to pay? What would be the point of my rowing if I didn’t expect you to do so also? What would be the point of using language if I didn’t expect the appropriate updating and/or behavior on your part?) I argue that for Brandomian, pragmatist reasons, we should extend the mutual belief requirement to include also this element of mutual expectation implicit in practice. We should not, of course, rule out cases of group action involving mutual belief­—for surely these exist. But most cases of group action will not involve mutual belief. Something, however, has to tie the group members together, and I suggest that this element of expectation plays the relevant role. So from now on, let us understand the mutual belief requirement as involving mutual belief or expectation (in this implicit, Brandomian sense). To re-­cast our example in terms of expectation rather than belief, suppose I speak to you using the word “bird” to mean bird. Implicit in my behavior is (a) a commitment on my part that “bird”s are •bird•s but also (b) an expectation on my part that “bird”s are •bird•s for you and (c) an expectation on my part that you have an equivalent expectation, i.e., an expectation that “bird”s are •bird•s for me. Thus, my commitment regarding the meaning of bird is a commitment in the we-­mode in the strong sense, as the mutual expectation of this commitment underlies the possibility of linguistic communication.7 That is, this element of mutual expectation is partially constitutive of linguistic norms, as the ability to communicate using linguistic tokens presupposes this element. Thus, the linguistic norms governing the use of “bird”­—that is, that constitute •bird•­—must be expressed in the strong we-­mode. I will return to this issue in Chapter 9. An individual’s norm-­following behavior will typically be parasitic upon a set of group attitudes. This is clearest in the case of norms intended to solve coordination problems, such as the norm of driving on the right side of the road. In following this norm, there is implicit in my behavior (a) a commitment to driving on the right side of the road, (b) an expectation that you will drive on the right side of the road, and (c) an expectation that you expect I will drive on the right side of the road. These implicit commitments are manifested in my behavior: I drive on the right side of the road, I don’t swerve onto the shoulder (as I might if I expected you to suddenly deviate from the norm and veer into my lane), and so on. Given the above element of mutual expectation, this norm, too, is underlain by a we-­expectation­—an attitude in the strong we-­mode­—which the norm presupposes for its very ability to function as the norm that it is. Again, these expectations are implicit in our behavior­—we generally don’t have an explicit attitude toward them. These beliefs only come to the psychological forefront in what Heideggerians call breakdown situations­—when the normal smooth

What Are Sellarsian We-­Intentions?  89 functioning of things is interrupted (as when someone is driving on the wrong side of the road). And since a shared expectation is a conceptual presupposition for my behavior, that means that my behavior is we-­ mode. Again, we see that social practices are constituted by behavior that must be described in terms of shared attitudes and therefore as we-­mode behavior. Thus, it seems like we have good reason to modify Tuomela’s account more in the direction of Gilbert’s, where the joint belief condition is weakened to something like joint expectation, joint commitment, or something less implicative of an occurrent attitude on the part of the agent. There is another way Tuomela’s account is too strong: His account requires acting for a group reason. The element of expectation I propose above is implicit and, in an important sense, presuppositional for group action. Here is another analogy: Crucial to Sellars’s anti-­foundationalist picture of knowledge is that a bit of language A can be conceptually and epistemically dependent on some theoretical commitments T, without A being inferred from T. Nevertheless, this dependence makes T epistemically prior to A in a way that keeps A from being foundational; the dependence is crucial to both the epistemic and conceptual status of A. (Nor, of course, is T foundational either; epistemic dependence is a two-­ way street for Sellars.) This is the role, for example, played by knowledge of standard viewing conditions in Sellars’s account of perceptual knowledge. This is also how we should understand the role of the above implicit expectation condition in group attitudes, such as in the deployment of linguistic norms or the utilization of social institutions like money. Tuomela insists that acting in the we-­mode in the strong sense requires acting for a group reason. But as I argued above, action in the we-­mode in the strong sense can instead involve mutual expectation, where these expectations are presuppositions of the action rather than reasons for the action. (In the language of Chapter 8, these expectations are “enablers”; they should not be seen as “premises” that would be appealed to if we were to reconstruct the agent’s reasons for action.) Thus, acting in the we-­mode in the strong sense does not seem to require either mutual belief or acting on a group reason. Rather, it seems to require the more modest requirement of mutual expectation or joint commitment, which is a less intellectualist, more Gilbertian view of collective action. To conclude this section, I will note that although Sellars does not extend his own account of collective attitudes beyond intentions to include beliefs and so on, there is good reason to regard such an extension as a friendly amendment to his account. First, Sellars seems to acknowledge the legitimacy of “predicates which ascribe ‘mental states’ to groups (e.g., Russia fears China)” (CSS, ¶2)­—although I admit that CSS is far from conclusive on the matter. Second, acting on a group intention presupposes the notion of acting for a group reason, and this may sometimes require appeal to group beliefs. Finally, the same applies

90  What Are Sellarsian We-­Intentions? to collective beliefs as to collective intentions: While collective attitudes are not reducible to personal attitudes, they do not presuppose anything more than individual attitudes. Thus, we need not posit a “group mind” or anything else ontologically suspect.8

4.2 How to Understand Sellarsian We-­Intentions As noted above, Gilbert and Tuomela reject singularism, which tries to construct group attitudes out of personal attitudes. They argue that group attitudes are irreducible, and that while they may entail the existence of dependent individual attitudes, they are not merely an agglomeration of personal attitudes. Such talk makes some naturalists uneasy, as it seems to imply the existence of a “group mind.” As noted earlier, however, it is a commonplace to attribute actions and attitudes to groups (“The Senate voted to cut the estate tax,” “This body believes that there is no just cause for war,” etc.), and a philosophical analysis of such talk should treat it as prima facie in order as is­—subject, of course, to a compelling reason to find such talk objectionable in principle. While Gilbert and Tuomela reject singularist accounts of we-­attitudes, they deny that proper ascription of such attitudes commits us to any ontological reality over and above individuals. Gilbert, for one, argues that we can talk about a plural subject, but the plural subject referred to in “We are doing A” is “the jointly committed individuals as such” (Gilbert 2014, p. 89). We have defined joint commitment above for Gilbert, and she defines what it is to collectively intend in terms of this: “Persons X and Y collectively intend to perform action A . . . if and only if they are jointly committed to intend as a body to do A” (Gilbert 2014, p. 88). So while joint intention cannot be reduced to personal intentions, it can be analyzed in terms of individuals and their relations. Thus, there is nothing spooky, nothing that should offend against the nominalism we have been developing here. Similarly, although Tuomela argues that “it makes (at least functional) sense to attribute mental states (e.g., acceptances, goals, intentions) to a group” (Tuomela 2007, p. 20), his analysis of group attitudes is also ultimately in terms of individuals and their relations and so should not offend against nominalism. Thus, while as a conceptual and semantic matter the attribution of group attitudes to groups is prior to the attribution of dependent individual attitudes to individuals, and while group attitudes are not reducible to personal attitudes, nevertheless, ontologically speaking we are not committed to anything over and above “the jointly committed individuals as such.” Thus, while semantically and conceptually group attitudes are prior to dependent individual attitudes, in speaking of group attitudes we need not ontologically commit ourselves to anything over and above individuals and their relations. (As I have repeatedly emphasized, Sellars holds that Φ’s being prior in the conceptual or epistemic order does not commit us to Φ’s being ultimate in the ontological order, the “order of being.”)

What Are Sellarsian We-­Intentions?  91 This is crucial for Sellars; as Olen and Turner write, “He wants an account of collective intentionality consistent with an ‘empiricist philosophy of mind,’ and he abjures appeals to ‘the existence of a group mind,’ capable of having beliefs and intentions, in a sense incompatible with empiricist principles” (Sellars, IILO, p. 203; Olen and Turner 2015, p. 960). On the present account, group intentions are not reducible to personal intentions but do not presuppose anything more than individual attitudes. However, as illustrated by the example of the Joneses (and their attitudes toward when Johnnie must return home), these individual attitudes cannot be understood apart from the collective attitude attributable to the Joneses. For someone like Sellars, there is nothing too puzzling in claiming that dependent individual attitudes conceptually presuppose group attitudes even though group attitudes commit us ontologically to nothing over and above individuals and their mental states. Dependent individual attitudes conceptually presuppose groups because the former cannot exist without (a) other agents and (b) the elements of mutual expectation. But this doesn’t mean that ontologically we need to be committed to anything over and above the beliefs and attitudes of agents. To think so is to commit a category error: Marriage, for example, requires (at least) two people. One person cannot individually marry. But from this fact, it does not follow that when Smith and Jones marry, you now have three ontologically real entities: Smith, Jones, and the married couple. So conceptual presupposition should be distinguished from ontological commitment. This follows generally from Sellars’s distinction (noted previously) between the order of being and the order of knowing. How, then, should we understand Sellarsian we-­intentions? If the above line of reasoning is correct, an individual’s we-­attitude presupposes the corresponding group attitude. It is therefore a dependent individual attitude. Thus, we can understand Sellarsian we-­intentions as what Gilbert would call dependent individual intentions. They presuppose a group attitude, which itself presupposes either a set of mutual beliefs or mutual expectations. The most straightforward evidence that Sellars is amenable to this reading comes from CSS, where Sellars writes, The moral point of view involves the form “We shall any of us do A, if C,” which entails “I shall do A, if C.” To flag its origin (or ground) in a logically intersubjective intention, I represent the latter as “Shall(we) [I do A, if C]” (CSS, ¶14).9

92  What Are Sellarsian We-­Intentions? The implication here is that an individual we-­intention is grounded in, and presupposes, a full-­blown collective intention. To describe this relation in terms of group intentions and dependent individual intentions is perhaps to attribute to Sellars’s account a level of sophistication beyond what Sellars himself developed, but that is the point of supplementing Sellars’s own account with more recent developments in the literature. Further, as Sellars would say, “Shallwe [I do A in C]” is binding on me because I am one of us, and we have (at least in the moral case) the (categorically reasonable) intention to do A in C. Thus, in the ideal case of moral action­—say, when I tell the truth­—I of course (a) do so intentionally but also (b) in the expectation that others generally have this intention and (c) in the expectation that it is mutually expected in the group that the members have this intention. Again, these expectations are implicit in my actions in the same way that my expectations about the solidity of the ground are implicit in my walking on the ground (and not over the ornamental pond)­—I converse with others, I convey information in the expectation that they will engage in updating (or at least Brandomian scorekeeping of my commitments, which presupposes they expect me to tell the truth about my commitments), and I update my own “scorecard” based on their assertions. All of this conveys a set of mutual expectations about truth-­telling. These expectations are all defeasible, of course, but are implicit in most conversational contexts in normal circumstances. All of this brings home the point that for Sellars, individual moral intentions presuppose collective attitudes and intentions and should be understood as dependent individual intentions. To be sure, I can certainly have a moral intention where I have little or no expectation that others share the same intention. But remember, again, how Sellars introduces the relation between group intentions and dependent individual intentions. Suppose Dick has moved from an “ego-­ centered” model of practical reasoning to the following model: Shall be [each of us have a new car] So, Shall be [I, Dick, will have a new car]. In his reasoning, so to speak, Dick has come into the picture as “one of us.” We might, therefore, reformulate his intention as Shall be [I, Dick, because I am one of us, have a new car]. . . . It is important to realize that because is just as much a part of the content of the above intention as it is a part of the belief content expressed by The Pirates should win because they are the better team and has the same rationale. (ORAV, §§155–6/p. 97) I have the dependent individual intention because we have the collective intention and I am one of us. Now, if morality is going to serve its

What Are Sellarsian We-­Intentions?  93 goals­—a chief one of which is solving collective action and coordination problems­—in the normal case, such intentions do in fact have to be shared among members of the relevant community. Further, it must be common knowledge that such intentions are shared­—otherwise members of the community would be unable to coordinate their behavior. Again, this knowledge might be embodied in the form of expectations implicit in behavior, but such behavior cannot be coordinated among agents without such common knowledge/expectation. Therefore, in the standard case of moral action, I share an intention that I possess because it is a communal intention, and I expect others to share this intention, and it is common knowledge (or commonly expected) that others will share/act on this intention. All of the elements for action in the we-­mode in the strong sense are present. However, Sellars clearly allows for the possibility that one can act on a moral intention that no one else shares. And clearly, people do behave in ways they take to be moral even though they know that their view is not shared by the community. At the risk of embracing a rather jarring form of relativism, we must make sense of how such non-­shared intentions can (a) be we-­intentions­—for this is Sellars’s view­—and (b) be genuinely moral intentions, although they are not shared by the larger community. I turn now to this task.

4.3 Relativism? If a Sellarsian we-­intention is a dependent individual intention, how can one have a we-­intention not shared by others? Doesn’t the present account commit us to the worst kind of relativism? Sellars certainly didn’t intend for we-­intentions to be so closely tied to practice. He claims, “An individual can have an intention of intersubjective form even if no one else in point of fact shares it, an individual rational being could have an intention of the above form even though few, if any, other rational beings had such an intention” (SM 7.XX.§143/p. 225). Sellars argues that even on his account, we can make perfect sense of a sentence such as “We disapprove of women smoking, but I don’t” (IILO, p. 203). Olen and Turner call this the “test sentence,” in the sense that “any Sellarsian account of collective intentionality must make sense of it” (2015, p. 954).10 Sellars’s account need not lead to relativism. Let me begin with a parallel Tuomela makes with institutional norms. Even in the case of “strongly social concepts,” like that of being a leader (i.e., in an organization), there are “constraints for correct collective acceptance” (2002, p. 147). Thus, if Smith is elected leader of Organization, but it turns out that a quorum wasn’t present at the meeting, then the election was not valid, and he is not in fact the leader even if people take him to be so. Therefore,

94  What Are Sellarsian We-­Intentions? someone could challenge his status as leader, demand a re-­vote, etc. Thus, I may be the only person in Organization to hold the attitude B. I BELOrg [Smith is not the leader] and yet my group attitude is correct­—though not shared by the organization as a whole. Notice a few things about this example. My possession of belief B, which I hold qua member of Organization, presupposes massive agreement with other members of Organization­—agreement about the structure of Organization (e.g., that there is a leader, about the relation of the leader to ordinary members, etc.), about the bylaws (voting procedures, procedures for conducting meetings, rules of quorum, rules for moving and seconding motions, etc.), and so on. Thus, I am only able to dissent qua member of Organization because we already share so many attitudes characterizable as “ATTOrg.” This is not merely true in institutional cases; this is true wherever we dispute over norms. A dispute over a semantic norm (e.g., the proper meaning of “beg the question”) presupposes massive agreement, without which we couldn’t even be characterized as having a semantic dispute. Nor could we have a specifically moral argument without a background of massive agreement about the subject matter of morality. Any discourse­—including dissent­—presupposes the kind of shared group attitudes that are at the heart of Tuomela and Gilbert’s accounts. And when Sellars argues that Jones can have a moral we-­intention that no one else possesses, one should understand this in the context of an ongoing social practice with massive agreement in the background. Essentially, Jones is making (in the moral sphere) the same move I made in Organization: Jones is arguing that given this background agreement (say, agreement on the formal end of morality, certain abstract principles of justice or fairness, and so on), we are in fact committed to doing Ai in Ci. Such a move can only be understood in the context of significant agreement over what morality requires, what can be appealed to as reasons, and so on. In general, the game of giving and asking for reasons presupposes general agreement in attitudes as a background for arguing over the remaining areas of disagreement.11 One might argue that the Organization case demonstrates that relativism does in fact follow from the group-­attitudes approach. After all, if Smith is elected leader of Organization (even though proper procedures aren’t followed) and no one objects (or no one listens to my objections), and all the (other) members of Organization treat Smith as the leader, then isn’t Smith in fact leader of Organization­—at least after a suitable amount of time? I offer two replies to this objection. First, one can in principle always recognize the illegitimacy of Smith’s election. One can undo the decisions

What Are Sellarsian We-­Intentions?  95 he has made, revoke his appointments, and so on. One might decide not to, on the grounds that such a decision would be costly­—but the point is that such a move is always available in principle. Second, while we can play rule-­governed “games” in a relativistic way, the decision about whether or not to do so is itself governed by reasons. As I argue at length elsewhere (Koons 2003), we can play games in a reified or non-­reified way. Thus, we can “reify” the rules of a game and declare that in this game such rules are “relativistically” true, in the sense that future rule changes are not retroactively applied but apply only to future “play.” It makes sense to reify certain sets of rules. For example, sports leagues often change their rules­—for example, basketball with the introduction of the “Wilt Chamberlain” rules. But these rules are not applied retroactively to games played in prior seasons. This would be unwieldy and, above all, unfair. It makes sense to “reify” the rules in games like this and therefore play these games in a relativistic way. But notice that the decision to reify is a decision based on reasons­— and the weight of reasons could press in the opposite direction. We can also treat rules as non-­reified and apply rule changes retroactively­—that is, non-­relativistically. Should we reify the rules of morality and play the moral “game” in a relativistic way? There are good reasons to say no. For starters, the majority often engage in activities that do not promote the formal end of morality, and individuals shouldn’t be compelled to act contrary to the welfare of rational beings merely because such actions are endorsed by others. Individuals ought to be permitted to appeal to deeper principles of morality. In one sense, this means acting on intentions not shared by the larger group­—but in another sense, it means acting on a more fundamental shared intention, such as the intention to promote the welfare of all rational beings or the intention to treat all alike. So again, even dissent presupposes a broader shared agreement; and this is what prevents the present account from sliding into relativism and what allows a principled opposition to reification of the moral “game.” (It is worth noting, though, that even in a non-­reified game, we wouldn’t be bound to apply every rule change retroactively; a decision about whether to do so in a particular case is itself reason-­governed.) There is one final objection to be addressed. On the present account, acting on a group intention requires the existence of a shared intention, which requires the element of shared belief (or mutual expectation, as outlined above). Suppose I endorse and act on the following intention, which I take to be a moral one: Shallwe (I do not eat animals) Suppose, further, this intention is not shared by the larger group. One might argue that because this intention is not shared by the larger group, it cannot count as a we-­intention. For even if we grant the fact

96  What Are Sellarsian We-­Intentions? of underlying agreement about various moral norms, there is no element of mutual belief or mutual expectation regarding this specific intention since I don’t believe that others endorse this intention and act for this reason even if others believe that I do. How, then, is this a we-­intention­—or even a moral intention­—on the Sellarsian view? I think we must recognize that when a person acts on a moral intention that others do not share, this represents a “defective” or “degenerate” case of acting in the we-­mode.12 The best model of this is what Tuomela calls the “weak notion of acting as a group member, also to be called I-­mode acting as a group member” (Tuomela 2007, p. 29). Tuomela defines weak action in a group mode as follows: (AGMW) An action X performed by member A of group g is an action performed as a group member in a weak sense if and only if A performed X in the group context in the core sense (X belonging to one of the action classes (1)-­(4) of section 3)13 in part for the purpose of promoting the ethos of g that she privately accepts and to which she is privately committed (and thus privately socially ought to promote). (2007, p. 29) Based on this, Tuomela gives a more general definition of what it is to have an attitude in the weak group sense­—what he calls the “progroup I-­mode”: (IM2) Agent A, a member of group g, has ATT(p) in the progroup I-­mode relative to g in situation C if and only if A has ATT(p) in part because she is functioning as a member of g in sense (AGMW) and is privately committed, in the ATT-­way, to content p at least in part for g in C. (Tuomela 2007, p. 53) As indicated above, I think a good case can be made that when I act on a moral intention that is not shared by others, this action should be understood as acting in the group mode in the weak sense (rather than the strong sense). To illustrate, consider the example where I am the only person in Organization to hold the attitude B. I BELOrg [Smith is not the leader] and yet my group attitude is correct­—though not shared by Organization as a whole. What makes this a we-­attitude? It is a we-­attitude for the following reason: I hold this belief because I believe it is supported by the laws and regulations that govern (and indeed partially constitute) Organization. Thus, there is this presuppositional relation between the “ethos” of

What Are Sellarsian We-­Intentions?  97 Organization and my particular belief, and it is in virtue of this relation that my belief counts as BELOrg. However, because my belief is not shared by others, it may not count as a strong we-­mode attitude. My intuition is that this belief is somewhat stronger than Tuomela’s conception of a progroup I-­mode attitude since my belief is based on a set of strong we-­mode attitudes­—namely, the bylaws of Organization, which we all mutually endorse (and there is an element of mutual belief or expectation of mutual endorsement, or at least compliance). But I will not insist on this. The chief point is that my belief at least counts as holding an attitude in the we-­mode in the weak sense. I hold this attitude qua member of Organization even if it is not a full-­fledged we-­mode attitude in the strong sense. A similar case can be made for holding a moral belief that is shared by few other members of your group. Durkheim writes, “A rebellion against the traditional morality you conceive of as a revolt of the individual against the collective . . . However, what I am opposing to the collective is the collective itself, but more and better aware of itself” (Durkheim 1965, p. 66). We may think of the issue this way: A crucial plank in the pragmatist, inferentialist position staked out by Sellars, Brandom, and others is that what makes an individual a language user is the ability to move to the metalanguage­—to offer reasons to justify the claims she is making. This doesn’t require that one be able to justify inferentially every claim she makes or every belief she holds; it only requires that one is able to engage in the game of giving and asking for reasons and is able to offer justifications for some suitable subset of these claims and beliefs. (I will have more to say about this in Chapter 9.) Now, it is a platitude about moral claims that they have a tight connection with reasons, and so one would expect, in the normal course of things, that one would be able to offer some kind of inferential justification for one’s (non-­basic)14 moral commitments. Thus, if asked to justify my moral commitment Shallwe (I do not eat animals) I might say, “Animals have interests” or “Animals feel pain” or even just “Animals have rights.” As noted above, even disagreement presupposes massive agreement. And in engaging in the game of giving and asking for reasons, what is appealed to is (ideally) a shared reason­—something that genuinely expresses a strong we-­mode attitude (such as “Shallwe [I do not cause gratuitous pain]”). Granted, we may have to dig down a bit before we find this level of shared attitudes, but whether I know it or not, when I act on a commitment like the one above­—one that is not shared by the rest of the group­—I am in fact promoting what I take to be the group ethos, in that I am promoting what I take it other members of the group ought to be committed to by the reasons underpinning my commitment, reasons that we share. Thus, as in the case of Organization, I am acting in the we-­mode, if only weakly.

98  What Are Sellarsian We-­Intentions? One might worry that I am departing too far from the spirit of the Sellarsian project. Sellars himself did not distinguish between different kinds of we-­intentions and seemed to think that one could have a proper, full-­fledged we-­intention that no one else shared. However, I think it is a virtue of the present account that it distinguishes between different kinds of we-­intentions. A primary function of we-­intentions is social coordination, and this is something that can only really happen when the group genuinely shares attitudes and commitments, enabling people to act in the we-­mode in the strong sense. When, for example, a person holds a commitment qua group member­—a commitment that is not shared by the group­—we should recognize that this is, in an important sense, a “deviant” case of we-­intending. Something has gone wrong with the group. Either the group is simply wrong about what it is committed to by its ethos­—and so fails to have the appropriate group attitude­—or the person in question is wrong and has, for example, failed to be properly socialized. Either way, we should understand this weaker case as a departure from the full and proper we-­attitude case where there is a shared group attitude and the members of the group have dependent individual attitudes that are conceptually dependent on this shared group attitude.15 But what if we don’t share any moral attitudes with the group in question? Isn’t this possible with, for example, Nazis or members of ISIS, or whatever? Doesn’t this mean that there are cases where there isn’t such massive agreement uniting us with the relevant community and generating the appropriate shared content? First, we should be hesitant to claim that we share few or no moral attitudes with a given group. I suspect that in many cases there is substantial overlap of moral principle but substantial disagreement with respect to whom we have moral obligations in the first place. Thus, the content of morality may overlap substantially; where there is disagreement is over the scope of the “we.” For Sellars, of course, the “we” is all rational beings. We will see in Chapter 13 whether a compelling argument can be made for so enlarging the scope of the “we.” Second, suppose for the sake of argument that there is little (if any) overlap between our moral worldview and that of another group. In that case, this seems like a reason to not necessarily exclude them from the “we” but merely to treat them in some specific respect as being in the space of causes, not reasons. After all, you don’t reason with Nazis; you fight them. Or, to consider a more empirically based example, consider the Mendota Juvenile Treatment Center (MJTC) in Madison, Wisconsin, which is a pioneer in the treatment of teenage male psychopaths­—a group previously considered to be more or less incorrigible. Inmates are treated well regardless of how badly they behave. As Caldwell and Van Rybroek note, “The treatment model used on MJTC is based on the notion that defiant behavior can become cyclic when the defiant response to a sanction is itself sanctioned, resulting in more defiance and

What Are Sellarsian We-­Intentions?  99 increasing sanctions” (2005, p. 626). The key is to break this cycle of defiant behavior. The focus at MJTC is not on punishing bad behavior but on rewarding good behavior. At the end of each shift, the staff dealing with each inmate assign the inmate points based on his behavior that day. As Caldwell et al. write, “The youth’s ratings on any given day determine his level of privileges the following day (the program is titled the Today-­ Tomorrow Program). A youth can obtain access to additional sets of privileges (e.g., computer game time, snacks, private music) by combining several consecutive days of positive ratings. In this way, the program offers readily available and rapidly increasing incentives for compliance” (2007, p. 577). A two-­year follow-­up demonstrated that inmates released from MJTC had dramatically lower recidivism rates across all types of offenses studied compared with the non-­MJTC comparison group.16 As with much psychiatric treatment, the “decompression treatment” used at MJTC walks a fine line between treating patients as being in the space of reasons and treating them as being in the space of causes. Inmates are treated with the respect due rational agents, but they are also subjected to a positive-­reinforcement regimen that incentivizes them to create stronger social ties, which themselves decrease the antisocial behavior observed in these youths. But perhaps there is a sense for Sellars’s Kantian project in which acting on a we-­intention really doesn’t require that we share moral commitments with others­—only rational agency. For there is perhaps a more basic sense in which a moral commitment­—even if not shared by others­— expresses a we-­intention (even if only in the weak sense). The Kantian notion of categorical reasonableness is of acting under autonomy­—it is of acting under a requirement that is internal to reason. In other words, the categorical imperative is a requirement of reason itself. As I will argue in Chapter 13, the moral point of view expresses, for Sellars, precisely such a requirement of reason. It is not rationally optional but is in fact necessitated by our nature as finite rational creatures. Thus, when I act on a moral requirement, I am acting on a requirement that I must take others to be committed to as well (even if they don’t recognize or act on this commitment). Thus, I am very much like the member of Organization: I take others to be committed­—by virtue of their membership in Organization and their corresponding commitment to its constitutive rules­—to recognizing Smith as president (but not Jones, whose election was illegitimate) even if they refuse to recognize this commitment. Similarly, if I judge that I should do Ai in Ci, I take others to be committed to do this as well­—by virtue of the fact that as finite rational agents, they are committed to the moral point of view, which dictates this action in these circumstances. Recall, for Sellars, how reasoning from the Kantian moral point of view works. As we noted earlier, “Shallwe [I do A]” amounts, essentially, to “Shall [I, because I am one of us, do A]”; and that the “because is just as much a part of the content of the above intention as it is a part of the

100  What Are Sellarsian We-­Intentions? belief content expressed by, ‘The Pirates should win because they are the better team,’ and has the same rationale” (ORAV §156/p. 97). Equally true, I think, it is the case that Shall [I do Ai in Ci] because Shall [Any of us does Ai in Ci] Thus, I ought to do Ai in Ci because any of us ought to do Ai in Ci (and I am one of us). Thus, even if I am acting on a principle no one else recognizes, from the Kantian perspective I am still acting on a group reason for the simple reason that I am doing Ai in Ci because it is what any of us ought to do in Ci. This is true even in radically defective moral circumstances: If I embrace a moral intention such as Shallwe [I assist escaped slaves on the Underground Railroad] or Shallwe [I hide this Jewish family from the Nazis] I am committed, by my actions, to anyone doing the same in my specific circumstances even if I recognize that few (if any) would in fact act similarly. (I am at least licensing a permission to do what I do.) This follows from the Kantian principle that my permission (or obligation) to do Ai in Ci follows simply from its being an instance of anyone’s doing (or being able to do) Ai in Ci. Ergo, although one’s intention is not shared by other members of the group, one is nevertheless doing Ai in Ci because one is a member of a group (namely, the group of rational beings)­—and indeed qua member of this group­—and so the action satisfies Tuomela’s definition of acting in the group mode in the weak sense. Again, though, such cases should be understood as acting in the we-­ mode in the weak sense. For although I am acting qua member of the group and am acting because I am a member of the group, the element of mutual belief/mutual expectation­—required for we-­mode action in the strong sense­—is missing. But again, this is in my view appropriate, as we should understand such we-­intentions (which are not shared by others) as, in a sense, a “defective” or “deviant” case of we-­intentions. One final point: One might worry that Sellars’s account threatens at a deeper level to collapse into relativism. Suppose everyone agrees that it is morally right to do Ai in Ci. On what basis can they be wrong? Again, I think the same answer applies as above: Relativism always assumes a completely consistent ethos. But a feature of moral argumentation is that

What Are Sellarsian We-­Intentions?  101 it generally involves appealing to one commitment to overturn another commitment. Thus, you might appeal to a person’s deeper commitment to equality in an attempt to convince him that a particular practice of his is unfairly discriminatory and therefore wrong. So it is with societies: Embracing a social practice account like Sellars’s doesn’t mean that we can’t appeal to some deeper commitment of a society’s to argue that a particular society’s practice is wrong. In any case, Sellars’s social practice account is no more vulnerable to charges of relativism than is any other social practice account of morality. And I don’t think that social practice accounts are themselves particularly vulnerable.17 Even if a culture different from ours has different practices, it does not follow that we cannot criticize them (or they us). Of course, we will judge by our lights that something they are doing is wrong­— but one of the chief points of a social practice account is to call off the search for a “view from nowhere,” a God’s-­eye viewpoint from which to adjudicate such disputes. All we have is our various socially embedded commitments and practices, and it is from this standpoint that we judge. A final word of caution, though. It might sound from the above as if social practice accounts commit us to chauvinism. If another culture is very different from ours, then of course by our lights they must be doing things wrong! Such a claim amounts, perhaps covertly, to an assumption of radical incommensurability and is therefore not plausible. To assume that in comparing our culture to another we could never find anything in our culture inferior amounts to the supposition that our culture and theirs have so little in common that nothing they can do or say could strike us as reason-­giving. Again, this strikes me as empirically implausible and perhaps conceptually impossible. Of course in observing another culture, even one quite different from ours, we might judge some practices as inferior. But we might judge some as superior and might consider their reasons for adopting these practices as offering us reasons for adopting similar practices. Social practice accounts of normativity need not entail relativism or chauvinism. They only entail the (obvious, I would hope) conclusion that judgment takes place using a conceptual system that is socially bound. It does not follow that we cannot recognize elements of rival systems as superior to our own. I will have occasion to revisit concerns about relativism a few times throughout the book. For an account of the present type, relativism­—like naturalism­—is a persistent worry, and I will address these worries as they arise in conjunction with elements of the ongoing presentation.

