Great Clarendon Street, Oxford, OX2 6DP, United Kingdom
Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries
All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above
You must not circulate this work in any other form and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press 198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2022940834
ISBN 978–0–19–284295–4
ebook ISBN 978–0–19–265461–8
DOI: 10.1093/oso/9780192842954.001.0001
Printed and bound in the UK by Clays Ltd, Elcograf S.p.A.
Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.
To the people who taught me philosophy at Harvard University, and at the University of Notre Dame
Acknowledgments
Introduction A Familiar Phenomenon
The Philosophy of Bias: Expanding the Playing Field
Apology
I. CONCEPTUAL FUNDAMENTALS
Diversity, Relativity, Etc.
Diversity
Relativity
Directionality
Bias about Bias
Biased Representation
Parts and Wholes
Pluralism and Priority
Explanatory Priority
Are People (Ever) the Fundamental Carriers of Bias? Processes and Outcomes
Unbiased Outcomes from Biased Processes?
Biased Outcomes from Unbiased Processes? Pluralism
II. BIAS AND NORMS
The Norm-Theoretic Account of Bias
The Diversity of Norms
Disagreement
The Perspectival Character of Bias Attributions When Norms Conflict
The Bias Blind Spot and the Biases of Introspection
The Introspection Illusion as a Source of the Bias Blind Spot
Why We’re More Likely to See People as Biased When They Disagree with Us
Is It a Contingent Fact That Introspection is an Unreliable Way of Telling
Whether You’re Biased?
How the Perspectival Account Explains the Bias Blind Spot, as Well as the Biases of Introspection
Against “Naïve Realism”, For Inevitability
Biased People
Biases as Dispositions
Bias as a Thick Evaluative Concept
Biased Believers, Biased Agents
Biased Agents, Unreliable Agents
Overcompensation
Norms of Objectivity
Some Varieties
Constitutive Norms of Objectivity
Following the Argument Wherever It Leads
Symmetry and Bias Attributions
Two Challenges
Norms without Bias?
Symmetry
Bias without Norms?
Pejorative vs. Non-Pejorative Attributions of Bias
Bias and Knowledge
III. BIAS AND KNOWLEDGE
Biased Knowing
Can Biased Beliefs Be Knowledge? Are Biases Essential to Knowing?
Knowledge and Symmetry
How and When Bias Excludes Knowledge: A Proposal
Knowledge, Skepticism, and Reliability
Biased Knowing and Philosophical Methodology
Are We Biased Against Skepticism?
Reliability and Contingency
A Tale of Three Thinkers
Bias Attributions and the Epistemology of Disagreement
On Attributing Bias to Those Who Disagree with Us
Main Themes and Conclusions
Five Themes Conclusions
Bibliography Index
Acknowledgments
Almost all of this book consists of previously unpublished material. The main exception to this is Chapter 6, §3, which updates material drawn from my paper “Following the Argument Where It Leads,” Philosophical Studies (2011). I utilize that material with the permission of Springer Publishers.
In addition to those thanked in that paper, I would also like to acknowledge a number of other individuals and institutions for their help along the way. Earlier versions of some of these ideas were presented in talks that I gave at Rutgers, Princeton, Notre Dame, Fordham, Union College, the Orange Beach Epistemology Workshop sponsored by the University of Alabama, and at Pacific and Eastern division meetings of the American Philosophical Association. I am grateful to the audiences present on those occasions for their feedback. Material for the book was presented in two graduate seminars, one at Princeton, the other at a joint Rutgers-Princeton seminar co-taught with Ernest Sosa. I am grateful to the participants in those seminars, and especially to Ernie for proposing our joint seminar and suggesting that I present some of my work on bias as my contribution to it. Two readers for Oxford University Press, one of whom was Endre Begby, the other of whom remains anonymous, provided timely, very helpful, and much appreciated sets of comments on an earlier version of the manuscript. In addition, I would like to acknowledge the following individuals for their help: Robert Audi, Alisabeth Ayars, Nathan Ballantyne, Lara Buchak, Pietro Cibinel, Brett Copenger, Silvia De Toffoli, Sinan Dogramaci, Felipe Doria, Andy Egan, Adam Elga, Jan Engelmann, Johann Frick, Samuel Fullhart, Fiona Furnari, Dan Garber, Jorge Garcia, Jeremy Goodman, Peter Graham, Alex Guerrero, Elizabeth Harman, Brian Hedden, Grace Helton, Thomas Hurka, Mark Johnston, Brett Karlan, Hilary Kornblith, Barry Lam, Tania Lombrozo, Harvey Lederman, Sebastian Liu, Errol Lord, Jon Matheson, Aidan McGlynn, Sarah McGrath, Tori McGeer, Philip Pettit, Ted Poston, Emily Pronin, Katie Carpenter Rech, Gideon Rosen, Peter Singer, Michael Smith, Joshua Smith, Roy Sorensen, Meghan Sullivan, Una Stojnić, Katia Vavova, Mike Veber, and Snow Zhang.
For some of the information about John Cook Wilson’s life used in the Introduction, I relied on Mathiew Marion’s entry on Wilson in The Stanford Encyclopedia of Philosophy. Peter Momtchiloff was an ideal editor, full of judicious advice and preternatural patience for my many missed deadlines, only some of which could plausibly be blamed on a global pandemic. I thank him for his encouragement and support of the project.
My greatest intellectual debts are to the people who taught me philosophy, to whom this book is dedicated. My love for philosophy was first ignited as an undergraduate at the University of Notre Dame in the 1990s when I encountered an unusually inspiring and encouraging group of teachers, a group that included Neil Delaney Sr., Alasdair MacIntyre,
Marian David, Vaughn McKim, David Solomon, Mike Loux, David O’Connor, Leopold Stubenberg, Karl Ameriks, Fred Freddoso, and Philip Quinn. Decades later, my memories of those early interactions continue to provide me with models for what a good teacher of philosophy should be like. At Harvard University, I had the further exceptionally good fortune to have as advisers Robert Nozick, Derek Parfit, and Jim Pryor, each of whom was almost unbelievably helpful and supportive of me in overlapping and complementary ways. When I finished my dissertation, each of the three advised me to publish it as a book, but I demurred, on the grounds that no one would want to read an entire book of philosophy by a previously unpublished graduate student. Although this book is not the one that they encouraged me to write, one of the standards that I kept in mind while writing it was to produce a book that they would have found worthwhile, and something that I would have been proud to present to them. I’m sorry that I’ll never have the chance to present a copy to Bob or Derek, or to discuss its ideas with them.
My greatest debts of all are to my family. Thanks to my wife Sarah McGrath for all that you did for me and for our children during the years in which this book was written. Thanks to Owen, Orla, and Hugh for being the source of most of my smiles, and for making things seem worthwhile. Finally, thank you to my parents, Tom and Toddy Kelly, for your unwavering and unstinting love and support in all things, for as long as I can remember.
