Explainable Artificial Intelligence (XAI) for Air Traffic Management

Explainable AI (XAI) AI in which humans can understand the decisions or predictions made by the AI system is becoming critical for many applications, and in particular, for the aviation domain. However, Machine Learning (ML) systems too often resemble black boxes that mysteriously convert incoming data into predicted outcomes and recommended decisions. This SBIR effort identifies a way to add explainability to AI and ML systems exploiting an XAI vocabulary that is unique to the Air Traffic Management (ATM) domain. The solutionleveragesrules, regulations, and policies that shape ATM activities, that are inherently learned within AI and ML systems. Given large quantities of historical 4-Dimensional (4D) Trajectory-Based Operations (TBO) data for training AI and ML systems, Intent Inference Learning (IIL) is used to establish XAI word labels on each historical data point. These labeled historical data points are used for the training of AI and ML systems, as well as for the explanations of the results of the AI or ML system. Explainable AI (XAI) – AI in which humans can understand the decisions or predictions made by the AI system – is becoming critical for many applications, and in particular, for the aviation domain. However, Machine Learning (ML) systems too often resemble “black boxes” that mysteriously convert incoming data into predicted outcomes and recommended decisions. This SBIR effort identifies a way to add explainability to AI and ML systems exploiting an XAI language that is unique to the Air Traffic Management (ATM) domain, leveraging rules, regulations, and policies that shape ATM activities, that are inherently learned within AI and ML systems. Perform Foundational Research - establish a vocabulary for XAI explanations based on aviation rules, regulations, and policies as well as basic statistical properties Demonstrate Concepts - leveraging Commercial Off The Shelf (COTS) Machine Learning (ML) algorithms, we demonstrate how XAI labels can be used to explain what the ML algorithms are "thinking"
Benefits: In-Time System-Wide Safety Assurance (ISSA) objectives Support intelligent “labeling” of large quantities of aviation data for NASA’s Digital Information Platform Support the needs of the Sky for All and Next projects at NASA Use of XAI in the test and evaluation of explainable ML and AI algorithms Applications: FAA ATCSCC Command Center FAA ANG-C, pursuing natural language processing and other AI techniques for ATM FAA AJR – SysOps systems performance analysis Air Navigation Service Providers (ANSP) around the world Airlines to improve post operations analysis with supporting explanations