Expectation generation and its effect on subsequent pain and visual perception

Bayesian accounts of perception, such as predictive processing, suggest that perceptions integrate expectations and sensory experience, and thus assimilate to expected values. Furthermore, more precise expectations should have stronger influences on perception. We tested these hypotheses using a within-subject paradigm that independently manipulated the mean, variance (precision), and skewness of cues presented as ratings from 10 prior participants. Forty-five participants reported their expectations regarding the painfulness of thermal stimuli or the visual contrast of flickering checkerboards. In a second session, similar (sham) cues were each followed by either a noxious thermal or a visual stimulus. Perceptions assimilated to cue-based expectations in both modalities, but precision effects were modality-specific: more precise cues enhanced assimilation in visual perception only, while higher uncertainty slightly increased reported pain. fMRI analysis revealed that the cues affected higher-level affective and cognitive systems–including assimilation to the cue mean in a neuromarker of endogenous pain processing and in the nucleus accumbens, and activity consistent with aversive prediction-error-like encoding in the periaqueductal gray during pain perception–but not early perceptual processing systems. Furthermore, behavioral and computational models of the expectation session revealed that expectations were biased towards extreme values in both modalities, and towards low-pain cues specifically. These findings suggest that predictive processing theories should be extended with mechanisms such as selective attention to outliers, and that expectation generation and its perceptual effects are mostly modality-specific and primarily influence higher-level processes rather than early perception, at least when cues are not reinforced.