Implicit Bias and Philosophy: Metaphysics and Epistemology, Volume 1

Abstract

Implicit Bias and Philosophy: Metaphysics and Epistemology is a broad, impressive, and interesting collection of essays on the nature of implicit bias. It presents various views on the kind of cognitive states that underlie implicit bias; how to interpret the results of methodological tools that purport to measure implicit bias; the relation to stereotypes, stereotype threat, and epistemic injustice; and application to philosophical skepticism, underrepresentation of women in science, and gendered stereotypes about philosophy. In this review, I shall not offer complete summaries of each chapter. I refer readers interested in that kind of information to the editors' introductory chapter. Instead, I will focus on how to think of and use this volume in light of the rapidly developing empirical and theoretical research on implicit bias. My task here is to explain the ways in which the literature has changed, and how this is (or in some cases is not) reflected in the contributions to this book. Overall, these chapters provide accessible, insightful perspectives on some of the enduring debates in the field of implicit bias, and the book promises to be a useful guide to these debates for years to come.As a point of history, the study of implicit bias grew out of the more general study of implicit cognition, which tends to be based on two-systems theories of cognition. Two-systems theories posit two distinct cognitive systems: a lower-level system that is associative, automatic, rapid, and unconscious, and a higher-level system that is rule-based, controlled, relatively slow, and conscious. On a strict two-systems view, explicit biases are conscious, self-reportable attitudes that are subject to reflective evaluation, whereas implicit biases are unconscious, automatic attitudes that are opaque to introspection and not subject to self-report or reflective evaluation. Contemporary empirical and philosophical study of implicit bias is historically anchored in this individualistic, unconscious, automatic conception of implicit bias.The literature on implicit bias has evolved substantially over the past few years. For instance, the two-systems account is no longer presumed to be the best way to understand implicit bias. In this volume, several theorists explicitly reject the two-systems analysis of implicit bias. For example, Bryce Huebner (1.2) argues that there are three learning systems underlying human cognition. On his view, implicit bias is a simple label for a complex, diverse set of dispositional traits that are produced and sustained by interaction between these three different learning systems and the environment. Louise Antony (2.1) similarly rejects two-systems views. In this philosophically rich, wide-ranging chapter, Antony argues that implicit biases result from and affect both careful, deliberative judgments and quick, casual inferences of various sorts, so we cannot circumscribe the problem of implicit bias in any neat way. This suggests, as Jennifer Saul maintains, that implicit bias likely affects the evaluation of philosophy students, philosophers, and philosophical theories, which generates an alarming skeptical problem for philosophy. Antony argues that to meet Saul's skeptical challenge, we must reject a misguided empiricist view of objectivity and understand that some cognitive biases are essential to successful cognition. Only then can we distinguish useful from harmful biases and effectively work to combat the pernicious biases that pose a threat to philosophy in particular and society more generally.Many theorists still employ two-systems theories to understand implicit bias, of course, though most no longer adhere to a strict division between types of cognitive systems. In this volume for example, Frankish and Mallon rely on looser conceptions of two-systems theory to understand implicit bias and stereotype threat. Keith Frankish (1.1) employs a nonstandard form of dual-process theory that eschews many of the hard and fast distinctions between system 1 and system 2 to account for implicit bias. He calls his theory a dual level theory and uses this framework to articulate conditions for overriding implicit bias. On Frankish's view, we can use various explicit, conscious methods of self-control to mitigate the manifestation of implicit bias. Ron Mallon (1.5) adopts a distinction between subpersonalist and personalist processes that mostly aligns with but does not exactly track two-systems views. He uses this framework to argue that subpersonalist accounts of stereotype threat are incomplete and fail to explain some of the basic facts about stereotype threat. On his view, individuals who experience stereotype threat are acting rationally in response to socially threatening situations. This rational behavior is a characteristic of personalist processes. To the extent that stereotype threat is a manifestation of implicit bias, this suggests that conceiving of implicit bias purely in terms of system-1/lower-level/subpersonal mechanisms is inadequate.This move away from strict two-systems accounts of implicit bias reflects both an evolution of two-systems theories in general and evolving views on implicit bias in particular. With respect to implicit bias, though, as I explain below, recent empirical evidence is sometimes difficult to interpret, one clear lesson from these data is that implicit biases are not literally automatic and not completely opaque to