Demonstration of an Explainable AI System Performing Human-Machine Teaming in a Real-Time Strategy Game

Moody, V., Druce, J., and Niehaus, J.

Educational Advances in Artificial Intelligence 2022 (EAAI-22) (February 2022)

In this paper we demonstrate the use of an Explainable Artificial Intelligence (XAI) system that enables users to team effectively with Artificial Intelligence (AI) agents in a real time strategy (RTS) game. Recent advances in deep reinforcement learning (DRL) approaches have created RTS agents with grandmaster level ability, winning matches against top human players. This performance on the complex RTS task enables new approaches addressing how users and domain experts can adopt AI agents as tools, determine best contexts in which to apply them, and delegate tasking appropriately to work alongside them in mixed-initiative scenarios. We present (1) an architecture for a generalizable XAI system, including model learning and analysis and user mental model training; (2) a working XAI user interface; and (3) example explanations generated by the system. Our demonstration features a playable system.

To learn more or request a copy of a paper (if available), contact Vanessa Moody.

(Please include your name, address, organization, and the paper reference. Requests without this information will not be honored.)