Reward prediction errors modulate attentional vigilance

Attention and learning are intertwined. While previous work has primarily examined how the focus of attention can shape learning, how the dynamics of learning might impact your attentional state on a moment-to-moment basis is an open question. Here we leverage reinforcement learning theory to investigate the influence of rewards and reward prediction errors on attentional vigilance. Specifically, we ask how trial-by-trial reward prediction errors, which are the currency of the RL system and the primary behavioral correlate of dopaminergic activity, affect attentional vigilance. Using a task that simultaneously assessed people’s attentional vigilance and RL performance, we demonstrate that attentional state is influenced by both the magnitude and valence of recent reward prediction errors, with attention covarying with signed reward prediction errors. These findings demonstrate that rewards – and surprising ones in particular – reduce lapses and improve attentional vigilance. By highlighting a robust interaction between core computations of the RL system and attentional dynamics, these findings may provide preliminary evidence for a potential role of dopamine in mediating the relationship between learning and attention.