Rule of Thumb: Explaining Artificial Intelligence Systems using Partial Information

RoT enables "black box" auditing without special access: The method can explain AI system decisions without needing access to the model's internal workings, making it possible to audit opaque commercial AI systems—a key requirement for responsible AI governance and regulatory compliance.

Works across diverse AI applications: Rule of Thumb explanations function effectively for large language models, scientific AI systems, and other opaque AI tools, suggesting broad applicability for understanding AI behavior across different domains and use cases.

Identifies the most relevant features for individual decisions: Rather than generic explanations, RoT pinpoints which specific information the AI actually used to make a particular prediction, providing more precise and actionable insights into how individual decisions were made.

Faster and more practical than existing alternatives: The method is substantially faster than competing XAI approaches while remaining model-agnostic, making it more feasible for real-world deployment in organizations seeking to explain their AI systems at scale.

Aligns with regulatory requirements: The approach meets specific requirements from leading AI regulations, helping organizations demonstrate compliance with emerging rules around AI transparency and explainability (such as the EU AI Act).