Phenomenology is Bayesian in its application to delusions.

Abstract

Sass and Byrom (1) argue that phenomenology “expands the range of testable hypotheses”. This resonates with our view that phenomenology “leads to neurobiological hypotheses, which can be tested experimentally” (2,3). It is also a welcome modification of Sass’ proposal (4) that phenomenology “serves an explanatory function”. If phenomenology “explains” schizophrenia by proposing its core essence as a disturbance of “hyperreflexivity/ipseity” (4), it claims knowledge about causal relationships without recourse to testing hypotheses about mechanism. The authors see a conflict between “enactive” or “embodied” approaches to cognition and “more intellectualistic sounding” prediction-error formulation. We suggest that this apparent conflict is related to a misunderstanding of the term beliefs in predictive coding accounts. In current accounts of Bayesian hierarchical predictive coding, a belief is considered merely a probability distribution over some unknown state and may or may not be consciously accessible (5). A central claim of hierarchical predictive coding models is that such beliefs are fundamentally embodied even at the lowest levels of sensory processing, clearly not implying intellectual conjecture and refutation. Accordingly, studies of patients with schizophrenia point to an alteration of predictive mechanisms at low levels of sensory processing. Behavioral and functional neuroimaging studies of illusory visual perception in schizophrenia patients have suggested impaired predictive mechanisms in early visual cortex (e.g., (6,7)). Similarly, mismatch-negativity (MMN), an electrophysiological signal that is thought to reflect the automatic registration of irregularities in sensory input, is reduced in patients with schizophrenia (8). The empirical evidence for altered predictive coding seems to contradict the authors’ assumption that the predictive mechanisms involved in delusion formation/maintenance necessarily implicate, or are limited to, cognitive or “intellectualistic” processes. Furthermore, the authors suggest that the exaggerated prediction-error signaling giving rise to hypersalience does not account for hyposalience and an associated “anything-goes” attitude, which they propose may be due to a dysfunction in the default-mode network. Apart from possible problems with “reverse inference”, we question the assertion that hyposalience as described by the authors is incompatible with the notion of prediction-error dysfunction. To the contrary, predictive coding accounts actually predict that the proposed exaggerated prediction-error signaling (or imbalance in the precision of prediction errors and prior beliefs) (5) results in an impaired distinction between normally expected and unexpected events. This is exemplified by reduced MMN amplitude in schizophrenia conceptualized as a consequence of altered prediction-error signaling. In this context, attenuated mismatch responses in schizophrenia patients may actually not reflect the failure to register surprising events, but rather the fact that each event is surprising (5,7). Hyper- and hyposalience are two sides of the same coin, accounted for by a single factor, prediction-error dysfunction (9). This is supported by Heidelberg psychiatrist Mayer-Gross’ (1932) observation of reduced anticipatory expectation in the “self disturbances”, due to the ongoing “interruption” of current goal-processing by the “made” or influenced perceptions, movements, thoughts, etc., which characterize those disturbances (10). There is only the compelling sensory evidence of now: “no temporal order prevails, each sensory impression is equally valued, replacing its predecessor”. This reduction in top-down, embodied perceptual expectation in the “self disturbances” observed by Mayer-Gross anticipates the predictive coding account of attenuation of visual illusions (e.g., the hollow-mask illusion) in schizophrenia and how this relates to delusions and related symptoms (as discussed by Corlett, Fletcher and Frith, and others). The phenomenological psychiatrist Binswanger also described the self in schizophrenia as captive in the present moment in a “temporal shrinking” of past and future which resembles dreaming (11). In his fiction, Kafka depicts the reduced expectation in dreamlike-hypnagogic experiences, where protagonists report “expecting” the very events that “surprise” them (12). This is not “bizarre-as-banal”, but the absence of banal altogether. It is also not “anything-goes”, but can be formalized in the Bayesian hierarchy as outlined above. Similarly, Binswanger describes a “monotonous” spreading of the delusion to the entire perceptual field in terms of a “loosening” of context from prior learning (2,11). Sass and Byrom's language suggests that “phenomenology” does the work of description and inference (e.g., “phenomenology is acutely sensitive”, “phenomenology is cautious”). Such phrasing may lead to the mistaken assumption that phenomenology is a body of finalized results articulated by one individual or group, rather than a rigorous method, which includes an ongoing process of dialogue, refinement, and consensus.