Cognitive and Computational Accounts of Delusions: Problems and Progress

Abstract

We discuss the evolution of a computational model of delusions, beginning with a background consideration of how computational psychiatry, with its roots firmly based in cognitive neuropsychiatry, seeks to develop descriptive and mechanistic models that reach across different levels of explanation in order to provide more comprehensive understanding of how neurobiological, cognitive, subjective and sociocultural factors may all contribute to complex psychopathology. This quest for bridging explanations – or “consilience” – across the levels is a shared goal of computational psychiatry and cognitive neuropsychiatry and is, we argue, crucial to explaining delusional beliefs. We outline how early computational models appealed to prediction error disturbances as a basis for understanding the early emergence of delusions and show that, despite empirical support, there have been certain explanatory limitations that make a simple prediction error account partially limited. Embedding the account within the increasingly influential hierarchical predictive processing framework subsequently offered a more powerful and comprehensive account, particularly by encouraging the consideration hierarchically-organized inference and its evolution over time. However, further limitations remain in its explanatory scope, most notably the fact that delusions can emerge rapidly and suddenly in a way that seems revelatory and convincing. This phenomenon is not easily encompassed by the standard predictive processing account which emphasizes an iterative process of optimizing inference. However, more recent development in the form of “Hybrid Predictive Coding” posits a complementary rapid inference mechanism. We discuss how this hybrid approach may be key to a more comprehensive computational understanding of delusions.