Social context matters - Characterizing adolescent cooperation strategies when perceiving the other as a peer versus computer partner

Cooperation is a crucial prosocial skill that is refined during adolescence and relies on trust. A substantial body of literature has shown that adolescence is a time of significant development in social decision making. It is unclear how this development affects adolescent decisions to establish trust and sustain cooperation with peers versus artificial agents. Given the current technological advances, adolescents have been immersed in virtual interactions with human and artificial intelligent (AI) agents, so they may establish similar levels of trust and reciprocity with human and non-human agents. On the other hand, adolescence is a time of increasing interest in peer relationships and, consequently, establishing trust and reciprocity with peers may be a more important goal. The current study directly tests whether adolescents differ in their cooperation tendencies when interacting with, or believing that they interact with peers, versus knowingly interacting with artificial partners in economic exchange games. Adolescents (N= 67) were invited into the laboratory to play various rounds of the multi-round trust game. Some participants (N= 38) also played version of the game with a fixed strategy algorithm by themselves on a separate visit. 40 participants returned for a second visit in pairs. Pairs were matched for race, sex, age, and IQ and played the multi-round trust game with each other. On some consecutive games, they were told that they continued to play with a peer (i.e., social condition) but they were actually playing with an adaptive algorithm, while in the remaining games, they were correctly informed that they were playing with an adaptive computer algorithms (i.e., non-social condition).In line with our hypothesis, we found that both partner adaptivity and social context (believing that they interacted with a peer) influenced cooperation. In line with our expectation, adolescents sustained cooperation more when interacting with adaptive versus fixed partners. They were also more generous when playing with an adaptive partner, particularly in the social condition. We employed fixed economic strategies, reinforcement learning (RL) models, and a combination thereof to describe adolescents’ cooperation. Across games, direct reciprocation, i.e., the economic tit for tat strategy, best captured their cooperation behavior. In the social condition, however, adolescents reinitiated cooperation more often after previously defecting. The winning model was a more generous strategy that combined tit for tat with an RL confidence term that prescribed repairing trust and reinitiating cooperation. Overall, we show that that both partner adaptivity and social context support sustained cooperation in adolescence.