罰は線条体の予測誤差を調節することにより報酬学習を強化する

People often make decisions in contexts where rewards and punishments co-occur, yet most human research still examines reward and punishment learning as independent processes. Here, across three studies, we address this gap by demonstrating that punishments amplify reward learning and its neural correlates in healthy human participants. In Study 1 (N = 102, 69 females and 33 males), participants performed a probabilistic learning task involving monetary rewards and punishments presented in either intermixed or separated contexts. In intermixed contexts, punishments enhanced reward learning accompanied by changes in computational parameters, including higher learning rates from reward prediction errors. In Study 2 (N = 26, 18 females and 8 males), fMRI revealed that punishments amplified reward prediction errors signals in the caudate. Study 3, an fMRI meta-analysis, confirmed that striatal reward responses are consistently stronger when punishments are present. Across studies, we found no reciprocal enhancement of punishment learning by rewards. Together, these findings demonstrate that punishments sharpen reward learning through striatal modulation and underscore the extent to which reward learning is influenced by its broader outcome context.Significance Statement A central question in cognitive neuroscience is how people learn to pursue rewards and avoid punishments. In everyday life, these processes rarely operate in isolation; they often co-occur and interact. For example, a win often feels more rewarding after a loss. Yet, human research has largely examined rewards and punishments separately, overlooking their interplay. Across three studies, we find that punishments enhance reward learning and its underlying striatal prediction-error signals, whereas rewards do not exert a comparable influence on punishment learning. This directional modulation shows that reward learning depends on its broader outcome context, with punishments acting as a key contextual factor. This work offers a theoretical framework for understanding punishment-reward interactions and a new benchmark for future research.