There is a close connection between video games and social life, and researchers are interested in whether and how video games shape aggression and prosocial behaviors. However, there are great inconsistencies across studies on this topic. These mixed results may be due in part to a focus on learning models that were relevant in research on traditional media like television but are less useful in research on video games. Unlike other media, video games are characterized by frequent game-player interactions and immediate feedback, and there is evidence that in-game rewards and punishments can shape aggressive or prosocial behavior inside and outside the game. We argue that reinforcement learning may help us to understand the effects of video games on aggressive and prosocial behaviors, and propose a conceptual model based on this argument.