Poster D35, Monday, March 27, 8:00 – 10:00 am, Pacific Concourse
Towards Optimal Competitive Behavior: Wins versus Losses Determine Model-based versus Random Choices in Competitive Task Switching
Atsushi Kikumoto1, Caitlin Corona1, Joshua Karpf1, Ulrich Mayr1; 1University of Oregon
In mixed strategy games, probability for equal-utility options should converge towards maximally unpredictable choices. Moving towards such optimal competitive behavior requires negotiating two conflicting goals: (1) being unpredictable for the opponent and (2) exploiting the opponent’s deviations from randomness by relying on a model of the opponent. We investigated how this balance is achieved using a voluntary task-switching paradigm, where typically (i.e., in a non-competitive context) strong deviations from randomness in form of a task-perseveration bias is observed. Across several experiments, participants played a task-switching version of the matching pennies task against human or computerized opponents that varied—either naturally or experimenter controlled—in switch probability. Generally, we found that task choices were much less predicable than in a non-competitive context and in particular showed only very small perseveratory tendencies. More importantly, after win trials participants exhibited task choices that indicated model-based adaptations to the opponents’ local and overall switch rate. In contrast, after loss trials choice behavior was close to random. In a final experiment, we probed the representations about the opponent using EEG. In multi-level analyses of the single-trial EEG signal, we found that after win feedback, signals at mid-central electrodes robustly represented information about the local and global strategy of the opponent. After losses, this information was nearly absent. Overall, these results suggest that win versus loss feedback triggers model-based or “randomizing” behavior, providing a process-level explanation to the question how people move towards optimal competitive behavior.
Topic Area: EXECUTIVE PROCESSES: Goal maintenance & switching