无锚点的忠实度：论 XAI 特征重要性忠实度度量的结构性错位
原标题：Faithfulness Without Anchors: A Position Paper on the Structural Misalignments of Explainable AI Faithfulness Metrics for Feature Importance
AI 摘要
这篇立场论文（发表于 IJCAI-ECAI 2026 XAI 研讨会）指出，可解释 AI（XAI）在提升信任方面整体未达预期，反而可能助长用户对错误预测的过度依赖。作者聚焦特征重要性（FI）这一最常见的 XAI 任务，认为现有忠实度（faithfulness）度量存在结构性错位，例如依赖迭代特征移除、受分布外数据影响、需要任意基线或重新训练而改变原模型性质。论文归纳出当前忠实度度量的五大缺陷，并提出四条研究路径以提升其可靠性。
正文节选
Faithfulness Without Anchors: A Position Paper on the Structural Misalignments of Explainable AI Faithfulness Metrics for Feature ImportanceThanks: Presented at the IJCAI-ECAI 2026 Workshop on Explainable Artificial Intelligence (XAI) Abstract Explainable Artificial Intelligence (XAI) tools have often been presented as methods to increase trust in machine learning models by making aspects of their algorithms more human-understandable. However, evidence accumulated over the last decade points to