The Problem of Old Evidence

Abstract

There is an underground stream in modern epistemology, starting from Pascal and Huygens in the seventeenth century. In ours, it is represented largely, but not wholly, by the writers who call themselves Bayesian. To follow Pascal and Huygens means only to represent opinion in terms of probabilities (possibly vague) and to insist that the subject of opinion cannot be understood in isolation from those of value or practice. The Bayesians add to this a specific attempt to reconstruct traditional epistemo-logical concepts in probabilistic terms. This attempt (for an overview, see Horwich [1982]) begins with the definition of confirmation as enhancement of subjective probability. I wish there were a catchy name for the three centuries old underground alternative as a whole. (Perhaps probabilism, a term that has some currency, could do.) For the specifically Bayesian addition, Clark Glymour raised a specific difficulty: the problem of old evidence.

The problem of old evidence is so simple to state, and elicits such complex responses, that is seemed a natural subject for a paper honouring Ed Gettier. I met Gettier in Pittsburgh while I was a graduate student there and he a visiting scholar; I immediately envied him his counterexamples, his smile, and the fun he got out of doing philosophy brilliantly.