Continuing on from the previous year, the lecture addressed a hypothesis which is currently the subject of intense theoretical and experimental exploration in the sciences : the idea that the human and animal brain contains statistical inference mechanisms approaching the normative equations of Bayesian inference.
In the previous lecture, we saw that this simple mathematical theory, which characterizes plausible reasoning in the presence of uncertainties, accounts for a wide variety of psychological and physiological observations. When our brain receives ambiguous input, it seems to reconstruct the most likely interpretation. This inference is hierarchical and provides access to abstract knowledge. Decision-making could result from combining this Bayesian calculation of probabilities with an estimate of the consequences of our choices. The architecture of the cortex could have evolved to perform Bayesian inference at high speed and in a massively parallel fashion. The algorithm used could explain how our brain anticipates the outside world and responds to novelty.
The aim of this year's lecture was to explore the hypothesis that all these elements are present in the very young child, from the very first year of life and perhaps from birth. Indeed, the human baby seems to be endowed with probabilistic reasoning skills. The child's brain makes predictions about the outside world, and seems to have a powerful algorithm for learning statistical regularities. Could the learning of faces, objects, words or linguistic rules be explained by Bayesian inference ?