Refutation of Shapley Values for XAI -- Additional Evidence

TL;DR

This paper provides additional evidence that Shapley values are inadequate for explainable AI, extending prior criticisms to non-Boolean and multi-class classifiers, and analyzing feature relevance in adversarial examples.

Contribution

It demonstrates the limitations of Shapley values across broader classifier families and in the context of adversarial examples, challenging their effectiveness in XAI.

Findings

Shapley values are inadequate for non-Boolean classifiers.

Shapley values fail for multi-class classifier explanations.

Features in adversarial examples do not include irrelevant features.

Abstract

Recent work demonstrated the inadequacy of Shapley values for explainable artificial intelligence (XAI). Although to disprove a theory a single counterexample suffices, a possible criticism of earlier work is that the focus was solely on Boolean classifiers. To address such possible criticism, this paper demonstrates the inadequacy of Shapley values for families of classifiers where features are not boolean, but also for families of classifiers for which multiple classes can be picked. Furthermore, the paper shows that the features changed in any minimal $l_0$ distance adversarial examples do not include irrelevant features, thus offering further arguments regarding the inadequacy of Shapley values for XAI.