Andrew Whalen (University of Edinburgh)
Abstract. Social learning and asocial learning are sometimes seen as two conflicting ways in which individuals make decisions and learn about the world around them. Increasingly research has found that instead of being two conflicting learning processes, individuals, including children, will combine social and asocial sources of information to make decisions. One approach to understanding how social information might be integrated into other learning processes is by studying already well established models of human and animal learning. We present work that tries to understand how social information might be integrated into models of reinforcement learning, particularly temporal difference (TD) learning. Using a series of simulations we demonstrate that, unlike other simulation studies that treat social and asocial learning separately, social learning is nearly always beneficial, and that small amounts of social learning allows for the formation of stable traditions of complex socially transmitted behaviors in artificial populations. To better understand how social and asocial information is combined in humans, we ran a series of experiments to analyze how social learning impacts reinforcement learning, and find that individuals use social learning both to choose the actions that they perform, and as a secondary reinforcer that alters the associations they build between actions and rewards. These results highlight the importance of understanding how social and asocial sources of information are integrated on fine temporal scales for understanding the evolution of social learning and human culture.