Investigating specificity of experimentally induced prior expectations in motion perception

The brain uses sensory information that is often uncertain in order to efficiently generate perceptual representations of the world. This observation has led to the Bayesian brain hypothesis, in which the brain combines internal expectations of the world with unreliable external sensory information in a nearly optimal probabilistic manner. Recent studies have suggested that to be true in statistical learning, a process by which people learn arbitrary associations between stimuli based on the statistics of interstimulus contingencies. Here, I employ a psychophysical experimental procedure to investigate the specificity of induced biases through fast statistical learning using moving stimuli with distinctly distributed motion directions, differentiated by color. I found that subjects exhibit sub-optimal behavior when estimating the direction of the stimuli by ignoring color information, and learning a single bimodal prior for both distributions of the stimuli. These results suggest that there was extensive transfer between the statistical properties of one distribution to the other. The biases in motion direction estimation were well explained by a computational model that assumed subjects used a Bayesian strategy, and extended variations of this model verified the formation of a single prior. Finally, I discuss the implications of these results, and propose interesting directions for future research.