Understanding, predicting, and learning from other people’sactions are fundamental human social-cognitive skills. Littleis known about how and when we consider other’s actionsand outcomes when making our own decisions. We developeda novel task to study social influence in decision-making: thesocial multi-armed bandit task. This task assesses how peoplelearn policies for optimal choices based on their ownoutcomes and another player's (observed) outcomes. Themajority of participants integrated information gained throughobservation of their partner similarly as information gainedthrough their own actions. This lead to a suboptimal decision-making strategy. Interestingly, event-related potentials time-locked to stimulus onset qualitatively similar but theamplitudes are attenuated in the solo compared to the dyadicversion. This might indicate that arousal and attention afterreceiving a reward are sustained when a second agent ispresent but not when playing alone.