The reality paradox: A unifying predictive coding theory of psychosis

We propose a predictive coding model of psychosis centred on precision-weighted Bayesian inference. Either over-precise priors or under-precise likelihoods result in posteriors strongly biased towards priors. When existing beliefs no longer can reliably discern the origin of stimuli in the world novel beliefs must arise to explain these changes and are themselves held with abnormal confidence.Changes in the world feel simultaneously familiar and strange, thinking about the changes makes my thoughts feel similarly paradoxical. If a Bayesian system under extremes of precision saw such a causational paradox, it would be forced to decide was that familiar but strange action done by me but controlled from afar or was that action from the environment and directed against me. Thus, the widespread positive symptoms become bayes optimal resolutions of the paradox as it pertains to thoughts, sensations, movement, and affect. A self-reinforcing process leads to a failure state when beliefs and perceptions become one; if a Bayesian system believes both my actions cause reality and reality causes my actions then it must do nothing, catatonia.The paper demonstrates how a healthy model could itself become high precision as an initially adaptive response to trauma. By doubling down on our beliefs about the world we can prevent trauma from changing the way we view it, but as the world keeps changing, a tsunami of high precision spreads from the initial schism.