Caroline Bévalot
National Institute of Health and Medical Research, Unicog, Gif-sur-Yvette
Sorbonne University, Paris
Our brain is constantly facing uncertainty. One way to reduce uncertainty is to build a representation of our environment, use this representation together with our sensory inputs and update it according to its confidence. However, subjects are more or less optimal in doing so and the difference is striking in psychiatric disorders. This difference can be characterized under the Bayesian framework which constitutes an optimum of the computations under uncertainty. Yet, up to now, characterization of subjects’ behavior in psychiatric disorders and especially schizophrenia is very heterogeneous. During my PhD, we analysed several factors of this heterogeneity. First, we studied whether alterations of the Bayesian inference could be compared between experiments using implicit or explicit priors. We found a dissociation in the way implicit and explicit priors are used and a difference in the computations they elicit. Secondly, we identified two main types of uncertainty about the prior representation. Each of them is related to distinct steps in the Bayesian inference : either decision, or learning. We studied how psychotic, anxious and autistic features were associated with an alteration at one of these steps. We showed that subjects with higher psychotic features tend to neglect the sensory likelihood at the decision step. We showed the opposite pattern in anxious disorders. Finally, using magnetoencephalography, we wondered whether the alterations of the Bayesian inference were a core alteration leading to symptoms in schizophrenia rather than constituting a side alteration.
View a recording of this session here.