Introduction
1. A Familiar Phenomenon
Which types of human beings best exemplify true courage?
When asked this question, some people might immediately think of soldiers on a battlefield or of firefighters rushing into a burning building. Others might think first of political dissidents or civil rights activists, individuals who knowingly risk grave harm in order to speak Truth to Power.
When the philosopher Plato posed this question over two thousand years ago, he offered us a surprising answer. True courage, Plato suggested, is most likely to be found among the philosophers. Even allowing for the significant differences between the philosophers of Plato’s time and the professors of philosophy of the 21st century, Plato’s answer to his own question does not exactly leap off the page at the reader as the most plausible thing he might have come up with to put it mildly. Moreover, coming as it does from Plato, the answer strikes us not only as implausible but also as suspicious. It’s very much the type of answer that we might expect a philosopher but no one else! to give. Although he offers us a characteristically complicated argument for his view, we naturally suspect Plato of bias when he tells us that it’s really the philosophers who are the most courageous of all.
Of course, if in fact Plato was biased about certain questions, he’s in good company. Presumably, people have exhibited biases in their behavior, as well as in their beliefs about what’s true for as long as there have been people. Concerns about bias usually, concerns about the biases of other people, and more occasionally, concerns about one’s own also have a long history. In philosophy, an interest in the topic of bias predates any of the figures or texts that are treated as canonical by the Western philosophical tradition: in the East, Confucius took the absence of bias to be one of the characteristic features that distinguishes the virtuous person from others.1
The word “bias” entered the English language in the 1500s. Its origins are instructive. One of its earliest uses and the one from which all current uses of the word that are relevant to this book derive was as a technical term in competitive sports. In the English sport of lawn bowling, or bowls, the object of the game is to roll your ball (or “bowl”) as closely as possible to another, smaller ball that has already been rolled somewhere on the field of play. (As that description suggests, the sport is a close cousin to the now somewhat better-known sport of bocce ball. The English rulers of the day generally disapproved of lawn bowling, on the grounds that its surging popularity might divert interest from archery, which they viewed
as better preparation for warfare.) Although the basic object of the sport is simple, much of its interest and challenge comes from an important twist. Each ball was constructed so as to have a certain “bias,” which distorts the direction in which it would otherwise travel when it is rolled. Originally, the bias was produced by making sure that one side of each ball was weighted more heavily than the other side. (Usually, lead was used to achieve this purpose.) This lack of balance ensured that the ball would travel, not in a straight line, but in a curved path in the direction of its bias. The general idea of a bias as that which distorts something in a certain direction, away from the natural path or direction that it would otherwise take, perhaps through a mechanism involving unbalanced weighting, survives to this day in many of the most prominent contemporary uses of the term “bias.”2
As the example of Plato suggests, intelligence is no guarantee against bias. The ease with which even the most impressive among us succumb to bias, and the perceived importance of avoiding it, sometimes makes its apparent absence especially celebrated. When, fifteen years after George Washington’s death, Thomas Jefferson sat down to explain what in his eyes made Washington a great leader, he gave pride of place not to Washington’s genius for military tactics, nor to his reputed personal honesty, but rather to his capacity for unbiased decisionmaking:
His integrity was most pure, his justice the most inflexible I have ever known, no motives of interest or consanguinity, of friendship or hatred, being able to bias his decision
Of course, Washington was not without his biases. Some of these were invisible to Jefferson, since he shared them, and they were common among those whom he regarded as his peers. Like Jefferson and Washington, we too no doubt have biases that are largely invisible to us, perhaps because they are shared by those around us whom we respect and admire, including some biases that will seem obvious to later generations. Like Jefferson, we too are much concerned with bias and its absence, both among our contemporaries and among those now dead.
Indeed, although bias has undoubtedly always been with us, our explicit concern with bias, and our tendency to see or conceptualize our most pressing social problems in terms of it, has perhaps never been greater than it is now. Notably, this tendency is shared by people who might otherwise seem to have little in common, such as those who occupy opposite ends of the political spectrum. For example, on the one hand, a prominent theme among social critics on the left is that radical changes to existing institutions are called for in response to pernicious and systemic racial biases, as manifested in the criminal justice system, or in the glaring disparities in wealth among different racial groups. On the other hand, conservative critics who deny that such changes are in order will frequently make much of what they regard as entrenched political biases in the media and in higher education. Moreover, on both sides, there is an emphasis on the various forms that biases can take. Even if bias was once thought of primarily as a characteristic of individual people, when the would-be reformer condemns the criminal justice system as racially biased, or the conservative claims that the mainstream media has a left-wing political bias, each is concerned with bias primarily as a characteristic not of individuals but rather of institutions, at least in the first instance. That
bias is the problem or at least, a big part of our problems is a common theme, even among people who agree about little else.
2. The Philosophy of Bias: Expanding the Playing Field
This book is a philosophical exploration of bias, in the sense of “bias” that we naturally suspect Plato of when he tells us that, contrary to what we might have thought, it’s actually the philosophers who are the most courageous of all. (Although as will become clear, the book is not exclusively concerned with biases exhibited by people.) I originally conceived of the project as a relatively wide-ranging exploration of a number of theoretically interesting issues about bias, issues that were largely independent of one another, as opposed to the development and defense of some more general, overarching theory of bias. For the most part, the book reflects that original vision. In particular, much of it consists of a series of independent proposals about different questions, and the individual chapters are largely selfcontained and can be read more or less independently of one another, in accordance with the reader’s interests.3
Nevertheless, despite my original intentions, and despite my awareness that the historical track record of philosophers who offer very abstract and general theories of this or that phenomenon is dismal to the point of embarrassment, as the project progressed I found myself increasingly attracted to a general theoretical framework for thinking about bias that seemed both fruitful and to illuminate many of the questions that had originally drawn me to the topic in the first place. I call this theoretical framework the norm-theoretic account of bias, and I put forward the basic ideas in Chapter 3. (Readers who wish to begin by considering that account might turn directly to Chapter 3. In particular, what I say there does not presuppose acceptance of or even acquaintance with the conclusions and arguments of the first two chapters, which are devoted to a number of fundamental conceptual questions about bias that I believe can be pursued independently of a commitment to any more specific theoretical framework.) Various later sections of the book are devoted to elaborating, refining, and defending that account, as well as to exploring some of its limitations. To the extent that I have a general “theory of bias” to offer, the norm-theoretic account is it.
In order to introduce some of the basic ideas of the account, an analogy is useful. Imagine three archers who take turns shooting at a target. One of the three archers is highly skilled and regularly hits the target; the other two are unskilled and typically miss it. However, the two unskilled archers differ from one another in a significant way. Although the first unskilled archer typically misses, her misses have no pattern: she is as likely to miss to the right as to the left, as likely to miss high as low, and so on. In contrast, the other unskilled archer has a hitch in his shooting form, which causes him to regularly miss low and to the left. Given the hitch in his shooting form, he is disposed to miss low and to the left, and this is something that is true of him even when he is not actually shooting, even if the main evidence that observers have that he is disposed in this way depends on his actually having missed low and to the left in the past. On the norm-theoretic account of bias, a biased person
is much like the archer who is disposed to miss in a particular direction, as opposed to an archer who consistently succeeds in hitting the target or an archer whose misses are random.
