Parsing the Mechanistic Heterogeneity of Hallucinations Using Theory- and Data-Driven Computational Approaches

Abstract

Behavioral task performance can be driven by latent factors which are not directly observed. Generative models of performance can reveal those latent factors, which may themselves help us understand the systems driving performance. Computational models of perception could identify the latent factors underlying hallucinations. Modeling of performance on an auditory conditioned hallucinations(ACH) task shows that hallucinations may arise from an over-weighting of prior beliefs during perception. However, it remains unclear whether this susceptibility reflects hallucination state or trait vulnerability, is specific to perceptual modality, and whether it is present in all types of hallucinations.