Abstract
To survive and thrive, animals must navigate risk by anticipating and avoiding potential losses while approaching potential gains. Although researchers have leveraged neuroimaging to predict risky choices in humans, consensus on unitary versus distinct underlying neural and psychological mechanisms remains elusive. Across four functional magnetic resonance imaging studies (combined n = 230), we tested univariate and multivariate models predicting trial-by-trial risky gambling choices in an original sample, replicated predictive features in an independent sample, and generalized predictive features to samples playing a different gambling task. Prechoice activity in distinct circuits predicted subsequent risky (i.e. nucleus accumbens [NAcc] and medial prefrontal cortex [MPFC]) versus safe (i.e. anterior insula [AIns]) choices. A triple dissociation analysis distinguished these neural predictors of risky choice from neural activity associated with sensory input and motor output. Prechoice NAcc and MPFC activity was further associated with individuals' preferences for risky choices, while AIns activity was associated with individuals' preferences for safe choices. Finally, prechoice AIns activity in response to risky gambles was associated with lower levels of debt in real life (controlling for demographic, behavioral, and self-report measures). Together, these convergent findings reveal distinct and replicable neural predictors of risky choice, which generalize across analyses, tasks, and individuals.