Proceedings of the Annual Meeting of the Cognitive Science Society

Attention-based models of categorization and associative learning have received considerable support from human learning phenomena in which multiple predictive cues compete for association with outcomes. Among these, several phenomena (e.g. the highlighting effect and inverse base-rate effect) lend strong support to models that propose attention is driven by the experience of prediction error, and is distributed strategically to minimize prediction error during current and future learning. Here we explore the possibility that attention is determined instead by a relatively simple combination of stimulus novelty and association strength. We apply the model to several key findings in the literature on the inverse base-rate effect and related phenomena. Overall, the model provides a surprisingly good account of complex behavioral biases.