Humans and animals adapt their behaviour based on the positive and negative feedback from the environment. This process is formally named as reinforcement learning (RL). This study investigated reinforcement learning in social versus nonsocial contexts, and examined whether individual differences of trait impulsivity moderates these learning processes.

42 participants completed both a social and a nonsocial version of the Probabilistic Selection Task and completed the Barratt Impulsiveness Scale (BIS-11). The learning processes were assessed using testing-phase learning accuracy and modelled  reinforcement-learning parameters. Reinforcement learning parameters were estimated using a Q-learning model with separate reward and punishment learning rates, as well as a stochasticity parameter (inverse temperature).

Results showed that learning differed across contexts. Participants showed greater punishment learning in the social condition compared with the nonsocial condition, whereas reward learning was better in the nonsocial condition. Testing-phase accuracy revealed a reduced advantage for reward learning over punishment learning accuracy in the social relative to the nonsocial context, primarily driven by improved punishment-avoiding performance. Trait impulsivity was found to interact with learning context and choice type in predicting learning accuracy. Exploratory analyses of inverse temperature further indicated greater choice consistency in the social condition.

Overall, the findings indicate that reinforcement learning is shaped by the social nature of the stimuli and feedback, and that the individual differences in impulsivity may influence the behavioural expression of learned values across contexts. Future research could incorporate more dynamic and diverse stimuli and extend these paradigms to clinical populations characterized by elevated impulsivity or social difficulties.