Research Group
in Analytic Philosophy

Abstract

Updating rules provide norms for a rational update from an initial probability function* to a new one** in light of new evidence. Updating rules may directly be turned into a theory of confirmation - in this context they describe the rational rules for an update from the prior probability of some hypothesis* to its posterior probability** in light of some evidence.

Three types of rules will be assessed: (i) standard Bayesian updating, (ii) Jeffrey-style updating, and (iii) minimizing Kulback-Leibler divergence. Both (i) and (ii) are based on conditional probability unlike (iii). Some advantages and disadvantages of these approaches will be addressed.  

One of the bigger threats for all of these probabilistic updating rules was laid out in Van Fraassen's Judy Benjamin Problem (1981). I will demonstrate how this threat may be overcome with some help of Bayesian networks. I will also assess a recent proposal that updating is essentially non-Bayesian but rather based on explanatory considerations (abduction).