Challenging and extending predictive coding as an account of brain function

Probabilistic models of brain function, which propose that the brain can be understood as implementing the principles of optimal statistical inference, have become extremely influential in recent years. Predictive coding is perhaps the most widely held, and best supported, of these models, particularly within cognitive neuroscience. However, current models of predictive coding and its neuronal substrates are still relatively simple, and do not explain how humans solve a number of fundamental problems. This limits their power to explain brain function. I propose a series of experiments designed to test how human subjects address a number of these core problems. I will use behavioural and neuroimaging data to develop and test extensions to current models (or, if necessary, provide an alternative framework for understanding brain function). The purpose of this is two-fold, to give insight into the computations that underlie cognitive function, and to provide understanding of the neurobiological processes that support those computations. The project will thus constitute a stepping stone towards developing a mechanistic model of how the brain implements cognition.