Better guesses

It has recently become popular to analyze scenarios in which we guess, in terms of a trade‐off between the accuracy of our guess (namely, its credence) and its specificity (namely, how many answers it rules out). Dorst and Mandelkern describe an account of guessing, based on epistemic utility theory (EUT), in which permissible guesses vary depending on how one weighs accuracy against specificity. We provide a minimal formal account of guessing that: (i) does not employ EUT, but rests on how such trade‐offs are treated in the sciences; (ii) is relatively parsimonious; and (iii) is consistent with a variety of more specific models that describe what an agent is doing when they (rationally) guess. Our account also recovers patterns of guessing and predictions about typical outcomes of guessing, as identified by Dorst and Mandelkern. Furthermore, we focus on how permissible guesses can be improved upon, via changes in an agent's credence distribution. Such better permissible guesses can be generated in solving Fermi problems—guessing problems of a type that has received almost no attention in the philosophical literature—which we also analyze. Our account strengthens the case for understanding guessing (now, very broadly considered) in terms of accuracy‐specificity trade‐offs.