In slogan form and to a first approximation, the core thought behind the norm-theoretic account is this: a bias involves a systematic departure from a genuine norm or standard of correctness. When people (or groups of people, or institutions) count as biased, this is because they systematically depart from certain norms, or because they are disposed to do so. I certainly claim no great novelty for this way of thinking about bias.4 Indeed, especially given how much attention has been devoted to the topic of bias in various contexts, I would regard with suspicion any account of it that did not strike us as relatively familiar, at least in broad outline. This core thought can usefully be regarded as a kind of generalization of or abstraction from the more specific notions of bias that are in play in various disciplines, as well as one that’s faithful to much that we say and think about bias in everyday life. To take one important example, consider the notion of bias as it is understood in statistics. In statistics, bias is the tendency of a statistical technique to either overestimate or underestimate the true value of some parameter. In this context, the relevant norm or standard is truth or accuracy, and something counts as biased in virtue of departing from the truth or the actual value in a patterned or predictable way for example, by consistently either underestimating or overestimating it as opposed to a way that’s unpatterned, unpredictable, or random. (As in a case in which someone repeatedly guesses wildly, and thus ends up missing the truth in a way that’s “all over the map.”) According to the norm-theoretic account of bias, the notion of bias that’s of interest to statisticians is actually a species or special case of a much more general phenomenon: the special case in which the relevant norm is truth, or accurate estimation.5
In the archery example, success consists in hitting the target. What is analogous to hitting the target when it comes to bias? As just noted, in many cases the relevant norm or standard of correctness is truth or accurate estimation. However, although truth is often an important norm, in many contexts the relevant norm that determines whether something counts as biased might have nothing to do with truth or accuracy. For example, many social scientists and philosophers think that an action is rational if and only if it maximizes expected value from the agent’s point of view. If in fact the norm governing rational action is that of maximizing expected value, then according to the norm-theoretic account of bias, a person might count as biased in virtue of their tendency to systematically depart from this norm. Consider, for example, status quo bias, the tendency to favor actions that preserve the status quo over alternative actions that are equally good or better options. According to the normtheoretic account of bias, a person who exhibits status quo bias counts as biased in virtue of being disposed to depart from the norm of maximizing expected value in a way that’s patterned, predictable, and systematic, as opposed to in a way that’s random or unsystematic. More generally, many behavioral biases can be understood as tendencies to systematically depart from the norm of maximizing expected value, just as many cognitive biases can be understood as tendencies to systematically depart from the norm of truth or accuracy.
One more example, for now. Consider the moral norm according to which we should treat other people with the respect to which they are entitled. A person who regularly fails to treat
others with the respect to which they are entitled violates an important moral norm and is properly subject to criticism for it. It doesn’t follow, however, that he’s biased. (Perhaps he’s simply an unbiased jerk, whose failures to treat others with respect don’t depend on their being a certain race, or sex, or age. He’s an “equal opportunity offender,” as it were.) Suppose, however, that he’s more likely to violate the norm with respect to women as opposed to men, or blacks as opposed to whites, or the old as opposed to the young. In that case, his departures from the relevant norm exhibit a systematic pattern, and it’s appropriate to attribute to him a racist, sexist, or ageist bias, as the case may be.
According to the norm-theoretic account of bias, biases typically involve systematic departures from norms. As is often noted, the term “norm” is ambiguous, in multiple ways.6 What sense of the term is relevant here? First, the relevant sense is not that of a statistical norm. A contemporary American family that has eight children is outside the norm in that respect, but that doesn’t mean that the family or any of its members is biased. Second, the relevant sense of “norm” is also not the sense which is in play in the academic literatures on “social norms” or “gender norms,” according to which norms are, roughly, a matter of a given society’s unwritten rules, conventions, or expectations about how people should behave. In that sense of the term, it’s a norm in many societies that women as opposed to men will do a majority of the housework and child-rearing. However, even if they belong to a society in which that norm holds sway, it doesn’t follow that a husband-and-wife pair who divides up the housework and child-rearing tasks evenly is biased. On the contrary, their egalitarian division of labor might very well reflect a lack of bias about such matters. Similarly, any pathbreaking artist, trendsetter, or iconoclast will systematically depart from prevailing norms in the sense of common expectations about behavior, but they are not biased on that account.
Contrast the sense of “norm” in which many of us think that it’s a genuine norm that people should be treated with respect. In this sense of the term, if it’s a genuine moral norm that people should be treated with respect, then we have good reason to treat them in this way, and we should. In this sense, the truth of the claim that treating people with respect is a genuine norm doesn’t depend on how often people are treated in this way in practice (even if it’s common for people to be treated with disrespect, that has no tendency to show that it’s not a genuine norm), or on how it fits with the unwritten rules and social expectations that prevail in a given society. (Even in a society in which people are expected to show disrespect to members of certain groups, this is consistent with the idea that those members should be treated with respect.) According to the norm-theoretic account of bias, when someone is properly subject to criticism as biased, this involves their systematically departing from a genuine norm in this sense.7
Although genuine moral norms provide a paradigm of the relevant kind of norms, they are a species of a more inclusive genus. As noted above, many economists hold that maximizing value from the agent’s point of view is the norm of rational action. If so, then one can count as biased by departing from that norm in certain systematic ways. Here, the relevant norm is, at least in the first instance, a norm of practical rationality as opposed to morality. More generally, there is an impressive diversity of norms relative to which someone or something
might count as biased, including moral norms, norms of justice, norms of accuracy, and norms of both theoretical and practical rationality. (For discussion of these, see Chapter 3, §1, “The Diversity of Norms.”)
Although the core idea that biases involve systematic departures from genuine norms might seem familiar, at least in some of its instantiations, I argue that when it is developed in the most plausible way it has far-reaching and radical implications. For example, I argue that both morality and rationality sometimes require us to be biased, in the pejorative sense of “biased” (Chapter 3, §4). The norm-theoretic account also has radical implications for the way we should view other people and our relationships to them. For example, in many cases of disagreement, we are rationally required to view those who disagree with us as biased, even if we know absolutely nothing about how they arrived at their views or about their psychologies, beyond the mere fact that they disagree with us in the way that they do. Appreciating why this is so allows us to offer a compelling explanation of a theoretically interesting and practically important empirical phenomenon: the fact that accusations of bias often inspire not only denials but also countercharges of bias, to the effect that the original accusation of bias is due to the fact that those who make it are themselves biased (Chapter 3, §3).
