Many aspects of human social behavior are likely to be shaped by punishment, threats of punishment, and avoidance of punishment. However, surprisingly little is known about the psychological mechanisms underpinning these influences. The goal of the research presented in this thesis was to investigate how basic reinforcement-based learning processes contribute to adaptive social behavior. To this end, we investigated how human behavior is shaped by punishment and threats of punishment in simple laboratory models aimed at capturing essential aspects of the social world. In order to probe the mechanisms underlying overt behavior, we tested the capacity of simple computational models to describe human behavior.
In Study I, we investigated how the capacity to control behavior was affected by threat of punishment for performance mistakes, motivated by the functional perspective that control is most needed when mistakes are costly, as common in social interaction. Furthermore, we investigated if activity in the facial corrugator supercilii muscle might function as a novel index of this control process. We found that performance was enhanced under threat of punishment relative to non-threat, and that activity in the corrugator supercilii responded to performance errors during threat of punishment, similar to the known electroencelographic correlate of performance errors, the error related negativity.
Studies II and III investigated how social stimuli, emotional and out-group faces, affected avoidance learning when serving as punishers (Study II) and cues for punishment (Study III). In both studies, we were interested in how pre-existing negative attitudes (racial bias) would affect avoidance learning, motivated by the fact that social groups and group animosity play an important role in modern societies. We found that individual differences in racial bias strongly modulated avoidance learning, by leading to an increased learning rate in response to out-group faces (Study II) and to more biased behavior when out-group faces served as cues for electric shocks (Study III). Furthermore, Study III revealed that several types of social and non-social stimuli, which previously have been shown to be preferentially associated with fear (snakes, threatening faces, out-group faces, and guns), all shared a common effect on avoidance learning relative to control stimuli, by promoting adaptive behavior when these stimuli were good predictors of aversive outcomes but corrupting behavior when serving as bad predictors of aversive outcome. This pattern could be explained by a computational model positing two competing valuation systems.
In Study IV, we asked how avoidance behaviors can be learned and transmitted socially by observing others ́ behavior in a setting with threat of punishment. We used agent-based simulations to show that the omission of expected punishment together with social learning can explain the generation, maintenance and transmission of avoidance behaviors that form “avoidance traditions” on the group-level. The assumptions underlying this model were corroborated in four experiments, which together showed that humans are exceedingly prone to copy the behavior of others when threatened with punishment.
Together, these four studies indicate that important aspects of human social behavior can be understood as involving mechanisms of basic reinforcement learning, and that simple computational models of these learning processes can provide important insights into human behavior.