Spontaneous convergence in cooperating dyads as explained by a reinforcement learning model of conformity

Conformity is one of the most basic forms of social influence and it has been studied under a variety of conditions. However, little is known about the mechanisms which explain how convergence emerges and changes on the bases of exposure to other’s actions. Most of the paradigms used to study this phenomenon induce behavioural convergence using a group rating task, a judgement task or game theory competition/cooperation frameworks. The aim of the current research was to create an experimental paradigm where cooperation emerges naturally and allowed to study the process of learning to adapt to other’s behaviour. First, 40 dyads performed a set of activities, either individually (I) or cooperatively (C). Then, participants performed a dyadic decision-making task in which they had to decide the position of a point and had the chance to change their guesses based on their peer responses. Importantly, convergence of responses was not explicitly instructed, nor rewarded. Responses were modelled using a Reinforcement Learning (RL) algorithm. Results showed a significantly higher level of convergence in C compared to the I group. In addition, participants in the C group reported a more trusting, satisfactory, rewarded, and synched overall experience. Also, RL Temporal Difference algorithm showed higher learning rate and explained better the convergence behaviour of the C group compared to the I group. Our study validates a new proposal to study conformity spontaneously appearing under cooperation.