Another set of applications of the norm-theoretic account concerns the so-called “bias blind spot,” or our tendency to see bias in other people in ways that we fail to see it in ourselves. The bias blind spot has been extensively documented and discussed by social psychologists in a fascinating body of research.8 I argue that, when developed in its most plausible form, the norm-theoretic account offers a compelling explanation of the bias blind spot, an explanation that improves upon the kinds of hypotheses favored by the psychologists (Chapter 4, §§4 and 5). This explanation depends on what I call the perspectival character of bias attributions. As I understand it, the perspectival character of bias attributions has both a psychological aspect and a rational aspect. As a psychological matter, our views about a topic for example, our views about politics naturally influence our higher-order judgments about who is biased about that topic (and vice versa), in predictable and familiar ways. But crucially, what holds as a matter of psychology holds with respect to rationality as well. Thus, our first-order views about a topic rationally constrain and influence our higherorder judgments about who is biased about it (including our higher-order judgments about ourselves), in systematic ways. Moreover, the perspectival account of bias attributions that I develop also provides insight into another notorious and well-documented psychological phenomenon: the fact that introspection is a highly unreliable way of recognizing one’s own biases (Chapter 4, §§3 and 4). I argue that the unreliability of introspection as a means for detecting our own biases is not a contingent fact of our psychology, but rather something that holds of necessity. To put it provocatively, if a bit too crudely: on the account that I offer, even God could not have made us creatures who reliably detect their own biases by introspection.
In typical cases in which someone counts as biased because they systematically depart from a norm, the norm from which they depart is not itself about bias. For example, norms like treat other people with the respect to which they are entitled, or maximize expected value, don’t say anything about bias, and they can be grossly violated by a perfectly unbiased
person no less than by a person who is biased. However, some norms are specifically concerned with bias, in one way or another. For example, a norm might tell us to correct for a bias, or to take steps to prevent its manifestation, or to be objective in certain ways. I call these norms of objectivity, and I explore them in detail in Chapter 6. There I distinguish between three fundamentally different kinds of norms of objectivity norms of preemption, norms of remediation, and constitutive norms of objectivity and I explore their characteristic features and the ways in which they interact with each other and with norms of other kinds (Chapter 6, §§1 and 2). I also offer an account of a venerable intellectual ideal closely associated with objectivity, the ideal of “following the argument wherever it leads” (Chapter 6, §3).
As some of the applications mentioned above suggest, the norm-theoretic account of bias has implications not only for questions about what bias is but also for related but distinct questions about the circumstances in which we attribute bias. Although that topic recurs throughout the book, it is explored most thoroughly in Chapter 7, “Symmetry and Bias Attributions.” Among the issues that I explore there is the pervasive role of symmetry considerations in guiding our thinking about bias. I also address what I take to be an important feature of our use of the term “bias”: namely, that while many uses of the term in both the sciences and in everyday life presuppose or imply that being biased is at least in some respect a negative thing, it is also true that people often use the term in both contexts in ways that don’t presuppose or imply this. I argue that, far from being a relatively uninteresting case of simple ambiguity, this is actually a significant fact about the way in which bias attributions work, and I offer an account of our practices that makes sense of it. The account developed there also proves useful in answering a number of objections to the norm-theoretic account, including objections that proceed from the facts that (1) we sometimes attribute bias even though we don’t think that any genuine norm has been violated, and (2) conversely, there are some systematic departures from genuine norms that we would not ordinarily count as biases.
As noted above, although the norm-theoretic account plays a prominent role in the book, much of what follows is independent of it. A major theme and the starting point of Chapter 1 is the sheer diversity of things to which we apply the labels “biased” and “unbiased.” Thus, we routinely describe people as biased, as well as many of their mental states (e.g. their beliefs or perceptions); their actions, including their linguistic behavior and its products (e.g. the testimony that they offer); and the larger groups to which they belong. (For example, just as we might describe an individual Supreme Court judge as biased, so too we might attribute bias to the Supreme Court as a whole.) In addition, we also attribute bias to some inanimate objects (e.g. dice and coins), many temporally extended processes (as in “the biased admissions process”), as well as to the outcomes of those processes (as in “the biased admissions decision”) and much else besides. How are these things related to one another? Are some of them more fundamental than others? If so, which?
For example, imagine that a biased judge arrives at a biased verdict by reasoning in a biased way. Here we have at least three different things that can accurately be described as biased: the verdict, the process by which it is reached, and the judge himself. Assuming the obvious point that it’s not a coincidence that all three things count as biased, what exactly is
their relationship? Does the verdict count as biased because it is produced by a biased reasoning process, or does the process count as biased because it produces biased verdicts? And what is it, exactly, that makes the judge himself count as biased the fact that he arrived at a biased verdict, or that he reasoned in a biased way, or both, or something else entirely? I explore questions of this general kind in Chapter 2, “Pluralism and Priority.” Among other conclusions, I argue that unbiased processes sometimes produce biased outcomes, and that biased processes sometimes produce unbiased outcomes. The general picture that I endorse there is a kind of robust pluralism about bias. The view is pluralistic in two different respects. First, a wide variety of different things (including things that belong to fundamentally different categories) are genuinely biased. The fact that in the course of everyday life we attribute bias more or less all over the place is not a result of overly causal speech or sloppy thought on our part. Rather, it reflects something deep about the nature of bias itself. (This is the “pluralism” in “robust pluralism.”) Second, not only are many radically different kinds of things genuinely biased, but no one of these is always fundamental or foundational compared to the rest. Rather, among the many things that can be biased, different types are fundamental in different contexts. This is the sense in which the pluralism that I endorse is robust.
Broadly conceptual questions about the nature of bias of this general sort are pursued throughout the book. Can one be biased in favor of something without being biased against anything, or vice versa? Or do “positive biases” and “negative biases” always come in complementary pairs (Chapter 1, §3)? What is the relationship between a whole’s being biased and bias among its parts? For example, what is the relationship between a group of people having or lacking a certain bias, and the individual people who make up the group having or lacking that bias? I argue that although a group’s being biased often depends upon (and is explained by) a critical mass of its members having that bias or some relevantly similar one, there is no necessity in either direction: a whole might be biased even if all of its parts or members are unbiased, and conversely, a whole can be unbiased even if its parts are biased (Chapter 1, §6). I explain how even a news organization that consistently offers perfectly unbiased coverage of every event that it covers might nevertheless be biased, and I consider the significance of that possibility.
In addition, I argue that whether a person is biased about a given issue is sometimes a relative matter (Chapter 1, §2); that biases of people are best understood as multiply realizable dispositions (Chapter 5, §1); and that two people might share the same bias for example, the same racial bias even if they share nothing else in common, in terms of their mental states, their behavior, or their dispositions (Chapter 5, §3). I also explore a number of striking facts about bias, including that fact that a severely biased person might be highly reliable if their bias dovetails with their environment in the right way (Chapter 9, §§3 and 4); that one common way of ending up biased is by trying not to be, as in cases of overcompensation (Chapter 5, §5); that even an account of a person or historical event that is wholly true and known to be so might nevertheless be biased (Chapter 1, §5); and that claims of bias often presuppose substantive and potentially controversial evaluative or normative claims about how it’s appropriate to think or act, claims that are not themselves about bias (Chapter 1, §5, Chapter 3, §3, Chapter 5, §2).
