Computational models of hallucinations

Abstract : Recent advances in theoretical neuroscience have provided new insights into information processing within large brain-like networks operating in an uncertain world. The computational framework can overcome some of the complexity within the object of study by predicting how basic changes in neural architecture may lead to systems-level changes that translate into changes in behavior. Computational models offer ways to unify basic neurochemical findings with data from more macroscopic levels and to start to apply these findings to cognitive sciences and psychiatry. Some of these approaches have been used to investigate the underlying mechanisms of subjective experiences, such as hallucinations, which can spontaneously emerge into consciousness in the absence of any corresponding external stimuli. This chapter describes some recent theoretical studies on four categories of positive symptoms of schizophrenia: neurodynamics, noise, disconnectivity, and Bayesian models of hallucinations. Results from simulations of these neural networks as well as the potential alterations leading to aberrant experiences are presented and discussed.