Notes   1 Olen and Turner argue that Sellars was leery of extending the notion of collective attitudes beyond mere collective intentions. They write, “According to Sellars himself, to argue . . . [that there are] collective desires, beliefs, and

102  What Are Sellarsian We-­Intentions? justifications specific to the collective . . . would amount to embracing an idea of a group mind inconsistent with empiricism” (2015, p. 973). I don’t see this hesitation in Sellars. Although he explicitly abjures the existence of a “group mind” (IILO 12/p. 203), he seems to regard as legitimate sentences attributing mental states to groups (e.g., “Russia fears China” [CSS, ¶2]). As we will see (particularly in Chapter 9), Sellars may need to expand his notion of collective attitude to include collective beliefs and so forth; and in any case, the tools that enable us to give a nominalist account of collective intentions can be extended to give a similar account of other kinds of collective attitudes.   2 For Gilbert, the obligation to act on a joint commitment isn’t a moral obligation per se. This seems appropriate since, for Sellars, most collective intentions are not moral intentions. But it seems plausible to regard our joint commitment to the formal end of morality (our commitment to promoting the welfare of rational beings generally) as giving rise to specifically moral obligations. And of course, for Sellars, it is only certain we-­intentions­— distinguished both by their content and their universal scope­—that qualify as moral intentions.   3 In a similar vein, Tuomela writes, “The collective aspect [of a group’s commitment to E] makes solo quitting reproachable in principle” (2007, p. 33).   4 Because Gilbert and Tuomela’s accounts rely on the claim that group action must stem from joint commitment and group ethos, I think they are not vulnerable to Searle’s “Harvard Business School” counterexample (Searle 2009, pp. 47–8). His counterexample­—where all the graduates set out individually “to try to benefit humanity by being as selfish as [possible] . . . in the mutual knowledge that others are [doing the same]” (Searle 2009, pp. 47–8)­—seems clearly to reflect Tuomela’s action in the we-­mode only in the weak sense. —where the business school graduates Searle’s second case (2009, p. 48)­ “make a solemn pact that they will each go out and try to help humanity by becoming as rich as they can and by acting as selfishly as they can” (2009, p. 48)­—does seem to be a case where the graduates are motivated by an ethos/joint commitment, and so this does seem to be a case of action in the we-­mode in the strong sense. Searle concludes that “the second case is a case of collective intentionality” (2009, p. 48).   5 Searle (2009, chapter 3) also focuses on collective recognition as the central element in the assignment of functions to objects and hence to the constitution of things like money.   6 Metaphor and other uses of language may seem to need a different account. But even conveying information via metaphor will require the use of commonly understood expressions and hence a range of group attitudes. Further, it will of course be the case that virtually no one shares all commitments regarding the use of any word (such as “bird”). As is often pointed out, communication would be all but pointless if we shared all commitments. This is where communication, Brandomian scorekeeping, and so on enter into the picture. But if you and I didn’t have some kind of shared, socially inculcated idea of what “bird,” “sentence,” “word,” and so on meant, then the project of communicating and scorekeeping couldn’t even get off the ground in the first place.   7 The picture of meaning and linguistic norms here runs counter to the one offered by, for example, Davidson in “A Nice Derangement of Epitaphs.” I don’t, of course, have the space to refute every competitor; all I can do is try to motivate the Sellarsian picture.   8 One final comment: I have been writing about linguistic and other sorts of norm-­governed behavior as instances of group action. This is something of

What Are Sellarsian We-­Intentions?  103 a misnomer for Sellars. As we will see in Chapter 9, our basic linguistic and other norm-­governed behavior does not, for Sellars, consist of actions. It is pattern-­governed behavior that is not rule-­following and consists of acts, not actions. Nevertheless, this behavior has the complex relation to mutual expectation outlined above. Therefore, as we will see in Chapter 9, we must understand this norm-­governed behavior as pattern-­governed behavior in the we-­mode in the strong sense.   9 See also ORAV (§179/p. 99). 10 Sellars makes other anti-­relativist comments, such as the following: “From the fact that Smith values something, X, as one of us, it doesn’t follow that We value X” (SM 7.XVIII.§127/p. 220). 11 The Organization case can be seen as an instance of agreeing on the rules but disagreeing on what follows from them. The moral case is more often a case of disagreeing about what the rules are in the first place. But again, such disagreement occurs against a larger background of agreement over other rules, values, and commitments, against which such disagreements may not be in principle unresolvable. 12 This is a conceptual claim, not an empirical one. The basic case of we-­mode action­—particularly moral action, which generally aims at coordination­—is when the intention is broadly shared. Even if non-­shared we-­intentions are empirically common, this must be seen as a “defective” social practice that is failing to achieve social coordination via we-­intentions. The claim is not that social practices should seek maximal conformity by all people on a broad range of behaviors. The claim is merely that on issues where coordination is required (particularly on issues of moral importance, such as protecting important moral interests), wider compliance is, all things considered, better than narrower compliance. 13 These “action classes” include actions one ought to perform in virtue of one’s occupying a particular group role, actions that uphold or are constitutive of a group norm or standard, and so on. See Tuomela (2007, p. 23). 14 This terminology is potentially misleading. Basic moral commitments will be commitments along the lines of “Causing needless pain is wrong,” “Killing innocent people is wrong,” etc. These are, in other words, claims that most people would find difficult to justify inferentially and that in most discursive contexts would successfully terminate a justificatory or argumentative chain. But of course, for Sellars, the fact that these claims can play such a discursive role does not mean that they are foundational or that their justification does not ultimately depend in some way or other on other justified claims. We will come back to the issue of foundationalism in moral theory in Chapter 12. 15 One reason I have relied little on Searle’s influential account is that I am unclear how he can account for Sellars’s “test case” since Searle seems to be offering in the first instance an account of collective intentionality that requires consensus: “Part of what it means to say that the intentionality is collective is that each agent has to assume that the other members of the collective are doing their parts” (Searle 2009, p. 52). 16 Caldwell and Van Rybroek (2005, p. 629). 17 See, for example, Koons (2003) and Wolf (2012).

5

Practical Reasoning and the Logic of Intentions

5.1 Expressivism and the Embedding Problem Sellars uses “shall” to mark the expression of an intention. So, “Shallwe [I do Ai in Ci]” expresses my we-­intention to act in a particular way when in a particular type of circumstance. Thus, Sellars’s metaethics might loosely be classified as a type of expressivism, whereby moral judgments express an attitude or other kind of state rather than involving the assertion of some factual content. The problems created by the unorthodox semantics of expressivist theories are well known. Sellars was not unaware of these problems and indeed seemed to develop much of his theory of intentional inference specifically with the problems of Hare’s emotivism in mind. It is a testament to the depth of Sellars’s insight that his solution evades some of the problems that continue to dog modern-­day expressivists. The chief problem faced by expressivist theories is the embedding problem, which has a number of manifestations. The embedding problem (also known as the Frege-­Geach problem) was originally presented by Geach in the early 1960s as a problem for various performative (as opposed to descriptive) accounts of assertion. Non-­ descriptivist accounts of morality have trouble accounting for the soundness of moral arguments since moral statements (when embedded within conditionals, for example) are no longer asserted. Geach asks us to consider arguments like the following, which on their face seem both valid and sound: (1) Lying is wrong. (2) If lying is wrong, then getting little brother to lie is also wrong. Hence, (3) Getting little brother to lie is wrong.1 The problem is that for this argument to be valid, “Lying is wrong” must mean the same thing in all occurrences; otherwise, the argument commits the fallacy of equivocation. However (continues Geach’s line of reasoning), “Lying is wrong” is asserted in the first premise but not in the second, where it is merely the antecedent of a conditional. Thus,

Practical Reasoning and the Logic of Intentions  105 whatever emotive or expressive meaning is present in the first occurrence of the phrase is absent in the second. Others have wondered if there can even be inferential relations between expressive statements. Consider Blackburn’s well-­developed theory of expressive inference. Blackburn introduces “expressive operators” H! (Hooray!) and B! (Boo!), which operate on descriptions and express approval or disapproval. Blackburn handles inferential relations among expressive statements by compounding them within an expressive operator using the compounding device “A;B” to denote the compounded (and hence inferentially related) statements. Thus, if I wanted to endorse the inference from “Lying is wrong” to “Getting little brother to lie is wrong,” I would express this as follows: H!([B!(lying)]; [B!(getting little brother to lie)])2 However, as Bob Hale writes, this merely presupposes the goodness of the inference from [B!(lying)] to [B!(getting little brother to lie)] without explaining how such an inference is possible (or sound) given the expressive nature of the statements involved. Blackburn’s theory involves reading “A;B” as “making B follow upon A”, and it seems to me that the projectivist is not entitled to this gloss, at least not without further explanation. The obvious one is that making B follow upon A is ensuring that if you have the attitude described by A, you also have that described by B. But this is surely unavailable; the problem was, inter alia, to explain how we come to speak conditionally of our commitments­—the theory loses all interest if it merely sneaks in conditionality in unfamiliar guise. (Hale 1986, pp. 73–4)3 Others have pointed out that expressivism has trouble handling even simple logical operations like negation. (See Schroeder 2008b, pp. 578– 80; Unwin 1999, 2001.) For example, one can construct a fairly convincing account of disagreement when it comes to belief: Smith believes p and Jones believes ~p. Both individuals have the same attitude toward logically contradictory contents. But what about moral negation? Suppose Smith believes that drug legalization would be morally good, and Jones believes it would be morally bad. Thus, Smith’s attitude is H!(d), whereas Jones’s is B!(d). Smith and Jones hold different attitudes toward the same content; no logical inconsistency has been created. One could, of course, create a logic of attitudes with axioms like “(x)H!(x) → ~B!(x).” This would generate a contradiction, but only in attitudes, not in contents. And besides, one suspects that this would invite the same kind of criticism raised by Hale and others­—that the expressivist is helping herself to a notion of expressive inference without explaining how such inference can be valid, which is really the task at hand.

106  Practical Reasoning and the Logic of Intentions Sinnott-­Armstrong holds that the problem for expressivism may be in-­principle and insoluble. Even using the panoply of tools at their disposal (such as a minimal theory of truth), deep problems remain for the theory. Sinnott-­Armstrong (2000) argues that even Gibbard (1990), who offers a sophisticated possible-­world semantics for his expressivist theory, cannot provide sufficient content to make sense of expressive inference. Roughly (and somewhat oversimplified for the present context), the meaning of a normative utterance for Gibbard is the set of possible worlds in which the utterance is true or is entailed by the facts and norms that are true in that world. Sinnott-­Armstrong offers Dreier’s (1996) “hiyo” argument as a challenge for the expressivist: Imagine that an “accostivist” tells us that the phrase “Hiyo, Bob” is used to greet Bob. He tells us that “Bob is hiyo” is also used to greet Bob. Appealing to a minimal theory of truth, our accostivist claims that “hiyo” can even be used in valid arguments: Bob is hiyo If Bob is hiyo, then so is his little brother Therefore, his little brother is hiyo As Sinnott-­Armstrong points out, if we generalized “hiyo” as a greeting for everyone and created a possible-­world semantics for “hiyo” (in terms of a set of “hiyormative factual worlds”), this wouldn’t help us understand the content of “hiyo”­—or why the above argument is valid­— any better than we did before. And this (he claims) shows that such a semantics doesn’t in itself show that expressive inference is any good: The content of the conditional “If Bob is hiyo, so is his little brother” rules out the intersection of the set of hiyormative-­factual worlds where Bob is hiyo and the complement of the set of hiyormative-­ factual worlds where Bob’s little brother is hiyo. The validity of the argument “Bob is hiyo. If Bob is hiyo, so is his little brother. So his little brother is hiyo.” can then be explained by showing that every hiyormative-­factual world that is ruled out by the conclusion is already ruled out by the premises. Nonetheless, this semantic story does not help us to understand “Bob is hiyo” any better than we did before going through all of this rigmarole. This lack of understanding shows that such formal semantics is not enough to explain what “Bob is hiyo” means. Consequently, analogous moves by expressivists are also not enough. (Sinnott-­Armstrong 2000, p. 690) The key challenge for the expressivist, then, is to develop a logic for expressive terms that is capable of meeting the above, very serious challenges.

Practical Reasoning and the Logic of Intentions  107

5.2 Sellars on the Logic of Intentions4 A reader of Sellars might wonder why Sellars devotes so much space to discussing the logic of intentions. (Sellars’s longest piece on the subject, IILO, weighs in at a substantial 60 pages.) One must understand, though, that debates about how to give an account of inference for non-­ descriptivist theories of ethics were front and center when Sellars was developing his own theory. One of the most influential of such accounts­— and an account frequently cited by Sellars­—is Hare’s own prescriptivist account. On Hare’s account, moral judgments primarily have prescriptive meaning and secondarily have descriptive meaning. Hare interprets this prescriptive meaning in terms of imperatives. Keen that his own ethical theory should capture intuitive features of morality, Hare developed a detailed imperative logic that (he hoped) would mirror deontic logic as it was understood at the time. His efforts were met with sustained criticism. For example, Geach (1958) criticizes several of the fundamental principles of Hare’s inferential system, concluding that “in any case deontic logic is not related as he thinks it is to imperative logic” (p. 56). Bernard Williams (1963), while admitting that “two imperatives may be said to be inconsistent,” argues that “there is not in general anything that can be called imperative inference. . . . I deny [that we can] apply the notion of inference to imperatives” (p. 30). In a reply to Williams, Geach casts doubt on whether imperatives can even be inconsistent in a way that matters to logicians. And, of course, the Frege-­Geach problem was developed in the early 60s in a series of papers by Geach. So Sellars is quite clearly aware of the demand facing his non-­ descriptivist theory of ethics. Thus, he realizes that it is incumbent upon him to develop a logic of intentions that will meet the various objections that have been raised against other, similar accounts. What we need to do now is examine the basic structure of his account and demonstrate how it avoids the objections (both contemporary and historical) against expressivism. To motivate Sellars’s account, let us start out with an analogy, one that Sellars himself mentions but does not rely on heavily. (See Sellars, VR; also, SM 7.III.) We have little trouble understanding how propositions can logically relate to each other: p and ~p are logically contradictory and, ergo, cannot both be true at the same time. If p and p → q are both true, then q must also be true, and so on. But what about our attitudes toward these propositions? How can these attitudes logically relate to each other? In particular, how can our beliefs about these propositions logically relate to each other? We know that p and ~p are logically contradictory; what account do we give of the logical relations between a belief that-­p and a belief that not-­p? Most philosophers do not find this question at all difficult to answer, I don’t think. We explain the logical relations among beliefs by adverting to

108  Practical Reasoning and the Logic of Intentions their propositional contents. Thus, a belief that-­p contradicts a belief that not-­p precisely because p contradicts not-­p. Similarly, inferential relations among beliefs are parasitic upon inferential relations among their contents. As Sellars writes, “Philosophers analyze the logical relations of belief-­contents by determining the logical relations of the factual statements which express them” (SM 7.III.§17/p. 183). Ceteris paribus, someone who believes that-­p and believes that p entails q should also believe that-­q. I say “ceteris paribus” because the literature on skepticism has produced many challenges to the closure principle. Also, people do not always believe as they rationally ought: As Sellars notes, “If ‘P’ implies ‘Q,’ then it is unreasonable to believe that P is the case without believing that Q is the case. (Though, as noted above, in point of fact one may well believe the former without believing the latter.)” (SM 7.III.§18/p. 183). But as I have argued above, let’s not take our cue as to how to deal with ordinary cases from the extraordinary ones: The latter example probably provides a good model of how to reason well in everyday cases, although it doesn’t invariably describe how people actually reason. In precisely parallel fashion, Sellars proposes to treat inferential relations among intentions as parasitic upon inferential relations among the contents of intentions. Thus, Sellars writes, “If ‘P’ implies ‘Q,’ then ‘It shall be the case that-­P’ implies ‘It shall be the case that-­Q’” (ORAV §47/p. 86). To be charitable to Sellars, one mustn’t assume that if one has the intention that P, one will as a matter of fact have the intention that Q. Rather, this must be construed as a rational principle, à la Kant’s principle that one who wills the end also wills the means. As, for example, Christine Korsgaard (1997) has so eloquently argued, this is not a causal principle but must instead be understood as a normative principle, a principle of what is required by practical reason. So, too, should we understand Sellars’s notion of practical inference. And, indeed, this is exactly the view Sellars defends: “If ‘It shall be the case that P’ implies ‘It shall be the case that Q’ it is unreasonable to intend that P be the case without intending that Q be the case. (Though, again, in point of fact one may very well intend the former without intending the latter and may even intend that the latter not be the case.)” (SM 7.III. §18/p. 183).5 Again, though, we saw the same fact with beliefs: Inferential relations among belief-­contents (i.e., propositions) provide a guide for rational belief-­formation and revision but not necessarily a guide to how people will actually form and revise their beliefs. To further simplify and regiment the logic of intentions, Sellars proposes to use “shall” as an operator, which operates on the content of intentions. Thus, inference among intentions is simplified: Shall-­P P entails Q Therefore, Shall-­Q

Practical Reasoning and the Logic of Intentions  109 Thus, we can hopefully see on the horizon the first part of Sellars’s answer to the challenge to expressivism: Moral expressions can be involved in inference and argument because inferential relations among moral judgments (we-­intentions) are derived from the inferential relations among their contents. Thus, let us again consider the argument rehearsed above, which was supposed to present so many difficulties for the expressivist: (1) Lying is wrong. (2) If lying is wrong, then getting little brother to lie is also wrong. Hence, (3) Getting little brother to lie is wrong. How might we represent this argument in terms of Sellarsian we-­ intentions? Let us begin with premise (1), “Lying is wrong,” which we might represent as Shallwe ~(Anyone lies) Recalling that inferential relations between intentions are parasitic on inferential relations among their contents, we can quite easily see that the contents of the antecedent and consequent in premise (2)­—and the logical relation connecting them­—are as follows: ~(Anyone lies) → ~(Getting little brother to lie) Straightforwardly, “~(Anyone lies)” and “Getting little brother to lie” (getting any of us to lie, for that matter) are inconsistent; the truth of one entails the negation of the other. Recalling Sellars’s principle (which he calls “S-­Imp”; ORAV, §47/pp. 85–6) that if P entails Q, then Shall-­P entails Shall-­Q, we can see that all of the tools are now in place to reconstruct the above argument: (4) Shallwe ~(Anyone lies). (5) ~(Anyone lies) → ~(Getting little brother to lie). (6) Shallwe ~(Anyone lies) implies Shallwe ~(Getting little brother to lie) [5, S-­Imp]. (7) Shallwe ~(Getting little brother to lie) [4, 6 MP]. Thus, using a parallel strategy as that employed with logical relations among beliefs­—namely, that logical relations among intentions are parasitic on logical relations among the contents of intentions­—Sellars is able to reconstruct, simply and elegantly, an argument that, we are told, resists plausible construal by the expressivist. A key element of Sellars’s picture is that logical operators appear only within the scope of the “shall” operator.6 Thus, “Shall (not-­p)” is

110  Practical Reasoning and the Logic of Intentions well-­formed, but “not-­Shall (p)” is ill-­formed. Nor do you have “Shall be (p) or Shall be (q),” but instead “Shall be (p or q).” Sellars himself is aware of the appearance that he is contradicting himself, writing, It might be claimed that I am inconsistent in denying sense to “~Shall [P]” and “Shall [P] • Shall [Q]” while admitting “Shall [P] implies Shall [Q].” But the latter, involving the relation word “implies,” which is not a truth-­functional connective, is in the metalanguage, and mentions rather than uses the shall-­sentence. (TA, p. 115, n. 4) Again, the analogy with belief is both apt and illuminating. When reasoning among beliefs, one must keep logical operators such as negation inside the scope of the belief operator. Thus, if p implies q, “Bp implies Bq” is a good rational principle of belief extension, but “~Bq implies ~Bp” is not. For this would commit us to something even stronger than the principle of closure: This would commit us to the principle that if you don’t have any belief, positive or negative, about some logical consequence of p, then you shouldn’t believe p. Thus, allowing negation outside of the scope of the belief operator means ultimately that one must believe all true propositions before one can believe any. This is obviously too stringent a requirement on belief. If I believe that q is false, then this logically commits me to other beliefs, but nothing need follow if I simply have no belief about the truth or falsity of q. Thus, when reasoning about propositional attitudes like belief and intention, logical operators must stay within the scope of the attitude operator. (We will have reason to revisit this requirement later, but for now, I will stick to a straight exposition of Sellars’s theory.) Sellars’s insistence that logical operators must remain inside the “shall” operator is principled. Failure to see this results, I think, from a confusion of ascriptions of intention and expressions of intention. Ascriptions of intention have both an internal and an external negation: I intend to go to the library today I intend not to go to the library today It is not the case that I intend to go to the library today But as Sellars says, “An expression of intention as an expression of intention can only be negative by virtue of having a negative content. ‘Not-­shall [I do A]’ is false in the radical sense of conceptually incoherent” (ORAV §57/p. 87). Sellars’s view seems to be while you can say of someone that they don’t have an intention to A, an expression of intention must either be to A or not-­A; you cannot express a non-­expression of intention (or whatever; it is difficult to state exactly what one would be doing when the negation is outside of the “shall”-­operator).

Practical Reasoning and the Logic of Intentions  111 Sellars does have a logical rule for conjoining expressions of intention, though. Sellars notes that when we have two separate sentences, such as A and B, these separate sentences imply the conjunction “A & B.”7 But if I have separate intentions, Sellars’s rule about logical operators rules out simply conjoining these together as we would ordinary assertions of propositions. Thus, Sellars introduces the rule CI (conjunction introduction), which he states as follows: ‘P1,’ ‘P2’ . . . ‘Pn’ together imply ‘P1 and P2 and . . . and Pn’ hence “Shall be [P1],” “Shall be [P2],” . . . “Shall be [Pn]” together imply “Shall be [P1 and P2 . . . and Pn]” (ORAV VI.§70/p. 88).8 Returning to the logic of belief gives us reason to reject Sinnott-­ Armstrong’s contention (noted above) that the embedding problem for expressivism may be in-­principle and insoluble. For if we were to conclude this, then we would also have to conclude that there is no possible logic for beliefs. Consider the following argument:   (8) I believe Fido is a dog.   (9) If I believe that Fido is a dog, then I should believe that Fido is a mammal. (10) Therefore, I should believe that Fido is a mammal. While this argument seems sound, it cannot be, if we accept Sinnott-­ Armstrong’s point. For it is a well-­known point that to assert sincerely “I believe that-­p” expresses one’s endorsement of p. This feature of belief-­ expression underlies Moore’s paradox, according to which it is absurd to say things like “I believe that it is raining but it isn’t.” But given this understanding of the expressive function of belief-­assertion, we can see that the embedding problem applies with regard to the above argument. Premise 8 asserts a belief, but this belief appears only as the antecedent of a conditional in premise 9; it is not asserted there. Ergo, there would appear to be an equivocation. It is doubtful, however, that many philosophers are keen to argue that we cannot reason among beliefs. (How would one argue for this?) Thus, the clear conclusion to draw is that we obviously can reason using propositional attitudes (like assertions of belief or expressions of intention) that serve an expressive function. The only question is how we can give an explanation of the underlying logical form of such reasoning. Not surprisingly, Sellars’s logic of intentions can be generalized to a logic of expressive assertions and used to analyze reasoning involving assertions of belief. In doing so, we need to adapt Sellars’s principle S-­Imp

112  Practical Reasoning and the Logic of Intentions from intentions to beliefs. A plausible (but highly defeasible) adaptation, which we can call B-­Imp, is this: B-­Imp: If p implies q, then Bel (p) implies Bel (q) In other words, if p entails q, then if you believe p, you ought also to believe q. Again, this is defeasible and contextual but will do for now. Given this principle, the above argument can be reconstructed using Sellarsian tools. (11) (12) (13) (14)

Bel (Fido is a dog). (Fido is a dog) → (Fido is a mammal). Bel (Fido is a dog) → Bel (Fido is a mammal) [12, B-­Imp]. Bel (Fido is a mammal) [11, 13 MP].

