Mixtape Middleware for Interactive XAI with Tree-based AI Performance Evaluation

Kitware, in partnership with Penn State University, is proud to propose this Phase II STTR effort to create explainable AI (XAI) middleware that can support the interactive explanation and visualization of AI decision-making systems. Our approach, MIXTAPE, will create modular and extensible software interfaces and implementations of different XAI tools for evaluating the performance of AIs in Multi-Domain Operations. For explanations, we will extend the tree-based visualizations we developed in Phase I to show perceived current states and anticipated future states of units at critical decision points. We will also develop novel explanations using neurosymbolic AI, decomposing complex reasoning strategies into more interpretable states representing agent sub-goals. To create a structured process that guides humans on how to use the provided explanations, we will extend our work on After-Action Review for AI. Our proposed tool will be validated by a set of user studies, with assessment along the axes of performance, usability, and appropriateness of trust. Additional forms of explanation and geospatial visualization can also be incorporated given the flexible nature of our framework. We will test our approach on the ARL Simple Yeho environment, but it can apply to different AI agents, wargaming scenarios, and AI testbeds.