David Danks
Institute for Human & Machine Cognition

"Theories of Human Causal Learning"

ABSTRACT: As of two years ago, there were three dominant psychological models of human causal learning: the Rescorla-Wagner model, conditional DP theory, and Causal power theory (sometimes called the Power PC theory). Unfortunately, these three theories were not directly comparable, since the Rescorla-Wagner model makes dynamic (i.e., step-wise) predictions, and the other two make predictions only about asymptotic behavior. In this talk, I first show how to compare the theories in two different ways: first by providing a general characterization of the long-run behavior of the Rescorla-Wagner model, and second by providing dynamical versions of the conditional DP and Causal power theories.

With these theoretical results in hand, we can readily see a crucial, common feature of these three theories: they all predict that C will be judged to cause E iff C and E are associated conditional on all other variables considered. There are certain causal structures for which this general rule will yield non-normative predictions, and I report the results of experimental work testing people's behavior when presented with these types of structures. I close by addressing a series of unresolved problems in this field (some suggested by the experiments just described), and possible extensions to other areas of cognitive psychology.

Back to Talks Page