Mark Schroeder (2008a) has argued that a solution like Sellars’s is one that expressivists should adopt to solve the embedding problem (although he does not mention Sellars, whose earliest foray into the logic of intentions was published in 1956). Schroeder distinguishes between two types of inconsistency one can appeal to as an expressivist. A-­type inconsistency obtains when one holds the same inconsistency-­ transmitting attitude toward inconsistent attitudes (2008a, p. 48). A-­type inconsistency is involved, for example, with the belief operator (e.g., in a belief that-­p versus a belief that not-­p). B-­type-­inconsistency obtains when one holds “two distinct and apparently logically unrelated attitudes toward the same content” (2008a, p. 48). For example, attempting to construct a logic of expressions using Blackburn’s “expressive operators” H! (Hooray!) and B! (Boo!)­—with principles of inference such as “(x)H!(x) → ~B!(x)”­—would involve asserting B-­type inconsistency. As we saw above, there are good reasons to doubt that B-­type inconsistency provides a genuine explanation of the inconsistency in question; and so Schroeder constructs an expressivist theory that relies on A-­type inconsistency. Schroeder argues that the basic strategy for the expressivist is to define some noncognitive attitude (such as “being for”) and treat being for as having an attitude that takes descriptive properties as its object; thus, “disapproval of murder is being for blaming for murder” (2008b, p. 589). Other moral judgments can be reconstructed in similar ways. Note the parallels to Sellars’s account and to the motivation I have ascribed to Sellars. Schroeder claims that by structuring his theory in this way, we can “reduce the problematic B-­type inconsistency between disapproval and tolerance to relatively unproblematic, because explicable, A-­type inconsistency in being for, and we can explain the inconsistency of non-­cognitive attitudes in terms of the inconsistency of their contents, in the same way as we do for beliefs” (2008b, pp. 590–1, emphasis added).9 Like Sellars, Schroeder restricts logical operators to within the scope

Practical Reasoning and the Logic of Intentions  113 of the “being for” operator. Beginning with negation (his basic case), Schroeder writes, “Our account so far tells us that if ‘N’ is an atomic normative sentence expressing being for n, then ‘~N’ expresses being for ~n. That is, the negation goes inside the state of being for, but outside everything else” (2008b, p. 591). Generalizing to other logical operators, Schroeder writes, If “P” expresses the belief that p and “Q” expresses the belief that q, then “P&Q” should express the belief that p&q and “P Ú Q” should express the belief that p Ú q. In each case, the connective gets located inside the general attitude, but outside whatever property is associated with the descriptive predicate. The analogous rules for normative language would have to look like this: &N: If “N” is a normative sentence expressing being for n and “O” is a normative sentence expressing being for o, then “N&O” expresses being for n&o. ÚN: If “N” is a normative sentence expressing being for n and “O” is a normative sentence expressing being for o, then “N Ú O” expresses being for nÚo. (Schroeder 2008b, p. 592) 5.2.1 An Interlude: Tense and Modals It is worth bringing up at this point that while Schroeder offers an expressivist program that is in many respects like Sellars’s, he also thinks that such a program is doomed to failure. There are a number of issues that (in Schroeder’s view) an expressivist account simply cannot handle. Prominent among these are moral statements that are tensed or that contain modal terms. Schroeder lists the following sentences, which (he argues) the expressivist cannot faithfully render: P1. Jon thought that murdering is wrong. P2. Jon thinks that murdering was wrong. P3. Jon thinks that having murdered is wrong. □1. Necessarily, Jon thinks that murdering is wrong. □2. Jon thinks that murdering is necessarily wrong. □3. Jon thinks that (necessarily murdering) is wrong.

(2008a, p. 69)

The first thing to notice is that since “thinks,” like “believes,” expresses a propositional attitude of endorsement, all of these sentences can be rewritten simply in terms of the contents of these propositional attitudes, i.e., P1'. Jon (at t-­n): “Murdering is wrong”

114  Practical Reasoning and the Logic of Intentions This method is parallel to Sellars’s device for reporting expressions of intentions by third parties: Smith: Shall be [the nations remain at peace] (ORAV, p. 96/§151). So, using this form to report Jon’s expressions of intention (rather than the cumbersome “thinks that” locution), let us see what progress we can make at rendering Schroeder’s sentences into “Sellars-­ese.” P1 seems straightforwardly to report an earlier expression of intention by Jon. Thus, we can write: P1S. Jont-­n: Shallwe (~M) P2, on the other hand, reports Jon’s expression of the view that murder was wrong at an earlier time. This we could render as follows: P2S. Jon: Shallwe (~M, if in C) [where C includes being at time t-­n] Now an apparent difficulty with P2S is that it seems to commit Jon to an intention to do something about the past, which is of course impossible. Schroeder’s criticism of the expressivist rendering of P2 is similar. And this would certainly be inconsistent if the intention in question were an individual intention. However, the intention in question is a collective intention, and such intentions have certain features that render cases like this less problematic. For example, members of a joint plan are often assigned different tasks in the completion of the plan. Thus, to use Bratman’s example, if you and I have a joint plan to paint the house, it might be my task to scrape off the old paint and yours to apply the new paint. Or, in American football, if the joint intention is “scoring a touchdown” or “earning a first down,” it might fall to the wide receiver to run a particular route (say, a slant) and for the quarterback to pass to him. Holding a joint intention serves as the basis for planning action, but it doesn’t follow that everyone will then perform the same action.10 With moral intentions, the community in question is not merely a team of housepainters, or a football team, but all of rational humanity. And my “task” in a particular project might be minimal, or nothing, since there are more moral imperatives in the world than I can reasonably act on. In many cases, my role may be solely that of condemnation of the act in question; more is not expected or possible. (Although we would clearly think poorly of someone who only condemned, say, injustice and cruelty and never took active steps to alleviate situations involving these.) I take it that moral condemnation of past actions is like these situations. In making an utterance like P2, Jon is expressing a we-­intention that murder not be allowed and taking the rational beings from time t-­n to be members of his moral community. In expressing the we-­intention, he is not individually committed to acting to prevent murders at t-­n. At most, he

Practical Reasoning and the Logic of Intentions  115 is committed to condemning such murders as actually took place at t-­n, praising those who prevented such murders, and so on. Such is the division of labor with joint intentions. Indeed, to express a group intention is not to be committed to acting on this intention individually­—it is not even to be committed to the possibility of acting on it individually since, as I noted, it is not possible to act on all moral imperatives in the world that are candidates for my action. (And, as we saw in our discussion in Chapter 4, one can express a we-­intention even if in fact other members of your community don’t endorse this commitment. In doing so, you are taking it that they are committed to it by other elements of the practice, the shared environment, and so on.) P3 can be rendered as follows: P3S. Jon: Shallwe ((t)~M at t-­n) Again, this should not be read as an individual intention to prevent something that has already happened. Rather, it should be understood on analogy with P2S. Thus, Jon considers rational humanity at t-­n part of his community and shares an intention with them. His part of the resulting collective action is (again) at most to condemn such murders as have taken place or to help see to it that those who have murdered are brought to justice, etc. Thus, expressing such intentions has a point, in that they serve as a basis for reasoning to further practical conclusions (e.g., “Shallwe be [Smith is prosecuted for murder]”). And such intentions only entail an impossibility if they are seen as individual intentions to do something rather than as a joint intention, which doesn’t entail that Jon needs to realize the content of the intention himself, only that Jon (may) have some action related to the intention to perform. Sellars can handle the modal claims as well. □1 we can set aside, for the modal operator is not modifying the moral component of □1; rather, it is operating on “thinks.” Thus, to give an account of □1, we would have to give an account of how modal operators work with propositional attitudes like “think” and “believe.” That is not our present task. Let us therefore move on to □2. This can be fairly unproblematically rendered as □2S. Shallwe (□~M) □3 is equally unproblematic: □3S. Shallwe (~□M) Thus, Sellars’s expressive account of moral terms is flexible enough to handle even these difficult cases. Schroeder also seems to think that an account based on A-­type inconsistency can avoid the van Roojen problem.11 As Schroeder puts the problem, expressivists of a certain variety­—particularly those like Blackburn

116  Practical Reasoning and the Logic of Intentions and Gibbard who endorse “higher order attitude” (HOA) accounts­— cannot distinguish between logical inconsistency and other kinds of inconsistency (such as pragmatic inconsistency). Schroeder offers the two following examples (Schroeder 2010, p. 120): P1. Murder is wrong P2a. If murder is wrong, then stealing is wrong C. Stealing is wrong This is a valid argument. The premise logically follows from the conclusion; one cannot endorse the premises and deny the conclusion without logical contradiction. And, of course, Sellars can reconstruct this argument as follows: Shallwe (~M) Shallwe (~M) implies Shallwe (~S) Therefore, Shallwe (~S) Schroeder’s second example is as follows: P1. Murder is wrong P2b. It is wrong to think that murder is wrong but not think that stealing is wrong C. Stealing is wrong This argument is not valid. Schroeder’s complaint about HOA accounts is that they reconstruct this argument in precisely the same way as the first one, thus conflating logical consistency/inconsistency with other kinds of consistency/inconsistency (such as Moorean inconsistency, as when one says, “It is raining and I don’t believe that it is raining”). How would Sellars reconstruct this argument? A lot turns on how we translate the phrase “S thinks that X is wrong.” In the examples involving tenses and modals, I noted that “thinks” and “believes” usually just attribute to someone an attitude of endorsement, and so we can usually move from Jon thinks that murdering is wrong to Jon: Shallwe (~M) This interpretive move will not obviously work in the case under discussion because this would require nesting “shall” operators. There are at least two reasons we don’t want to nest “shall” operators. First, when

Practical Reasoning and the Logic of Intentions  117 translated back into natural language, this construction seems (to me) incoherent: “It ought to be the case that it ought to be the case that murder is wrong.” Second, nesting shall operators don’t accurately translate what premise P2b says. It is easy to overlook this because this second argument of Schroeder’s makes it very easy to conflate the expression of an intention with the attribution of an intention, though these are crucially different acts in Sellars’s theory. If we translate “to think that lying is wrong” as “Shall ~(Lying),” then what we have is instead “that lying is wrong,” not “to think that lying is wrong,” which is straightforwardly a mistranslation of P2b. Thus, we should interpret the first part of premise P2b (“It is wrong . . .”) as expressing a view. However, the remaining clauses in premise P2b­—the ones that begin “to think that . . .”­—are attributing a view, not expressing the wrongness of any action. Thus, reminding ourselves of Sellars’s critical distinction between an expression of an intention and an attribution of an intention, we must conclude that this phrase attributes an intention (albeit a we-­intention) to an agent. Thus, Sellars would translate the above argument as follows: Shallwe (~M) Shallwe ~(Willwe [~M]) & ~Willwe [~S]) Therefore, Shallwe (~S) Now, accepting the premises of this argument while rejecting the conclusion probably involves some kind of inconsistency. But it is probably a Moorean-­type inconsistency of the sort involved in expressing an intention while refusing to self-­attribute the same intention. Thus, the inconsistency is not logical but is perhaps pragmatic or something else. Thus, Sellars need not assimilate this argument to the model of logical validity. He can accept Schroeder’s contention that the first argument is logically valid and the second one is not. As Schroeder argues, “To distinguish valid arguments from invalid arguments . . . [expressivists] need to appeal to the very kind of clash that obtains between beliefs with inconsistent contents” (Schroeder 2010, p. 122). But, of course, this type of inconsistency­—what Schroeder calls A-­type inconsistency­—is precisely the kind that Sellars relies on in his logic of intentions. Thus, Sellars’s expressivism seems to evade the van Roojen problem.12

5.3 The Logic of Intentions and Intensional Logic As I have noted, the basic machinery of Sellars’s logic of intentions is ingenious and prefigures similar views (like Schroeder’s) by half a century. However, even if the basic insights are solid, many of the specific details may need to be revised in light of subsequent developments in

118  Practical Reasoning and the Logic of Intentions expressivism. One immediate problem­—a problem that Sellars does not address­—arises with making intentional inference parasitic on the inferential relations among intentional contents. Sellars notes that the same strategy is used to explain inferential relations among beliefs. However, things are not so simple with beliefs. For example, consider the following inference: (15) Oedipus believes that Jocasta is beautiful. (16) If Jocasta is beautiful, then Oedipus’s mother is beautiful. (17) Therefore, Oedipus believes that his mother is beautiful.13 Similar problems lurk with inference among intentions: (18) Oedipus intends to marry Jocasta. (19) If Oedipus intends to marry Jocasta, then Oedipus intends to marry his mother. (20) Therefore, Oedipus intends to marry his mother.14 Sellars is clearly aware of the issue and mentions it in several places,15 but he doesn’t seem to think it sufficiently serious to offer much in the way of a solution. Sellars’s own view seems to be this: He explicitly offers his account not as a descriptive account of how people in fact reason but as a normative account of how people ought to reason.16 Thus, it is open to Sellars to embrace the above inferences, on the condition that we interpret the conclusions as prescriptions about how one ought to believe or intend. So assuming Oedipus has a firm and settled intention not to marry his mother, by modus tollens Oedipus ought to intend not to marry Jocasta­— and presumably this is the intention he would form if he had the relevant information. A second option would be for Sellars to modify his logic of intentions so that it relied on an underlying intensional logic.17 Intensional logic will have the usual elements: a set of worlds (G), an accessibility relation (R), and so on. But it will have some additional elements as well: DO. a domain of objects x1 . . . xn, y1 . . . yn. object variables DI. the intension domain, containing a set of functions from G to DO f1 . . . fn, g1 . . . gn. intension/concept variables I. an interpretation matching members of DO to object variables and members of DI to intension variables Now, as Melvin Fitting (who has worked out one of the best developed systems of intensional logic) has noted, a logical system like this one

Practical Reasoning and the Logic of Intentions  119 allows us to make sense of certain natural language intuitions that we have. For example, “the morning star” and “the evening star” designate the same object in the actual world but might have designated distinct objects. Thus, suppose f is the intension of “the morning star” and g is the intension of “the evening star.” Fitting’s intensional logic delivers the result that “if the objects denoted by f and g are identical, these objects are necessarily identical” (Fitting 2004, p. 182). However, you do not get the result that “if f and g designate the same object in the present world, they will designate the same object in all accessible worlds” (Fitting 2004, p. 182). Of course, the framework will still have to be relativized to persons because different individuals will associate different intensions with different objects. Thus, an ancient Babylonian might be unaware that “the morning star” and “the evening star” co-­refer, but a 20th century astronomer would of course know that they do. Thus, the following inferential pattern could represent the beliefs of the latter but not the former: I believe that the morning star is Venus The morning star = the evening star Therefore, I believe that the evening star is Venus I think that instead of relying on a formal model of intensional logic, it would be more helpful for us to rely on the informal model of deontic scorekeeping developed by Brandom (1994). As Brandom notes, different speakers associate different auxiliary commitments with the same claim. And yet we still must communicate with others, gain information from them, and so on. Thus, “if you want to understand what I say, you have to be able to associate with it a sentence that in your mouth expresses a claim that the sentence I uttered expresses in mine” (Brandom 1994, p. 510). Ergo, communication presupposes deontic scorekeeping ability. To express what we are doing when we engage in such scorekeeping, Brandom introduces the de re and de dicto locutions. A de re locution specifies the attributor’s own commitments and is marked off by the preposition “of.” The de dicto locution marks the commitments that are attributed to someone else (but that are not necessarily endorsed by the attributor) and is marked off by the conjunction “that.” Thus, we may say Oedipus believes of his mother that she is beautiful The de re (“of . . .”) clause expresses our commitment (that Jocasta = Oedipus’s mother). The de dicto (“that . . .”) clause attributes to Oedipus the commitment that this individual is beautiful (a commitment we may or may not share).

120  Practical Reasoning and the Logic of Intentions Thus, in interpreting the behavior of someone, we rely on logical relations among contents that we attribute to them in our deontic scorekeeping. These contents are, in the above model, expressed in the de dicto clause of our interpretive commitments. For example, in interpreting the behavior of Oedipus, we attribute to him the following commitment or belief: Jocasta ≠ my [i.e., Oedipus’s] mother Indeed, the tragedy of the play revolves around his holding this belief. Thus, we attribute to Oedipus the following inferential patterns: (15) Oedipus believes that Jocasta is beautiful. (16′) [Jocasta ≠ Oedipus’s mother]. . . . I place the contents of 16′ in brackets only to emphasize that these are contents we attribute to Oedipus rather than contents we ourselves endorse. Nothing logically follows from 15 and 16′, so using only the contents attributable via deontic scorekeeping in our major premise (premise [16′]), we can avoid attributing to Oedipus a belief he is not capable of inferring. Similarly, on the practical side, we can attribute to Oedipus the following reasoning: (18) Oedipus intends to marry Jocasta. (19′a) Oedipus intends not to marry his mother. (19′b) [Jocasta ≠ Oedipus’s mother]. (19′c) [Therefore, marrying Jocasta ≠ marrying Oedipus’s mother]. (20′) Therefore, Oedipus intends to marry Jocasta and does not intend to marry his mother. One might worry that this solution departs too far from the spirit of Sellars’s logic of intentions. In particular, Sellars indicates that logical relations among intentions should be parasitic on logical relations between their contents. But doesn’t the modified view I am endorsing here make logical relations among intentions parasitic on logical relations among beliefs we attribute to an agent? So, to return to our example of belief, we wanted to model logical relations among beliefs along the following lines: (21) Bel (Benjamin Franklin is the inventor of bifocals). (22) The inventor of bifocals = the first postmaster general of the United States. (23) Therefore, Bel (Benjamin Franklin is the first postmaster general of the United States).

Practical Reasoning and the Logic of Intentions  121 However, it seems like premise 23 must involve a commitment or belief we are attributing to you. Thus, it seems like the argument must in fact have the following form: (21) Bel (Benjamin Franklin is the inventor of bifocals). (22′) Bel (The inventor of bifocals = the first postmaster general of the United States). (23) Therefore, Bel (Benjamin Franklin is the first postmaster general of the United States). However, I think that our account of intensional logic via deontic scorekeeping should be interpreted differently. For the major premise doesn’t (necessarily) record the beliefs we are attributing to the agent in question. Rather, it records the commitments we attribute to the agent we are interpreting. What is the difference? Brandom sometimes talks about commitments as though they are beliefs­—propositions that the agent in fact endorses or is disposed to endorse. Other times, Brandom seems to offer a more normative characterization of commitment­—as propositions that the agent ought to endorse given the other things the agent is committed and entitled to. In the spirit of treating the logic of intentions and belief as normative enterprises, I propose to treat commitment in the latter sense­—not as beliefs held by the agent but as propositions or inferences we think the agent ought to endorse (or inferential patterns she ought to conform to or conclusions she ought to draw, etc.). Thus, if I interpret you as having a commitment to treating all religions with respect, I am not saying that you endorse this claim­—but I am saying that it follows from other things that you do endorse (or are committed/entitled to), and that you ought reasonably to endorse this as well. (The latter clause is necessary, as it isn’t reasonable to demand of people that they endorse all of the consequences of all of their commitments.) Thus, for example, consider Nathan’s rebuke to David in 2 Samuel 12. We may attribute the following beliefs and commitments to David: (24) Bel (The rich man of Nathan’s story has done great evil and deserves to die). (25) [David’s behavior with respect to Uriah is morally the same as the rich man’s behavior]. (26) Bel (David has done great evil). David himself expresses the belief in (24). When Nathan cries, “You are the man!” he is trying to elicit David’s recognition that David should recognize a commitment to (25)­—and therefore ought to believe that he (David) has done great evil. Now, in fact, David might refuse to acknowledge the truth of (25) for any number of reasons­—moral blindness,

122  Practical Reasoning and the Logic of Intentions self-­deception, pride, or whatever. In this case, he would refuse to form the belief indicated in (26). But we can still say that given the similarities between his treatment of Uriah and the rich man’s treatment of the poor man­—similarities with which David is familiar­—David’s other commitments in fact commit him to endorsing (25). Thus, the major premise (premise [25] in the above argument) doesn’t model the beliefs of the agent you are interpreting. Rather, we should think of this premise as ranging over a domain of propositions that are suitable contents for beliefs of the agent we are interpreting. Some of these beliefs will be ones the agent actually endorses. Some will be beliefs she doesn’t endorse but which we think she is committed to endorsing based on her other commitments. As noted above, although this domain will include propositions that the agent is committed to by other claims she endorses, it will not include all such propositions; it is not closed under entailment. Thus, the domain doesn’t simply consist of the beliefs that the agent actually holds. Nor does it consist of the beliefs the agent holds plus all the material or logical consequences of these beliefs. Rather, it consists of something in between­—perhaps the agent’s beliefs plus the consequences of these beliefs we think the agent ought to hold, that is, the ones we hold the agent accountable for. Now, this requirement is fuzzy and not very specific, but as Aristotle says, “We must not admit more precision than the subject-­matter admits of” (NE I.3). And in any case, this fuzzy requirement accords with our common interpretive practice. For example, although the rules of basic arithmetic entail infinitely many propositions, I don’t think you rationally ought to believe all of these propositions. That is, I don’t attribute to you a commitment to all of them even if I hold you to be proficient in arithmetic. However, even if your commitments don’t run to the infinity of all true arithmetic propositions, I do take your commitments to outstrip your beliefs. Thus, for example, if I know that you are reasonably adept at addition, I will take you to be committed to the claim that “144 + 347 = 491,” in the sense that it follows from your other commitments and you can be expected to infer according to this principle if called upon to do so (in both the predictive and normative sense of “expected”). Similar comments will apply to inference among intentions. If I am interpreting Oedipus’s intentions, I may properly attribute to him the intentions and commitments (18), (19′a-­c), and (20′). These are in fact all commitments that Oedipus does endorse. But we can also engage in the following kind of interpretation: (27) Smith intends to collect his daughter from school at 1:30 p.m. (28) [Smith collects his daughter from school at 1:30 p.m. implies Smith leaves work by 1 p.m.]. (29) Therefore, Smith should intend to leave work by 1 p.m.

Practical Reasoning and the Logic of Intentions  123 Again, the statement in premise (28) appears in brackets to signify that this is a commitment we are attributing to Smith. However, this doesn’t mean that Smith endorses this commitment. Smith might think he can leave at 1:10 and still get there on time. But we might say that Smith is still committed to (28): Say, he knows from experience that Smithfield Road is heavily congested between the hours of 1 and 2, and that he will never arrive on time if he leaves at 1:10, but this fact has temporarily slipped his mind (due to distraction or absentmindedness, or whatever). Maybe Smith would even deny (28), arguing that he can somehow manage despite the traffic. What we are willing to insert as our major premise­— and how far removed it is from things the agent actually endorses­—will depend on a number of other questions in philosophy, from the internal versus external reasons debate to how much responsibility we wish to impute to people for reasoning out the consequences of their occurrent commitments, and so on. I shall not even attempt to settle these questions here. I will rest satisfied with showing (as I hope I have) that the notion of deontic scorekeeping can be pressed into service to rescue intentional inference from the problem of intensional contexts. The above comments on intensional logic relate, I think, to certain of Anscombe’s comments on intention. Anscombe famously writes that we intend something under a certain description. Of course, there is not a unique description under which something is intended, but there are various epistemic and intensional constraints on how we may describe a person’s intention. Thus, suppose I have the following intention: Shall (I pump water) It seems as a factual matter to be true that “I pump water” entails that “I move such and such muscles (which I cannot even name) in such and such a manner” Thus, it seems that we may conclude (via S-­Imp): Shall (I move such and such muscles in such and such a manner) This is a problematic conclusion because, in pumping water, I will ordinarily have no such intention. Fortunately, our intensional logic allows us to explain why this inference is not a good one and why we are not forced to conclude that I have this implausible intention. Anscombe identifies intentional actions as “ones to which a certain sense of the question ‘why?’ has application” (1963, §6/p. 11). But she qualifies this statement by stating that “this question is refused application by the answer: ‘I was not aware that I was doing that’” (1963, §6/p. 11). Recall, also, from Chapter 3 that, for Anscombe, to describe

124  Practical Reasoning and the Logic of Intentions an intention is to place it in a calculative order. So there are two kinds of constraints on the logical consequences we are allowed to infer in intentional inference: (1) “calculative order” constraints and (2) epistemic/ doxastic constraints. Let me illustrate each one. Consider the example of the calculative intentional order from Chapter 3: He is moving the pump handle up and down Why? In order to pump water to the house Why? In order to poison the inhabitants of the house Etc. Now, where in this calculative order can we place the following claim? He is moving such and such muscles in such and such a manner. In the calculative order, this claim would belong first, before “He is moving the pump handle up and down.” But we run afoul of the epistemic constraint on intentions: He does not know that he is doing it, and hence it is not one of the descriptions under which we can place his intentional action. Thus, while “He is moving the pump handle . . .” might materially imply “He is moving such and such muscles . . .” the latter is not an allowable consequence of the former in our intensional logic, which restricts the inferential consequences of any claim based on the agent’s conception of that claim, his collateral commitments, and so on. What about a claim like “He is casting a shadow on a rockery such that it looks like a face were looking out of the rockery” (cf. Anscombe 1963, §23/p. 37)? Now, the man might notice that he is doing this, and so the above epistemic constraint doesn’t apply. But our other constraint applies here: This claim doesn’t fit in the calculative order. That is, we cannot complete the sentence “He is casting a shadow on the rockery in order to . . .” in a way that fits into the above rational calculation, and so this is not an allowable antecedent or consequence of any of the other claims in our above calculative sequence. Ordinary propositional logic places no in-­principle limits on the consequences of any proposition, but intensional logic­—particularly the logic of intentions­—restricts inferential consequences according to the above calculative and epistemic constraints. This also offers Sellars a way to rescue the doctrine of double effect, which seems jeopardized under the logic he proposes.18 Given S-­Imp, it seems as though we must intend any consequence of our expressed intentions. Thus, on Sellars’s logic of intentions (as he originally proposes it), there is no such thing as a foreseen but unintended consequence of a volition. However, modifying Sellars’s logic of intentions so that it relies on an intensional logic rather than a traditional propositional logic allows us to deny that we intend every consequence of our intentions. This can be wed, I think, to Anscombe’s account (in Chapter 3) of intention-­attributions as placing actions within a calculative order. Thus, by moving the pump

Practical Reasoning and the Logic of Intentions  125 handle up and down, I intend to pump water to the house, but I do not intend to cast a particular shadow on the rock. Not everything I do, in doing X, is part of my intention on such an account. This is the upshot of tying our logic of intentions to an intensional logic, which (as I said) fits nicely with Anscombe’s particular account of intention-­attribution. No doubt certain refinements to the account are in order, but double effect is a sticky problem for anyone, and it is not obvious to me that Sellars’s account (as presently modified) has more difficulty handling it than do competing accounts. (I will briefly return to the doctrine of double effect at the end of the chapter, after a few more tools are at hand.)

