Modeling social influence as a reinforcer reveals varying individual learning strategies and the group's structure

Abstract

In a social context, knowledge can be gained through observation and imitation, but in a numerous group, social influence can also be contradictory and confusing. The mechanisms of social learning in groups have not been fully characterized, partly due to the lack of adequate mathematical description. Using the known reinforcing property of social influence we introduce a reinforcement learning model which can account for social effects at the group and individual level. We use it to reveal pairwise influence relations and the overall learning strategies in a reversal learning task applied to a cohort of mice housed in an Intellicage. Animals make an efficient use of social influence when the goal of the group agrees with their own, but switch to an individual learning scheme when these goals are misaligned. They also exhibit varied decision rules depending on their own motivational state, and a selectivity in assimilating social information depending on the state of their conspecifics.