The effectiveness of feature attribution methods and its correlation with automatic evaluation scores
This work addresses a critical gap in explainable AI for high-stakes applications by revealing that current evaluation practices are misaligned with human needs, urging a shift towards more rigorous testing.
The paper tackles the problem of evaluating feature attribution methods for AI explainability by conducting the first user study on human-AI team performance in image classification tasks, finding that attribution maps are not more effective than showing nearest training examples and can even hurt performance in fine-grained tasks, with automatic evaluation metrics correlating poorly with human outcomes.
Explaining the decisions of an Artificial Intelligence (AI) model is increasingly critical in many real-world, high-stake applications. Hundreds of papers have either proposed new feature attribution methods, discussed or harnessed these tools in their work. However, despite humans being the target end-users, most attribution methods were only evaluated on proxy automatic-evaluation metrics (Zhang et al. 2018; Zhou et al. 2016; Petsiuk et al. 2018). In this paper, we conduct the first user study to measure attribution map effectiveness in assisting humans in ImageNet classification and Stanford Dogs fine-grained classification, and when an image is natural or adversarial (i.e., contains adversarial perturbations). Overall, feature attribution is surprisingly not more effective than showing humans nearest training-set examples. On a harder task of fine-grained dog categorization, presenting attribution maps to humans does not help, but instead hurts the performance of human-AI teams compared to AI alone. Importantly, we found automatic attribution-map evaluation measures to correlate poorly with the actual human-AI team performance. Our findings encourage the community to rigorously test their methods on the downstream human-in-the-loop applications and to rethink the existing evaluation metrics.