Investigating Distributed Decisions Using Bandit Problem Environments

Research in cognitive psychology regarding sequential decision-making usually involves tasks where an observer must make successive decisions among alternatives in order to determine the best choices to maximize payoffs. However, there exists little empirical work on how such observers might behave in a task where actions at each decision point require the allocation of resources over a distribution of alternatives rather than simple selection of a single option. Here, we present some preliminary work into how humans may make distributed decisions using a variant of bandit problem environment where multiple tokens are assigned on each trial. Early results show that participants are generally able to understand the bandit problem task. While not performing as well as an optimal model, the best subjects display a wide variety of general strategies in their progress through the task. Discussions of current and future study goals are also provided