Social Influence, Temporal Discounting and Active-Passive Gap in Explore-Exploit Dilemma

Abstract

In this dissertation we aim to further investigate the factors affecting explore-exploit decision making in humans. Exploratory behaviour refers to a very broad category of behaviors engaging in gathering information about environment and there are so many factors at play. Factors from specifics of situation and stimuli to social features. In the current work, we studied a few of these factors using a version of bandit task, called the Horizon Task. These factors are temporal discounting, social influence and being active versus passive in receiving information. The explore-exploit dilemma describes the trade off that occurs any time we must choose between exploring unknown options and exploiting options we know well. Implicit in this trade off is how we value future rewards. In theory there should be a tight connection between how people discount future rewards - `temporal discounting' - and how likely they are to explore, with less temporal discounting associated with more exploration. In the first experiment, we tested whether this theoretical prediction holds in practice. We used the 27-item Delay-Discounting Questionnaire to estimate temporal discounting and the Horizon Task to quantify two strategies of explore-exploit behavior: directed exploration, where information drives exploration by choice, and random exploration, where behavioral variability drives exploration by chance. We found a clear negative correlation between temporal discounting and directed exploration and no relationship between temporal discounting and random exploration. In the second experiment, we investigated the effect of social information on explore-exploit decisions. Social information was shown in the form of choices that (participants were led to believe) were made by another person. Given this social information there are three ways for subjects to behave: to copy what the other person did (herding), to do the opposite (diversification), or to ignore the other person’s response completely. We found the social effect depends crucially on the interaction of two factors: the number of choices that subjects would make in the future (the horizon) and whether the outcome of the other person’s choice would ultimately be revealed or not. In the third study, we took a closer look at a failed replication from our lab. In particular, we examined our failure to detect evidence for horizon-dependent changes in directed and random exploration in a passive version of the Horizon Task which was also used in the second experiment here. We hypothesized that this failure to replicate may have been due to the change between active and passive versions of the task. In both a between-subjects and within-subjects design we found that participants showed increased ambiguity aversion and reduced horizon-dependent directed and random exploration in the passive version of the task. This clearly demonstrates how behavior in the Horizon Task is critically dependent on the nature of the forced-choice trials, and that our failed replication was caused by the seemingly minor shift from active to passive versions of the task.