Neural Measures of Human Decision-Making Track Evidence Accumulation in Learned Space

Abstract

Neural decision-making flexibly integrates evidence across sensory, mnemonic, and semantic domains. Yet prior demonstrations have focused on evidence that is either directly available in stimuli, retrieved from established representations, or computed relative to fixed perceptual frameworks. A fundamental question remains: does neural evidence accumulation extend to decisions based on evidence that must be computed through learned representational transformations? Here we show it does. We recorded scalp EEG while participants classified continuously oriented visual stimuli into discrete categories defined by an experimenter-imposed boundary. Category-level evidence was operationalized as category coherence or the angular distance between each stimulus and the learned boundary. We predicted that if neural decision mechanisms are truly domain-general, the centroparietal positivity (CPP)—a scalp EEG potential indexing evidence accumulation—should scale with category coherence, and individual differences in CPP sensitivity should correlate with computational drift rates. Both predictions were supported: CPP slopes measured from human volunteers (both sexes) increased monotonically with category coherence, and individual differences in CPP slopes correlated with individual differences in drift rates across participants. These findings reveal that the brain's decision machinery treats evidence identically regardless of its representational origin—whether externally available, preexisting, or computed through learned transformations.