Issues with post-hoc counterfactual explanations: a discussion
This work addresses reliability issues in interpretability methods for machine learning practitioners, but it is incremental as it critiques existing approaches without introducing new solutions.
The paper discusses the limitations of post-hoc counterfactual explanations for blackbox classifiers, highlighting that they can be unreliable due to assumptions about data and classifiers, and it identifies three properties—proximity, connectedness, and stability—that these approaches may fail to satisfy.
Counterfactual post-hoc interpretability approaches have been proven to be useful tools to generate explanations for the predictions of a trained blackbox classifier. However, the assumptions they make about the data and the classifier make them unreliable in many contexts. In this paper, we discuss three desirable properties and approaches to quantify them: proximity, connectedness and stability. In addition, we illustrate that there is a risk for post-hoc counterfactual approaches to not satisfy these properties.