5.4 Further Refinements to Sellars’s Account of Inference Among Intentions Another headache for expressivist accounts has been mixed inferences, namely those that involve both factual and normative premises. Hare placed significant restrictions on such inferences, adding the following two principles to his system of imperative inference: (1) No indicative conclusion can be validly drawn from a set of premises which cannot be validly drawn from the indicatives among them alone. (2) No imperative conclusion can be validly drawn from a set of premises which does not contain at least one imperative. (Hare 1952, p. 28) Geach (1958) presents compelling counterexamples to both principles. What most concerns Sellars is how to get “relevant beliefs . . . into the scope of practical reasoning” (ORAV §66/p. 88). After all, our practical reasoning involves not just our preexisting intentions (such as my intention to fly to Chicago) but our relevant beliefs about the state of the world (e.g., I need to arrive two hours before departure, the airport is a one-­hour drive away, etc.). However, one might worry that Sellars has made it impossible to create a mixed inference since he has explicitly denied that one can place logical operators outside the scope of the “shall” statement. Thus, one cannot reason “Shall (I fly to Chicago) • [I have a ticket for the 2:35 flight] • [I need to arrive at the airport at 12:35] • [If I am to arrive at the airport at 12:35, I need to depart at 11:35] . . .” Sellars’s solution is to introduce the principle “So-­be-­it,” which allows us to introduce the agent’s factual beliefs into the scope of the intention so that they may be taken up in practical reasoning. Sellars states So-­be-­it as follows: So-­Be-­It: “Shall be [φ]” and ‘p’ imply “Shall be [φ and p]” where ‘φ’ is a formula which may or may not be logically complex. (ORAV §66/p. 88)

126  Practical Reasoning and the Logic of Intentions Thus, I actually reason as follows: “Shall (I fly to Chicago • [I have a ticket for the 2:35 flight] • [I need to arrive at the airport at 12:35] • [If I am to arrive at the airport at 12:35, I need to depart at 11:35] . . .).” Sellars illustrates So-­be-­it with the following example. He bids us consider the following text: “I shall go downtown tomorrow, if the weather is clear. The weather will be clear. If the weather is clear, the streets will be crowded. If the streets are crowded, I will be miserable” (ORAV §83/p. 89). Sellars writes, This text implies, by virtue of Shall-­Be,19 So-­be-­it, and CI, the conjunctive intention Shall be [I go downtown tomorrow, if the weather is clear and the weather will be clear and if the weather is clear, the streets will be crowded and if the streets are crowded, I be miserable]. . . . The above conjunctive intention, in turn, implies, by three subordinate uses of modus ponens, the intention Shall be [I go downtown tomorrow and be miserable]. I ask myself if I accept this intention. If I reject it, I am logically committed to rejecting either Shall be [I go downtown tomorrow] or Shall be [I be miserable tomorrow]. Presumably I will reject the former. (ORAV §§83–85/pp. 89–90) Now, one question that immediately arises about this solution is this: By placing factual beliefs within the scope of the intention, does that make these beliefs (or their objects) the object of our intention? Do I now intend not only that I fly to Chicago but also that the flight depart at 12:35 or that the airport is a one-­hour drive from my house? It seems, on the face of it, that So-­be-­it makes all of the factual states of affairs that we consider in our practical reasoning an object of our intending­—even though many of these states of affairs are manifestly outside our control. To understand Sellars’s answer to this question, one must bear in mind that like all logical and normative rules, Sellars’s logic of intentions is not intended as a description of how agents practically reason. Rather, it is intended as a guide for how rational agents ought to reason. Here, I will let Sellars speak for himself, as he is crystal clear on his rationale: To swallow this principle, one must bear in mind that to intend Shall be [p and α]

Practical Reasoning and the Logic of Intentions  127 a person need not desire that p be the case nor even like the idea of p’s being the case. The point of the principle is simply that the state of affairs gets into the logical framework of practical thinking. So by virtue of So-­be-­it, an agent who has the intention that α be the case, Shall be [α] and has decided that p is (or will be) the case, is, if he is rational, confronted by the alternatives, Shall be [p and α]? Shall be [p and not-­α]?

(VR §§55–6/p. 59, final emphasis added)

An agent who is rational will understand that p is given and thus realize that she is choosing between [p and α] and [p and not-­α]. Including p within the scope of the intention reflects this fact while allowing the agent to include p in her reasoning as to whether to intend α or not-­α. Thus, in natural language, the above would be more perspicuously rendered as a choice between [α given p] and [not-­α given p] even though “given” is rendered “and” in logical notation. Thus, an intention will have “intention-­factors . . . which are viewed sub specie ‘up-­to-­me,’ and those which are viewed as given, i.e. those which are there by virtue of So-­be-­it” (ORAV §79/p. 89); and Sellars points out that for an intention to count as an intention, every intention “must have at least one ‘up-­to-­me’ constituent” (ORAV §80/p. 89). Intentions are concerned with doing, and so by their nature intentions must have at least one element that is (at least as far as we can discern) under our control and not simply a given and unchangeable state of the world providing a background against which we choose. This move allows us to address certain problems that might be thought to arise in Sellars’s logic of intentions. While space­—and the reader’s patience­—does not permit a discussion of every paradox of deontic logic, there are some that might be thought to be particularly problematic for Sellars’s view; let us discuss a couple of these. For example, many views about the logic of moral and evaluative terms have difficulties with cases like the Gentle Murder Paradox:20 It ought to be the case that if I kill Smith, I kill him gently. This is true, I suppose; if you are going to do something bad, you should do it in the least bad way. But suppose I know that I am going to kill Smith. This information enters the intention via So-­be-­it, so we get the following complex intention:21 Shallwe (If I kill Smith, I kill him gently and I will kill Smith)

128  Practical Reasoning and the Logic of Intentions Using a single operation of modus ponens, we get the following we-­intention: Shallwe (I kill Smith gently) This is a conclusion many have found highly problematic. Sellars addresses issues in this neighborhood in CDI. There, he is addressing a challenge issued by Chisholm: Deontic logic must be able to deal with contrary-­to-­duty imperatives, which take the form “You ought to do a, but if you don’t do a, you must, by all means, do b” (CDI, p. 303). The essay is (in true Sellarsian form) long and almost painfully detailed (and the formalizations often contain elementary and confusing errors, such as missing negation symbols or incorrect variables), but I will try to extract some of the essential lessons from it. The key to the solution, for Sellars, lies in recognizing that “obligations are relative to circumstance specification” (CDI, p. 325)­ —that is, an obligation only holds conditional to a set of grounds­—and so an argument that you have such and such an obligation cannot be made in the presence of a ground that is inconsistent with your having that obligation. Sellars approaches this issue by noting that implication relations, of course, hold relative to a set of background assumptions. Thus, α implies β relative to γ. If γ is the case, then α implies β. But given γ, and given therefore that α implies β, a plausible principle of deontic logic is as follows: P2. If α implies β then O(α)′ implies O(β)′ The prime symbol (′) indicates that the entire bracketed content is the “target” of the obligation. This allows Sellars to represent different ways of understanding conditional obligation. We might understand it in either of the following ways: CO1. O(s → b′) CO2. O(s → b)′ In the second case (CO2), the entire conditional is the target of the obligation: You are obligated to bring it about that (if stealing, then paying reparations). In the first case, only b is the target of the obligation. Thus, if you steal, then you are obligated (to pay reparations). As Sellars notes on CO1, the conditional is “transparent to the indicative propositional expression ‘s’” (CDI, pp. 310–11). This means that given the first interpretation, we can infer according to the following principle of deontic inference: Ia. If O(p → q′) and p, then O(q)

Practical Reasoning and the Logic of Intentions  129 Thus, understanding conditional obligation in terms of CO1 allows for factual detachment­—an indicative statement can be used to “detach” the conclusion of the deontic conditional. CO2 is not transparent in this way, though, and thus only supports deontic detachment: Ib. If O(p → q)′ and O(p), then O(q) Now, it is well-­known that allowing for factual detachment can lead to the sort of contradictions Chisholm complains about­—and to the type of paradox with which we began this discussion. (And it is precisely for this reason that Sellars ultimately rejects factual detachment and endorses a suitably modified and restricted version of Ib­—namely, the principle P2, which I presented above, although I have yet to discuss the restriction Sellars places on its deployment.) So consider Chisholm’s original version of the paradox, where s represents “stealing” and b represents “paying reparations”:22 (1)  O(~s) Premise Premise (2) s Add. (3)  ~s implies (~s ∨ ~b) (3), Def. “→” (4)  ~s implies (s → ~b) (4), P2 (5)  O(~s) implies O(s → ~b) (6) O(s → ~b) (1), (5), MP O(~b) (2), (6), Ia (7)  As noted above, the problem seems to arise directly from allowing factual detachment. If you disallow factual detachment, then the problem vanishes. Sellars suggests that inferential principles like Ia gain plausibility by their resemblance to modus ponens but ought to be set aside because (a) they generate paradoxes and (b) there are other principles, sufficiently similar, that do the required work in deontic logic. (Sellars has in mind, of course, P2.) To further shore up his account of deontic logic against the possibility of paradox, Sellars argues we must place restrictions upon deontic arguments, including upon the deployment of conditional obligation in such arguments (which in turn limits the applicability of P2): We must “require that the assumption on which the implication rests be normatively cotenable with the ground of O(α)′” (CDI, p. 334). What this means, concretely, is that the argument cannot contain premises that normatively contradict the grounds or premises of the obligations that are asserted in the argument. Thus, if you have a conditional obligation­—say, an obligation to do [A → B]­—the grounds, or conditions, of the obligation to do [A → B] must be normatively compatible with doing A at the relevant time. But this requirement will rule out the paradoxical cases, argues Sellars. Thus, in the argument above, the total circumstances include

130  Practical Reasoning and the Logic of Intentions O(~s)­—the obligation not to steal. So the circumstances or grounds cannot support an obligation to do (s → ~b) because the grounds are incompatible with executing the antecedent of this conditional­—these grounds are normatively incompatible with stealing. Thus, as Sellars writes, “As long as he has not yet stolen, the obligation not to steal entails the obligation to not-­steal-­or-­whatever, and hence the obligation to-­if-­steal-­then-­ whatever” (CDI, p. 332). Thus, premise (5) involves an illicit appeal to P2, an appeal ruled out by this restriction. (The restriction doesn’t forbid the paying of reparations. As Sellars continues, if one has already stolen, then this becomes a circumstance factor that is relevant to one’s prima facie obligation to pay reparations. The restriction only forbids one from executing the conditional obligation of stealing and then paying reparations or stealing and then not paying reparations, or whatever.) Does this help with the example with which we began? Let us formalize it, where k represents “kill” and g represents “kill gently”: (1) O(~k) Premise Premise (2) k Premise (3) O(k → g) (4) O(g) (2), (3), Ia It should be clear on what grounds Sellars would reject this argument. The total circumstances include O(~k), so the circumstances cannot support an obligation to do (k → g) because the grounds are incompatible with executing the antecedent of this conditional­—these grounds are normatively incompatible with killing. CDI is curious, in that it discusses deontic logic in a way that doesn’t at all touch on Sellars’s logic of intentions, something that was very well-­ developed by CDI’s 1967 publication. However, since P2 exactly mirrors S-­Imp, I think we can easily take the above restriction on P2 as a plausible restriction on S-­Imp and thereby avoid the Gentle Murder Paradox. Thus, I might have the following commitments: Shallwe (If I kill Smith, I kill him gently) I will kill Smith23 Sellars’s requirement that factual premises involved in practical reasoning be placed inside of the shall-­operator reflects, in a sense, a rejection of factual detachment: “Shallwe (I kill Smith gently)” does not follow directly from these two premises. However, one might think this is a distinction without a difference since once the factual premise is pushed inside the shall-­operator, the inference can go ahead anyway. But not so fast! For another of our commitments is Shallwe (I do not kill Smith)

Practical Reasoning and the Logic of Intentions  131 This commitment forms part of the total conditions for my action and thus cannot support the conditional obligation “Shallwe (If I kill Smith, I kill him gently)” because it is normatively inconsistent with the performance of the antecedent of this conditional obligation. I think that, in general, even if you think consequences like “I ought to kill Smith gently” are detachable on Sellars’s deontic logic, Sellars’s theory has certain features that make this conclusion less damaging than it might otherwise be. In particular, the fact that Sellars allows factual information (including information about what one will intend, such as “I will kill Smith”) inside the scope of the intention may be a hidden advantage of his account when it comes to examples like this. Recall our discussion when I introduced So-­be-­it. Sellars wrote, So by virtue of So-­be-­it, an agent who has the intention that α be the case, Shall be [α] and has decided that p is (or will be) the case, is, if he is rational, confronted by the alternatives, Shall be [p and α]? Shall be [p and not-­α]?

(VR §§55–6/p. 59)

The point of So-­be-­it is to allow a person to settle on an intention given certain factual information. Thus, using the model provided in the quote by Sellars, one can be seen as deciding among the following options: Shallwe (Given that I am going to kill Smith, I kill him gently)? Shallwe (Given that I am going to kill Smith, I kill him brutally)? Presumably, one decides on the former course of action; and thus, So-­be-­it allows for a fuller description of one’s action to be represented as follows: Shallwe (Given that I am going to kill Smith, I kill him gently) which to my ear doesn’t sound greatly implausible. At any rate, I think that Sellars’s deontic logic handles this as well as do competing accounts. So-­be-­it and S-­Imp might be thought to create other headaches for Sellars.24 For example, can’t we use disjunction introduction to produce counterintuitive results? Consider the following: Shallwe (I help the poor) “I help the poor” implies “I help the poor or light them on fire” Therefore, Shallwe (I help the poor or light them on fire)

132  Practical Reasoning and the Logic of Intentions Counterexamples like this, though, treat Sellars’s logic of intentions as though it were underlain by a standard propositional logic instead of (as I have argued) an intensional logic that places significant constraints on the allowable consequences of claims. These constraints are, as I said, both calculative and epistemic. Intentional action is intentional under a certain description, meaning that it plays a certain role in a calculative order. Thus, allowable contents­—and allowable inferential consequences­—are constrained by this calculative order. And a claim like “Shallwe (I help the poor or light them on fire)” cannot play a role in the calculative order of a person who intends to help the poor and who intends to do so under that description. Thus, let us modify our above calculative order as follows: Shall (I move the pump handle up and down or sing “Fair moon, to thee I sing” from H.M.S. Pinafore) Why? In order to pump water to the house Why? In order to poison the inhabitants of the house Etc. Immediately, the calculative order has broken down after the first step, as satisfying the second disjunct is not a means of fulfilling the further descriptions of the action and is in fact completely unrelated to the other actions in the calculative order. Thus, we can see that disjunction introduction is limited in its scope by the above listed constraints. I don’t think, in any case, that Sellars would have objections to modifying his logic of intentions in this way. Already, his logic of intentions departs in important ways from traditional rules of logic. As Sellars himself notes, the proscription on negation outside the “shall” operator means that “‘Shall P’ implies ‘Shall Q’” has no contrapositive. “If it is countered that an implication which has no contrapositive is not an implication, the answer is that this simply reflects the extent to which the theory of implication has been dominated by the study of reasoning concerned with mathematical relationships and empirical matters of fact” (TA, p. 112).

5.5 Compositionality A key requirement on an expressivist account is that such accounts are compositional, i.e., natural language expressions can be reconstructed in a systematic and non–ad hoc way out of the basic tools offered by the account. Can Sellars’s expressivism capture the difference between other, subtly different natural language expressions? For example, Jack Woods has noted that there is a difference between the following expressions: Either murder is wrong or murder is not wrong and

Practical Reasoning and the Logic of Intentions  133 Murdering or not murdering is wrong and he expresses the concern that Sellars must give the same translation of both: Shallwe (M ∨ ~M) I think there is a way out of this. However, the most natural way to resolve this difficulty­—and preserve compositionality by translating the above sentences in the most intuitive manner­—may require us to adopt a solution explicitly rejected by Sellars25 (something I try to avoid in this book if at all possible). That solution is to allow a certain class of logical operators to appear outside of the “shall” operator.26 Sellars’s most explicit argument against allowing logical operators outside of the “shall” operator regards the negation operator. As noted above, Sellars argues that it is incoherent to have an external negation of the expression of an intention. One may grant this but wonder how it follows that one cannot join expressions of intention using other operators. After all, Sellars explicitly allows expressions of intention to be joined using “implies,” on the understanding that “implies” is understood metalinguistically and the expressions in question are mentioned and not used. Why, if expressions of intention can be related by implication, can they not be related by conjunction or disjunction? I suspect the answer traces back to the denial of an external negation for expressions of intention. (As I noted above, Sellars admits implication among intentions but correspondingly denies contraposition, which would introduce an external negation.) Sellars may be worried that if we join expressions of intention with other kinds of logical operators, the most natural way to reason using complex premises containing multiple intention-­expressions will be to use external negation. For example, how would we reason from (30) to (32)? (30) Shall (p) ∨ Shall (~p) (31)  . . . (32) Therefore, Shall (p) The most natural premise to insert is (31) ~Shall (~p) which is ruled out by Sellars’s insistence that expressions of intention have no external negation. However, a consistent­—and non–ad hoc­— solution is already present. If you are convinced that expressions of intention have no external negation, and if you agree with Sellars that

134  Practical Reasoning and the Logic of Intentions inferential relations among intentions are parasitic on inferential relations among their contents, then the above inference can be reconstructed without external negation and by using S-­Imp. (30) (31a) (31b) (31c) (32)

Shall (p) ∨ Shall (~p) Shall ~(~p) ~(~p) implies p Shall ~(~p) implies Shall (p) [24b, S-­Imp] Therefore, Shall (p)

This move­—allowing conjunction and disjunction of expressions of intention­—allows us to give straightforward translations of the above natural language sentences. Either murder is wrong or murder is not wrong Becomes Shallwe (~M) ∨ Shallwe (M ∨ ~M)27 (For the time being, I am interpreting permissibility as indifference, i.e., “Shallwe (p ∨ ~p).” This is just a placeholder, and I will revisit the topic of permissibility shortly.) On the other hand, Murdering or not murdering is wrong Becomes Shallwe ~(M ∨ ~M) The latter, of course, expresses a moral dilemma and in logical form is a contradiction. Whether such claims are allowable in deontic logic is not a claim I will explore. However, the point is that the natural language claim has a straightforward translation in Sellars’s idiom. Thus, Sellars’s expressivism seems fully compositional as long as we are willing to deviate from some of the constraints Sellars originally laid out for his logic of intentions. But how far must we deviate? I have already introduced an intensional logic (in place of the propositional logic that seemed to underlie his logic of intentions), and I have already pushed back against his contention that “shall”-­expressions cannot be joined by conjunction or disjunction. It seems, though, that we may have to reject one of Sellars’s key contentions­—that shall-­statements have no external negation. To see this, we must see how Sellars would reconstruct traditional —and the corresponding notions of deontic logic. Sellars’s notation­ logic of intentions­—offers us a straightforward translation for some key

Practical Reasoning and the Logic of Intentions  135 concepts of deontic logic. To say that Jones is obligated to tell the truth, we can say, Shallwe (Jones tells the truth) To say that Jones is forbidden to lie, we then say, Shallwe ~(Jones lies) The concept of permissibility might seem to resist easy resolution on Sellars’s account, however. In most systems of deontic logic, permissibility is defined in terms of obligation as follows: ◇P ↔ ¬□¬P where ◇ stands for “permissible” and □ stands for “obligatory.” However, since Sellars does not allow logical operators outside the scope of the “shall” operator, this reconstruction of permissibility is not open to us. That is, we cannot reconstruct “P is permissible” as follows: ~ Shallwe ~(Jones lies) How, then, to reconstruct the notion of permissibility in Sellars’s system? One suggestion arises out of Sellars’s definition of “indifference”: “‘X is indifferent to G with respect to K’ has the sense of ‘X prefers neither G • K to ~G • K nor ~G • K to G • K’” (TA, p. 115). This suggests we might understand “It is permissible for Jones to grow carrots in his garden” as “Shallwe ([Jones grows carrots in his garden] or ~[Jones grows carrots in his garden]).” Of course, indifference from the standpoint of we-­intending doesn’t mean personal indifference. I might have strong personal objections to carrots and carrot-­growing while nevertheless recognizing that from the moral point of view, I cannot object to Jones’s carrot-­growing. Unfortunately, understanding permissibility this way does not preserve the intuitive relations of deontic logic. For example, if something is not permissible, then it is forbidden: ~ ◇p → □~p How would we represent this inference in Sellars’s ethics? We would have to represent it as follows: Shall ~(p ∨ ~p) implies Shall (~p) which (per S-­Imp) is true just in case ~(p ∨ ~p) implies ~p

136  Practical Reasoning and the Logic of Intentions Now, the latter implication is true in the vacuous sense that a contradiction entails anything. Thus, if we reconstruct permissibility along the above lines, we could also conclude that if something is not permissible, it is obligatory! However, if we reconstruct permissibility in the traditional way, then the entailments follow normally: ~ Shallwe ~(Jones lies) = It is permissible for Jones to lie ~~ Shallwe ~(Jones lies) = It is not permissible for Jones to lie The double negation cancels out, of course, leaving us Shallwe ~(Jones lies) = It is obligatory for Jones not to lie, i.e., it is forbidden for Jones to lie Now, to adopt this solution­—external negation­—we have to explain why our original reasons for rejecting external negation don’t hold. First, I noted that external negation (for example, with belief) actually commits us to something far stronger than closure: For example, if P implies Q and you don’t have any belief at all regarding Q, then you should not believe P. This means that we must affirmatively believe all logical consequences of our beliefs or withhold belief. We must therefore, it seems, believe everything before believing anything. However, the move from a propositional logic to an intensional logic to underlie the logic of beliefs and intentions may solve this problem for us. Consider our two principles of inference: B-­Imp: If p implies q, then Bel (p) implies Bel (q) S-­Imp: If p implies q, then Shall (p) implies Shall (q) On Sellars’s original picture, the allowable consequences of p would be anything materially implied by p (or perhaps even all things logically implied by p). But on our intensional logic, which is wed to a Brandomian account of deontic scorekeeping, the domain of allowable consequences of p is constrained by the commitments of the person we are interpreting. Thus, even if q materially follows from p, it may not be an allowable consequence of p under B-­Imp or S-­Imp. Thus, Jocasta might be Oedipus’s mother, but that doesn’t mean that “‘Oedipus is married to Jocasta’ implies ‘Oedipus is married to his mother’” is an allowable substituend for “p implies q” when we are interpreting Oedipus. A practical consequence of this is that belief is not closed under entailment and one need not have any commitments at all regarding all of the consequences of one’s beliefs. Obviously, many substitutions are allowed (and rationally required): We can substitute “‘Jocasta is beautiful’ implies ‘Oedipus’s wife is beautiful,’” for “p implies q” when we are interpreting Oedipus and expect him to make inferences corresponding to this principle. But

Practical Reasoning and the Logic of Intentions  137 allowable substitutions are determined in part by the commitments and entitlements we attribute to the agent we are interpreting. Similar comments will apply to external negation for intention, and intention is not closed under entailment either. What about Sellars’s claim that an external negation of an expression of intention is “conceptually incoherent” (ORAV §57/p. 87)? I think it is coherent if you conceive of the act simply as the expression of the absence of an intention. The analogy with belief should make external negation for expressions of intention more palatable. Plausibly, there are more than two doxastic attitudes to hold toward a proposition (belief or disbelief). There is also suspension of belief, and expression of suspension of belief is perfectly coherent. It is for this reason that there is the category of agnosticism between theism and atheism. Suspension of judgment is often a rational response to inconclusive evidence, and thus neither belief nor disbelief is rational­—but rather suspension of judgment, which implies an external negation on the belief operator: ~Bel(p) and ~Bel(~p). If belief-­expression can have an external negation, then there is no principled reason for thinking that there is anything incoherent about expanding external negation to include intention-­expressions. Further, I would argue that ordinary usage supports a difference in meaning between internal and external negation of an expression of intention, as do some central debates in ethical theory: The doctrine of double effect (DDE) relies on a notion of ~Shall(p), meaning that I don’t intend to p but perhaps foresee that p will result from my action, and I am perhaps even willing to bring about p. This clearly has a different meaning from Shall(~p), which indicates I intend for p not to occur. This is explicit in criticisms of the DDE: Simply not to intend the death of civilians is too easy. . . . What we look for in such cases is some sign of a positive commitment to save civilian lives. (Walzer 1977, p. 155; quoted in French 2003, p. 40) It’s not enough not to try to kill civilians; you have to try not to kill them. (Lichtenberg 2001, p. 7; quoted in French 2003, p. 41) There is a clear difference between saying, “I do not intend to kill the patient (but foresee that this dose of painkiller, given with the intention of relieving her suffering, might prove fatal)” and “I intend to not kill the patient (and hence will be conservative in my treatment of her pain).” The above considerations demonstrate that (a) there is, in ordinary language, such a thing as expressing not having an intention (as opposed to expressing an intention not to do something); (b) this notion of expressing suspension of attitude is perfectly coherent; (c) there is therefore a

138  Practical Reasoning and the Logic of Intentions difference in meaning between internal and external negation of expressions of intention; and (d) both types of expressions serve a legitimate purpose. Thus, we can reconstruct an individual’s reasoning using our intensional logic in a way that respects the doctrine of double effect. As noted above, Anscombe’s point about intention-­ascriptions placing items in a calculative order puts constraints upon allowable consequents in the conditionals upon which S-­Imp operates. Thus, “I pump water to the house” implies “I move the pump handle up and down” because the latter is a means to the former; these are related in the calculative order. But “I move the pump handle up and down” does not allowably imply “I cast a certain shadow on the rock” since the latter is obviously not a means by which the former is achieved. Similarly, for our hypothetical physician who foresees the possible death of her patient, we may use the following conditional in reconstructing her reasoning: “I relieve Smith’s pain” implies “I administer Smith n mg of painkiller” But we may not use either of the next two conditionals in reconstructing her reasoning: “I administer Smith n mg of painkiller” implies “Smith dies” or “I relieve Smith’s pain” implies “Smith dies” for the simple reason that the consequent isn’t in either case (for our physician) a means of achieving the antecedent. This wedding of Anscombe and Sellars achieves the nice result of meshing with the standard DDE requirement that the bad effect may not be a means of achieving the good effect if it is to be allowable. (Now, one may object to Sellars’s account and argue that we are responsible for the causal consequences of our intended actions even if they aren’t intended means of achieving our ends. But this would just devolve the argument into the standard set of debates over the DDE and is not specifically an argument over the details of Sellars’s account.) I will close by noting that Sellars’s requirement­—that logical operators must remain inside the scope of the “shall” operator­—is not a requirement common to logics of belief and intention. For example, Segerberg (1982) allows negation outside of the “Int” operator, Singh and Asher (1993) allow negation outside of both the “Intends” and “Believes” operators, and standard epistemic logics allow negation to operate on the K (“knows that”) operator. Sellars has a principled objection to external negation in the case of intention, but it seems clear to me that he needs external negation in order for his logic to be fully compositional.28

Practical Reasoning and the Logic of Intentions  139

Notes   1 This specific version of the argument is from Hale (1986, p. 71), although I have reordered the premises.   2 Hale (1986, p. 73).   3 See also Zangwill (1992, p. 181), who endorses Hale’s critical reading of Blackburn’s account of inference.   4 Stovall (manuscript, forthcoming) argues that a modified version of Gibbard’s plan-­theoretic expressivist semantics can be offered “to defend Sellars’ contention that moral reasoning is a species of collective intentionality” (Stovall forthcoming) and to develop an account of the logical relations among moral judgments. This is an interesting development, but I wish to see how far we can get by sticking (more or less) with Sellars’s original semantics and logic of intentions.   5 See also ORAV, §§37–41/p. 85.   6 “Logical operators do not occur outside expressions of intention, or, to put it in other words, expressions of intention do not occur within the scope of logical constants and quantifiers.” (ORAV §56/p. 87).   7 See ORAV §§63–4/p. 87.   8 In TA, Sellars claims that “there is no valid move from Shall [A], Shall [B], to Shall [A • B]” (TA, p. 114). He admits in VR that “it was not clear why I did so” (VR II.§39/p. 56) and embraces CI as a principle for making precisely this inference both there and in ORAV.   9 Of course, as we saw in Chapter 3, only we-­intentions can, strictly speaking, be inconsistent with each other in virtue of their contents. 10 Finally, as Tuomela has written, many groups are hierarchical and therefore have agents who act on behalf of the group­—that is, who are “operative,” in Tuomela’s terminology. Other members of the group share the intention that the group do A, but only the members who are operative for the group can, as it were, actualize this intention. But in doing so, they fulfill the collective intention of the group. See, for example, Tuomela (2007, chapter 6). 11 See van Roojen (1996). 12 Baker and Woods (2015) argue that the van Roojen problem can be solved with appeal to B-­type inconsistency. However, to appeal to this type of inconsistency would involve too radical a departure from the basic mechanics of Sellars’s theory, and as it stands I will already be forced to make fairly substantial revisions to the theory, as we will see shortly. 13 Lucas Thorpe raised this objection; this example invalid inference is also his. 14 An example along these lines is suggested in Segerberg (1982). 15 See, for example, SM (7.III.§§15–18/pp. 182–3), FCET (II-­III.§§7–11), and ORAV (IV.§§37–41/p. 85). 16 For example, “An ideally rational being would intend the implications of his intentions, just as he would believe the implications of his beliefs” (SM 7.III.§16/p. 183). More radical is Sellars’s suggestion in ORAV: “In most contexts in which I refer to intentions, the latter are to be construed as intention candidates or intendibles. . . . Roughly, intentions stand to intendings as propositions to believings” (ORAV III.§38.n. 5/p. 85). This suggestion cannot be correct, though: Propositions obey (of course) ordinary propositional logic, and so if intentions were analogous to propositions, we wouldn’t need a special “logic of intentions” in the first place. 17 A particularly well-­developed intensional logic­—and the one I rely on in this paragraph­—is laid out in Fitting (2004, 2015). 18 Lucas Thorpe raised this concern. 19 This is Sellars’s principle, introduced in Chapter 3, according to which ought-­ to-­be’s imply ought-­to-­do’s.

