A characterization of probability-based dichotomous belief revision (Q2118977)

The paper concerns the problem of how best to define belief sets and their revisions in probabilistic terms. Notoriously, identifying the belief set associated with a probability function with those propositions that have probability 1 does not work for various reasons, one of which is that classical probability theory lacks the means to alter the probability of a proposition once it has reached 1. Setting a threshold lower than 1 creates problems such as those illustrated by the lottery and preface paradoxes. The author's approach is to work with non-standard probability functions taking as values hyperreals from 0 to 1, identify the belief set associated with such a function with the propositions whose probability is infinitesimally close to 1, and envisage revision of the belief set in terms of a suitable kind of conditionalization on the underlying function. The ratio definition of conditionalization will not do in this context, so a form of Jeffrey conditionalization is used. The properties of the kind of belief set revision thus defined are worked out in detail and compared with the familiar AGM ones.