GLARE Offers Natural Language Access to AI Model Explanations

This research presents GLARE, a new interactive system designed to make global explanations for vision models more accessible. Global explanations, while vital for understanding how AI models make decisions across various contexts, are often complex and difficult for users to explore effectively. GLARE addresses this by providing a natural language interface, allowing users to ask specific questions about a black-box image classifier's behavior. At its core, GLARE utilizes a Large Language Model (LLM) to interpret user questions and convert them into structured SQL queries. These queries then operate on underlying local explanation data, enabling flexible data aggregation without requiring users to understand low-level technical details. The system responds with clear, statistics-enhanced natural language answers and relevant visualizations, significantly improving the usability of explainable AI (XAI) for human users.