introspection. Situational context, task, and a subject's goals moderate implicit bias activation, and subjects are able to estimate their implicit biases with reasonable accuracy if asked the right kinds of questions. This complicates the simple story about implicit bias.Many chapters in this volume discuss the diversity and malleability of implicit biases. For example, Huebner (1.2), Holroyd and Sweetman (1.3), and Madva (2.2) each devote significant space to explaining these features of implicit bias, though they offer different perspectives on these features. Huebner takes these data as evidence that implicit biases result from complex interaction between various cognitive systems and our social niches, and he suggests interventions that appropriately reflect the complexity of the mechanisms of implicit bias. Jules Holroyd and Joseph Sweetman make a powerful case for the heterogeneity of implicit biases. They argue that implicit biases exhibit a variety of functional differences with respect to both content and underlying cognitive structure, which defies any simple, unified categorization. This has significant implications for empirical work, philosophical theorizing, and practical interventions. I encourage philosophers who are relatively new to the implicit bias debate to carefully read this important chapter. The authors do not advocate a particular view of the cognitive mechanisms underlying implicit biases or argue for a particular conception of implicit biases, which makes it an excellent primer on the topic. Finally, Alex Madva distinguishes simply knowing a stereotype from activating a stereotype and argues that implicit biases result from stereotype activation, which is highly context sensitive and malleable. Madva concludes with several concrete strategies for manipulating our cognitive practices to block the activation of problematic stereotypes. In various ways, these chapters aim to account for the persistence, diversity, and malleability of implicit biases.Unfortunately, some contributions to this volume still unreflectively treat implicit bias as automatic, unconscious, and relatively uniform. For instance, Catherine Hundleby (2.4) argues that we should conceive of implicit bias as a status-quo fallacy, that is, a tendency to regard the status quo as justified. In doing so, her argument treats the status quo as stable across contexts and implicit bias as an unconscious, automatic, and uniform reflection of the status quo (238, 242–43). The empirical evidence conflicts with this picture of implicit bias. Other chapters define implicit bias as automatic and unconscious but go on to describe contextual moderation of implicit bias activation, for example, Lee (2.5) and Dei Bella, Miles, and Saul (2.6). In each of these cases, the overall arguments do not depend on implicit biases being strictly automatic and unconscious. This is likely an oversight or reliance on outdated empirical data. One could fairly easily revise these chapters to be perspicacious about the nature of implicit bias.Another change in this literature concerns how to think of the various methods for testing implicit bias. The most robust and frequently employed tests of implicit bias are the Implicit Association Test (IAT), the Affect Misattribution Procedure, and various semantic priming tasks. (See the editors' introductory chapter for a description of how these tests work.) These days, many theorists shy away from treating these tests as diagnostic of individuals' biases for three reasons. (1) An individual's score on one type of implicit bias test only weakly correlates with his or her score on another type of implicit bias test. (2) An individual's score on one type of implicit bias test only weakly correlates with his or her score on that same test at a later date. (3) An individual's scores on each of these implicit bias tests only weakly predicts his or her behavior. These are the empirical data that are difficult to interpret. Theorists in this book and the broader literature more generally offer various perspectives on 1–3, but one point of consensus is that these data indicate that results of these tests are not diagnostic of an individual's implicit biases. That is, my score on a race IAT score does not tell me how racist I am. (Despite the consensus about this claim, it has not penetrated the popular narratives about measures of implicit bias, so it is worth repeating here.)Beyond this point of consensus, there is little agreement about how to interpret the weak correlational data and poor predictive ability of measures of implicit biases. Almost every chapter in this collection mentions the messy empirical data, and some chapters devote substantial space to explaining these data. For example, Holroyd and Sweetman (1.3) use these and other data to make a strong case that there are various cognitive structures underlying what we label as implicit bias. They do not venture to articulate the specific cognitive mechanisms underlying implicit biases or speculate about the nature of implicit biases except to say that they are a motley collection of cognitive processes and more empirical and theoretical work remains to be done.Huebner (1.2) also takes on the task of explaining the messy empirical data on implicit bias, arguing that implicit bias tests are measuring the influence of multiple causal factors. On Huebner's view, implicit biases are stable dispositional traits produced and sustained by complex and dynamic interaction between learning systems and the environment. This