Models and deductive rationality

Rationality means different things to different people. During its long history as a theoretical concept, it has referred to logical thought, to a prudent choice among alternative courses of action, to a wise allocation of resources, to a capacity to arrive at truth, and much else besides. We intend to address the question: are human beings rational? But, before we can do so, we need to choose a yardstick against which to measure performance. One strategy, which has been adopted by Anderson (1990 ), is to assume that rational behaviour is whatever is optimized to the structure of the environment. This strategy calls for the almost impossible task of characterizing the structure of the environment in a way that is independent of human conceptions, so that one can then specify what would count as optimal behaviour. Another strategy, which has been pursued by students of decision making, is to propose a set of a priori principles. The ‘rational man’ of classical economics and modern decision theory, however, is a singularly implausible psychological individual. The burden of recent investigations is that human beings make decisions not by choosing among alternatives according to their expected utilities — the sine qua non of such theories — but by constructing reasons for one choice as opposed to others (see Shafir, 1991 ; Tversky and Shafir, 1991 ). Rationality in still other domains, such as inductive and probabilistic inference, depends on technical knowledge. Anyone who had known the calculus of probabilities in Roman times, as Hacking (1975 ) has remarked, would soon have won all of Gaul. To conclude that Romans, and others unfamiliar with the calculus, are therefore irrational seems to miss the point.