Building and Dismantling Trust: From Group Learning to Character Judgments

AbstractTrust is central to social behavior. In interactions between strangers some information about group affiliation is almost always available. Despite this, how group information is utilized to promote trust in interactions between strangers is poorly understood. Here we addressed this through a two-stage experiment where participants interacted with randomly selected members of two arbitrary groups and learnt their relative trustworthiness. Next, they interacted with four novel individuals from these two groups. Two members, one from each group, acted congruently with their group’s previous behavior while the other two acted incongruently. While participants readily learnt the group-level information in the first phase, this was swiftly discounted in favor of information about each individual partner’s actual behavior. We fit a reinforcement learning model which included a bias term capturing propensity to trust to the data from the first phase. The bias term from the RL model predicted participants’ initial behavior better than their expectations based on group membership. Pro-social tendencies and individuating information can overcome knowledge about group belonging.

Return to previous page