Abstract

According to the Lockean thesis, rational belief corresponds to rational credence above a certain threshold. A version of the Lockean thesis can be derived from the assumption that rational agents maximize the expected utility of their beliefs, given their credences (Hempel 1962, Easwaran 2016, Dorst 2019).

This talk presents parts of work aimed at generalizing the utility-theoretic Lockean thesis to cases where rational credences are imprecise, in the sense of being given by a set of (probability) functions (overview: Bradley, 2019). In these cases, beliefs are not guaranteed the type of precise expectation values required for maximization, but must be deemed rational on the grounds of conforming to some more general decision rule.

I will formulate some basic desiderata for such decision rules, and use this in an assessment of two well-known rules from imprecise decision theory: E-admissbility (Levi 1980) and Gamma-maximin (Gilboa & Schmeidler 1989). I will also consider the prospects of using an aggregative decision rule (classic majority voting), after showing that the domains of the relevant decision problems are Arrow consistent.