Proceedings of the Annual Meeting of the Cognitive Science Society

Spatial coincidences allow us to infer the presence of latent causes in the world. For instance, an unusually large clusterof ants allows us to infer the presence of a food source. The leading cognitive model for such inferences is Bayesian,but the Bayesian algorithm is computationally taxing. Humans likely employ a more efficient, approximative algorithm.To characterize the cognitive algorithms used, we had subjects judge whether a set of dots was drawn from a uniformdistribution or from a mixture of a uniform and a gaussian source (tending to produce clusters). Responses systematicallydeviate from Bayesian optimality: as the number of dots increase, subjects more often report a latent cause where noneexists. The bias is accounted for by a Bayesian clustering algorithm that cumulatively considers the next-nearest dot to aputative source. This finding helps characterize our tendency to perceive causal patterns where none exist.