Abstract by Najma Mathema
Modeling Human Behavior in Repeated Games
Modeling other agents is important for longitudinal Human-Robot Interaction (HRI). Our work improves HRI by forming a model of belief and behavior of humans interacting with algorithms. Our work builds on prior work on repeated games with cheap talk, which enables HRI and human-human interaction. A Bayes Filter has been implemented for modeling human behaviors in repeated games. The filter predicts the next actions for the modeled agent. The overall goal is to enable a relationship narrative of the agents’ interaction, which would improve long-term HRI. On the Prisoners Dilemma, the filter accurately predicted 91.2% of next actions when modeling S# algorithm(robot), and 88.9% when modeling humans. Future work will use the Viterbi algorithm to suggest the most favorable actions for modeling long-term interactions.