Bayesian nonparametric models characterize instantaneous strategies in a competitive dynamic game

Kelsey R. McDonald, William F. Broderick, Scott A. Huettel, John M. Pearson

Abstract: Previous studies of strategic social interaction in game theory have predominantly used games with clearly-defined turns and limited choices. Yet, most real-world social behaviors involve dynamic, coevolving decisions by interacting agents, which poses challenges for creating tractable models of behavior. Here, using a game in which humans competed against both real and artificial opponents, we show that it is possible to quantify the instantaneous dynamic coupling between agents. Adopting a reinforcement learning approach, we use Gaussian Processes to model the policy and value functions of participants as a function of both game state and opponent identity. We found that higher-scoring participants timed their final change in direction to moments when the opponent’s counter-strategy was weaker, while lower-scoring participants less precisely timed their final moves. This approach offers a natural set of metrics for facilitating analysis at multiple timescales and suggests new classes of experimental paradigms for assessing behavior. Game theory typically models strategic human behavior using scenarios with decision constraints that poorly represent real-world social interactions. Here, the authors show it is possible to model dynamic, real-world strategic interactions using Bayesian and reinforcement learning principles.

External IDs:doi:10.1038/s41467-019-09789-4