The development of perceptual priors

Abstract

Bayesian inference has come to be regarded as the best, statistically optimal, way to deal with the sensory uncertainty inherent in our natural environment. One way to cope with such uncertainty is to incorporate our pre-existing knowledge about the world. However, we know very little about the circumstances in which human observers integrate sensory information with prior knowledge in a way that is close to optimal. We understand even less about how the developing brain adapts to the environmental statistics, learns to use them efficiently, and what factors may underlie the development of this critical perceptual skill. We addressed these questions though a series of psychophysical experiments, in which adults and 6- to 11-year-old children estimated the location of unseen targets based on a noisy sensory cue and a prior distribution that can be learned over the course of the experiment.
In Chapter 2, we showed that adult observers weighted sensory and prior information by their reliabilities but were far from optimal and struggled to generalise to untrained reliabilities in complex situations. The findings of Chapter 3 showed that 6- to 8-year-olds also weighted priors in proportion to their reliability, but they were slow to tune their behaviour to the statistics over time and remained furthest from optimal. Six- to -eight-year-olds’ performance reached adult-like levels when the priors were explicitly shown. Conversely, when the decision rule was made more complex, 6- to 8-year-olds’ abilities to distinguish between the priors broke down and adults’ performance became more child-like. These findings prompted us to investigate whether individual differences, specifically in working memory, may predict performance in adults. The distance from optimal was not predicted by working memory capacity, beyond general cognitive abilities.
Together, these studies offer fresh insights into the capacity and limitations both adults and 6-11-year-old children have in learning and efficiently using novel environmental statistics.