Towards consistency of rule-based explainer and black box model -- fusion of rule induction and XAI-based feature importance

Rule-based explanations may not accurately reflect how AI systems actually make decisions: The paper reveals that when researchers use simple, human-readable rules to explain complex AI models, these rules often don't match the actual decision-making process of the original system—a critical gap for trustworthy AI that hasn't been thoroughly studied before.

A new method improves alignment between explanations and actual AI behavior: The researchers developed an approach that combines rule generation with feature importance analysis to create rule-based explanations that more faithfully reproduce how the black box model operates, tested across 30 different datasets.

Both global and local explanations can be made more consistent: The proposed solution can generate explanations at different levels—either explaining overall patterns in how an AI system works or explaining individual decisions—while maintaining consistency with the underlying model's actual behavior.

Consistency matters for responsible AI deployment: By ensuring that simplified explanations accurately reflect what complex AI systems are actually doing, this work addresses a fundamental requirement for responsible AI: users and regulators need explanations they can trust to represent genuine decision-making processes, not just plausible-sounding approximations.