essay on learning and motivation

Classical Bayesian epistemology is often criticized for being too subjective. Bayesian theory requires only that inquirers have coherent prior probabilities (de Finetti, 1974) and that these probabilities be updated by conditionalizing on evidence. Because it is rationally permissible, on this view, to adopt any probability measure whatsoever as one's prior, the theory allows widespread disagreement between different rational agents' probability judgments.1 In response to this, the criticism goes, Bayesian theory cannot explain why some probability judgments are rationally superior to others. For example, it cannot explain science's claims to rationality and objectivity.2 In response to charges of excessive subjectivity, Leonard Savage and subsequent theorists have shown that, under mild assumptions, the probabilities of two Bayesian agents who learn the same sequence of propositions are almost sure to eventually reach a consensus (Savage, 1972, Sections 3.6, 3.7), and they are almost sure to converge to the truth