140  Practical Reasoning and the Logic of Intentions 20 Jack Woods suggested that the following kind of case might be particularly problematic for Sellars. 21 For Sellars’s account of how one reasons with So-­be-­it and the resulting complex intentions, see ORAV §§83ff/pp. 89ff. 22 This formalization is somewhat different from any version appearing in CDI but is simplified for clarity. Also, in CDI, Sellars uses b to represent “not paying reparations,” but I reversed this to make it less confusing. 23 It is important, in this example, to understand this as a prediction or claim about what I am going to do rather than as an expression of an intention. 24 Jack Woods pressed me on this objection. 25 See, for example, ORAV §56/pp. 86–7, where Sellars explicitly rejects the idea that expressions of intention might be connected by disjunction or conjunction. 26 Bana Bashour suggested that one might restrict Sellars’s rule to negation and not to logical operators (like conjunction and disjunction) linking two propositions. 27 I am understanding “Murder is not wrong” as “Murder is permissible” and translating this as I suggested permissibility be translated above. If we understand the sentence as saying “Either murder is wrong or murder is obligatory,” then it would translate as “Shallwe (M) ∨ Shallwe (~M).” 28 Having said all this, it is possible that Sellars himself accepted external negation only for the expressions of we-­intentions that are the constitutive state expressed by moral judgments. As Preston Stovall (forthcoming) notes, Sellars makes the following claim (note the use of the word “ordinary” in the first sentence): “I suggest that the fact that ought-­statements, unlike ordinary shall-­statements, have a proper negation is built on the shared intending expressed by ‘ought.’ In other words, the syntactical intersubjectivity of ought-­statements which permits Dick to contradict Tom [regarding what ought to be done] and which consists in the existence of the form ‘not-­ought [X doing A]’ in addition to the form ‘ought [X not doing A]’ rests on the intersubjectivity of the intention expressed by ought-­statements. Notice that I am not saying merely that the existence of shared formally universal intentions is a sociological condition of the existence of intersubjective ought-­talk. I am saying that intersubjective ought-­talk contains within itself the ‘symbolic form’ which is the very existence of intending-­as-­one-­of-­us” (IILO 12/p. 204). Also, as Stovall notes, Sellars claims repeatedly that the moral ought has both an internal and an external negation, and if a we-­intention (of a specific sort) is the constitutive state expressed by the moral ought, the implication is that this type of we-­intention also has an external negation. Klemick (forthcoming) also seems to interpret Sellars as allowing external negation for moral we-­intentions due to their intersubjectivity. Nevertheless, Sellars never explicitly says that moral we-­intentions have an external negation, so I am reluctant to conclude that this was his view.

6

Material Practical Inference

6.1 Practical Reasoning and Nomologicals Although Sellars’s account of inference among intentions is sophisticated, it does involve some oversights and oversimplifications. One such oversimplification is that he and his commentators often treat inference among intentions as though intentional contents were related by nomological (or even logical) necessity. (As we will see, this treatment leads to a flawed account of practical inference.) Thus, when he introduces inference among intentions, Sellars says that “Shall-­P” entails “Shall-­Q” just in case P entails Q. The plausibility of this simple account is aided by the fact that in the examples Sellars uses, the performance of Q is generally a necessary condition for the performance of P, and so P really does entail Q. Thus, when Sellars uses an example such as “(Shall [Jones’s shortly taking all the boxes to the station) implies (Shall [Jones’s shortly taking this box to the station])”1 (IILO, p. 173), the implication is made quite plausible for the simple reason that “Jones takes all of the boxes to the station” logically entails “Jones takes this box to the station.” Sellars too often reinforces this mistaken understanding of practical reasoning. In his analysis of hypothetical imperatives from SM (7.VIII), Sellars proposes to analyze the “nomologicals” that underlie hypothetical imperatives, the nomologicals that “concern the causally necessary conditions for bringing about a certain kind of state of affairs in a certain kind of circumstances” (SM 7.VIII.§51/p. 194–5, emphasis added). He writes, “They have the form, ‘Doing Ai if [in] Cj is causally necessary to the realization of Ek’” (SM 7.VIII.§51/p. 195). Put in terms of “causal implication,” this amounts to the following: “‘x brings about Ek’ implies ‘x does Ai if in Cj’. . . . Transposed into practical discourse as a general hypothetical imperative, [this becomes] ‘I shall bring about Ek’ implies ‘I shall do Ai, if in Cj’” (SM 7.VIII.§51/p. 195).2 Again, one sees the (mistaken) thought that practical reason involves reasoning from an end (embodied in an intention) to a means that is a necessary condition of achieving this end. Practical reasoning seldom displays this feature, however, as there is almost always more than one means available for

142  Material Practical Inference achieving a given end. That is, performance of any one means is almost never a necessary condition on the achievement of a given end. Sellars barely retreats from this extreme position in his later writing. In ORAV, he writes, “Strictly speaking, the core of ‘If Jones wants S, he ought to do A’ is ‘If Jones wants X he must do A,’ for that is the proper force of the implication on which it is grounded, which tells us that Jones doing A is a necessary condition of Jones getting X” (ORAV §107/p. 93). An all-too-minuscule degree of stance-­softening appears in the next paragraph, where Sellars writes, “The ‘must’ can be weakened by a ceteris paribus clause to allow for optional alternatives, but this root idea remains that of a necessary condition” (ORAV §108/p. 93).3 Matters are not helped by the fact that commentators on Sellars often use examples with similar features. For example, Solomon (1977, p. 166) rehearses the previously offered example of inference among intentions: “ShallJones (Jones will take all of the boxes to the station) implies ShallJones (Jones will take this box to the station)” As noted above, this is obviously a good inference among intentions on Sellars’s account because “Jones will take all of the boxes to the station” logically entails “Jones will take this box to the station,” assuming the additional premise “This is one of the boxes” (a premise Solomon states explicitly). —and like Sellars’s­ —are fundaHowever, examples like Solomon’s­ mentally misleading about the nature of practical inference. With most inferences involving intentions, inferential relations among the contents of intentions are not so straightforward. “I eat dinner” doesn’t entail­— logically or nomologically­—“I make a quiche lorraine” or “I make a cheese sandwich” or “I get Chinese takeout from Royal China.” There is almost always more than one means to an end, and so performance of no one means is a necessary condition of the accomplishment of the end. Thus, practical inference doesn’t typically involve relations of strict logical or nomological necessity between end and means; more often, it shows valid paths of practical inference available to the agent­—paths from an end to satisfactory means for achieving this end. Thus, an agent might be hungry and form the intention Shall [I eat dinner] Keeping in mind that inferential relations among intentions are parasitic on inferential relations among their contents, it is clear that “I eat dinner” in no way formally entails, for example, “I get Chinese takeout from Royal China.”

Material Practical Inference  143 Bruce Aune (1977) is one commentator who recognizes that practical inference is not as straightforward as Sellars (and many of his expositors) has made it out to be. Writing about the expression of the intention “Shall (I to do A),” Aune notes that Sellars ties intention to action through the following principle: S(p) if SD (I do what I can to make it the case that p) where S is an older operator meaning “It shall be the case that . . .” and SD means “Shall do.” However, Aune describes this way of connecting intention to action as “unsatisfactory”: “If the right hand side means ‘I’ll do whatever I can do to make it the case that p,’ it is patently unsatisfactory, since my available means to a given end are typically incompatible (I can do one thing only by not doing others)” (1977, p. 21). There is a more satisfying account of practical inference available to us. However, to develop it, we must first explore a notion that was central to much of Sellars’s philosophy and that will prove central to much of this book­—the notion of material inference. A consequence of this exploration is that we can bring Sellars’s practical philosophy­—and his account of practical inference­—more into harmony with the rest of his philosophical picture, in particular by taking the notion of material inference, which is so central to Sellars’s project, and putting it to work in his ethical theory.

6.2 Material Inference As noted above, the examples often used­ —both by Sellars and by commentators on Sellars­—to illustrate the notion of inference among intentions are misleading, in that they use logically necessary­—indeed, deductively valid­—inference as a model for practical inference. But as we saw, practical inference is seldom so neat and tidy. In any case, though, formally valid inference was never the type that most interested Sellars. For him, material inference was prior in the order of explanation. It is worth dwelling for a moment on the priority of materially to deductively valid inference, as the notion of materially valid inference will play a crucial role in several chapters. Sellars’s argument for why material inference rules are essential to language (in that they determine the meaning of a language’s descriptive terms) defies short summary; his argument is spread out over several extraordinarily difficult articles (for example, CIL, CDCM, and IM). At the core of this argument is the idea that in using any descriptive concept, you undertake commitments with modal force. (This claim is what Brandom calls the “Kant-­Sellars thesis”; see Brandom 2015.) Thus, if you claim or believe something is copper, you are committed to its

144  Material Practical Inference being metal. If you believe something (a tone) is A above middle C, you are committed to its not being red (synesthetes excluded, I suppose). As Sellars argues in CIL, if these “material invariances” didn’t hold, then these universals (“red,” “copper,” etc.) simply would be different universals. The laws that govern the use of descriptive terms make them the descriptive terms they are. A material rule of inference­—“If x is A above middle C, then x is not colored”­—merely makes explicit a commitment one has already undertaken by using the relevant terms in the first place. To say, of course, “If x is A above middle C, then x is not colored” is not merely to talk about actual tones or to assert that in point of fact all As above middle C are not colored­—it is to talk about possible tones as well as actual ones. Anything that could count as A above middle C could not be colored. Thus, the claim embodies a sort of necessity. And, as Sellars argues, In the case of [logical, modal and normative predicates], it is obvious that their conceptual meaning is entirely constituted by their “logical grammar,” that is, by the fact that they are used in accordance with certain syntactical rules. . . . Why is it obvious (once we escape from the mental eye) that the conceptual meaning of a modal or normative term is constituted by its logical grammar? Because it is obvious that it cannot be constituted by the term’s being a learned response to a class of extra-­linguistic particulars. A modal or normative property (if we permit ourselves to speak of them as such) cannot significantly be said to be exemplified by a particular (or pair of particulars). (IM, p. 334) If modal claims aren’t meaningful with reference to extralinguistic particulars, then how do we account for their meaning? Sellars’s suggestion is that the sentence “If x is A above middle C, then x is not colored” conveys “the same information as a logical rule permitting the inference” of “x is not colored” from “x is A above middle C” (IM, p. 323). It follows, then, that the meaning of a descriptive term­—such as “copper” or “A above middle C”­—embodies a set of rules governing material inferential proprieties. Rules of the form “x is not colored” may be inferred from “x is A above middle C,” and “x is metal” may be inferred from “x is copper.” Thus, Sellars holds that material rules of inference are covertly metalinguistic, in that they issue tickets to infer one sentence or claim from another. (We will return to these themes in Chapter 9.) One of the most eloquent advocates for the priority of material inference is, of course, Brandom. Brandom (2000, p. 88) asks us to consider an inference of the following sort: The match has been struck Therefore, the match will light

Material Practical Inference  145 A feature of material inferences of this type is that they appear to be non-­monotonic. What this means is that adding more premises can actually defeat the inference: R may follow from P but not from P and Q. Thus, if we add the premise The match is wet the inference is defeated; the conclusion doesn’t follow. By contrast, deductively valid inference is monotonic: That is, adding more premises can never get you fewer conclusions. Thus, if P entails R, then P and Q also entail R. As Sellars (IM, p. 313) and Brandom both note, there is a strong tendency among philosophers (particularly those who are empiricist-­ minded) to regard supposedly material inferences like the one above as enthymemes, possessing suppressed premises. The added premise contradicts the suppressed premises, explaining the defeat of the inference. Thus (goes the argument), the above argument properly reads The match has been struck (If a dry, well-­made match is struck, then it will light) (The match is dry and well made) Therefore, the match will light We may call philosophers who reinterpret material inferences as enthymemes formalists, as they interpret all valid inferences as deductively valid. Which interpretation of theoretical inference should we prefer? Should we interpret theoretical inferences as material (and hence non-­monotonic) or enthymematic (and hence deductively valid)? Brandom argues that we should take the former route. For one thing, we don’t know how to spell out the enthymematic premises for most such inferences. As Brandom writes, “The problem is not just that we would need an infinite list of the conditions being ruled out­—though that is true. It is that the membership of such a list would be indefinite: we do not know how to specify in advance what belongs on the list” (Brandom 2000, p. 88). To illustrate, Brandom gives the following example of a material inference that “oscillates” from defeat to non-­defeat and back again: (1) If I strike this dry, well-­made match, then it will light (p → q). (2) If p and the match is in a very strong electromagnetic field, then it will not light (p & r → ~q). (3) If p and r and the match is in a Faraday cage, then it will light (p & r & s → q). (4) If p and r and s and the room is evacuated of oxygen, then it will not light (p & r & s & t → ~q).

146  Material Practical Inference At this point, notes Brandom, the formalist is tempted to insert a ceteris paribus clause, saying something like “Ceteris paribus, q follows from p.” But Brandom argues that this is equivalent to saying that “q follows from p unless there is some infirming or interfering condition. But this is just to say that q follows from p except in the cases where for some reason it doesn’t” (Brandom 2000, p. 88). The ceteris paribus clause is not shorthand for a list of defeaters; it cannot serve this role for reasons Brandom has explained. Essentially, Brandom argues, “Ceteris paribus clauses should be understood as explicitly marking the nonmonotonicity of an inference, rather than as a deus ex machina that magically removes its nonmonotonicity” (Brandom 2000, pp. 88–9). In general, Brandom argues, one should “prefer an order of explanation that begins with what is implicit in practice (what people do) and proceeds to an account of what they explicitly believe or say, over one taking the opposite tack” (Brandom 1994, p. 101).4 If one is not already able to see that “The match will light” is a good conclusion to draw from “This dry, well-­made match has been struck,” then insertion of the major premise­— “If I strike this dry, well-­made match, then it will light”­—will scarcely help matters. The major premise merely makes explicit what one already knew. The formalist claims that if you have an argument such as P. Fido is a dog R. Therefore, Fido is a mammal the insertion of a major premise explicitly linking the premise to the conclusion (e.g., “If Fido is a dog, then Fido is a mammal”) is necessary to explain both (a) our psychological ability to transition from premise to conclusion and (b) the normative goodness of such a transition. However, rather than explaining the possibility and goodness of such a transition, the formalist account in fact makes such an explanation impossible. Only the Sellars/Brandom account leaves open a possible route of explanation. I think this was the point Lewis Carroll was making in “What the Tortoise Said to Achilles.” There, Achilles is trying to explain inference to Tortoise. Consider the following truncated version of the argument presented by Tortoise: B. The two sides of this Triangle are things that are equal to the same Z. The two sides of this Triangle are equal to each other The formalist will say that this argument isn’t valid because there is no premise linking B and Z. For this reason, no doubt, the version of the argument presented by Tortoise includes the following premise: A. Things that are equal to the same are equal to each other

Material Practical Inference  147 The problem is that if you consider that an argument cannot go through if there is no premise linking premises and conclusion, then the addition of A doesn’t solve the problem­—because now there is no explicit link between A and B and the conclusion (Z). With this in mind, Tortoise slow-­wittedly asks Achilles why he should infer Z from A and B. Achilles explains that Z follows from A and B. Tortoise treats this as another hypothetical and adds it to the argument: C. If A and B are true, Z must be true This, of course, means that there is no premise linking the new set of premises (A and B and C) to the conclusion (Z), and so on the assumption that there is a requirement that premises be linked to the conclusion by an explicit hypothetical, Tortoise asks Achilles for a new hypothetical linking A and B and C to Z. Of course, each time Achilles adds a new hypothetical, Tortoise professes not to see how the conclusion follows from the new set of premises. A frustrated Achilles argues: If you accept A and B and C, you must accept Z. But why must I? Because it follows logically from them. If A and B and C are true, Z must be true. You don’t dispute that, I imagine? If A and B and C are true, Z must be true, Tortoise thoughtfully repeated. That’s another Hypothetical, isn’t it? And so Tortoise adds another premise: D. If A and B and C are true, Z must be true and another: E. If A and B and C and D are true, then Z must be true and so on ad infinitum (Carroll 1895). As I said, Carroll’s argument is an ingenious illustration of Sellars’s two points. First, if one is not psychologically able to infer Q from P, then adding the major premise doesn’t help explain one’s ability. Rather, our ability to understand and follow the major premise presupposes a prior ability to infer in the first place. Indeed, if you think our ability to infer a conclusion must be explained by a hypothetical that explicitly links premise and conclusion, then you are in trouble because adding a hypothetical does not satisfy that explanation. It merely creates a new demand: How is one to infer Q from the new set of premises­—namely, the original premise (P) plus the new hypothetical (P → Q)? One needs,

148  Material Practical Inference now, an additional hypothetical [(P & (P → Q)) → Q] to connect the new set of premises to the conclusion, and the infinite regress is off and running. To be sure, “Fido is a mammal” follows from “Fido is a dog” and “If Fido is a dog, then Fido is a mammal.” But “Fido is a mammal” also follows simply from “Fido is a dog,” and someone who knows what a dog is and knows what a mammal is but who is unable psychologically to infer the conclusion from the simpler premise needs more than logic lessons (and arguably doesn’t actually know what “dog” and “mammal” mean).5 Second, and for precisely parallel reasons, the normative authority of the inference does not rely on the presence of the major premise. If the normative authority of any inference from premise to conclusion relies on a major premise linking the two, then a vicious regress looms for exactly the reasons just rehearsed. The normative authority to derive “Fido is a mammal” from “Fido is a dog” derives not from any major premise but instead from the meaning of “dog” and “mammal”­—that is, from the inferential proprieties constituting the meanings of the relevant terms. This argument­—that our cognitive capacities cannot be explained (psychologically or normatively) in terms of the following of a formal rule­—famously appears in Wittgenstein before reappearing in Sellars. I will return to this argument in Chapter 9. I will also have more to say about material inference in Chapter 8, where I discuss in detail contemporary accounts of defeasible inference.

6.3 Material Practical Inference Brandom argues that the above account of theoretical inference can be extended to practical inference. Thus, while an inference like The traffic signal is red Therefore, I shall apply my brakes is non-­monotonic and subject to defeat in various circumstances, that does not mean that it is an enthymeme or an otherwise disguised or truncated formally valid inference. The impulse to argue that it is only a good inference with the insertion of an intermediate premise to the effect that I desire to stop at red lights results from the same formalist prejudices that pushed us in the direction of claiming that only deductively valid theoretical inferences can be good theoretical inferences. (We shall return to this issue­—practical inference and the Humean philosophy of mind­— at greater length in Chapter 10.) What bearing does that have on the present topic? As noted above, in the examples that are often given to illustrate the logic of intentions (including the examples Sellars himself gives), there is often a relation

Material Practical Inference  149 of necessary or strict logical entailment between antecedent and consequent. Recall Solomon’s example: “ShallJones (Jones will take all of the boxes to the station) implies ShallJones (Jones will take this box to the station).” Again, though, examples like this, where the contents of the first intention logically entail the contents of the other, encourage subsumption of practical inference under the formalist, deductivist model. But as we have already noted, practical inference is seldom so straightforward. Practical reasoning takes two different forms that are of interest to us. The first type of practical reasoning is where a means of achieving the end is not a necessary condition for achieving the end, but it is nevertheless plausible to represent the inference from end to means of achieving the end as a material inference. That is, given the end and the circumstances at hand, a means presents itself to the agent as the obvious or most plausible one. Thus, the agent can move directly from the end to a means without a need for intermediate deliberation. For example, if someone e-­mails me asking for a copy of my paper on classical foundationalism, I would first adopt an end, represented by the intention Shall (I send my paper on classical foundationalism to Jones) Without deliberating on method, I immediately—as a direct consequence of adopting the first intention­—adopt the next intention, which represents a means of fulfilling this end: Shall (I hit “reply” and attach the pdf of my paper on classical foundationalism) Now, despite Sellars’s claims to the contrary, there is no relation of nomological necessity between “I send my paper on classical foundationalism to Jones” and “I hit ‘reply’ and attach the pdf of my paper on classical foundationalism.” Were I to stop and think about it, I could imagine a number of other ways of getting my paper to Jones: I could print out a copy, look up Jones’s mailing address, and mail a copy to her. I could open a copy on my computer, laboriously copy it out by hand, and hire (at great personal expense) a courier to hand-­deliver it to Jones. I could send it to Jones in a series of text messages or carrier pigeon messages. These options range from suboptimal to ridiculous, and of course I never even consider any of them. I simply move from end to means even though the end did not necessitate this specific means. But, as I hope the previous section has made plausible, this is quite frequently the way it goes with material implication: Even when it is defeasible­ —that is, even when the premise doesn’t logically (or nomologically) necessitate the

150  Material Practical Inference conclusion­—the move from premise to conclusion is materially valid. Its validity does not require the insertion of other steps, and crucially, we need not explain the agent’s ability to make this inference by positing any intervening deliberative steps on the part of the agent. Thus, in the first form of practical reasoning, the move from the end (represented by an intention) to a means (represented by the intention adopted to fulfill the end) is best seen as direct, without any intervening deliberative steps. It is best seen as a material practical inference. However, there is a second form taken by practical reasoning, and this second form resists this simple treatment. The second form taken by practical reasoning is when there are numerous, incompatible means of satisfying one’s intention, none of which is obviously indicated, and one must choose among the various means. Thus, Shall [I eat dinner] doesn’t strictly entail any particular means of satisfaction; nor is it plausible to claim that a single means (among other possible available means) is usually materially inferable from this intention. Thus, unlike in the first form of practical reasoning, it is not entirely plausible to represent the inference from Shall [I eat dinner] to Shall [I go to Royal China] as a material inference. There is an intuitive difference between   (5) Shall (I send my paper on classical foundationalism to Jones).  (6) Shall (I hit “reply” and attach the pdf of my paper on classical foundationalism). and   (7) Shall [I eat dinner].   (8) Shall [I go to Royal China]. Plausibly, (6) follows from (5) via material implication. But there is more distance between (7) and (8). Even if we reject formalism, it seems like some kind of reasoning or deliberating is required to get us from the premise to the conclusion. Thus, I do not think it is plausible to represent the move from (7) to (8) as a material inference­—although, as we will

Material Practical Inference  151 see, practical reasoning of this sort will typically still involve material inference. So, if the move from (7) to (8) is not a material inference and the contents of these intentions are not related by logical or nomological necessity, then how do we reason from one intention to the next? Does Sellars offer us any guidance? A helpful suggestion is offered by Bratman. On Bratman’s picture, one’s intention sets a problem (eating dinner) and also constrains “admissible solutions to that problem” (1999, p. 35). However, the intention itself doesn’t determine a unique solution to the problem. While rejecting the belief-­desire model of intentional action (both as a description of human action and as a general normative account of reasons), Bratman says that here is where one’s belief-­desire reasons can help one sort among available solutions to a problem. While my intention itself doesn’t determine a unique means of satisfaction, my intention plus my relevant beliefs and desires may do so (and can even help decide between equally satisfactory options­—“Buridan cases,” as Bratman calls them). Thus, I might reason as follows: Let φ represent the complex array of desires and factual beliefs entered into the scope of the intention via So-­be-­it; φ will run, roughly, “[Quiche lorraine sounds good, but I would have to make it from scratch, and I don’t feel like it] • [I don’t want a cheese sandwich] • [Chinese food sounds good to me, and Royal China is open, and I don’t mind driving there],” etc.   (9) Shall [φ • I eat dinner]. (10) “φ • I eat dinner” implies “I go to Royal China.” (11) Shall [φ • I eat dinner] implies Shall [I go to Royal China]. (12) Shall [I go to Royal China]. A few comments are in order about this argument. First, we see that the notion of material implication has made a reappearance in premise (10). As I noted (and lamented) above, Sellars repeatedly claims that the entailments that underlie hypothetical imperatives of the sort expressed by premise (11) embody nomological necessities. Clearly, though, the implication in (10) expresses no such necessity. Even given the various beliefs, desires, and other conditions laid out in φ, “I go to Royal China” doesn’t follow from “φ • I eat dinner” as a matter of necessity. But one infers “I go to Royal China” as a sensible means of satisfying the various constraints laid out in the antecedent, not as the strictly necessary means. Thus, it is an instance of (highly) defeasible material implication. Second, we can see from the above a concrete way in which, as Bratman argues, intentions are elements in partial plans.6 The intention “Shall [I eat dinner]” doesn’t entail, as a matter of logical or nomological necessity, any particular means of satisfaction. But it “sets the stage” for planning.

