The joint estimation of uncertainty and its relationship with psychotic-like traits and psychometric schizotypy.

Abstract

Learning involves reducing the uncertainty of incoming information-does it reflect meaningful change (volatility) or random noise? Normative accounts of learning capture the interconnectedness of this uncertainty: learning increases when changes are perceived as meaningful (volatility) and reduces when changes are seen as noise. Misestimating uncertainty-especially volatility-may contribute to psychotic symptoms, yet studies often overlook the interdependence of noise. We developed a block-design task that manipulated both noise and volatility using inputs from ground-truth distributions, with incentivised trial-wise estimates. Across three general population samples (online Ns = 580/147; in-person N = 19), participants showed normative learning overall. However, psychometric schizotypy and delusional ideation were linked to non-normative patterns. Paranoia was associated with poorer performance and reduced insight. All traits showed inflexible adaptation to changing uncertainty. Computational modelling suggested that non-normative learning may reflect difficulties inferring noise. This could lead one to misinterpret randomness as meaningful. Capturing joint uncertainty estimation offers insights into psychosis and supports clinically relevant computational phenotyping.