In addition to work by other philosophers, this book draws on research by psychologists, economists, political scientists, historians, and theorists of law. It aspires to be of interest to a relatively broad audience. Nevertheless, in various respects it is very much a work of philosophy and reflects the fact that it was written by an analytic philosopher, for better or for worse in its choice of topics, its manner of addressing them, and the style in which it is written. Moreover, even when compared to other works of analytic philosophy, the questions pursued here reflect my own biases or at least, my professional training and knowledge base. One of the fascinating things about bias as a topic of philosophical reflection is the way that it cuts across so many of the various subfields of the subject, including the philosophy of mind and psychology, social philosophy (especially feminist philosophy and the philosophy of race), the general theory of knowledge, and moral and political philosophy, as well as others. Although the book addresses issues that are relevant to each of these subfields, a relatively large portion of it is devoted to questions that belong to the theory of knowledge, in addition to the kinds of quite general metaphysical and conceptual questions about bias that are center stage for much of it.
In particular, Part III, “Bias and Knowledge” is devoted to exploring connections between bias and various topics that belong to the theory of knowledge, including skepticism, the underdetermination of theory by evidence, cognitive reliability, the epistemic significance of disagreement, as well as the connections between bias and knowledge itself. Among other conclusions, I defend the possibility of what I call “biased knowing” (Chapter 8, §1), and I argue that its possibility has significant implications both for philosophical methodology and for skepticism (Chapter 9, §§1.and 2). I explore the commonsense idea that a belief is not knowledge if its formation is the manifestation of a bias (Chapter 8, §2). I also take up the seemingly incompatible idea that, far from excluding knowledge, biases are actually essential to knowing, and are in fact deeply implicated in paradigmatic instances of knowledge acquisition and exemplary human reasoning, an idea that might be plausibly taken to be a central lesson of some of the most interesting science and philosophy of the last seventy years (Chapter 8, §3). I argue that, properly interpreted, there is a sense in which both of these ideas are true, and I show how they can be consistently reconciled. (See Chapter 8, §4 and especially §5, “How and when bias excludes knowledge: a proposal.”) In Chapter 9, “Bias Attributions and the Epistemology of Disagreement,” I devote extended attention to an extremely common but also extremely suspicious habit, our tendency to attribute bias to well-credentialed and seemingly formidable people on “the other side” of controversial issues, in a way that serves to minimize the perceived pressure to reconsider or revise our own views about those issues. I propose that both the presence of bias and its absence are subject to strong “the rich get richer, and the poor get poorer” effects, as initially biased believers tend to fall into cognitively vicious cycles while initially unbiased believers tend to benefit from cognitively virtuous ones.
This focus on broadly epistemological questions about bias, particularly in the later parts of the book, is not based on a conviction that these questions are more important than any number of other philosophical questions about bias. Indeed, I do not even claim that they are as important as some other philosophical questions about bias that are not pursued here. (Still less do I claim that they are as important as many of the questions about bias that are
typically pursued outside of philosophy, e.g. questions how best to reduce or eliminate some pernicious bias in certain concrete contexts.) But they seemed like the issues that I was best prepared to address given my philosophical background and knowledge base, and the ones about which I was most likely to say something worthy of the consideration of others.
At least compared to the amount of attention that bias has received from psychologists, it’s fair to say, I think, that the topic has been relatively underexplored by philosophers. In recent years, that has begun to change. Much of this recent philosophical work concerns the theoretically interesting and practically important topic of implicit bias.9 Notwithstanding the interest and importance of that topic, this book is not concerned with implicit bias in particular as opposed to the more general phenomenon of which it is a species. (Although as that suggests, I would regard it as a good objection to the claims made here if they failed to apply to the phenomenon of implicit bias.) Inasmuch as a good number of the issues that I pursue have not been directly addressed in the philosophical literature, one aspiration of the current work is to expand the philosophical playing field and to put more questions into play, if only by proposing inadequate answers to them that can then be criticized and improved upon by others.
3. Apology
John Cook Wilson, the founder of a philosophical movement known as “Oxford Realism,” held one of the most prestigious professorships at Oxford University from 1889 until his death in 1915. Although it is not among the views for which he is remembered today, Cook Wilson thought that he had identified an important bias that amounts to a kind of occupational hazard for authors who publish their work. In the century since his death, psychologists and others have conducted countless studies of an ever-increasing list of biases. Despite this, I do not believe that the specific bias which concerned Cook Wilson has ever been systematically investigated by anyone. In fact, subsequent history has paid so little attention to this bias that it does not, as far as I know, even have a name.
Cook Wilson’s idea was that, once authors commit to a view in print, they would, more often than not, be disposed to continue to believe the view and defend it even if later investigation turned up good reasons for thinking that it is mistaken. They would thus be inclined to engage in fruitless and unproductive exchanges with their critics, even when their critics were right. By contrast, an author who has not yet published their view will be at least relatively open to changing their mind in response to compelling objections. (Contrast the well-known phenomenon of confirmation bias. In the case of confirmation bias, the important distinction is between views that the believer already holds and views that they do not already hold. In the case of Cook Wilson’s bias, the important distinction is between views that one holds but has not yet published and views that one has already published.)
In part because of this concern, Cook Wilson published little philosophy during his lifetime. Instead, he privately circulated pamphlets containing the latest statements of his views among his colleagues and students, pamphlets that he was constantly revising. Only
after many years did he sit down to finally compose his magnum opus. Unfortunately, by then he had only a short time left to live, and he never came close to finishing the task. It was not until eleven years after his death that the two-volume treatise that bears his name, Statement and Inference, finally appeared. Still, it is clear that this treatise is very far from the book that Cook Wilson would have written had he lived: it was cobbled together by a colleague from Cook Wilson’s lecture notes, the circulated pamphlets, and some letters that he had left behind.
I have no reason to think that I am immune to the scholarly bias that so worried Cook Wilson. Like him, I have no appetite for engaging in unproductive exchanges with critics. On the other hand, I also have a very strong preference for this book to appear while I am still alive as opposed to dead. (And even if I did not have that preference, I have no disciples to whom its posthumous publication could be entrusted.) Given these considerations, it seemed on balance to make sense to proceed and place the present volume before the public, including potential critics. I thereby run the risk of losing a level of objectivity about its weaknesses that I might otherwise have retained, and of being drawn into unproductive exchanges in which I attempt to defend the indefensible. I apologize in advance for any annoyance or frustration that might be caused in these ways.
