Sifter: Explainable AI for Improved Multi-Domain Operation Wargaming

R-DEX Systems (R-DEX) proposes to team with its strategic partner to develop Sifter, a unique and innovative technology for explainable AI (XAI). This proposed STTR leverages the team’s unique experience in deriving representations of reinforcement learning policies and the team’s extensive work in machine learning quality assurance (MLQA). The proposed Sifter will provide multiple levels of explanation including the higher-level strategy as well as the lower-level tactical behavior of the policy network. The sum total of this project will be a suite of tools that provides many different views of XAI. All of these views provide unique insights into the policy’s behavior, allowing human users to select the views that align with the aspects of explainability they wish to prioritize.