Abstract

One school of theory choice is Bayesianism, which advocates assigning subjective prior probabilities over relevant statements and changing these beliefs by conditionalizing on new evidence. Bayesianism handles many circumstances adequately but appears to suffer from two problems. Firstly, nothing in Bayesian formalism prefers simple theories to more complicated ones. Secondly, Bayesianism seems to have trouble evaluating theories based on evidence acquired before the theories were formulated. ;This dissertation argues that neither of these problems is very serious. Although Bayesianism doesn't formally distinguish between simple and complicated theories, it isn't clear that a theory of confirmation or theory choice should do so. In particular, Kolmogorov complexity theory and Akaikean statistics do not show that simpler theories predict better or are more likely to be true, and unmodified Bayesianism seems to obtain the right results when comparing simple and complicated theories. ;Finally, Bayesianism has resources which have been overlooked in the debate over the problem of old evidence. Instead of suggesting that Bayesianism is fundamentally flawed, the problem of old evidence points toward areas in which Bayesianism could be extended.