Explainable Artificial Intelligence (XAI)
Abstract
The Explainable Artificial Intelligence (XAI) program is developing a new generation of machine learning techniques that are able to explain their rationale, characterize their strengths and weaknesses, and convey an understanding of how they will behave in the future. AI is a critical enabler for U.S. military systems that will perform increasingly complex and sensitive missions. However, in order for developers, users, and senior leaders to feel confident enough to deploy and use AI-enabled systems, these systems must be able to explain their rationale, and their recommendations, decisions, and actions must be delivered in a way that military users can understand and trust. Today, most machine learning systems provide no explanations, or provide explanations that are at the wrong level of abstraction, not meaningful to a human user, or inconsistent with the full range of behaviors of the AI system. XAI is developing the tools necessary to build explainable AI systems, specifically with: (1) new machine learning techniques that produce human-interpretable models and (2) user interfaces that generate explanations from those models that are meaningful to end-users, using natural language, saliency maps, and other representations. XAI implementations will be developed and demonstrated in next-generation data analytics and autonomous systems.
Document Details
- Document Type
- Accomplishment
- Publication Date
- Oct 01, 2023
- Source ID
- 8c320d1ab32064a1457b59c458b970c8