Abstract for Russell’s talk

According to the orthodox Bayesian picture, a rational agent pushes credences around some set of epistemically possible worlds. But certain kinds of epistemic progress don’t fit well with this picture. Instead of shifting our spread of credences over a fixed set of worlds, it seems better to say that we are changing which worlds we have credences in at all, by collapsing distinctions or adding new ones. I’ll discuss a technical perspective on probability theory (the “monadic” perspective which is influential in computer science) that gives us natural ways of modelling this sort of conceptual change.