GATSBY COMPUTATIONAL NEUROSCIENCE UNIT

Rational Models and Heuristic Approximations

Rational models aim to understand human cognition in terms of optimal behavior for a given task and environment. However, many rational models are arguably not psychologically plausible (at least, as models of conscious cognition) since they require quite complex computations. One common response is to argue that people are closely approximating the behavior of the rational model using various heuristics, and these approximations are "good enough" (in some vague sense). In this talk, I will focus more closely on the relationship between rational models and proposed heuristic approximations. Two examples of relevant questions are: When does the explanatory power of a rational model (i.e., providing an explanation for why cognition has the form it does) transfer to the heuristic approximation? And: Under what conditions are we justified in claiming that some process theory is actually a heuristic approximation to a particular rational model? In addition to exploring these connections at a relatively high level, I will illustrate them with a number of examples from rational models of causal learning.