Dissociable encoding of evolving beliefs and momentary belief updates in distinct neural decision signals

Abstract

Making accurate decisions in noisy or otherwise uncertain environments requires integrating evidence over time. Using simple tasks requiring rapid evaluation of a stationary sensory feature, two human neurophysiological signals have been found to evolve with similar integration dynamics, with one (centroparietal positivity; CPP) appearing to compute the running integral and continuously feed it to the other (motor beta lateralisation; MBL). However, it remains unknown whether and how these signals serve more distinct functional roles in more complex scenarios where information arrives discontinuously and requires more computational steps to form appropriate belief updates. We employed a volatile expanded judgement task that permits dissociation of the encoding of raw sensory information in each of a series of discrete stimuli ( objective evidence), the appropriately transformed belief updates from that information ( effective evidence), and the evolving belief itself (accumulated evidence). Whereas MBL traced the evolving belief across stimuli, the CPP was found only to locally encode the effective evidence associated with each individual stimulus. Furthermore, fluctuations in CPP amplitude after each single stimulus could account for variability in the encoding of accumulated evidence in MBL. These results suggest a flexible computational hierarchy in which effective evidence can be computed sample-by-sample at an intermediate processing level to drive downstream belief updates for protracted decisions about discrete stimuli.