152  Material Practical Inference The further elements of the plan are then cashed out by one’s other intentions (“Shall [I never darken the door of Bill’s Burger Barn again]”), one’s circumstances (Royal China is open), one’s desires (I don’t want a cheese sandwich), and so on. This model of intention more accurately represents practical reasoning (and the role of intention in practical reasoning) than does Sellars’s original, extremely rigid nomological-­based model. The third comment to make about the above argument (premises 9–12) regards its consistency with Sellars’s philosophy of mind. Bratman’s suggestion is that the gap between end and means can be filled in with the agent’s beliefs and desires. What role do desires play, however, in Sellars’s practical psychology? In TA, Sellars gives a technical definition of desire according to which “desires are relatively long-­ term dispositional intentions” (TA, p. 117). Desires have some other unique features (such as their conceptual connection to enjoyment when contemplating the satisfaction of the desire in question), but the main thing to point out is that desires are defined in terms of intentions and so, it seems, cannot play the role in Sellars’s theory envisioned by Bratman. Sellars similarly defines other technical notions (such as “preference” and “favorably disposed”) in terms of intentions, and so there is no help from that quarter either. But perhaps we have been too hasty in concluding this, for Sellars quickly admits that “the word ‘desire’ has in ordinary usage a more specific meaning than the technical usage I have in mind” (TA, p. 117). Toward the end of TA, he writes, Let me emphasize once again that not all current uses of the term “desire” can be explicated along these lines. As far as I can see, the term is sometimes used in such a way that desires constitute a subspecies of the sort of thing I have been talking about, so that it would remain to spell out the differentia. Sometimes, on the other hand, the term is used to refer to what I would call “hungers” or “appetites”— i.e. dispositions not to intend, but to enjoy or disenjoy Y-­ing. (TA, p. 130) Thus, Bratman’s solution­—with Sellars’s gloss on it­—seems like a very plausible solution to the problem of multiple means. If I form the intention, “Shall (I eat dinner),” additional thoughts along the lines that I would enjoy Chinese but would not enjoy a cheese sandwich and do not feel like making quiche lorraine seem, ceteris paribus, to give me a reason (conditional upon my original intention) to go to Royal China. Fourth, the above account is compatible with a model on which choices like these (where there are multiple, roughly equal means of satisfying an end) are chosen via a satisficing strategy­—a strategy that is itself rational, as it minimizes the cost of deliberation.7 Thus, when deciding where to go out to eat, we may be faced with several options and seek an option that

Material Practical Inference  153 meets a certain minimum standard rather than necessarily seeking the maximizing option. But this can be reflected in φ, which will lay out the options (“[Bob’s Burger Barn and Royal China and Tandoori Garden and Del Rio are open and nearby]”) but will also lay out a set of constraints or desires or other relevant factors (“[I had burgers yesterday and don’t feel like them today, and I don’t feel like Indian food, and Royal China is closer than Del Rio, and besides, we won’t have to wait to be seated there]”). From these considerations, we get (11) Shall [φ • I eat dinner] implies Shall [I go to Royal China]. where the “imply” is read not in the sense of logical or nomological necessity but as a weak and defeasible kind of implication. One final comment: From the fact that material inferences are not enthymemes, it does not follow that all material inference is immediate. Surely, some are; the inference from “Shall (I send my paper on classical foundationalism to Jones)” to “Shall (I hit ‘reply’ and attach the pdf of my paper on classical foundationalism)” is immediate. But there are also chains of material inference. For example, I might reason as follows: “It is raining; therefore, the streets are wet; therefore, I will take my car rather than my motorcycle, which has poor traction when the streets are wet.” In this chain of material inferences, the first inference is theoretical and the second practical. Much of our reasoning exhibits this pattern; it is not immediate but follows a chain of reasoning to a conclusion, with some links of the chain being theoretical, some practical. In saying that practical reasoning from an end to a means satisfying this end involves material inference, I am not committed to the claim that it involves a single inference­—or only practical inferences­—rather than a chain of potentially mixed inferences.

6.4 Objections One objection is that Sellars actually does recognize that there are multiple means of fulfilling ends, and this recognition is built into his account in a way that preserves the underlying nomological necessity.8 In our example above, filling out φ will involve a fair amount of ruling out alternatives. Thus, one might argue that once various alternatives are ruled out and the content of my intention “Shall [φ • I eat dinner]” is fully spelled out, we will see that there is a strong logical connection between “φ • I eat dinner” and “I go to Royal China.” Support for this interpretation can be seen, for example, in IILO (§102/p. 92), where Sellars writes that “Jones gets X” implies “Jones does A” on “a proper spelling out of just what it is that Jones wants” (emphasis added). I think we should reject this reasoning for the same reason that Brandom rejects the attempt to render all theoretical reasoning enthymematic and

154  Material Practical Inference deductively valid. It is an attempt to reassimilate practical reasoning to the formalist paradigm and should be rejected for the same reasons we rejected formalism in theoretical reasoning. A key reason is that there is in the paradigm case of practical reasoning no way to establish a strong logical relationship between antecedent and conclusion for the simple reason that there is no way to rule out all of the various alternatives. And again, this isn’t because there are many alternatives­—rather, it is because the list of alternatives is indefinite. What are all the alternative methods of my getting my paper on classical foundationalism to Jones? I could take advantage of the perpetual coolness of all things retro, crowdfund a new telegraph network, and send it to Jones via Morse code. I could have a skywriter write it in the sky over Jones’s house. The list of alternatives is open-­ended, and reasoning from antecedent to conclusion in practical reason will almost never involve ruling out all alternatives. Trying to assimilate practical (or theoretical) reasoning to the formalist, deductive paradigm should be seen as a failed project. (I will return to this theme in Chapter 8, when I discuss defeasible rules and non-­monotonic reasoning in more detail.) Second, one might wonder if material practical inference counts as material inference on Sellars’s account.9 In the theoretical realm, material inference links concepts internally in terms of their content. For example, the inference from “Fido is a dog” to “Fido is a mammal” is a materially good inference because of the content of “dog” and “mammal.” Similarly, the inference from “The dry, well-­made match is struck” to “The match will light” is a materially good inference because of the content of the central terms. A match just is the kind of thing that lights when struck. But if I want to light a fire, and I decide to strike a match, how can this be represented as an inference? How can striking a match be fairly represented as part of the content of “I light a fire” or “Shall (I light a fire)”? The key to answering this question lies in Anscombe’s insight that “if we should want to give conditionals which are logical truths, which we might think of as giving us the logically necessary connexions which ‘stand behind’ the inferences, they will be exactly the same conditionals for the practical and for the corresponding ‘theoretical’ inferences” (Anscombe 1995). The only difference, Anscombe argues, is the use to which the conditional is put. Thus, consider the conditional If this dry, well-­made match is struck, then it will light Anscombe notes three uses we can make of this conditional (though there may be more). For example, we can use it in theoretical reasoning. That is, we can start with a premise (“I strike this dry, well-­made match”) and predict a conclusion (“The match will light”). In this case, we reason from antecedent to consequent. Or, we can use this conditional in

Material Practical Inference  155 explanatory reasoning. Here, we reason in the opposite direction­—from consequent (“The match is lit”) to antecedent (“The match was struck”). Finally, we can use this conditional in practical reasoning. Again, we reason from consequent to antecedent:10 Shall (I light a fire) I light a fire implies I strike a match Shall (I strike a match) Although in practical reasoning we reason from consequent to antecedent, the practical inference is underlain by the same conditional that provides conceptual content to the concepts that are in play in the theoretical inference. This content-­constituting conditional is the “common core” underlying theoretical, explanatory, and practical reasoning and is the reason why all three are properly treated as inferences. One final comment is appropriate at this point. Technically, both of the following are inferences for Sellars, as the transition from antecedent to consequent is licensed by the contents of the relevant terms (and hence can be expressed by a material rule of inference): It is raining Therefore, the streets will be wet There is a dangerous axe murderer running amok Therefore, shall [I flee] It is a bit jarring, however, to hear the latter, in particular, described as an inference, in that it over-­intellectualizes our exercise of practical judgment. In fact, Sellars is keen to emphasize that all kinds of linguistic transitions­—be they language-­entry, language-­language, or language-­ exit­—are not actions at all. Sellars writes, “Such central kinds of thought as perceptual takings, inferences, and volitions are not actions for the simple reason that they are not the sort of thing which can be done intentionally or that one can decide to do. One can decide to look in the next room, but not to take there to be a burglar in the next room. Of course there are mental actions, thus, working on a mathematical problem or pondering what to wear. But . . . they consist of chains of thoughts which are not themselves actions” (NAO 4.II.§16/p. 66). This Sellars quote brings out two features of practical reasoning and judgment. First, much of our practical lives involves skillful engagement with the world, which involves transitions from judgment to intention to volition, and these transitions must be regarded as acts, not actions (in that they are not themselves executed intentionally­—on pain of vicious regress, it cannot be a general requirement that intentions be formed intentionally). Second, we do of course engage in practical deliberation that is properly characterized as intentional. We often carefully reason about what to do

156  Material Practical Inference before deciding on a course of action. But at the basis of our analysis of such actions must lie acts that are themselves not done intentionally­— again, on pain of infinite regress. Finally, whether an intention (and the corresponding volition) results from intentional deliberation or is the output of a transition characterizable only as an act, the resulting action will be, of course, an action since it will be done intentionally. (Thus, we must distinguish, as Sellars does, between whether a transition is an action versus whether the output of the transition is an action.) Looking Ahead I have outlined the two types of intentions in Sellars’s theory of practical reason and shown how Sellars’s theory is not only plausible but also answers a number of pressing objections to expressivist accounts of moral language. There are even more ways, however, that Sellars’s account of intentions can be wed to contemporary philosophical developments to provide a deeper understanding of moral reasoning. In the next chapter, I will argue that there are two distinct kinds of reasoning­— individual reasoning and cooperative reasoning­—and that the distinction between acting on an individual intention and acting on a we-­intention is best captured by this distinction between individual and cooperative rationality. I will also argue that intending from the moral point of view involves we-­intending, and that these moral we-­intentions issue from a distinct type of reasoning­—cooperative reasoning­—which is not reducible to or replaceable by individual reasoning.

Notes   1 Sellars seems to have borrowed this example from Hare (1952).   2 Sellars is clear that the “implies” is used here “without qualification to mean causally or physically implies” (SM 7.VIII.14/p. 181); he distinguishes this notion of implication carefully from the “‘p → q’ of truth functional logic” (SM 7.VIII.§14/p. 180).  3 Sellars is of course not wholly unaware of this issue and mentions it in ­correspondence. In particular, in CSA (23 June 1975), Sellars acknowledges that a rational agent will select a means whose performance is a sufficient (but perhaps not a necessary) condition of accomplishing her intended end. But this recognition doesn’t appear in a meaningful way in Sellars’s published work­—even, as I note, his later published work, such as ORAV.   4 And, of course, Brandom argues that that you can construct an account of logically valid inference from materially valid inference but not vice versa. See, for example, Brandom (2000, pp. 104–5).   5 As Olen and Turner write, the formal rules of a system “only exhibit a kind of ‘internal’ sense of normativity, the normativity of the rules of permissible transformation within the system itself” (Olen and Turner 2016, p. 2066). The normative force licensing various inferential transitions exists within the practice and does not reside in the rules that express these preexisting normative proprieties.   6 See, for example, Bratman (1999, p. 29).

Material Practical Inference  157   7 Lucas Thorpe argued that such decisions often follow such a model.   8 I owe this objection (or one in the neighborhood; the actual elaboration of the objection is mine) to Arudra Burra.   9 This objection was raised by Dionysis Christias. 10 I’m not sure the order of reasoning (antecedent to consequent or consequent to antecedent) always aligns so tidily with the type of reasoning being used. But I’m not sure this point is of great importance. The more significant point is that different types of reasoning are underlain by the same content-­ constituting conditionals, which are merely put to different types of use in the different types of reasoning.

7

Cooperative Rationality and We-­Intentions

7.1 Cooperative Rationality Many discussions of rationality assume that all rationality must be ­conceived of as individual: Individual agents are acting on intentions with contents that might at most be parallel (in Sellars’s terminology) but cannot be shared. To use Sellars’s example, Tom and Dick may each have an intention of the form “It shall be the case that the war ends,” but this merely reflects an “agreement in attitude,” and “agreement in attitude is not an identity of intention” (FCET, §49). It is not clear, though, how one can get an account of moral reasoning out of individual rationality, as the ship of morality always runs aground on the rocks of free rider problems and the like. Consider a traditional problem for utilitarianism: Suppose there is a water conservation order in effect during a drought. Plausibly, it is utility-­maximizing for you to violate the order since it produces positive utility for you and no negative utility for anyone else. Your violation of the water conservation order will not contribute in any appreciable way to the city’s water shortage; after all, you are just one person! So an attempt to build morality on the foundation of individual rationality seems to ground morality in sand. The problem is not merely (or not even) that people are selfishly motivated­—the problem arises even in a utilitarian system that counsels impartial benevolence. Each person individually doing what is utility-­maximizing leads to an overall outcome that is not utility-­maximizing. Assuming the existence of some altruistic preferences does not help ground morality in individual rationality. As Elizabeth Anderson writes about what she calls the “altruistic evasion,” the “key feature of the payoffs that generates prisoner’s dilemmas is not the fact that people care only about the payoffs to themselves. It is the fact that any single person’s action, considered in isolation, has an expected marginal payoff of zero, or close to zero, with respect to the socially desirable outcome” (2001, pp. 25–6). But defection can have a non-­trivial payoff in terms of one’s personal utility (and therefore in terms of overall utility), and so it is difficult to defend cooperation on rational grounds. Anderson provides, as

Cooperative Rationality and We-­Intentions  159 a familiar case, voting (2001, p. 26). Suppose we agree that democracy is a public good. Suppose some subgroup of us wishes to elect candidate A, which can (of course) only happen if a substantial number of us go out and vote for A. But I may reason as follows: “The contribution of my vote to the election of A is negligible, and the probability of my vote determining the outcome of the election is for all intents and purposes nil. However, voting inconveniences me, and it may even inconvenience others, such as those in line behind me to vote. So I should not vote.” Obviously, if this reasoning­—which is perfectly rational from an individual standpoint­—is adopted by all members of our subgroup supporting A, then our collective goal is defeated. But how else are people supposed to reason? Sellars’s claim is that essential to the moral point of view is that we intend not individually but qua members of the larger moral community. Thus, there is a fundamental difference between intending from the individual and moral points of view. This suggests the further idea that there is also a fundamental difference between reasoning as an individual and reasoning as a moral agent­—that is, reasoning as a member of a group. When I reason morally, I am not just reasoning about what I should do in a particular case. Rather, as Sellars puts it, “Moral principles primarily concern the consequences of anybody acting in a certain way in a certain kind of circumstances” (SM, 7.XIII.84/p. 207). The idea that there is a notion of cooperative, or team, rationality not reducible to individual rationality is advanced by Margaret Gilbert (1989, 2001), Robert Sugden (2000), and others. We can show that there is a viable notion of cooperative rationality if we can give an example where it is clear that rationality can only be construed cooperatively, not individually. Sugden presents such an example with what he calls the Footballers’ Problem. Suppose A wishes to pass the ball to B so B can score. A can pass (and B can run) either left or right, but both must decide simultaneously, without communicating. Both can see that the chance of scoring is slightly better if both choose right instead of left (say, 11 percent versus 10 percent). If one chooses left and the other right, the pass fails and the chance of scoring is zero. The Footballers’ Problem can be presented in a traditional decision matrix (Table 7.1):

Table 7.1 player B

player A

left right

left

right

10,10 0,0

0,0 11,11

160  Cooperative Rationality and We-­Intentions What should the two players do? “The answer seems obvious: each should choose right,” writes Sugden. “But paradoxically,” he continues, “this obvious answer cannot be generated by the theory of individual rationality, as used in game theory” (Sugden 2000, p. 179). On the assumptions of traditional game theory, A will presume the rationality of B and will go right if he thinks it is rational for B to go right. But B is in precisely the same decision situation! He will go right if he thinks it is rational for A to go right. Thus, as Sugden concludes, “We have entered an infinite regress: what it is rational for a player in a situation like A’s to do depends on what it is rational for a player in a situation like A’s to do” (Sugden 2000, p. 181). Others have noted that the attempt to solve coordination problems from the perspective of individual intention generates circularity or regress problems. Tuomela writes, The paradigm case is a coordination problem with two or more equally good equilibria. There is a dilemma when the agents lack a reason to deconditionalize their conditional intentions to arrive at the same choice alternative. . . . Suppose that we wish to meet and that this can take place either by both going to the station (X) or to the church (Y). Our meeting of course requires that we go to the same place. Accordingly, choosing X, the agents’ rational formation and satisfaction of the intentions expressed by “I will do X” and “you will do X” are interdependent on conceptual grounds. . . . The interdependence can be formulated . . . in terms of a biconditional, respectively, “I will do X if and only if you will do X” and “I will do Y if and only if you will do Y.” (Tuomela 2007, p. 74)1 The solution to the problem is for A and B to think not about what is rational for each of them to do but what is rational for the team to do. And it is clearly rational for the team (meaning A and B) to go right. Again, though, this solution cannot be reached from the perspective of individual rationality; it can only be a product of team rationality. And so we have an example where rationality is irreducibly cooperative: It is a question of what we ought to do and cannot be reduced to separate questions about what I ought to do and what you ought to do.2 One might argue that a solution to the Footballers’ Problem can be reached from the perspective of individual rationality.3 Many decision theories make available to one, before one makes a decision, credences about which action one will perform in single-­player games. If you are player A, a natural thing to do (since you don’t know what player B will do) is give yourself a credence of .5 on either side. (Player B will, of

Cooperative Rationality and We-­Intentions  161 course, do the same.) Since R, R has a collectively greater value than L, L, the result will be that when the above credences are taken into account, each player will choose­—individually­—to go right. To see where the above objection goes wrong, it is best to outline the premises of the argument.4 The argument seems to proceed as follows: (1) I don’t know what the other player will do. (2) So, as a rational player, I attach a probability of .5 to his playing R. (3) Given this belief, rationality requires me to play R. (4) So, as a rational player, I will play R. So far so good. Notice, though, that the argument could continue as follows, where I represent the other player’s reasoning: (5) The other player doesn’t know what I will do. (6) But he is rational, so he will attach a probability of .5 to my playing R. (7) Given this belief, rationality requires him to play R. (8) So, as a rational player, he will choose R. One problem that becomes immediately apparent is that (8) contradicts (1), as now the argument claims both that I don’t know what B will do and that I do know (or at least can confidently predict) what B will do.5 Thus, instead of using individual reasoning to reason from “I don’t know what the other player will do” to a conclusion about what I ought to do, the argument instead seems to operate as a straightforward reductio of the claim that I don’t know what B is going to do! Thus, the argument seems to vindicate team reasoning rather than substitute for it. In general, Sugden writes, This kind of bootstrap reasoning (i.e., from the premise that I don’t know what the other players will do, I infer knowledge about what he will do) can’t be justified as an implication of the classical rationality and common knowledge. . . . I have presented the Footballers’ Problem as a problem for “classical” game theory, which I take to assume only that the players have common knowledge of the payoffs of the game and common knowledge that each of them is rational. In this framework, it is illegitimate to appeal to ideas about what it is “natural” for people to do. We already know that most people think it natural to play R­—the problem is to show that this is an implication of the classical assumptions!6 But aside from this, the very move from premise (1) to premise (2) is suspect. As Sugden writes, “This isn’t a case in which one can appeal to

162  Cooperative Rationality and We-­Intentions symmetry to (try to) justify equal probabilities­—the two strategies are clearly different. In general, it isn’t legitimate to go from ‘I don’t know whether x is true or false’ to ‘my subjective probability for x is 0.5.’”7 Some thinkers try to construct the group viewpoint from the tools of individual rationality. J.J.C. Smart and, most famously, David Gauthier have argued that defection (e.g., breaking a water conservation order) is irrational for the following reason: Suppose it would maximize utility for you to break the water conservation order. If it is rational for you to do so, then it is rational for all others to do so as well. But since all others doing the same would result in overwhelmingly negative consequences, it is therefore not rational for you to break the water conservation order. However, Gauthier and Smart are trying to sneak team rationality in through the back door. If we insist on individual rationality, then we can ask only what is rational for you to do. And if your breaking the conservation order would not cause others to do the same, then it is utility-­maximizing for you to break the order and irrational for you not to do so. Thus, their argument fails. But when we move to a model of cooperative rationality, we can see that it is rational for us to obey the water conservation order, and hence to restrict our water usage. One of the strongest arguments that can be made for cooperative rationality is as follows: We use different models of reasoning in different circumstances. For example, Rawls famously argues that while maximin reasoning is inappropriate for a wide range of cases, the original position displays a range of features that make maximin reasoning appropriate (Rawls 1971, pp. 154–6). But if a particular model of reasoning is unable to secure one’s instrumental goals­—whereas a different model of reasoning is able to secure these goals­—then it is simply irrational to employ the former model of reasoning in preference to the latter. Natalie Gold and Robert Sugden make a similar point: “Following Bacharach, we define a mode of practical reasoning for a given class of games as valid if it is success-­promoting: given any game in that class, it yields only choices which tend to produce success, as measured by game payoffs. . . . Such reasoning is instrumental in that it takes the standard of success as given; its conclusions are propositions about what the agent should do in order to be as successful as possible according to that standard” (2007, p. 121). By this measure, individual reasoning is invalid in a variety of circumstances/games, as (a) it fails to produce the best outcome for agents and (b) there is an alternate mode of reasoning­ —namely, cooperative reasoning­ —which produces better outcomes. And there are cases where individual reasoning is simply unable to produce the clearly rational answer. We have already seen the Footballers’ Problem; Gold and Sugden also offer the traditional Hi-­Lo game (Table 7.2):

Cooperative Rationality and We-­Intentions  163 Table 7.2 player 1

player 2

high low

high

low

2,2 0,0

0,0 1,1

(Outcomes are listed as payoffs rather than preference orderings.) As Gold and Sugden point out, in traditional game theory, there is no unique Nash equilibrium, and the overwhelmingly obvious solution­—that you should pick “high”­—can only be achieved from within traditional game theory by assuming the irrationality of the players (Gold and Sugden 2007, pp. 119–20).8 Clearly, cooperative rationality is a valid and, what is more, irreplaceable mode of practical reasoning. Indeed, it is really only possible to make sense of the rational promotion of public goods if we approach their pursuit in terms of collective reasoning. Recall Elizabeth Anderson’s example above: Voting is a public good; electing our candidate is a good for us (for a suitably defined “us”). Yet it is a familiar refrain that “my vote won’t make any difference (and voting is inconvenient), so it is not rational for me to vote.” Fortunately for democracy, many people are perfectly willing to behave “irrationally” and show up at the polling station. Of course, describing this behavior as “­irrational” presupposes that we can only understand rationality individually. Fortunately, people are quite capable of cooperative reasoning, which is why the solution to Hi-­Lo is so obvious to most people. And cooperative reasoning succeeds in solving collective action problems precisely because it focuses not on the consequences of individual action, but rather, as Sellars would say, on the “consequences of anybody acting in a certain way in a certain kind of circumstance” (SM 7.XIII.§85/p. 207)­— that is, of us adopting rules of the sort “For all values of ‘x’ . . . it shall be the case that x does Ai if x is in Ci” (SM 7.XV.§109/pp. 214–15). As Sugden writes, “A cooperative morality enjoins each individual to do her part in achieving outcomes that are good for all . . . the individual does not ask whether her own actions, considered in isolation, yield preferred outcomes. Instead, she acts on rules which, when generally followed, yield preferred outcomes” (1993, p. 72). Striking a similar note, Elizabeth Anderson writes, “In regarding themselves as members of a single collective agency, the parties are committed to acting only on reasons that are universalizable to their membership. The universalization principle rules out the principle of maximizing expected utility (individual preference satisfaction) as an acceptable principle of rational choice for members of a collective agency who constitute the parties to a

164  Cooperative Rationality and We-­Intentions prisoner’s dilemma” (2001, p. 29). Thus, the voter’s excuse is not valid­— it is an illegitimate appeal to individual utility when the question at hand is one of cooperative goals. (This is not to say that individual utility can never trump collective goals; if one is quite ill, then this might of course excuse participation in voting. We will have more to say about the interaction between individual and group rationality later.) Of course one’s vote doesn’t in itself determine the election. But our votes do­—just as the actions of players A and B determine whether a goal is scored, the actions of fishermen determine whether the fishery declines or is preserved, and the actions of herders determine whether the commons are overgrazed. It is an instance of a goal that can be achieved only by collective action, and so cooperative­—not individual­—reasoning is called for. To criticize participants in collective action for irrationality because they may fail in this instance to maximize personal utility­—or because their individual action doesn’t affect the action of the group­—quite misses the point and myopically insists on viewing all action through the scope of the individual, not the group. As Sugden writes, The person who acts in an instrumentally rational way evaluates actions in terms of cause and effect: for him, an action is rational to the extent that it causes good outcomes. The person who acts as a member of a team evaluates actions as parts of a whole that is made up of the actions of all the members of the team: for her, an action is rational to the extent that it is part of a set of actions which, taken together, cause good outcomes. As Hurley puts it, “[It is wrong] to assume that when someone acts as a member of a group he must have made some mistake about the causal consequences of his act on the acts of other members of the group, or have engaged in magical or superstitious thinking of some kind; such an individualistic assumption in fact distorts familiar forms of human motivation and agency.” (Hurley 1989, p. 148; Sugden 1993, p. 86)

7.2 Two Types of Intention, Two Types of Reasoning 7.2.1 Wedding Cooperative Rationality to We-­Intentions I want to suggest that there is a deeper difference than mere logical form between the two kinds of intention in Sellars’s ethics. Rather, each kind of intention involves a different kind of reasoning. Elaborating these two types of intentions in terms of these two types of reasoning provides us with an excellent tool for further elaborating a truly Sellarsian ethics. Consider: Individual intention (shallI) is, as Sellars points out, inherently I-­centered. Two people can have parallel intentions, but the logical form of their intention does not, as Sellars would say, reflect membership in a group. Individual rationality is similarly I-­centered: It essentially asks what it is rational for me to do. And this egocentric perspective on

Cooperative Rationality and We-­Intentions  165 rationality misses something crucial about the moral point of view­—its inherently communal nature. Thus, it is difficult to see why it is individually rational, in the absence of external sanctions, for me to follow a water conservation order during a drought. My conservation of water doesn’t benefit me. Worse, it doesn’t benefit anyone else either­—again, I am just one person, so the contribution made by my cooperation here is negligible, as would be the cost of my defection. Thus, it is also difficult to justify this behavior from the perspective of altruism. It is this communal perspective that we-­intentions were introduced to capture. When one we-­intends (shallwe), one intends not merely as an individual but as a member of a larger group. And, of course, it is precisely this sort of group intention that should issue from an exercise of group rationality. If we have a general shared intention (such as that our general welfare be promoted), it follows that it is rational for us to conserve water. Thus, it is rational­—collectively rational­—for each of us to adopt the following intention: I shallwe conserve water. These two types of rationality­—individual rationality and cooperative rationality­—can be wed without significant difficulty not only to Sellars’s account of intention and to his notion of intentional operators but also to his account of inferences involving intentions. Taking into account some of the lessons about the logic of intentions from Chapter 5, let’s turn now to how our account of cooperative rationality fits with Sellars’s account of practical inference involving we-­intentions. Suppose, again, that the following intention is categorically valid: It shallwe be the case that our welfare is maximized As in the case of individual intentions, practical reasoning involves reasoning from ends­—i.e., the contents of particular intentions­—to means of satisfying these ends. Given the end of maximizing our welfare, we can determine certain means of promoting that end. But, of course, these means are means that we must promote in concert­—as a group. Thus, “Our welfare is maximized” will materially entail as means “Each of us refrains from harming others,” “Each of us keeps our promises,” etc., each with a ceteris paribus clause attached. Then, we conclude various derivative intentions, such as “Shall(we) be [Each of us refrains from harming others].” It is this intermediate step­—from end to means­—that our account of cooperative rationality is particularly well suited to explain. This is for (at least) two reasons. First: As we noted in the case of individual reasoning, the path from end to means is not usually one of strict entailment; it is one that requires practical reasoning to find a means that will satisfy the end given the various constraints in play. Thus, if we have an end (e.g., prevent the reservoir from running dry) and a set of constraints (the reservoir is critically low, and we are in the middle of a drought), collective rationality can suggest an appropriate means of

166  Cooperative Rationality and We-­Intentions satisfying this end: Each of us shall conserve water. Again, any number of different means­—trucking in water, at great expense, from a distant, water-­rich location; killing off half the town’s population to reduce demand for water; and so on­—would achieve the end of preventing the reservoir from running dry. There is no strict entailment from end to means of achieving this end. Second: Cooperative rationality, as deployed to reason from end to means, captures an essential feature of moral reasoning that Sellars repeatedly emphasizes: “Members of our community, in so far as they value the general welfare . . . [do so] not from a personal point of view­— external benevolence­—but as one of us” (SM 7.XVII.§130/p. 221). We have seen already how it is difficult, from the individual point of view, to reason from ends such as “prevent the reservoir from running dry” to means such as “I will conserve water.” But the perspective of cooperative rationality contributes precisely what was missing­—reasoning as a member of a group to a conclusion about what each of us ought to do. This is precisely the content that Sellars claims a moral intention must have. Such intentions have, schematically, the categorical form “It shallwe be the case that each of us does Ai if in Ci.”9 Thus, wedding our account of collective rationality to Sellars’s account of reasoning with intentions, we can reason as follows: (1) Shallwe be (Our welfare is maximized). (2) (Our welfare is maximized) implies (Each of us refrains from harming others if in C1). (3) Shallwe be (Our welfare is maximized) implies Shallwe be (Each of us refrains from harming others if in C1) [2, S-­Imp]. (4) Shallwe be (Each of us refrains from harming others if in C1) [1,3 MP]. (5) Shallwe be (I refrain from harming others if in C1) [4].10 (6) Shallwe be (I refrain from harming others if in C1) implies Shallwe (I refrain from harming others if in C1) [5, Shall-­be]. (7) ∴ Shallwe (I refrain from harming others if in C1) [5,6 MP]. (8) (Our welfare is maximized) implies (Each of us keeps our promises when in circumstances C2). (9) Shallwe be (Our welfare is maximized) implies Shallwe be (Each of us keeps our promises when in circumstances C2) [8, S-­Imp]. Etc. Notice that material implication has made a reappearance here in premises 2 and 8. The link between antecedent and consequent in premises 2 and 8, while seeming fairly obvious to us, nevertheless involves a fair amount of understanding of the social world in which we live, most of which we are probably unable to make fully explicit. All of this, of course, is the hallmark of a flexible, material inferential practice.