1 “The virtuous man can see a question from all sides without bias. The small man is biased and can see a question only from one side ” The Analects, 2 14 (Confucius 1979)
2 As is often the case when it comes to the development of the English language, Shakespeare seems to have been a key figure in the relevant history He uses the word “bias” eleven times in eight of his plays In some cases, the word is used in talking about the game of bowls. But in other cases, his characters use it figuratively, extending the concept familiar from the sport in order to describe their own situations, and in a way that’s recognizably akin to common contemporary uses
3 The proposals are summarized in the eleventh and final chapter of the book, “Main Themes and Conclusions.”
4 For example, among psychologists, compare the characterization offered in Baron (2012:1).
5 As the statistics example suggests, the sense of “systematic” that matters for the norm-theoretic account is the sense which contrasts with “random,” as opposed to the sense that means “done or acting according to a fixed plan or system.” Notoriously, many biases operate in ways that don’t involve anyone’s acting in accordance with a fixed plan or system For an intellectually engaging account of the contrast between bias and random error or “noise,” and an attempt to redress the usual imbalance of attention towards the former at the expense of the latter, see Kahneman, Sibony, and Sunstein (2021)
6 See, e.g., Brennan et al. (2013):2–4. Compare Wedgwood (2018):23–4.
7 The relevant sense of “norm” is thus the same as that which figures in “normative economics” (as opposed to “positive economics”), “normative jurisprudence,” or, in philosophy, “normative ethics” and “normative epistemology” Even once statistical norms and social norms are set aside for current purposes, one might still think that any attempt to understand bias in terms of systematic norm violations will inevitably yield an overly inclusive account, one that incorrectly attributes bias even where none exists For example, a serial killer who consistently employs a particular method of execution violates a genuine moral norm against harm in a way that is non-random, but it doesn’t follow that he’s biased. (Imagine that he selects his victims without regard to their race, ethnicity, gender, etc.) Conversely, one might also think that any such account will also be underinclusive and fail to count some genuine cases of bias as such For example, we frequently attribute bias to inanimate objects such as coins and dice, and a coin that counts as biased in virtue of being disposed to land heads rather than tails seems to have something important in common with a judge who counts as biased in virtue of being antecedently disposed to rule in favor of the prosecution rather than the defense However, coins are not subject to genuine norms, in
anything like the way judges (or more generally, people) are subject to genuine norms. For a discussion of such considerations and how they can be accommodated within the kind of approach explored here, see especially Chapter 7
8 See especially Pronin, Lin, and Ross (2002); Pronin, Gilovich, and Ross (2004); Ehrlinger, Gilovich, and Ross (2005); Armor (1999); and Ross, Ehrlinger, and Gilovich (2016), as well as the further references provided in Chapter 4
9 For a sampling of the literature, see Beeghly and Madva (2020), Johnson (2021), Mandelbaum (2016), Gendler (2011), Schwitzgebel (2010); the useful surveys by Brownstein (2017), Holroyd (2017), Kelly and Roedder (2008), and Holroyd, Scaife, and Stafford (2017); and especially, the two volume collection edited by Brownstein and Saul (2016). The point of departure for this philosophical literature is the explosion of empirical research on implicit bias in recent decades within social psychology. See, e.g., the landmark study by Greenwald and Banaji (1995) and, for a popular overview, Greenwald and Banaji (2013).
PART I
CONCEPTUAL FUNDAMENTALS
1
Diversity, Relativity, Etc.
1. Diversity
Let’s start with a very general question: What types of things can be biased or unbiased?
Clearly, not everything is of the right type. The number 17 has various properties, such as the property of being odd as opposed to even. But it doesn’t seem to make much sense to claim that the number 17 is biased or unbiased, and the same seems true of every other number. Similarly, most of the things that natural scientists theorize about, ranging from the very large (e.g. planets and universes) to the very small (e.g. subatomic particles) aren’t the kinds of things that we would naturally describe as either biased or unbiased. Thus, the mere fact that something lacks the property of being biased doesn’t mean that it’s unbiased, in the sense in which “unbiased” is more or less synonymous with either “objective” (if what’s under consideration are, e.g. judges) or with “fair” (if what’s under consideration are coins). Biased and unbiased are contraries, not contradictories. Although the fact that something is unbiased entails that it’s not biased, the fact that it’s not biased does not entail that it’s unbiased. (My office desk chair lacks the property of being biased, and is therefore not biased, but it’s not unbiased.) Roughly and to a first approximation: something is unbiased if and only if it’s the sort of thing that might have been biased, but isn’t.1
Nevertheless, although many things are neither biased nor unbiased, it’s striking how many things do routinely get described in these terms. It will be helpful to briefly survey some of the more important categories:
We frequently predicate bias of people or particular individuals. (We say: “He’s biased” or “She’s biased” with respect to some issue or cluster of issues.)
We predicate bias of individuals in their social roles, as in “the biased judge.”
Similarly, we attribute bias to groups or collections of people (e.g. “the biased committee.”)
We attribute bias to inanimate objects (e.g. “the biased dice,” or “the biased coin”).
We often predicate bias of things that can play the role of evidence. For example, we talk about biased samples, biased testimony, biased surveys, biased data, biased information, biased studies, biased research, and biased tests. Relatedly, we attribute bias to sources of information or putative information (e.g.
•
• • •
• “Fox News has a conservative bias,” “MSNBC has a liberal bias.”)
We sometimes attribute bias to temporally extended processes, practices, and procedures, as in “a biased admissions process” or “a biased job search.”
An especially important category for both philosophers and psychologists consists of mental states: we regularly speak of biased perceptions, biased beliefs, biased judgments, biased opinions, and so on.
Overlapping with the previous category, we frequently attribute bias to the outcomes of deliberative processes: thus, we talk about biased verdicts, biased decisions, biased evaluations, biased grades, and so on.
We often predicate bias of broadly linguistic phenomena, as when we speak of biased narratives, biased texts, biased labels, biased descriptions, biased testimony, biased interpretations, biased presentations, biased discussions, biased accounts, and biased reports.
Recently, there has been significant discussion of biased algorithms. 2
This list is far from a complete inventory.3 Nevertheless, it suffices to make the point that many different things can be biased, at least if we take ordinary thought and talk at anything like face value.
Generally speaking, to describe a person or thing as biased is not to make a neutral statement about that person or thing. For example, in any ordinary context, the claim that a particular judge is biased would naturally be understood as a criticism, and if the claim were made in the judge’s presence, we would naturally expect them to deny it. Similarly, when someone claims that a particular interpretation of an historical event or a text is biased, we naturally take them to be disputing that interpretation; someone who claims that a scientific study or political poll is biased is naturally understood as suggesting that its putative findings shouldn’t be accepted at face value, and so on.
