Challenges in Evaluating Explanation Methods for Static and Evolving Data

2026-08-06

AI Summary

This work addresses the insufficient evaluation of existing explainable artificial intelligence (XAI) methods in dynamic data environments, particularly their neglect of concept drift and human-AI collaboration. Focusing on the DetoxAI image recognition system, the study proposes a dynamic explanation adaptation framework tailored for evolving data streams, which jointly models the co-evolution of data, models, and explanations. It introduces a human-centered evaluation mechanism grounded in counterfactual reasoning, integrating counterfactual explanations, human-grounded assessment, and concept drift detection. The research exposes critical limitations of current XAI approaches in dynamic settings, demonstrates the adaptive potential of counterfactual explanations under concept drift, and establishes a novel paradigm for evaluating XAI systems in evolving real-world scenarios.

Abstract

This paper addresses the limitations of Explainable Artificial Intelligence (XAI) with respect to insufficient evaluation. They are illustrated through the DetoxAI image recognition system for bias detection and concept unlearning. Then, an example of a human-grounded evaluation of methods for explaining image classification is presented. The paper further explores methods for adapting explanations to evolving data streams with concept drift. Experiences with adapting counterfactuals for this problem are discussed. Finally it is related to the challenges of tracking the co-evolution of data, models, and explanations.\footnote{This paper has been accepted for a publication in J.Nalepa (ed) Explainable AI in Space. Proceedings of EASi 2026 Workshop at IJCAI-ECAI 2026 Bremen, Springer CCIS vol 3107 (2016).}

Problem

• Explainable AI
• evaluation
• concept drift
• evolving data
• co-evolution

Innovation

• Explainable AI
• concept drift
• counterfactual explanations
• human-grounded evaluation
• co-evolution