A Learning-Based Method for Automatic Operator Selection in the Fanoos XAI System

Fanoos is an explainability system that can adjust how detailed its explanations are: The system can make descriptions of AI systems more or less abstract based on user requests, allowing people to get explanations at the right level of detail for their needs.

The system now learns from experience rather than relying on manual rules: Instead of requiring engineers to hand-code which explanation techniques to use, Fanoos can now learn which approaches work best in different situations, making the system more adaptive and potentially easier to maintain.

Simulated users help bootstrap the learning process: The researchers created artificial users to help Fanoos learn general principles before interacting with real people, similar to how students practice with textbooks before real-world application—this approach could reduce the burden on actual users during system development.

The system balances exploration with practical constraints: Fanoos learns which explanation operators to apply while considering both trying new approaches and using proven methods, plus it can incorporate expert knowledge when available—an important consideration for responsible AI systems that need to be both innovative and reliable.

This work addresses a practical challenge in explainable AI: By automating operator selection, the system becomes more scalable and user-responsive, which supports the broader goal of making AI systems more transparent and understandable to non-technical users.