White-box Induction From SVM Models: Explainable AI with Logic Programming

Logic programs can make "black box" AI models interpretable: The paper demonstrates that complex SVM models can be translated into human-readable logic rules (if-then statements), making it possible for non-experts to understand why an AI system made a particular decision.

Focusing on the most important data points improves explanations: Rather than analyzing all training data, the new approach identifies "support vectors"—the data points that most influenced the model's decisions—resulting in simpler, more focused explanations with fewer rules needed.

AI explanation tools like SHAP can guide which features matter: By using SHAP (a specific explainability technique), the algorithm automatically identifies which input features are most relevant to explain the model, rather than requiring humans to manually decide what to look at.

This method outperforms existing explanation techniques: The approach generated fewer logical rules while maintaining or improving classification accuracy compared to FOIL, a traditional algorithm for extracting rules from models, suggesting it produces more efficient and practical explanations.

Better explanations support responsible AI deployment: By making SVM models' decision-making logic transparent and understandable, this work helps organizations audit AI systems for bias, errors, or unfair patterns before deploying them in high-stakes applications like hiring, lending, or criminal justice.