Kenny Easwaran (Texas A&M University)
There are two major families of proposals for probabilities conditional on events of measure zero. One family (associated with Popper, De Finetti, Dubins, and their followers) fixes a conditional probability for every pair of events. The other (associated with Kolmogorov, and standard in mathematical probability theory) requires a third argument of a partition from which the conditioning event is drawn. I argue that partition sensitivity is not a problem, because every context in which conditional probability is relevant has a natural partition, whether in learning, confirmation theory, decision theory, causal modeling, or otherwise.
