Anomalous Perceptions and Beliefs Are Associated With Shifts Toward Different Types of Prior Knowledge in Perceptual Inference

Abstract

Psychotic phenomena manifest in healthy and clinical populations as complex patterns of aberrant perceptions (hallucinations) and tenacious, irrational beliefs ( delusions). According to predictive processing accounts, hallucinations and delusions arise from atypicalities in the integration of prior knowledge with incoming sensory information. However, the computational details of these atypicalities and their specific phenomenological manifestations are not well characterized. We tested the hypothesis that hallucination-proneness arises from increased reliance on overly general application of prior knowledge in perceptual inference, generating percepts that readily capture the gist of the environment but inaccurately render its details. We separately probed the use of prior knowledge to perceive the gist vs the details of ambiguous images in a healthy population with varying degrees of hallucination- and delusion-proneness. We found that the use of prior knowledge varied with psychotic phenomena and their composition in terms of aberrant percepts vs aberrant beliefs. Consistent with previous findings, hallucination-proneness conferred an advantage using prior knowledge to perceive image gist but, contrary to predictions, did not confer disadvantage perceiving image details. Predominant hallucination-proneness actually conferred advantages perceiving both image gist and details, consistent with reliance on highly detailed perceptual knowledge. Delusion-proneness, and especially predominance of delusion-proneness over hallucination-proneness, conferred disadvantage perceiving image details but not image gist, though evidence of specific impairment of detail perception was preliminary. We suggest this is consistent with reliance on abstract, belief-like knowledge. We posit that phenomenological variability in psychotic experiences may be driven by variability in the type of knowledge observers rely upon to resolve perceptual ambiguity.

Fig. 1.

(a) Theoretical function determining probability of generating percepts based on sensory evidence explained by predictions (perceptual hypotheses). At (i), no predictions explain enough evidence in a two-tone to generate a percept. At (iii), templates are recognized using existing prior knowledge. At (ii), prior knowledge increases probability of generating percepts. Generating a percept when less evidence is explained (leftward-shifted function), which we hypothesized occurs in early psychosis and psychosis-proneness, might readily generate percepts that sometimes misrepresent the environment, giving rise to anomalous inferences. (b) An example of a two-tone image used in the study (see figure 2c for template). (c) Global condition. (d) Local condition.

Fig. 2.

(a) Observers discriminated whether a dot was on/off a figure embedded in two-tone images in “two-tone trials.” Observers gained prior knowledge about two-tones from natural images in “template trials.” (b) Trials were grouped into two-tone and template blocks. Two-tone blocks contained 24 trials with 6 two-tones, each presented 4 times (all combinations of dot on/off, Global/Local condition). Per two-tone block, observers had viewed templates for 3 images (Post-Template) but not the others (Pre-Template). Pre-template and Post-Template two-tones were interleaved. Template blocks contained presented 3 templates, corresponding to Pre-Template two-tones from the preceding block. There were 30 two-tone/template images and 10 of each block. (c) An example of a template image that corresponds to the two-tone in figure 1b.

Fig. 3.

(a) Mean (SE) discriminability (d′), drift rate (v), and decision threshold (a) across task conditions. Template prior knowledge allowed participants to extract evidence faster (higher drift rate, v) and make perceptual decisions based on less evidence (lower decision threshold, a), with an overall improvement in discrimination (higher d′). (b) Δd′ and Δv correlated in both conditions, supporting that prior knowledge improved performance by enhancing extraction of information from two-tone images.

Fig. 4.

Univariate regressions showed hallucination-proneness (CAPS) was associated with greater improvement in discriminability (Δd′) with template exposure (a) but this was not reflected in change in drift rate (Δv) (b). Delusion-proneness trend-predicted lower Δd′ (c) and predicted lower Δv in the Local condition (d). Multiple regressions showed a different pattern of results. Predominant hallucination-proneness predicted greater Δd′ in the Global condition (e) but also greater Δd′ and Δv in the Local condition (e and h). In contrast, predominant delusion-proneness predicted lower Δd′ and Δv in the Local condition (g and h). R-values are equivalent correlations. CAPS, Cardiff Anomalous Perceptions Scale.