A Computational Model of the Motivation-learning Interface

Saggar M., Markman A.B., Maddox W.T., Miikkulainen R. 2007. Proceedings of the Cognitive Science Society Meeting (2007)

Abstract

This paper presents a computational model of how motivation influences learning, elaborating on the empirical study of Markman, Baldwin and Maddox (2005). In a decision criterion learning task with unequal payoffs, the subjects were more likely to maximize the reward when their motivation was in line with the reward structure (i.e. when they were in a regulatory fit), whereas they were more likely to maximize accuracy when their motivation did not match the reward structure (i.e. when they were in a regulatory mismatch). The model accurately replicates this pattern of results, and also accounts for the individual subject’s behavior. In addition, the model makes the novel prediction that regulatory-fit subjects who are near the reward threshold will shift their strategy toward maximizing accuracy, whereas regulatory-mismatch subjects who are far from the reward threshold will shift their strategy toward maximizing reward. When the original data was reanalyzed, this model-driven prediction was confirmed. These results constitute a first computational step towards understanding how motivation influences learning and cognition.