Study Questions Link Between XAI Explanation Correctness and Human Understanding

A recent investigation published on arXiv (2603.25251) disputes the prevalent belief in Explainable AI (XAI) that enhanced functional correctness of explanations equates to improved human comprehension. In their user study involving 200 participants, researchers varied explanation accuracy across four levels (100%, 85%, 70%, 55%) within a synthetic time series classification task. Participants were not permitted to use domain knowledge, with correctness measured against a known ground truth. Findings indicated that while correctness influences understanding, the correlation is complex; accuracy in forward simulation declined in a non-linear fashion. This study raises doubts about the reliability of functional correctness metrics as indicators of human interpretability, emphasizing the need for user studies to assess explanation efficacy.