Asymmetric Shapley values: incorporating causal knowledge into model-agnostic explainability
This addresses the problem of enhancing AI explainability and fairness for developers and users by providing a more flexible and rigorous method, though it is incremental as it builds on the existing Shapley framework.
The paper tackled the limitation of Shapley values in ignoring causal data structure by introducing Asymmetric Shapley values (ASVs), a model-agnostic framework that incorporates causal knowledge to improve explanations, test for unfair discrimination, enable time-series explanations, and support feature selection without retraining.
Explaining AI systems is fundamental both to the development of high performing models and to the trust placed in them by their users. The Shapley framework for explainability has strength in its general applicability combined with its precise, rigorous foundation: it provides a common, model-agnostic language for AI explainability and uniquely satisfies a set of intuitive mathematical axioms. However, Shapley values are too restrictive in one significant regard: they ignore all causal structure in the data. We introduce a less restrictive framework, Asymmetric Shapley values (ASVs), which are rigorously founded on a set of axioms, applicable to any AI system, and flexible enough to incorporate any causal structure known to be respected by the data. We demonstrate that ASVs can (i) improve model explanations by incorporating causal information, (ii) provide an unambiguous test for unfair discrimination in model predictions, (iii) enable sequentially incremental explanations in time-series models, and (iv) support feature-selection studies without the need for model retraining.