Can irrationality be intelligently discussed?

that can be pressed into service for many tasks, sometimes appropriately, sometimes not. So Cohen attacks heuristics. But is this attack necessary, even from Cohen's own point of view? Why not simply regard the heuristics as tools or subroutines available to more specific mechanisms? I wish to comment also on Cohen's discussion of Bayes' theorem. As I see it, the fundamental difficulty with the Bayesian analyses Cohen criticizes is that they do not allow for an assessment of the reliability and relevance of "base rates. ' We are not certain that the predominance of blue cabs in the city as a whole is fully relevant to the particular situation. In a recent paper (Shafer 1982) I discuss how base rates can be discounted using the theory of belief functions; it turns out that one obtains the Bayesian answer if the discount rate is sufficiently low, but that even a moderate discount rate can sharply reduce the influence of the base rate in the face of conflicting evidence. Cohen seems to feel that base rates should always be totally discounted, and this is very difficult to sustain. In Cohens medical story, for example, the Bayesian analysis seems compelling if there is no reason to discount the relative rarity of disease B.