Philosophy Research Seminar - Updating Imprecise Probabilities - Jason Konek (Bristol)
Cotham House G2
Abstract: On the Bayesian view, belief is not an on-off matter. Bayesians model agents not as simply categorically believing or disbelieving propositions, but rather as having degrees of belief or credences in those propositions. Rather than flat out believing that your Kimchi Jjigae will turn out splendidly, you might, for example, be 70% confident that it will turn out splendidly. Or you might have less precise opinions. You might be more confident than not that it will turn out splendidly, but not have any precise degree of belief on the matter. In that case, we say that your credences are imprecise. The standard method for updating imprecise credences in the face of new evidence is called pointwise conditionalization. Unfortunately, pointwise conditionalization lacks a compelling justification. While there are various extant justifications for updating precise credences by traditional Bayesian conditionalization, these justifications do not extend in any obvious way to pointwise conditionalization. The aim of this talk is to sketch an epistemic utility argument for pointwise conditionalization. Roughly put, you ought to update by pointwise conditionalization because if you fail to do so, then on any reasonable measure of the epistemic value for imprecise credences, you will be guaranteed to accrue less epistemic value over the course of your epistemic life than you could have otherwise.