We care not only about how people treat us, but also what they believe of us. If I believe that you’re a bad tipper given your race, I’ve wronged you. But, what if you are a bad tipper? It is commonly argued that the way racist beliefs wrong is that the racist believer either misrepresents reality, organizes facts in a misleading way that distorts the truth, or engages in fallacious reasoning. In this paper, I present a case that challenges this (...) orthodoxy: the case of the supposedly rational racist. We live in a world that has been, and continues to be, structured by racist attitudes and institutions. As a result, the evidence might be stacked in favour of racist beliefs. But, if there are racist beliefs that reflect reality and are rationally justified, what could be wrong with them? Moreover, how do I wrong you by believing what I epistemically ought believe given the evidence? To address this challenge, we must recognize that there are not only epistemic norms governing belief, but moral ones as well. This view, however, is at odds with the assumption that moral obligation requires a kind of voluntary control that we lack with regard to our beliefs. This background assumption motivates many philosophers to try to explain away the appearance that beliefs can wrong by locating the wrong elsewhere, e.g., in an agent’s actions. Further, even accounts that accept the thesis that racist beliefs can wrong restrict the class of beliefs that wrong to beliefs that are either false or the result of hot irrationality, e.g., the racist belief is a result of ill-will. In this paper I argue that although the these accounts will capture many of the wrongs associated with racist beliefs, they will be only partial explanations because they cannot explain the wrong committed by the supposedly rational racist. The challenge posed by the supposedly rational racist concerns our epistemic practices in a non-ideal world. The world is an unjust place, and there may be many morally objectionable beliefs it justifies. To address this challenge, we must seriously consider the thesis people wrong others in virtue of what they believe about them, and not just in virtue of what they do. (shrink)

Historical patterns of discrimination seem to present us with conflicts between what morality requires and what we epistemically ought to believe. I will argue that these cases lend support to the following nagging suspicion: that the epistemic standards governing belief are not independent of moral considerations. We can resolve these seeming conflicts by adopting a framework wherein standards of evidence for our beliefs to count as justified can shift according to the moral stakes. On this account, believing a paradigmatically racist (...) belief reflects a failure to not only attend to the epistemic risk of being wrong, but also a failure to attend to the distinctively moral risk of wronging others given what we believe. (shrink)

Data-driven, decision-making technologies used in the justice system to inform decisions about bail, parole, and prison sentencing are biased against historically marginalized groups (Angwin, Larson, Mattu, & Kirchner 2016). But these technologies’ judgments—which reproduce patterns of wrongful discrimination embedded in the historical datasets that they are trained on—are well-evidenced. This presents a puzzle: how can we account for the wrong these judgments engender without also indicting morally permissible statistical inferences about persons? I motivate this puzzle and attempt an answer.

A number of findings in the field of machine learning have given rise to questions about what it means for automated scoring- or decisionmaking systems to be fair. One center of gravity in this discussion is whether such systems ought to satisfy classification parity (which requires parity in accuracy across groups, defined by protected attributes) or calibration (which requires similar predictions to have similar meanings across groups, defined by protected attributes). Central to this discussion are impossibility results, owed to Kleinberg (...) et al. (2016), Chouldechova (2017), and Corbett-Davies et al. (2017), which show that classification parity and calibration are often incompatible. This paper aims to argue that classification parity, calibration, and a newer, interesting measure called counterfactual fairness are unsatisfactory measures of fairness, offer a general diagnosis of the failure of these measures, and sketch an alternative approach to understanding fairness in machine learning. (shrink)

According to principles of probability coordination, such as Miller's Principle or Lewis's Principal Principle, you ought to set your subjective probability for an event equal to what you take to be the objective probability of the event. For example, you should expect events with a very high probability to occur and those with a very low probability not to occur. This paper examines the grounds of such principles. It is argued that any attempt to justify a principle of probability coordination (...) encounters the same difficulties as attempts to justify induction. As a result, no justification can be found. (shrink)

The requirement that moral theories be usable for making decisions runs afoul of the fact that decision makers often lack sufficient information about their options to derive any accurate prescriptions from the standard theories. Many theorists attempt to solve this problem by adopting subjective moral theories—ones that ground obligations on the agent’s beliefs about the features of her options, rather than on the options’ actual features. I argue that subjective deontological theories suffer a fatal flaw, since they cannot appropriately require (...) agents to gather information before acting. (shrink)