However, although attributions of bias often involve a negative judgment about a person or thing, this is not always the case. In fact, in both everyday life and in the sciences, it’s also common for people to attribute bias where no negative judgment is intended; in such cases, the attribution functions much more like a neutral description. Consider a couple of examples of this. When cognitive scientists construct models of human cognition in order to study how human beings manage to learn about the world, they sometimes speak of our “inductive biases.”4 In such contexts, there is no suggestion that inductive biases are a bad thing, or that it would be better if we didn’t have them. On the contrary, it’s assumed that such biases play an indispensable role in human reasoning at its best, and that without them, learning from experience would be impossible.5
Compare the following remark from Tyler Burge:
[T]here is a methodological bias in favor of taking natural discourse literally, other things being equal. For example, unless there are clear reasons for construing discourse as ambiguous, elliptical, or involving special idioms, we should not so construe it. (1979/2007:116)
Even once it’s acknowledged that many uses of “bias” are not intended to convey a negative assessment, one might still have expected that, when the term is used in the context of a discussion of proper methodology (as when one speaks of “methodological biases”), it would carry at least a negative connotation. But as the quotation from Burge shows, even that much isn’t true.
Thus, in some cases, talk of “bias” involves a negative assessment while in other cases it doesn’t.6 As is sometimes noted, this is true even when what is at issue are uses of the term “bias” in the same academic discipline,7 or within the same scholarly literature.8 Given this, I regard it as a good question and an open one how much the two cases have in common. Is it simply a coincidence that we use the same word “bias” to do quite different jobs, in anything like the way it’s simply a coincidence that we use the same English word “bank” to talk about both financial institutions and river banks? For reasons to be explored later, I think that that suggestion is implausible. On the other hand, if, as I believe, the two uses have something important in common, then a good theory of bias, or a good theory of our practices of attributing bias, should make clear what that is. I offer my own account of this in Chapter 7. That having been said, as my list of examples suggests, in this book I’ll be primarily concerned with bias in the non-neutral, evaluative sense (or in its non-neutral, evaluative uses), and the norm-theoretic account of bias that I offer in Chapter 3 is intended as an account of bias in this sense. As the list also suggests, even with that narrowing of focus, the sheer range and diversity of things to which we attribute bias is striking.
In the rest of this chapter, I explore a number of basic structural features of bias that should, I believe, be recognized independently of a commitment to any specific theory of bias. In §1, I explore and defend the idea that whether a person or thing counts as biased can be a relative matter. In §3, I explore the fact that biases characteristically have a direction, and the related fact that they typically come in complementary pairs (a bias in favor of something typically comes paired with a bias against some salient alternative). §4 and §5 offer preliminary discussions of the important phenomena of higher-order bias (or bias about bias itself) and biased representation. Finally, §6 takes up some of the metaphysical and conceptual questions that arise from the fact that both wholes (e.g. groups of people) and their component parts (e.g. the individual people who make up the group) can be biased or unbiased. Each of these discussions is self-contained and can be read independently of the others.
2. Relativity
Often, a person or thing might count as biased under one description but as unbiased under another. Relatedly, whether they count as biased or unbiased might very well depend upon our interests and purposes in raising the question of whether they’re biased in the first place, or on which question we seek to answer by consulting them. Imagine a presidential approval poll in which people are asked whether they currently approve or disapprove of the job the American president is doing, but in which those conducting the poll deliberately only survey
residents of the state of Louisiana. If the poll is an attempt to gauge the President’s level of support among Americans, then the poll counts as biased indeed, both the sample of voters, and the procedure used to generate that sample, count as paradigms of biased samples and biased sampling procedures, respectively. On the other hand, if the question at issue is the President’s support among residents of Louisiana, then the identical poll, sample, and sampling procedure might all be unbiased, and exemplary in other respects as well. The same point holds for texts. An account of social life in mid-Victorian England that dwelled on the characteristic moral shortcomings of that era while omitting any mention of its more positive features might justly be accused of bias. On the other hand, the same charge would be misplaced when directed at the same text if it purported to be an account of the limited opportunities available to women at the time, although such an account might still count as biased against the Victorians for other reasons (e.g. if it failed to provide a certain kind of historical context, or suggested that that moral failing was in some way unique to Victorian society).
It’s obvious enough that whether someone or something counts as biased might depend on what question is at issue. But whether something counts as biased can be relative in more interesting and surprising ways as well. In particular, even when we hold the question or issue fixed, whether a person or thing counts as biased might still be a relative matter. Consider, for example, the following case:
ALREADY CONVINCED: Before a criminal trial begins, a person announces that she’s already firmly convinced that the defendant committed the crime of which he stands accused
If the person is in the pool of candidates to serve on the jury for the trial, she will be rightly excused from jury duty on the grounds that she doesn’t qualify as an unbiased juror. Similarly, if she’s a judge and would otherwise be in line to preside over the trial, the trial will be assigned to one of her colleagues: given that she’s already made up her mind, she doesn’t qualify as an unbiased judge. However, suppose that the reason why she’s firmly convinced that the accused committed the crime is this: she personally witnessed him commit the crime, and thus knows that he did. Let’s stipulate that the witness saw the event under more or less ideal viewing conditions, and that she was not in any way a biased observer or perceiver: if someone else had committed the crime rather than the actual defendant, then the witness would believe that that person, and not the actual defendant, committed the crime, and so on. Although such a person wouldn’t qualify as an unbiased juror or judge, she would count as an unbiased witness. If the prosecution calls her as a witness, and the defense attorney objects on the grounds that she’s biased against his client, it’s the defense attorney who speaks falsely. (Contrast a case in which she is a biased witness, who mistakenly believes that she saw the accused commit the crime, but where that mistaken belief is a product of racial or ethnic animus.)
If, notwithstanding her prior belief in the defendant’s guilt, the protagonist in ALREADY CONVINCED somehow did make it on to the jury or serve as the judge, the later discovery of this fact would constitute a compelling basis for the appeal of a guilty verdict on the grounds that the defendant didn’t receive a fair trial: the judge or a member of the jury was biased against him. In contrast, the fact that a witness was firmly convinced before the trial
begins that the defendant committed the act of which he’s accused provides no basis at all for an appeal on the grounds of bias. Indeed, in the case of the witness, things seem to run in the opposite direction: if it’s later revealed that, on the eve of the trial, the witness lacked the belief that the defendant committed the act of which he’s accused, and only arrived at that belief over the course of the trial, then that would seem to undermine the credibility of her allegedly eyewitness testimony, and cast severe doubt on the claim that she’s an unbiased witness. With respect to the issue that’s before the court, the same person (with the same mental states, and so on) would count as biased if they were playing the role of judge or juror but as unbiased if they were playing the role of witness. More generally, whether a person counts as biased or unbiased with respect to an issue might be relative to which social role they occupy.
3. Directionality
We often say things of the form “So-and-so is biased.” However, such statements should generally be understood as elliptical. Unless the context already makes their answers clear, the bare, unadorned claim that “So-and-so is biased” naturally invites the questions: “Against whom or what?” or “In favor of whom or what?” Something that’s biased isn’t biased simpliciter or tout court. Rather, it’s biased in favor of x, or against the ys, or…Any bias has a direction or valence.