Cooperative Rationality and We-­Intentions  167 As a final note, wedding cooperative rationality to we-­intentions fits well with Sellars’s goal of enabling “us to see how ‘teleological’ and ‘deontological’ themes are harmonized in Kant’s ethics” (SM 7.XXI.§147/ p. 226). Sellars writes, Moral principles primarily concern the consequences of anybody acting in a certain way in a certain kind of circumstance, and it is only because of the subtle involvement of what others do in the relevant description of an agent’s circumstances that it is a conceptual truth that if the consequences of anyone doing A in a certain fine-­grained kind of circumstance are good, the consequences of everyone doing A in precisely that kind of circumstance must also be good. With these careful qualifications, the move is simply that from “any” to “all.” (SM 7.XIII.§85/p. 207) As I will argue in Chapter 13, Sellars doesn’t think that a purely formal requirement (like Kantian universalizability) will generate concrete moral content, and so the universalizability requirement must be supplemented with a material end (such as welfare maximization). But in Sellars’s hands, the universalizability requirement just becomes a consideration of what would maximize the general welfare were any of us to do likewise in similar circumstances. Universalizability allows the inference from “any” to “all,” and so what this amounts to is a question of what would be utility-­maximizing were we to do it. 7.2.2 Relativism, Again As I noted in Chapter 6, Sellars represents these premises (connecting ends to means) as nomologicals, which express necessary conditions on fulfilling the end in question. However, I argued that there are excellent reasons for rejecting this account. Although on Sellars’s account material implication can (and often does) express nomological necessity, it also finds its home in a more flexible inferential practice where inferences are defeasible and non-­monotonic. How should we interpret premises like 2 and 8? Are the consequents necessary conditions of achieving the antecedents? Or are these inferences more like inferences from “I shall eat dinner” to “I shall go to Royal China”­—that is, reflective of a more flexible inferential practice where ends do not necessitate any particular means? (Or are 2 and 8 like inferences from “The match is struck” to “The match will light,” where they are materially valid but defeasible?) Whether the consequents of premises 2 and 8 are necessary conditions of the antecedents is essentially a question of whether some form of relativism is true. Relativism can threaten at the global or local level (although I suspect this is a difference of degree, not of kind). Sellars

168  Cooperative Rationality and We-­Intentions maintains that “Shallwe be (Our welfare is maximized)” is intrinsically categorically valid­—binding on all rational beings. If this is true, then global relativism is false­—all rational beings are bound to maximize our welfare where “our” refers to all of us. However, premises 2 and 8 connect this general intention to specific intentions that are its concrete realization. If the consequents of 2 and 8 are not necessary conditions on the antecedents, then local relativism is true. That is, even though all rational beings are obligated to maximize the general welfare, they are rationally permitted to adopt different (perhaps radically different) practices to do so. We will discuss in Chapter 13 whether the general intention “Shallwe be (Our welfare is maximized)” is categorically valid. But what about local relativism? Assuming, for the moment, that this general intention is categorically valid, is local relativism true or false? There are two things to be said about local relativism as a philosophical thesis. The first is that we should hope that Sellars is committed to some kind of local relativism. Virtually everyone is committed to some kind of local relativism. As Lucas Thorpe has emphasized to me, even a hard-­nosed realist like Aquinas held that while the natural law is the same for everyone in its general precepts, it will differ from place to place in its application.11 Jack Donnelly has argued that radical universalism­— the thesis that “culture is irrelevant to the validity of moral rights and rules, which are universally valid” (Donnelly 1984, p. 400)­—is wildly implausible. On such a view, not only would some right (such as political participation) be universally valid, but there would not even be permissible local variations in the implementation of this right (e.g., parliamentary versus presidential democratic systems, differences in the frequency of elections, etc.). Thus, Donnelly embraces what he calls “weak relativism,” which allows for universal rights but a range of permissible local variations in the implementation of these rights. And this, of course, seems precisely the correct way to think about moral obligation more generally­—it should allow for this kind of weak or local relativism. Any moral system that doesn’t allow for this is implausible in the extreme. Second, Sellars’s constructivism allows for local relativism, generating universal thin precepts with permissible local variations. As Westphal explains, “The key to providing objectivity within a constructivist moral theory is not to appeal to subjective states [like sentiments, passions, affective responses, etc.] . . . but instead to appeal to objective facts about our form of finite rational agency and circumstances of action which are basic to the human condition” (2016, p. 21). In this way, the rules of morality can be shown to be, as Hume put it, “artificial, [but] not arbitrary” (1740/2000, 3.2.1.19). The reasoning embodied in premises 2 and 8 involves inferences about the consequences of certain types of collective action­—of permitting (or requiring) anybody to act in a certain way when she finds herself in “a certain fine-­grained kind of circumstance”

Cooperative Rationality and We-­Intentions  169 (SM 7.XIII.§85/p. 207). But once we view the matter this way, we see that worries about relativism need not be acute. For it is unlikely that it would in fact be the case that in one community, say, widespread lying promoted the general welfare, whereas in another community truth-­telling promoted the general welfare; or that in one community, refraining from arbitrary killing promoted the general welfare, whereas in another community, such killing did promote the general welfare. (As always, it is crucial to distinguish between what a community believes versus what is in fact the case.) For this reason, Hume included promise-­keeping as one of his laws of nature­—a rule that (although artificial) is required for any society to subsist. For Sellars, we must remember that “ethical truths are the projections of ‘matter-­of-­factual’ truths . . . into the framework of intentions and purposes” (SM 5.I.§3/p. 117)­—specifically, truths about what means will in fact promote our overall welfare. As Sellars comments when discussing the “craft of the citizen” in Plato, we are faced with “the following general schema” (RAL, p. 15): If a citizen wants to contribute to making and/or maintaining a city ordered to the general welfare, then If a citizen is in Ci, he ought to do Aj The crucial thing to note here, for Sellars, is that given “a specific conception of the [general] welfare . . . the distinction between truth and falsity is relevant” to these derivative principles. Thus, “given such a specific concept [of the general welfare], that a specific plan of action is required to order a city to welfare, thus conceived, is an objective matter for rational determination” (RAL, p. 15). Sellars, of course, here implies that certain means are necessary to the end in question, a move I have resisted. But even if we move from necessary to sufficient conditions, whether a means will in fact promote a particular end is “an objective matter for rational determination” and admits of truth or falsity. However, given that we have moved to a picture of means as sufficient (but perhaps not necessary) conditions on the achievement of the end of general welfare promotion, there may be permissible local (to time and place) variations on the means chosen to promote the general welfare. And perhaps even Sellars in his strictest mood would admit this since Sellars does admit of “fine grained” specification of circumstances. Consider an example of a culture that is compelled by circumstances to adopt a moral rule different from one adopted by a culture to which most of us belong. To use an oft-­cited example, consider the killing of the elderly and sick among the Inuit during times of hardship. Rolf Kjellström (1974–5) cites a number of such examples reported by anthropologists, writing, “Even though the old and disabled persons normally were treated very well, nobody hesitated to abandon these unproductive members of the society

170  Cooperative Rationality and We-­Intentions if the margins of survival for the group grew narrow. . . . Those who left their relatives dying were not blamed appreciably, especially not during an acute period of crisis” (p. 117). The reasons behind such behavior are not far to seek: “When [the aged and infirm] became a burden during moving and hunt[ing] or a threat during hard periods of famine or similar pressing situations they were the first ones to be sacrificed in order to save the able-­bodied” (1974–5, p. 120). Examples like this are often cited as evidence for some kind of relativism but should not be interpreted as such. As Sellars writes (SM 7.XIV.§96/p. 211), a moral rule of the form “I shall do A” is categorically reasonable relative not only to the conditional intention from which it is derived (“I shall do A if I am in C”) but also to the circumstances (“I am in C”). The conditional intention is binding on us all, Sellars would argue, even if we never find ourselves in C (and therefore never find ourselves obligated to do A). Thus, some individuals (like the Inuit) might find themselves formulating and acting on conditional intentions that are specific to circumstances they face. But this is not a troubling sort of relativism: If a group of Inuit were to move to a more forgiving environment, where such life-­threatening hardships were absent, then it would no longer be permissible for them to practice senilicide, for they would no longer find themselves in the appropriate circumstances. Similarly, if a group from a wealthy society were to find themselves in such a harsh environment, they might find senilicide or invalidicide permissible (whereas it would be rightly considered murder if they were home and surrounded by plentiful resources). Consider a real-­life case: During the summer (Southern Hemisphere) of 1911–12, Robert Scott tried to lead the first party to reach the South Pole (famously arriving five weeks after Roald Amundsen). On the return trip, his team encountered harsh conditions and faced many difficulties. On 17 March 1912, four of the original five-­man team remained alive, but Lawrence Oates was suffering from severe frostbite and feared that his condition was jeopardizing the survival of the remaining three members of the party. Scott reported in his diary that Oates said, “I am just going outside and may be some time,” and walked out of the party’s tent to his death. “We knew that Oates was walking to his death,” Scott wrote, adding that “it was the act of a brave man and an English gentleman” (Flynn 2011). (Despite Oates’s sacrifice, the remaining three party members froze to death in late March.) In the England of the early 20th century, no such sacrifice would have been necessary on Oates’s part, as no such hardship would have arisen. In those circumstances, Oates’s colleagues would, I presume, have had a prima facie duty to intervene had he indicated to them that he had an intention to end his life. But when these Englishmen were placed in different conditions, with different and more demanding circumstances, they found a different set of rules relevant to their circumstances­—just as the Inuit, living in their particular demanding circumstances, must often appeal to a

Cooperative Rationality and We-­Intentions  171 different set of rules. This is not relativism; it is merely a reflection of the variety of circumstances in which human activity is conducted.

7.3 Conflicts Between Team and Individual Rationality A persistent question that arises in the context of team rationality is what to do when there is a conflict between what is rational for me to do qua individual and what is rational for me to do qua member of a group. Gilbert’s view seems to be that being involved in a joint commitment will, ceteris paribus, give one sufficient reason to act, and that exculpatory reasons­—while possible­—will always indicate that the situation in question is an exception and non-­standard. Gilbert writes, Participation in collective agency­—in particular, subjection to the underlying joint commitment­—. . . gives me sufficient reason to act in a certain way, reason I cannot remove at will . . . it does not . . . deprive me of my capacity to reason and act according to my own best judgment. I may break away from a collective action in progress at any time­—sometimes this may be rationally required, sometimes at least rationally permitted . . . but if I do break away, there will always be something amiss­—something not quite right about the situation­—that is more than a violated commitment of the will. I will not have given the other participants what I owed them qua parties to the underlying joint commitment. (Gilbert 2014, p. 91) Tuomela thinks it depends on the nature of the group­—some groups may give people only “pro tanto” reasons to act, whereas other group reasons may be preemptive. He argues, though, that “in the case of strictly obeyed normative requirements (concerning what one ought to do) I-­mode utilities have no place” (Tuomela 2007, p. 168). And, indeed, Tuomela argues that in fully socialized individuals, we-­mode reasons will take precedence: “The first view is to take a group member to be well socialized, so that the we-­mode goals, preferences, and utilities will overwhelmingly be dominant reasons for action. His social identity is fully fixed, we may say” (Tuomela 2007, p. 168). Perhaps the more common situation, though, is where the individual has, in Tuomela’s phrase, a “flexible social identity” (2007, p. 169), and so we-­mode and I-­mode utilities will sometimes conflict. Then we are faced with the “switch problem, that is, a switch from (dominantly) thinking, including preferring, and acting as a private person (or, for that matter, as a member of another group) to (dominantly) thinking and acting as a member of g” (2007, p. 169). As I hinted at in Chapter 4, and will argue at length in Chapter 13, there is no meaningful sense in which a person can think other than as

172  Cooperative Rationality and We-­Intentions a member of the community (for some specification of “community”), and this imposes a kind of rational requirement on the individual to act for the welfare of the community. This is, as we will see, the payoff of Sellars’s hint that if we could demonstrate the reality of the community, we could demonstrate the categorical validity of the formal end of morality. But let us set this aside for the moment. If we assume that the perspectives of group and individual rationality are independent and competing points of view­—and Sellars sometimes writes in this way­—then it would be helpful to have some way of assessing the pull of each when they conflict. As Tuomela notes, one way to do this is to represent both the agent’s I-­mode preferences and her we-­mode preferences as I-­mode preferences (Tuomela 2007, pp. 162–3). Thus, an agent’s I-­mode utility for an action X is represented as uij(X), and her we-­mode utilities are represented as ugj(X). We then represent the agent’s weighting of each utility with a modifier (0 ≤ w ≤ 1) and add the weighted utilities together to get the agent’s total utility utj(X) as follows: utj(X) = wiuij(X) + wgugj(X) This would yield a rational action that took both types of utilities (I-­mode and we-­mode) into account. Tuomela, however, seems to prefer a somewhat more sophisticated method that takes into account a number of factors that will affect the agent’s utility (Tuomela 2007, pp. 169–74). Suppose E is some outcome two agents are producing and u is the reward (utility) each agent receives when E is produced. Tuomela argues that when agents are a member of a group, a number of different factors will contribute to u: j (“jointness effect”): this can include things such as “the increase in trust, networking, skills,” and so forth gained from the exercise, as well as the costs of cooperation f (“free rider incentive”): the amount an agent can gain from defecting, or free riding s (“sucker effect”): the (dis)utility of cooperating while your “partner” defects m (“moral sanction”): this is the negative normative sanction that is in place given that there is an expectation (social or moral) that one will cooperate m* (“mutual defection utility”): the “moral sanction” utility in the event that both players defect; Tuomela imagines that the sanction might be less (or nonexistent) in this case Thus, Tuomela can present the payoffs for player A (the row player) in a single-­shot prisoner’s dilemma as follows (Table 7.3):

Cooperative Rationality and We-­Intentions  173 Table 7.3

C D

C

D

u+j u+f+m

u+s m*

So what is it rational for A to do? This will of course depend on the weight we assign to the above variables. But Tuomela seems to think that it can be rational for A to cooperate, even if A has a “flexible social identity,” once we take A’s we-­mode attitudes into account. As Tuomela notes, “Forming such . . . we-­mode intention and group-­social reason[s] will increase the value of the jointness parameter j” (2007, p. 173); group members will have a “standing or falling together” attitude. Further, m and m* will have bigger negative values given the presence of we-­ mode preferences; this will be due to “the agent’s feeling of guilt . . . or shame. . . . Furthermore, there may be objective social sanctions” (2007, p. 173). Finally, the sucker effect s might be zero or even positive; as Tuomela notes, one might have the attitude “I am proud to be the sole person to act for the group” (2007, p. 173). Even without these assumptions, Tuomela demonstrates that an individual’s utilities can easily align behind cooperation. Adding these assumptions, it is clearer how cooperation can be rational. And to head off accusations that this is philosophical speculation that is at odds with how people actually behave, it is worth noting­—as Tuomela does­—that in fact people behave cooperatively more often than would be predicted by individual rationality game theoretic models. The suggestion is that people do form joint commitments and do act on them. To add these commitments to an account of what is rational to do is realistic; it is the assumption that people reason only as individuals that does not do justice to the facts. The above argument doesn’t displace the need for cooperative rationality, though. What it may show is individuals with a “flexible social identity” may nevertheless find that it makes sense in terms of I-­mode utilities to adopt a we-­intention. The resulting action, then, is still action in the we-­mode. And, as I argued at length, agents can accomplish things using we-­intentions and cooperative rationality that cannot be accomplished from the perspective of individual intentions and individual rationality; this was the chief lesson of, for example, the Footballer’s Problem. But reasoning from personal preferences to the adoption of a we-­ intention would represent a defective case of an imperfectly socialized individual. The regulative ideal is of an individual who is socialized to have dependent individual commitments (parasitic on group commitments),

174  Cooperative Rationality and We-­Intentions and to base moral action upon these. If an individual consistently asks, “What’s in it for me?” when confronted with the moral point of view, then their inculcation into the moral point of view has already gone awry. So while Tuomela may show that it can be I-­mode rational to adopt a we-­intention and cooperate, this doesn’t mean that as a general account we should try to justify we-­intentions and joint commitments in terms of personal commitments and preferences. Tuomela’s account of the relation between personal utility and the moral point of view is a useful adjunct to ethical theory, but it would be misleading to think that the moral point of view needs to be justified in terms of personal utility or that the two points of view are necessarily in competition. How ought we conceive of the relation between group commitments and personal commitments? As noted in Chapter 4, joint commitments have a number of features that personal commitments don’t have­—for example, joint commitments are not individually rescindable. This suggests that, ceteris paribus, an individual cannot individually decide to forgo satisfying a joint commitment in favor of satisfying a personal one. Indeed, treating all of our preferences as commensurable in this way­— indeed, treating all of our reasons for choice as preferences­—deeply misrepresents practical reasoning and the way in which people value. The joint commitments we have qua rational beings­—our specifically moral commitments­—are granted, in practical reasoning, a status that differentiates them from personal preference. They play a correspondingly different role in practical reasoning and are not merely another set of preferences to be thrown on the scale and measured against I-­mode utilities. Amartya Sen, in particular, has been critical of the “one size fits all” model of human behavior that reduces all choice to preference (and, by extension, to self-­interest). As Sen notes, traditional game theorists measured utility in terms of preference satisfaction; preferences are identified by choices. Preference satisfaction, in turn, is often further identified with welfare, thus making rational agents egoists more or less by definition.12 Among his many concerns with the traditional view is the assumption that “there is no choice-­independent way of understanding someone’s attitude toward alternatives” (Sen 1977, p. 323). For we can make perfect sense of someone acting contrary to their welfare­—indeed, contrary to their own preferences­—out of commitment, as Sen calls it. Indeed, Sen writes, “One way of defining commitment is in terms of a person choosing an act he believes will yield a lower level of personal welfare to him than an alternative that is also available to him” (1977, p. 327). Now, I think this is an excessively pessimistic way of viewing the notion of commitment­—and, indeed, Sen goes on to note that there might be cases where “a person’s choice happens to coincide with the maximization of his anticipated personal welfare, but that is not the reason for his choice. If we wish to make room for this, we can expand the definition

Cooperative Rationality and We-­Intentions  175 of commitment to include cases in which the person’s choice, while maximizing anticipated personal welfare, would be unaffected under at least one counterfactual condition in which the act chosen would cease to maximize personal welfare” (1977, p. 327). The point is that we don’t treat the commitments we have as rational beings as such as just another set of preferences, commensurable with the preferences I have qua individual and to be thrown on the scale together with the latter and weighed indifferently. We treat them, as Sen notes, as commitments over against personal preferences. That is not to say that the latter can never trump the former. But it is to say that the former are not merely preferences to be weighed against the latter. In a sense, then, the discussion of weighing I-­mode versus we-­mode utilities misses the point when the we-­mode commitments in question are the we-­intentions that constitute the moral commitments we have as rational beings. These latter are commitments, in Sen’s sense, and something we owe to other rational beings. While personal preference can trump this, an agent must have a powerful reason for overriding the force of such a commitment. Thus, I have misgivings about Tuomela’s attempt to show that it can be individually rational to adopt a we-­intention. (It will then be incumbent upon Sellars to show some other categorical basis for the we-­intention that constitutes the moral point of view. Again, this is the task for Chapter 13.) I will conclude this section by addressing two related objections (although it might not be immediately obvious how they are related).13 First, from the point that something is collectively rational, it doesn’t follow that it is individually rational. Thus, even if we can demonstrate that a course of action is collectively rational, it doesn’t follow that any individual thereby has a reason to perform this action if no personal benefits accrue to the agent from said performance. As I have already indicated­—and will argue at length in Chapter 13­— this objection rests on the dogma that “rational” means “self-­interested,” and that ultimately the only kinds of reasons we could have for doing anything are egoistic reasons. As I will argue in Chapter 13, rational egoism is self-­undermining, as an individual’s self-­interest is not conceptually separable from the welfare of the community­—indeed, the individual’s very identity and constitution are not separable from the community’s institutions, norms, and roles. Thus, as I will argue, promotion of the community’s welfare­ —adoption of the moral point of view­ —is not optional, nor need it be justified in terms of self-­interest. The second objection is this: If Sen and others are correct that there are irreducibly moral commitments, and these are not merely reducible to preferences (or aggregates of preferences), then what need is there of an account of cooperative rationality? If each of us is morally required to do X, then cooperation follows from this requirement alone; a separate account of cooperative rationality is otiose.

176  Cooperative Rationality and We-­Intentions Again, though, these moral commitments aren’t quite irreducible. We have them because we are members of a community and because our identities are constituted by such communal membership­—an argument I will advance in greater detail in Chapter 13. As we will see, Sellars advances the Hegelian thesis that moral obligation arises out of communal membership and is essentially social, communal, and intersubjective. Thus, these commitments are rationally mandatory precisely because of the necessity of the group perspective. So these moral commitments presuppose rather than replace the team perspective.14 More importantly, perhaps, this objection overlooks the degree to which these moral commitments are somewhat underdetermined. If irreducible moral commitments could be determined in isolation from others, then we wouldn’t need to coordinate our behavior with others­— we could just determine our moral obligations and act. Everyone else doing the same would result in coordination without anyone actually coordinating. However, as noted earlier in the chapter, any moral theory must admit of local relativism. Thus, while very thin values might be universal, their local specification will vary from time to time and place to place. Thus, acting on the concrete specification of a thin, universal value will require (a) agreeing on a specific strategy to promote this universal value and (b) acting in concert to realize this cooperative strategy. Consequently, the underdetermination of concrete moral practices means that we have to coordinate with others­—we have to act as a team. We can no more derive specific, timeless, universal moral practices in isolation from other individuals than a football player can a priori derive specifically which run to make without taking into account what the other players on the field are doing. Cooperation is needed­—and hence team reasoning.15 7.3.1 Supererogation One final issue we haven’t addressed is supererogation. How do we make sense of this within the present framework? My suggestion begins with the following observation: If you have a team strategy for pursuing the general welfare, it will (a) establish defeasible moral rules, violation of which is (prima facie) immoral, and in doing so (b) limit act-­teleological constraints that are too demanding. Supererogatory acts are those that bypass constraint (b) while still being bound by (a). And it makes pragmatic sense to allow voluntary violations of (b) but not of (a). Constraint (a) sets, as it were, the minimum for agents: Don’t lie, don’t steal, give time and/or money (occasionally, as you can) to a worthy cause, etc. We have seen that acting contrary to (a) in an act-­utilitarian way turns out to be, paradoxically, not welfare-­promoting. But allowing people to abide by (a) while forbidding violations of (b) would be all cost and no gain and

Cooperative Rationality and We-­Intentions  177 so pointless from a pragmatic perspective. Thus, if people observed the constraints of (a), but then also performed actions that were not required by (a) in virtue of running afoul of the demandingness requirements of (b), then there would be good reason to permit this. Thus, in addition to acting as above, one might spend many hours per week volunteering for a charity or donate a significant percentage of one’s income to charity, or one might go “above and beyond the call of duty” for a friend in need. Again, as long as the requirements of (a) are not violated, there is no good reason to forbid actions that go beyond the normal limitations imposed by (b). They cannot be required as one reason to move from an act-­based to a rule-­based teleological system to avoid objections about overdemandingness.16 But they should be permitted (and indeed encouraged and praised)­—and it is this class of actions that is supererogatory.

7.4 The Moral Point of View I will conclude this chapter by emphasizing an aspect of Sellars’s moral philosophy that is easily overlooked­—even by some of Sellars’s most astute commentators. David Solomon, one of the few people to write about the topic in any detail, wrote an early paper discussing “Sellars’s Defense of Altruism.” But I think this is the wrong way to conceptualize what Sellars is doing.17 For thinking of the moral point of view as the point of view of altruism inexorably leads one to think in terms of individual rationality­—of juxtaposing the individual welfare against the welfare of others. But the profound insight contained in Sellars’s work is that to occupy the moral point of view is to occupy the standpoint of the community­—to think and reason from the point of view of the community. Consider the analogy of team sports, where the phenomenon of the selfish player is well-­known. This is the player who keeps control of the ball, refuses to pass, takes low-­percentage shots (rather than passing to a teammate) in pursuit of personal glory, etc. But it is out of place to describe the actions of the consummate team player as a­ltruistic. The player who passes when appropriate, plays defense, and generally contributes as she ought to the overall goals of the team isn’t being altruistic­—she is being a team player. Similarly, to think morally isn’t in the first instance to think altruistically. It is to think as one of us, for the benefit of us. This is a crucial point to be kept in mind throughout this work. As Olen and Turner emphasize, Sellars takes the following view: “‘To value from a moral point of view is to value as a member of the relevant community’ [SM 7.XVIII.§124/p. 220]. The novelty of his revision of Kant is to distinguish ‘the impartial love of one’s fellow men’ [IILO 13/p. 210] from the ‘recognition of each man everywhere as one of us’ [IILO 13/p. 210]” (Olen and Turner 2016, p. 2068).

