Clinton Castro

When agents insert technological systems into their decision-making processes, they can obscure moral responsibility for the results. This can give rise to a distinct moral wrong, which we call “agency laundering.” At root, agency laundering involves obfuscating one’s moral responsibility by enlisting a technology or process to take some action and letting it forestall others from demanding an account for bad outcomes that result. We argue that the concept of agency laundering helps in understanding important moral problems in a number of recent cases involving automated, or algorithmic, decision-systems. We apply our conception of agency laundering to a series of examples, including Facebook’s automated advertising suggestions, Uber’s driver interfaces, algorithmic evaluation of K-12 teachers, and risk assessment in criminal sentencing. We distinguish agency laundering from several other critiques of information technology, including the so-called “responsibility gap,” “bias laundering,” and masking.

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.

We offer an ethical assessment of the market for data used to generate what are sometimes called “consumer scores” (i.e., numerical expressions that are used to describe or predict people’s dispositions and behavior), and we argue that the assessment has ethical implications on how the market for consumer scoring data should be regulated. To conduct the assessment, we employ two heuristics for evaluating markets. One is the “harm” criterion, which relates to whether the market produces serious harms, either for participants in the market, for third parties, or for society as a whole. The other is the “agency” criterion, which relates to whether participants understand the nature and significance of the exchanges they are making, if they can be guaranteed fair representation, or if there is differential need for the market’s good. We argue that consumer scoring data should be subject to the same sort of regulation as the older FICO credit scores. Although the movement in the 1990s that was aimed at regulating the FICO scores was not aimed at restraining a market per se, we argue that the reforms were underwritten by concerns about the same sorts of problems as those outlined by our heuristics. Therefore, consumer data should be subject to the same sort of regulation.

Impermissivists hold that an agent with a given body of evidence has at most one rationally permitted attitude that she should adopt towards any particular proposition. Permissivists deny this, often motivating permissivism by describing scenarios that pump our intuitions that the agent could reasonably take one of several attitudes toward some proposition. We criticize the following impermissivist response: while it seems like any of that range of attitudes is permissible, what is actually required is the single broad attitude that encompasses all of these single attitudes. While this might seem like an easy way to win over permissivists, we argue that this impermissivist response leads to an indefensible epistemology; permissive intuitions are not so easily co-opted.

One of the main goals of Bayesian epistemology is to justify the rational norms credence functions ought to obey. Accuracy arguments attempt to justify these norms from the assumption that the source of value for credences relevant to their epistemic status is their accuracy. This assumption and some standard decision-theoretic principles are used to argue for norms like Probabilism, the thesis that an agent’s credence function is rational only if it obeys the probability axioms. We introduce an example that shows that the accuracy arguments for Probabilism given by Joyce and Pettigrew fail, and that Probabilism in fact turns out to be false given Pettigrew’s way of conceiving of the goal of having accurate credences. Finally, we use our discussion of Pettigrew’s framework to draw an important general lesson about normative theorizing that relies on the positing of ideal agents.