Given that everything that’s biased is either biased in favor of something or biased against something else, is everything that’s biased both? Or is it possible to be biased in favor of something without being biased against anything (or vice versa)? Consider the simple paradigm of the biased coin. Any coin that’s biased in favor of heads is ipso facto biased against tails, and vice versa. (In that case, the property of being biased in favor of heads and the property of being biased against tails are literally, for once! two sides of the same coin.) Does what’s clearly true of biased coins hold for bias in general?
Consider the following case:
DESPISED CANDIDATE: At a given time, fifteen candidates are vying for their party’s presidential nomination. A news network is biased against one of the candidates because of her staunchly anti-interventionist foreign policy views, but it’s scrupulously neutral among the other fourteen.
Is this a case in which the network has a bias against someone without a corresponding bias in favor of someone else? In a context in which what’s salient is a one-to-one comparison between the anti-interventionist candidate and any one of the other fourteen, the network counts both as biased against the anti-interventionist and also as biased in favor of the candidate with whom she’s being compared; here the negative bias or bias-against is paired with a corresponding positive bias, or bias-in-favor. What about a context in which what’s salient is the entire field? In that context, it will still be correct to describe the network as biased against the anti-interventionist, but at best misleading to describe it as biased in favor of any of the other fourteen, for that seems to overstate its support for any of the others.
However, even in the more inclusive context, the negative bias against the anti-interventionist carries in its wake a corresponding positive bias in favor of another entity, namely, the rest of the field. It might seem then, that whenever it’s correct to describe a person or thing as biased against something, it will also be correct to describe them as biased in favor of something else (and vice versa) even if the “something else” is simply the complement of the relevant set.
However, we should not generalize too quickly from examples like that of the biased coin or DESPISED CANDIDATE. On any given flip, the coin will land either heads or tails but not both; its landing heads precludes its landing tails, and vice versa. Similarly, a contested presidential primary is a competitive context in which a scarce good the party’s presidential nomination is being allocated; any one of the candidates’ receiving the nomination precludes any of the other candidates receiving it. We should also test the hypothesis that positive and negative biases always come in pairs in cases in which this feature is absent. For this purpose, consider the following case:
OPEN-ENDED ADMISSIONS: An admissions process is open-ended, in the sense that there is not a finite number of spots. Applicants for admission are thus not in competition with one another, and any applicant who satisfies the relevant standard for admission is automatically admitted, regardless of how many others already have been admitted under that standard. Approximately 2 percent of the applicants have feature F. The admissions process is biased in their favor: they are either automatically admitted, or else the standard for admission that they must meet is significantly lower than the usual standard that applies to the other 98 percent However, the fact that there are some Fs in the pool who get admitted because the standard that they must satisfy is lower in no way adversely affects the chances that any particular not-F will be admitted Not only does the admission of some Fs by way of the lower standard not reduce the spots open to any of the other applicants, but the fact that the Fs are favored hasn’t in any way altered the probability that any other applicant will be admitted, inasmuch as if there hadn’t been any Fs in the pool, the standards that not-Fs would need to satisfy in order to be admitted would be exactly as they are as things actually stand. (Although it’s true of at least some non-Fs who are rejected that: if they had been F rather than not-F, then they would have been accepted rather than rejected, since they then would have been evaluated according to a lower standard than the one that they failed to meet )
Is OPEN-ENDED ADMISSIONS a case where we see a bias-in-favor without any corresponding bias-against? In presenting the example to students and colleagues, I’ve found that people’s intuitive judgments about it are quite mixed. Some think (and sometimes, adamantly insist) not only that the process is biased in favor of Fs but also that it’s biased against not-Fs, while others deny this, and still others confess to having no clear intuitions about the example. It’s unsurprising that the case inspires mixed reactions. Notice that even the question “Does the fact that the process has a pro-F bias ever explain why a particular person who isn’t F got rejected?” will elicit mixed responses, depending on how one thinks about explanation. On the one hand, it’s stipulated in the example that the fact that the process has a pro-F bias is probabilistically irrelevant to any not-F’s chances of admission: if there had been no pro-F bias, the applicant’s chances of admission would have been no higher. Thus, if one takes probabilistic relevance as necessary for explanation, the pro-F bias never explains why any not-F applicant fails to be admitted. On the other hand, consider some particular not-F who isn’t admitted, but who would have been if they had been F (since in that case they would have met the standard for admission). Notably, any such individual will satisfy the traditional “but for” criterion of discrimination as understood in American
discrimination law: they would have been admitted, “but for” the fact that they are not-F rather than F.9
Here I won’t attempt to resolve things one way or the other but simply note that the example seems to be something of a marginal case. Notice also that as soon as we alter the example so that Fs and not-Fs are competing for a finite, limited number of spots, it seems much more natural and straightforward to describe it as a case in which the process is both biased in favor of Fs and therefore biased against not-Fs.10
Setting aside certain arguable cases such as OPEN-ENDED ADMISSIONS, negative biases and positive biases typically come paired with one another in a complementary package. However, in some cases it’s the positive bias which is fundamental while in other cases it’s the negative bias which is fundamental. A father who nepotistically hires his own child for a job over another applicant whom he would otherwise have hired is both biased in favor of his child and biased against the other applicant. Here it’s the father’s positive bias in favor of his child that’s fundamental. Because of this, it’s the presence of the positive bias that explains the negative bias, and not vice versa. (It’s true to say that the father is biased against the other applicant because he’s biased in favor of his child, but false to say that the father is biased in favor of his child because he’s biased against the other applicant.) On the other hand, if he hires a white applicant over a black applicant because he wrongly believes that blacks make poor workers, then he’s both biased against the black applicant and biased in favor of the white applicant; but in this case, it’s the negative bias which is fundamental and which explains the positive bias.11
Suppose that a person is biased in favor of X and against some salient alternative Y, and that the bias manifests itself in a judgment about the relative merits of X and Y. In paradigmatic cases, the content of the judgment will itself favor X over Y. However, this is not always the case. For example, a bias in favor of X over Y might give rise to a judgment whose content favors neither X nor Y, as in the following example:
BIASED PARENT: A parent is biased when it comes to evaluating their child’s musical performances When the child performs at a recital, the parent’s bias leads them to conclude that the performance was equal to the best student’s, whose performance was in fact objectively superior to all of the others and was recognized as such by the other parents
Here the parent’s bias in favor of their own child manifests itself not in a judgment of superiority but in a judgment of equality or parity. Similarly, a bias against Y might manifest itself in a judgment or decision that doesn’t disfavor Y relative to salient alternatives. (A bias against the most talented performer might lead one to conclude that they performed on a par with the rest of the field, whereas in fact their performance was objectively superior.) In general, there is no safe inference from the fact that “So-and-so’s judgment/decision did not go against Y” to the conclusion that “So-and-so’s judgment/decision was not a manifestation of bias against Y.”
Suppose that the performance of the biased parent’s child was in fact the worst of all, but the parent’s bias causes them to judge that it was in fact on a par with the rest. Notice that the parent’s judgment that my child’s performance was just as good as any of the other children’s