178  Cooperative Rationality and We-­Intentions

Notes   1 Tuomela notes, of course, that in practice people have devised a number of ready solutions to such coordination problems; solutions are most readily available when communication is possible (as it is not in the Footballers’ Problem). However, Tuomela notes that such solutions are rational only in a very broad sense and not strictly speaking r-­rational (i.e., individually instrumentally rational); “a solution is not entailed if only the situation of coordination is given and mutual belief or knowledge concerning it is assumed” (Tuomela 2007, p. 76). And, of course, Sugden’s contention isn’t that solutions to these coordination problems are impossible. His contention is rather the opposite: The solutions to these problems are often glaringly obvious, and so it is a serious defect of the perspective of individual rationality that these solutions are simply unreachable using its tools.   2 As Sugden points out, from the point of individual rationality, it will not help to try to understand this problem by looking for a Nash equilibrium. Not only are there three different Nash equilibria in this problem but also, and more fundamentally, “the fact that a particular combination of strategies is a Nash equilibrium (even a unique Nash equilibrium) gives neither player a reason to choose to play his part in that equilibrium. All we can say is that, if it were the case that each player expected every other player to play his part in a certain (strict) Nash equilibrium, then each player would have reason to do the same. And that provides us with no escape route from the infinite regress of reasons” (1993, p. 182).   3 Jesse Holloway raised the following objection.   4 I am grateful to Robert Sugden for discussing this objection with me in personal correspondence.   5 Amir Saemi has suggested there is no contradiction, as the reasoning involved in this argument is diachronic­—it involves Bayesian updating. But it is not accurate to represent this argument as involving Bayesian updating. As Sugden writes, “In Bayesian updating, you start with a prior based on your existing information (e.g. the prior is: the probability it will rain today is 0.5) and then you receive additional information (e.g. you see dark clouds) and so you update your probability. But in this case, you are not receiving new information, you are working out the implications of the proposition that your belief is correct. The standard principles of propositional logic apply: if p implies not p, then p is false.” In other words, the argument serves as a reductio of the claim that I don’t know what B will do because assuming this leads to a contradiction (it leads to the conclusion that I do and do not know what B will do), requiring us to reject the initial assumption. (The quoted passage is from Robert Sugden, personal correspondence, 10 May 2018.)   6 Robert Sugden, personal correspondence, 21 June 2015.   7 Robert Sugden, personal correspondence, 21 June 2015.   8 Sugden develops this argument in detail in his (1993).   9 See, for example, Sellars (SM 7.XX.§138/p. 224). 10 Sellars doesn’t name the principle that allows the inference from an intention naming each of us to an intention naming one of us but thinks that the essence of a we-­intention is that it supports the move in scope from each to I. See, in particular, ORAV, §§153ff/pp. 96–7; also, “Letter to David Solomon.” 11 Comments made at the CIRS workshop on the present book manuscript, 5 November 2017. 12 We will return to this “formal” argument for egoism­—and why it fails­—in Chapter 13. 13 Amir Saemi pressed me on both of these objections.

Cooperative Rationality and We-­Intentions  179 14 As I noted in Chapter 4, for Gilbert, the obligation to act on a joint commitment isn’t a moral obligation per se. Thus, one might worry about my pressing into service her notion of joint commitment for Sellars’s specifically moral project. But I don’t think this is problematic, for there are a range of types of joint commitments appropriate to a range of uses. For Sellars, most collective intentions are not moral intentions, and so these would arise out of joint intentions that don’t necessarily give rise to moral obligations. But it seems plausible to regard our joint commitment to the formal end of morality (our commitment to promote the welfare of rational beings generally) as giving rise to specifically moral obligations. And, of course, for Sellars, it is only certain sets of we-­intentions­—distinguished by both their content and their universal scope­—that qualify as moral intentions. Thus, joint commitment as such doesn’t give rise to a moral obligation, but a certain class of joint commitment does. 15 The underdetermination of specific moral commitments by the supreme principle of morality (“Shallwe maximize the general welfare”) can be made sense of in terms of Sellars’s distinction (which he discusses in RAL) between the principles of a craft and conventions. The principles of the craft of the citizen are those principles that, independent of convention, factually promote the general welfare (e.g., “Tell the truth in circumstances Ci”). Conventions are “enforced imperatives”­—rules such as “Drive on the right side of the road.” But a person practicing the art of the citizen will include conventions in the circumstances of application when applying her art: Given, for example, that the convention is to drive on the right side of the road, maximization of welfare requires that she do so. Thus, conventions aren’t the same as principles, but conventions may fill in the details that are underdetermined by the supreme principle of morality, allowing for coordination and allowing for culturally specific patterns of welfare-­promoting practices. 16 I argue at length for the superiority of rule-­based moral systems over act-­ based ones and reply to various objections in chapter 3 of my (2009). 17 “The point of view of benevolence . . . is not the moral point of view, though, as Kant saw, it is easily confused with it. Even generalized and embracing benevolence is, so to speak, an external point of view. If I were to say, in all candor, ‘Would that language-­using featherless bipeds generally led satisfying lives,’ I would be expressing a friendly sentiment, which might be very strong indeed, with respect to language-­using featherless bipeds” (SM 7.XIV.§86/ pp. 207–8). But this “friendly sentiment” is still not the moral point of view. using featherless bipeds (and other rational beings) as To view language-­ members of the moral community is not to treat them as external objects of benevolence, but to take the internal point of view­—to see them as one of us.

8

Defeasible Rules and the Particularist Challenge

Rules play a central role in Sellars’s philosophy. As Sellars writes in LRB, “When God created Adam, he whispered in his ear, ‘In all contexts of action you will recognize rules, if only the rule to grope for rules to recognize. When you cease to recognize rules, you will walk on four feet” (p. 298). However, the centrality of rules to human conduct and rationality has come under sustained attack in recent decades. Perhaps most famously, the anti-­generalist project has been carried out by particularists in ethics. Many particularists are motivated by the sort of holism displayed by moral reasons. As McKeever and Ridge write, “Defenders of particularism have, with near unanimity, sought support from a doctrine they call ‘holism in the theory of reasons’” (2005, p. 93), and cite as examples Dancy (1993), Little (2000), and others. For example, David McNaughton writes, “Whether the fact that an action causes pleasure is a reason for or against doing it is not something that can be settled in isolation from other features of the action. It is only when we know the context in which the pleasure will occur that we are in a position to judge” (1988, p. 193). Other arguments against the centrality of rules for human conduct have come from criticisms of early AI programs. Hubert Dreyfus has long argued that rule-­following is characteristic of the novice, whether this person is a novice driver, chess player, or moral agent. What characterizes the expert is “absorbed, skillful coping,” something that cannot plausibly be represented by the expert agent following a set of rules. Dreyfus deploys any number of examples throughout his writing on this issue to illustrate this point; consider his example of the tennis player. As a beginner, one will be taught to keep one’s eye on the ball, to hold one’s racket at a particular angle, to keep one’s feet positioned like so, and so on. The poor performance of the beginner is explained in part by the difficulty of keeping all of these rules in her mind while performing, and also by the fact that she is not yet trained in these movements. But the expert, when involved in “absorbed, skillful coping,” cannot plausibly be portrayed as representing in her mind all of the various things that she is doing (holding her racket at this angle, holding her feet like this, moving her

Defeasible Rules and the Particularist Challenge  181 arm along this curve, etc.). Rather, as Dreyfus writes, “I cannot represent how I should turn my racket since I do not know what I do when I return the ball. I may once have been told to hold my racket perpendicular to the court, and I may have succeeded in doing so, but now experience has sculpted my swing to the situation in a far more subtle and appropriate way than I could have achieved as a beginner following this rule” (2002, p. 379).1 More generally, the Kantian idea that what distinguishes sapient­—i.e., rational­—beings from nonrational ones is that we are bound by rules runs afoul of major strands in the pragmatist tradition. In this tradition (most prominently influenced by Wittgenstein, but also by Heidegger, and thus implicit in Dreyfus’s comments above), to be a rational creature is to be a creature bound by norms. However, being bound by norms cannot be equated with rule-­following, for this conception of sapience leads to a vicious regress. As Wittgenstein famously argued, no rule (such as a rule for addition) is self-­interpreting; nor does any rule necessitate its own correct interpretation. If we need a rule for correctly interpreting the first rule, then we are confronted with the same problem about interpreting the second rule, and so are off on an infinite (and vicious) regress. Wittgenstein concludes that we cannot understand being bound by norms as, in the first instance, following rules; instead, we must explain norms as appropriatenesses that are implicit in practice. As Brandom writes, “The conclusion of the regress argument is that there is a need for a pragmatist conception of norms­—a notion of primitive correctnesses of performance implicit in practice that precede and are presupposed by their explicit formulation in rules and principles” (1994, p. 21). All of the above considerations seem to present formidable challenges to the Sellarsian idea that rules are central to rational agency generally and to practical reason specifically. Taken together, the above considerations may seem to constitute an insurmountable obstacle to Sellars’s ethical project. However, I think a case can be made for the centrality of rules to ethics, and indeed to human rationality. In this chapter, I will argue that while radical particularism is not very plausible, it is incumbent upon the Sellarsian to develop a notion of a rule that has genuine force but admits of exceptions that can serve the role of Sellars’s important material inferential rules. Fortunately, recent years have seen the emergence of well-­developed accounts of defeasible rules of inference. We will examine both formal and informal accounts of such rules and see how they can be pressed into service for Sellars’s project. In the next chapter, we will address Wittgensteinian and neo-­Heideggerian challenges to the role of rules in human rationality. This chapter will give us an important tool for the remainder of the book. As noted, defeasible material rules of inference play a central role in Sellars’s philosophy, and they will reappear many times throughout this book. Giving a philosophically sound account of such defeasible

182  Defeasible Rules and the Particularist Challenge rules is crucial to shoring up the foundations of Sellars’s philosophical project. Further, to the extent that we want to understand intermediate moral principles (e.g., “Lying is wrong”) as defeasible principles, we need to develop an account of this defeasibility relation. I will be understanding intermediate moral principles as moral material inferential principles, and so developing an account of defeasible material rules of inference fills in another piece of the Sellarsian ethical theory. (I will return to the —the intermediate moral role of moral material inferential principles­ principles­—and their relation to the supreme universal principle of morality in Chapter 14.) In this chapter I will be offering a general account of defeasible rules and so will often talk in correspondingly general terms (and not specifically about moral rules). The reader should keep in mind the above points, however. (Although there is a method here, as Sellars needs a general account of defeasible rules given that his own account of material rules of inference is perfectly general.)

8.1 The Particularist Challenge As noted above, one of the most common arguments for moral particularism is that moral reasons function holistically. I cited David McNaughton —from above; McKeever and Ridge (2005) write that this argument­ holism to particularism­—is offered by Jonathan Dancy (1993, 2000) and Margaret Little (2000) and attribute it also to John McDowell. However, McKeever and Ridge argue that “holism provides no support for particularism in any of its guises” (2005, p. 95). Consider an oft-­cited example in support of particularism: While the fact that an action causes pleasure is usually at least a prima facie reason in favor of that action, the pleasure a sadist feels in torturing his victim carries no positive moral weight at all. Thus, one cannot formulate any invariant rule regarding the moral valence of pleasure. McKeever and Ridge demur, arguing that a codification of morality can itself reflect the holism of moral reasons. “Consider the following moral theory,” they invite us: (U) The fact that an action would promote pleasure is a reason to perform the action if and only if the pleasure is non-­sadistic. The fact that an action would promote pain is a reason not to perform the action. An action is morally right just in case it promotes at least as great a balance of reason-­giving pleasures over pain as any of the available alternatives; otherwise it is wrong. (2005, p. 96) Such a moral theory “gives no weight to sadistic pleasures,” and hence reflects the sort of holism that is supposed to support particularism. The only way that holism could support particularism is if one understood holism as the conjunction of two theses: (a) reasons are context-­sensitive

Defeasible Rules and the Particularist Challenge  183 and (b) this context-­sensitivity is uncodifiable. Following Jackson, Pettit, and Smith, they call this “unrestricted holism.” However, McKeever and Ridge point out that this unrestricted holism entails particularism only because the second thesis just is a version of particularism. Thus, they conclude, “unrestricted holism cannot figure in an argument for particularism without gratuitously begging the question” (2005, p. 101). Now, I am inclined to think that the list of defeaters for any given moral rule is, in fact, uncodifiable, and so moral rules like U above are not stateable without either (a) being false or (b) containing unexplained moral terms or (c) being trivial in virtue of having a ceteris paribus clause or some other defeasibility marker. Thus, McKeever and Ridge seem to be operating under the assumption that a rule, in order to count as a rule, must not admit of exceptions that cannot themselves be codified. This is a common understanding of how rules work, and it is an understanding that inhibits our ability to see how rules can play the role they need to play in Sellars’s system. Nevertheless, perhaps McKeever and Ridge’s article is helpful, in that it shows that there is no easy route from holism to the sort of radical particularism embraced by some ethicists. A more direct attack on particularism has been carried out by philosophers such as Brad Hooker (2000). Hooker’s main dialectical opponent is Jonathan Dancy, who embraces something like McKeever and Ridge’s “unrestricted holism.”2 On such a view, any feature of a situation might have any moral valence, depending on context. Further, there is no way to codify these contextual features in a way that would allow for a generalist conception of morality. Hooker attempts to rebut a number of pro-­particularist arguments. But he also attacks particularism directly, offering as his central example that of Patty the particularist. Imagine that you and Patty find yourselves in the same predicament: Each of you needs help harvesting your crops or half of them will rot and you will go bankrupt. Patty promises that if you help her harvest her crops, she will help you. All that you know about Patty is that she is a committed particularist, and that there will be no way to use legal or other coercive means to make her fulfill her promise: “The only thing that could possibly make her keep her promise is her moral outlook” (2000, pp. 17–18). The question is, should you trust Patty to keep her promise? It is not obvious that you should. As Hooker writes, “As a particularist, she can and might attach no weight whatsoever to the promise when the time comes to keep it. And, as a particularist, she can’t point to any general considerations that mark off the situations in which a promise would be morally binding from the situations in which it wouldn’t” (2000, p. 18). There are really two issues at stake here. The first is a sort of pragmatic one. As I have argued in this book, morality is in large part a tool for solving a certain sort of social coordination problem. As Michael Bratman writes, “Policies facilitate coordination” (1999, p. 88). That is,

184  Defeasible Rules and the Particularist Challenge we can coordinate with others only if we can reliably predict what they are going to do. And particularism makes this difficult to impossible. The second reason is a more fundamental one about the nature of moral concepts. As Hooker writes, “We might think that someone would have to misunderstand the very nature of a promise in order to think that an informed uncoerced promise whose content is morally innocent could lack moral force” (2000, p. 18). That is, if you don’t understand that a freely given promise is prima facie binding, then you don’t understand what it is to make a promise. If particularism says that an uncoerced promise has no intrinsic moral valence or that one incurs no prima facie moral obligation by making such a promise, then particularism seems to misrepresent the nature of moral concepts.3 Particularism seems to face other significant challenges. Some worry that if radical particularism were true, morality would be unlearnable. Principles can be learned, but if there is nothing general at all shared between different contexts and different moral situations, how could we possibly learn to apply moral predicates correctly? (Jackson, Pettit, and Smith 2000) Further, the particularist has a hard time explaining why certain features (that something was painful, or that something was a promise, or that something was cruel) are relevant in their own right, whereas certain other features (that it involved broccoli, or that her shoelaces were white, or that it occurred in the month of June) are never significant in their own right but are only derivatively so in virtue of their connection to the former sorts of features.4 For these reasons and others, radical particularism doesn’t seem a promising route. What is the alternative to particularism? Hooker seems to regard the most obvious alternative, absolutism, as “mistaken” and “implausible” (2000, p. 21). Rather, Hooker wants to endorse a “Ross-­style generalism [that] gives promises only pro tanto force” (2000, p. 19). On such a view, promise-­breaking is always a minus, but it is a minus that can be overridden by sufficiently weighty considerations. In general, Hooker thinks that the particularist challenge leads to two possible responses on the part of the generalist. Either one can argue that some feature (say, breaking a promise or harming others) is always a “pro tanto moral minus,” or the generalist can “fine tune their principle” to try to accommodate the various defeaters (Hooker 2000, p. 11). This way of thinking seems to embody a fairly rigid view of how principles operate: Genuine principles do not admit of exceptions. If they did, they would not be principles. If two principles come into conflict, one might outweigh the other, but the defeated principle remains “in force”­— that is, the consideration named in the principle retains its moral salience even if it is outweighed. Thus, in Hooker’s preferred view, moral considerations do not switch valence. Promise-­breaking is always a moral minus. If we must admit that the sadist’s pleasure is a moral minus, then we must do so as a further specification of the moral rule regarding the

Defeasible Rules and the Particularist Challenge  185 moral valence of pleasure (i.e., the relevant principle must read something like “Nonsadistic pleasure is always a moral plus” [Hooker 2000, p. 8]). We saw a similar understanding of principles in McKeever and Ridge. They attempted to accommodate holism by building the list of defeaters into the principle in question so that the principle could still remain, in a sense, exceptionless­—if only because all exceptions were already mentioned in the principle itself. This understanding of principles rests on a dichotomy, one clearly seen in, for example, J.J.C. Smart’s classic work “Extreme and Restricted Utilitarianism”: A rule can only be an exceptionless principle or a rule of thumb­—that is, a generalization over individual cases but not a rule that has any authority in itself. On the latter conception, rules may be thought of as statistical generalizations, e.g., “In most cases, we have found that doing A in C maximizes utility,” but they are just that­—­generalizations with no independent explanatory or justificatory force. As Smart writes, “In themselves [the rules] do not justify any action” (1956, p. 348). Rawls (1955) calls this the “summary conception of rules.” This dichotomy represents an impoverished understanding of the forms that rules can take. Crucially, it seems to leave no room for the kind of rule that plays a central role in Sellars’s philosophical project­—and in this book­—namely, that of a defeasible material rule of inference, such as “If this dry, well-­made match is struck, then (ceteris paribus) it will light.” Such rules are genuinely rules­—and yet they admit of exceptions, exceptions that are arguably not codifiable. As I noted in Chapter 6, Brandom argues that the “ceteris paribus” phrase here essentially operates as a non-­monotonicity marker­—marking that the inference is one that we may make in the absence of infirming or defeating conditions. Replacing the “ceteris paribus” with an explicit list of infirming or defeating conditions is generally not possible, however. The problem, of course, “is not just that we would need an infinite list of the conditions being ruled out­—though that is true. It is that the membership of such a list would be indefinite: we do not know how to specify in advance what belongs on the list” (Brandom 2000, p. 88). So it seems that if we want to make good on Sellars’s claim that there are such things as defeasible material rules of inference, we need an account of rules that (a) have genuine authority as rules, (b) admit of exceptions, and (c) are not mere rules of thumb or generalizations. Thus, the solution is not to flee from exceptionless principles all the way into the arms of the radical particularist. The solution, as I have noted previously, is to attempt to discover some conception of principles according to which the principles are genuinely justification-­conferring but still allow of exceptions. Sellars calls such rules ­defeasible rules5 and cites H.L.A. Hart (1948–49) as an early expositor of the notion of defeasibility. Thus, defeasible rules are like what Bratman calls policies. Bratman writes that a policy “differs from an intention in favor of a necessary

186  Defeasible Rules and the Particularist Challenge means,” in that “if I were to balk at intending the necessary means I would have to abandon the intention in favor of the end. But this is not quite so when I apply a general policy to a particular case: the defeasibility of general policies makes it possible to block the application of the policy to the particular case without abandoning the policy” (Bratman 1999, p. 90). Thus, Bratman concludes, “I agree with Wilfrid Sellars that ‘it is essential not to assimilate action on policy to using means to achieve an end’” (Sellars TA, p. 136; quoted in Bratman 1999, p. 185n15).

8.2 Defeasible Rules We have good reason to be optimistic that there is a defensible conception of such rules, as a number of promising contemporary accounts have been developed. One such account is due to Mark Lance and Margaret Little. According to Lance and Little, a defeasible generalization is “a kind of generalization that is both genuinely explanatory and ineliminably exception-­laden” (Lance and Little 2004, p. 441). Lance and Little reject the idea that, for example, pain is only accidentally and contingently connected to moral badness. However, they agree with the particularist that such nonmoral properties can “switch valence.” Thus, they write, Pain is always bad-­making­—well, except when it’s constitutive of athletic challenge; intentionally telling a falsehood is prima facie wrong­—well, not when done to Nazi guards, to whom the truth is not owed, or when playing the game Diplomacy, where it’s the point of the contest. Pleasure always counts in favor of a situation­—well, except when it’s the sadist’s delight in her victim’s agony, where her pleasure is precisely part of what is wrong with the situation. (Lance and Little 2005, pp. 306–7) On the face of it, these two requirements­—that, say, pain is necessarily connected to moral badness but can sometimes switch valence­—are incompatible. But the point of Lance and Little’s account of defeasible generalizations is that it allows them to sketch such a necessary connection while making sense of the cases in which the valence changes. Let us examine this account in more detail. Generalizations like “Fish eggs turn into fish” and “Causing pain usually contributes to the immorality of an action” are true, other things being equal. “Other things being equal” does not mean that these generalizations are true in the majority of cases, for this need not be true. For example, only a small minority of fish eggs survive to become fish. However, there is also a sense in which these generalizations are necessarily true. That is to say, such generalizations “in some way are meant

Defeasible Rules and the Particularist Challenge  187 to capture the nature of the object in question” (Lance and Little 2005, p. 310). Thus, while few fish eggs ultimately turn into fish, there is a crucial privileging of cases in which they do. One classifies the fish egg as being the kind of biological organism it is by reference to its “standard” or “normal” development. There are of course an infinite number of trajectories that fish eggs could take, from developing into fish, to being enucleated with sheep DNA and becoming a sheep, to breaking down into nutrients for a turtle, to being irradiated and turning into a strange and dysfunctional pile of flesh. Nonetheless, we elevate one such trajectory as a “natural” one, viz., one that does not call for explanation (at least at this level of theory); and in this sense, we circumscribe some developments as expressions of an organism’s “nature.” (Lance and Little 2004, p. 446) Thus, in two senses, such generalizations are necessary. First, these generalizations are necessary, in that they are particularly revealing of the nature of some object (e.g., a fish egg) or some property (e.g., that of causing pain). Here are Lance and Little: Sometimes . . . when we issue a generalization to the effect that something has a certain feature, what we really want to say is not that such a connection always, or even usually, holds, but that the conditions in which it does hold are particularly revealing of that item’s nature, or of the broader part of reality in which the item is known. We might put it by saying that we’re asserting what happens in “normal” conditions, except that the notion of “normalcy” is so freighted with misleading connotations. Better put, then, we are taking as privileged, in one way or another, cases in which the item has the feature specified. Such generalizations tell us about the nature of something, not by eliminating exceptions to the connection, but by maintaining and demarcating their status as exceptions. (Lance and Little 2005, pp. 310–11) Second, in this privileged class of cases, the generalization is necessarily true. Lance writes, “To commit oneself to a defeasible moral generalization, in the sense that we develop, is to embrace a particular explanatory strategy, one that characterizes the moral structure of a given situation in terms of its relation to a morally and explanatorily basic type of context which itself is governed by exceptionless moral principles. Thus, to fully grasp a defeasible generalization one must understand first the ‘privileged’ contexts in which the generalization is true without exception, and then to possess the practical ability to make sense of non-­privileged situations as ‘riffs’ on the privileged” (Lance manuscript, n.p.). Thus,

188  Defeasible Rules and the Particularist Challenge every failure of the generalization to hold­—every exception­—must be explained in terms of some deviation from the privileged context or set of conditions. All exceptions are, in this sense, occasioned by “deviant” circumstances.6 Thus, lying is bad-­making, in general, but lying to the Nazis is morally permissible. However, the very moral permissibility of lying in this case can only be explained by the fact that the situation is deviant and morally defective: The Nazis are looking to round up innocent victims to be murdered. If a fish egg fails to develop into a fish, there will be some explanation (perhaps unavailable to us) for why its development was interrupted, whether this explanation is environmental, genetic, or what have you. If a particular bird is an exception to the generalization “Birds fly,” then there will be an explanation of why this instance is “deviant”­—either in terms of how this bird is injured or too young, or perhaps an evolutionary story about how the species as a whole lost its ability to fly (as with ostriches and penguins). None of this is to say that we can rebuild this into an account of monotonic inference by specifying what the privileged conditions are or by exhaustively listing the infirming conditions. As I have already argued, we can’t list the infirming conditions (the set of “deviant” conditions) because we simply don’t know them all (and because the list is open-­ended). And the set of privileged conditions is defined all but trivially as the conditions in which everything goes as it should and none of the infirming conditions obtain. For example, a biologist in a laboratory could construct a set of conditions under which a fish egg would reliably develop into a fish, but these would not thereby count as the privileged conditions, as fish eggs develop into fish under a much wider variety of conditions­—a wide variety that is, again, open-­ended and resistant to specification.7 Lance and Little’s account makes sense of Sellars’s insight that even defeasible material inferential principles are underlain by nomologicals; Lance and Little give a deeper explanation of why this should be and how it works. Further, I think this account sheds some important light on further aspects of Sellars’s project. Because these nomologicals hold without exception in privileged conditions and are revealing of the nature of the object or property in question, they are ideal to play the role envisioned by Sellars­—constituting the meaning of the terms in question. To delve a bit deeper into Lance and Little’s account, this account of the necessity of defeasible generalizations makes sense of the cases in which these generalizations don’t hold. Indeed, in many cases, it is only because of the truth of the generalization that we can make sense of deviant cases. As Lance and Little write, “In athletics for instance, it is only because pain is paradigmatically something to be avoided that the notion of physical challenge has the meaning, and the status of constitutive good, that it does. It is only because pain is normally bad-­making, then, that we can understand its good-­making instances” (Lance and Little 2005, p. 313). Thus, again, the account of defeasible generalizations they give explains how such

Defeasible Rules and the Particularist Challenge  189 generalizations are necessary while allowing us to make sense of the cases in which these generalizations don’t apply. This case also illustrates the fact that “non-­standard cases . . . while always perforce deviant, are not always defective” (Lance and Little 2004, p. 307). The Nazi case is a case where the circumstances are morally defective; the pain-­as-­constitutive-­of-­ athletic-­challenge case is merely deviant but not thereby defective. Formal accounts have been offered of the defeasibility of the reason relation. One of the best developed is Horty’s (2012), which explicates the notion of a reason in terms of default inference rules. The basic building blocks of Horty’s theory are fairly simple. A default is not a formula; it is a specific rule of inference, such as δ1: “If Tweety is a bird, then Tweety can fly” However, various default rules conflict with each other. Thus, δ1 conflicts with δ2: δ2: “If Tweety is a penguin, then Tweety cannot fly” Thus, as Horty notes, there is “a priority ordering < on these default rules” (2012, p. 22). In the above case, we would represent this priority ordering as follows: δ 1 < δ2 This means that “the default about penguins has higher priority than the default about birds” (2012, p. 23), and so δ2 is the binding default in this particular case. In the simplest case­—that of a fixed priority default theory, a theory in which “all priorities among default rules are fixed in advance” (2012, p. 22)­—we have the following elements: W: “A set . . . of ordinary propositions, taken to represent the agent’s fixed background information.” (2012, p. 22) D: “A set . . . of default rules.” (2012, p. 22)