This paper develops a precise understanding of the thesis of moral encroachment, which states that the epistemic status of an opinion can depend on its moral features. In addition, I raise objections to existing accounts of moral encroachment. For instance, many accounts fail to give sufficient attention to moral encroachment on credences. Also, many accounts focus on moral features that fail to support standard analogies between pragmatic and moral encroachment. Throughout the paper, I discuss racial profiling as a case study, (...) arguing that moral encroachment can help us identify one respect in which racial profiling is epistemically problematic. (shrink)

As machine learning informs increasingly consequential decisions, different metrics have been proposed for measuring algorithmic bias or unfairness. Two popular “fairness measures” are calibration and equality of false positive rate. Each measure seems intuitively important, but notably, it is usually impossible to satisfy both measures. For this reason, a large literature in machine learning speaks of a “fairness tradeoff” between these two measures. This framing assumes that both measures are, in fact, capturing something important. To date, philosophers have seldom examined (...) this crucial assumption, and examined to what extent each measure actually tracks a normatively important property. This makes this inevitable statistical conflict – between calibration and false positive rate equality – an important topic for ethics. In this paper, I give an ethical framework for thinking about these measures and argue that, contrary to initial appearances, false positive rate equality is in fact morally irrelevant and does not measure fairness. (shrink)

We offer a new argument for the claim that there can be non-degenerate objective chance (“true randomness”) in a deterministic world. Using a formal model of the relationship between different levels of description of a system, we show how objective chance at a higher level can coexist with its absence at a lower level. Unlike previous arguments for the level-specificity of chance, our argument shows, in a precise sense, that higher-level chance does not collapse into epistemic probability, despite higher-level properties (...) supervening on lower-level ones. We show that the distinction between objective chance and epistemic probability can be drawn, and operationalized, at every level of description. There is, therefore, not a single distinction between objective and epistemic probability, but a family of such distinctions. (shrink)

Predictive algorithms are playing an increasingly prominent role in society, being used to predict recidivism, loan repayment, job performance, and so on. With this increasing influence has come an increasing concern with the ways in which they might be unfair or biased against individuals in virtue of their race, gender, or, more generally, their group membership. Many purported criteria of algorithmic fairness concern statistical relationships between the algorithm’s predictions and the actual outcomes, for instance requiring that the rate of false (...) positives be equal across the relevant groups. We might seek to ensure that algorithms satisfy all of these purported fairness criteria. But a series of impossibility results shows that this is impossible, unless base rates are equal across the relevant groups. What are we to make of these pessimistic results? I argue that none of the purported criteria, except for a calibration criterion, are necessary conditions for fairness, on the grounds that they can all be simultaneously violated by a manifestly fair and uniquely optimal predictive algorithm, even when base rates are equal. I conclude with some general reflections on algorithmic fairness. (shrink)

I argue that there are non-trivial objective chances (that is, objective chances other than 0 and 1) even in deterministic worlds. The argument is straightforward. I observe that there are probabilistic special scientific laws even in deterministic worlds. These laws project non-trivial probabilities for the events that they concern. And these probabilities play the chance role and so should be regarded as chances as opposed, for example, to epistemic probabilities or credences. The supposition of non-trivial deterministic chances might seem to (...) land us in contradiction. The fundamental laws of deterministic worlds project trivial probabilities for the very same events that are assigned non-trivial probabilities by the special scientific laws. I argue that any appearance of tension is dissolved by recognition of the level-relativity of chances. There is therefore no obstacle to accepting non-trivial chance-role-playing deterministic probabilities as genuine chances. (shrink)

According to a common conception of legal proof, satisfying a legal burden requires establishing a claim to a numerical threshold. Beyond reasonable doubt, for example, is often glossed as 90% or 95% likelihood given the evidence. Preponderance of evidence is interpreted as meaning at least 50% likelihood given the evidence. In light of problems with the common conception, I propose a new ‘relevant alternatives’ framework for legal standards of proof. Relevant alternative accounts of knowledge state that a person knows a (...) proposition when their evidence rules out all relevant error possibilities. I adapt this framework to model three legal standards of proof—the preponderance of evidence, clear and convincing evidence, and beyond reasonable doubt standards. I describe virtues of this framework. I argue that, by eschewing numerical thresholds, the relevant alternatives framework avoids problems inherent to rival models. I conclude by articulating aspects of legal normativity and practice illuminated by the relevant alternatives framework. (shrink)