Most General Explanations of Tree Ensembles (Extended Version)

Explanations work better when they apply broadly: The paper demonstrates that "most general" explanations—which describe ranges of input values rather than single specific cases—are more useful for understanding AI decisions. For example, instead of explaining why a specific loan application was rejected based on exact numbers, a general explanation might say "any applicant with income between $X-$Y and debt ratio above Z would receive the same decision," making the logic clearer and more applicable to similar situations.

Tree ensemble models can now be explained more rigorously: The research provides a formal method to find the single best explanation for decisions made by tree-based AI systems (like Random Forests), which are widely used in banking, healthcare, and hiring. This addresses a practical problem: when multiple valid explanations exist for why a model made a decision, the method identifies which one covers the most ground and is therefore most trustworthy.

Broader explanations increase trust and fairness: By finding explanations that apply to the widest range of inputs possible, this approach helps ensure that similar cases are treated similarly. This is critical for responsible AI, as it makes discrimination easier to detect—if the explanation for rejecting one applicant doesn't logically apply to another similar applicant who was approved, that's a red flag for bias.

One clear explanation is better than many confusing ones: The paper emphasizes that providing a single, most general explanation to humans is preferable to offering multiple possible explanations. This practical insight recognizes that decision-makers and affected individuals need clarity, not complexity, when understanding why an AI system made a particular choice about them.