Contingency Learning and Causal Reasoning
Psychology Press eBooks · 2000
F rom at least Ferster and Skinner&s;s (1957) early work docu mented in Schedules of Reinforcement there has been a great deal of interest in contingencies of reinforcement. The relation be tween the conditional stimulus (CS) and the unconditional stimulus (US) in Pavlovian conditioning and between the response and the out come or reinforcer has long been known to exert subtle control over ani mal behavior. Moreover, students of causal reasoning consider the correlation or contingency between cause and effect to be one of the cru cial empirical determinants of the causal reasoning process (e.g., Einhorn & Hogarth, 1986). Recently there have been attempts to analyze how humans judge contingencies in causal reasoning scenarios and in categorization tasks in terms of contingency sensitivity. Much of this re search has been motivated by associative models derived directly from experim ents on anim al learning (e.g., Baker, Mercier, ValléeTourangeau, Frank, & Pan, 1993; Shanks, 1991; Wasserman, Elek, Chatlosh, & Baker, 1993) It is puzzling, that, following a burst of interest generated by Rescorla&s;s (1968) seminal work on contingency learning in Pavlovian conditioning, there has been relatively little analysis of the mechanisms by which animals represent contingencies. This has partly been due to what we believe to be a rather naive empirical and theoretical analysis of the issue. This simplistic analysis has led many to deny that the corre lation or contingency between the CS and the US is even an important concept in animal learning, or even if it is, that there is strong prima fa cie evidence against the argument that animals learn these contingen cies (see Papini & Bitterman, 1990). Much of this chapter deals with an empirical and theoretical response to this position. In fairness it must be recognized that the motivation behind this "simpler" analysis of the mechanism underpinning learning in animals is at least partly driven by the traditional desire for parsimony. The argument that animals might actually learn the correlation or contingency between events and then retrospectively base their decisions on this information seems to imply a homunculus located inside the animal that has within itself all the cognitive capacities we wish to explain with our science. We argue that a modern theoretical analysis maintains the parsimony of the tradi tional associative position yet still has a place in it for the representation of, or sensitivity to, event contingencies or correlations.