This is co-authored work with my friend Boris Babic (Caltech and INSEAD).
We address the idea that, in cases in which you have information about the base rate of a socially disvalued trait within a certain social group and you then meet somebody who is a member of that group, you face an inescapable conflict between the requirements of epistemic rationality and those of morality: rationality requires that your credence that this individual has the socially disvalued trait matches the base rate in the group, while morality prohibits this. We bring the good news that this idea is mistaken. Moreover, we need not adopt a conception of epistemic rationality that is unfriendly toward data and probabilistic reasoning in order to secure this result. On the contrary, one of the most technical and statistics-friendly ways of thinking about epistemic rationality entails that no particular attitude is required in cases like this. The good news follows from a version of accuracy-based epistemology according to which epistemic agents are rationally required to form and revise their credences so as to maximize expected epistemic utility, where epistemic utility is measured by a loss function known as a “scoring rule” which embodies the agent’s attitude toward epistemic risk. The key point is that epistemic rationality permits a wide range of attitudes toward epistemic risk, allowing for a wide range of ways of thinking about the relative disvalue of high credence in a falsehood and low credence in a truth. This is all we need in order to show that the allegedly-inescapable conflict is readily escapable; there are plenty of options on which morality smiles and at which rationality shrugs, and thus no inevitable conflict between the requirements of these two normative domains. We close by discussing the extent to which our argument generalizes to other kinds of reasoning besides predictive inference. Results are mixed.
This paper is promised to Oxford Studies in Normative Ethics, but it won’t be out for a while. Here’s the latest draft.