Predictive Processing, Source Monitoring, and Psychosis

Abstract

A comprehensive understanding of psychosis requires models that link multiple levels of explanation: the neurobiological, the cognitive, the subjective, and the social. Until we can bridge several explanatory gaps, it is difficult to explain how neurobiological perturbations can manifest in bizarre beliefs or hallucinations, or how trauma or social adversity can perturb lower-level brain processes. We propose that the predictive processing framework has much to offer in this respect. We show how this framework may underpin and complement source monitoring theories of delusions and hallucinations and how, when considered in terms of a dynamic and hierarchical system, it may provide a compelling model of several key clinical features of psychosis. We see little conflict between source monitoring theories and predictive coding. The former act as a higher-level description of a set of capacities, and the latter aims to provide a deeper account of how these and other capacities may emerge.

Keywords: predictive coding; psychosis; schizophrenia; source monitoring.

Figure

This toy example illustrates how sensory evidence is shaped by expectations, and how ensuing inferences govern further evidence-gathering biased towards confirming initial hypotheses. It shows, in simplified form, a recursive stream of processing leading from ambiguous sensory experience to an inference that one is being threatened. Several levels of inference are depicted, with top-down expectations based on past experience (referred to as “priors”) shaping the interpretation of bottom-up inputs from each previous level. Thus, the very first impression of the distant silhouette (“is it a human or a scarecrow?”) will be affected by the context (a scarecrow perhaps becomes more probable in an isolated rural setting), and guided by top-down predictions about which bottom-up signals will most reliably discriminate between likely candidate hypotheses. (Here for example, the “human” and “scarecrow” hypotheses generate very different predictions about the presence of agentic movement – motion signals are therefore highly informative (precise), and are upregulated accordingly by top-down gain control mechanisms including gaze direction and covert attention) Importantly, the winning inference (here, human) generates modified expectations for ensuing bottom-up input (for example, movement may perhaps be more readily perceived and attended to when the over-arching hypothesis is “human”). Put simply, the inference that the causal source of the sense data is “human” means that further inputs are processed in a way that tend to confirm “human” as the correct hypothesis. Moreover, the winning hypothesis generates specific predictions about which aspects of the percept are behaviourally relevant, constraining the array of hypotheses entertained at the next level of inference (for example, the attribution of social significance becomes relevant if the existence inference includes “human” and “movement”). And, at this higher level, different priors play a role in inferring the social significance. Importantly, as well as leading to a further updating of ever higher-level inferences, the interpretation of social intent provides further confirmatory evidence that the lower-level inferences (“movement” and “human”) are correct and meaningful. After all, if the percept has social significance it confirms that the figure is human, and if it is human, movement is highly probable.