The Explanation Game -- Rekindled (Extended Version)

2025-01-20

AI Summary

Existing SHAP values suffer from theoretical deficiencies in explainable AI (XAI), violating causal consistency and explanation fidelity—thereby misleading decision-makers. Method: This work first systematically identifies the root cause: semantic inconsistency in the conventional SHAP formulation. We propose a novel, semantically rigorous, and causally grounded SHAP definition; reconstruct the Shapley value’s semantic foundation via cooperative game theory; design a hierarchical feature perturbation scheme; and integrate Monte Carlo variance reduction for efficient, high-precision estimation. Contribution/Results: Experiments demonstrate substantial improvements in attribution accuracy and alignment with human judgment. Our method explicitly exposes and corrects systematic biases of standard SHAP across critical scenarios—including model-agnostic benchmarks—establishing a more reliable, trustworthy foundation for XAI-based attribution.

Technology Category

• Humans and AI: Explainable AI (XAI) for Human Understanding
• Reasoning under Uncertainty: Causality
• Machine Learning: Causal Learning

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Recent work demonstrated the existence of critical flaws in the current use of Shapley values in explainable AI (XAI), i.e. the so-called SHAP scores. These flaws are significant in that the scores provided to a human decision-maker can be misleading. Although these negative results might appear to indicate that Shapley values ought not be used in XAI, this paper argues otherwise. Concretely, this paper proposes a novel definition of SHAP scores that overcomes existing flaws. Furthermore, the paper outlines a practically efficient solution for the rigorous estimation of the novel SHAP scores. Preliminary experimental results confirm our claims, and further underscore the flaws of the current SHAP scores.

Problem

• Shapley Values
• Artificial Intelligence Explanation
• XAI Misguidance

Innovation

• SHAP Score Refinement
• Computational Methodology
• Enhanced XAI Accuracy