view is compatible with the heterogeneity posited by Holroyd and Sweetman but differs in arguing for a specific view of the cognitive mechanisms that realize implicit bias.Machery (1.4) takes the messy empirical data on implicit bias to yield a more skeptical conclusion. He argues that the weak, variable correlational data and the weak predictive ability of measures of implicit bias show that conceiving of implicit attitudes as mental states is inappropriate. The phenomena that implicit bias tests are tracking are too mercurial to be mental states. (This view is in conflict with Frankish's view [1.1], which holds that implicit biases are beliefs.) On Machery's view, implicit biases are best understood as dispositional traits, which are broad-based psychological dispositions to act, emote, and behave in certain ways in certain contexts. Dispositional traits are neither implicit nor explicit, so on this view it makes no sense to talk of implicit bias. (In this way, Machery's view conflicts with the conception of implicit bias presented in many of the other chapters of this book.) The view of implicit biases as dispositional traits is compatible with Huebner's account, but it differs in a few crucial ways. In contrast to Huebner's chapter, Machery does not offer a mechanistic account of implicit biases, situate implicit biases in our social niches, or suggest interventions based on his account of implicit biases. Moreover, Machery's argument extends far beyond implicit biases to the whole field of research on implicit attitudes more generally. His claim is that implicit attitudes in general are not mental states.The contributions to this book offer different explanations of the weak correlations and poor predictive ability of measures of implicit bias. In addition to the perspectives on the messy empirical data articulated in this book, I recommend that those using this book as a research or teaching tool supplement it with articles expressing other perspectives on the data. The research on this specific topic moves very quickly, and there are now several very interesting perspectives not represented in this volume. For example, Keith Payne, Heidi A. Vuletich, and Kristjen B, Lundberg (2017) argue that the best explanation of the messy empirical data focuses more on where an individual lives than on what an individual believes. In the so-called Bias of Crowds model, individuals' performance on implicit bias tests is more of a reflection of their geographical location than what they truly believe. This model purports to explain the noisy data for individuals, the fact that children and adults from the same geographical regions display similar implicit bias scores, the stability of aggregate data on implicit bias, and the strong correlation between aggregate implicit bias scores and disparate social outcomes.Another perspective on the messy data that is not represented in this book argues that these messy empirical data are to be expected. Such arguments flip the script on the critical discussions and argue that we have prior theoretical reasons to expect these empirical measures of implicit social bias to give us these kinds of results. Views like this do not take the messy data to challenge the notion of implicit bias (or implicit attitudes more generally), nor do they imply that the focus of the implicit bias literature is fundamentally mistaken. Rather, such perspectives argue that we should expect to see these particular patterns in the data. See Brownstein, Madva, and Gawronski (in progress) and Del Pinal and Spaulding (2018) for two different expressions of this kind of view. Supplementing the book with articles offering these different perspectives will give a broader, more updated view on this debate about interpreting the messy empirical data.The final theme I shall discuss in my review is the applications and normative recommendations with respect to implicit bias. Almost all of the chapters in this volume include recommendations for decreasing implicit bias or mitigating its effects. Some of these recommendations focus on individuals (Frankish, Holroyd and Sweetman, Madva, Hundleby), whereas others focus more on the social environment (Huebner, Goguen, Lee). The recommendations reflect the views of implicit bias articulated in the chapters and thus vary widely. Taken as a whole, however, they show various concerns about this literature to be wrongheaded. Some people worry that the focus on implicit bias obscures work on many significant causes of discrimination and prejudice, like explicit bias, which is clearly not a feature of bygone eras, and institutional or systemic biases. The chapters in this book do an admirable job of addressing both individual and systemic causes and effect of implicit bias. The chapters by Carole Lee (2.5) and Huebner (1.2) are especially adept at addressing the broader social issues.Some people also worry that discussions of implicit bias have outpaced the empirical evidence and therefore recommending specific interventions on implicit bias is premature. Obviously, I cannot speak for all discussions of implicit bias. However, I hope to have shown that although the implicit bias literature moves very quickly, the contributions to this book are, for the most part, based on up-to-date empirical data and theoretical reasoning. Thus, I think there is very little reason to worry that the ideas in this book are premature and imprudent. To the contrary, I suspect that this book will be relevant for a very long time.