The Dangers of Post-hoc Interpretability: Unjustified Counterfactual Explanations
This work addresses a critical issue for practitioners and researchers in machine learning who rely on interpretability tools, highlighting a significant flaw in current methods that could lead to misleading explanations.
The paper tackles the problem of post-hoc interpretability methods generating unjustified counterfactual explanations that may stem from model artifacts rather than true data knowledge, and finds that this risk is high across several datasets, with most state-of-the-art approaches failing to distinguish justified from unjustified examples.
Post-hoc interpretability approaches have been proven to be powerful tools to generate explanations for the predictions made by a trained black-box model. However, they create the risk of having explanations that are a result of some artifacts learned by the model instead of actual knowledge from the data. This paper focuses on the case of counterfactual explanations and asks whether the generated instances can be justified, i.e. continuously connected to some ground-truth data. We evaluate the risk of generating unjustified counterfactual examples by investigating the local neighborhoods of instances whose predictions are to be explained and show that this risk is quite high for several datasets. Furthermore, we show that most state of the art approaches do not differentiate justified from unjustified counterfactual examples, leading to less useful explanations.