Faithfulness Without Anchors: A Position Paper on the Structural Misalignments of Explainable AI Faithfulness Metrics for Feature Importance

Explain Like I'm Normal

Researchers argue that Explainable AI (XAI) metrics are structurally flawed and often fail to represent how models actually make decisions. Instead of building trust, these oversimplified 'explanations' can lead to overreliance or total distrust, functioning more as anecdotal debugging tools than true windows into the model's logic. The paper calls for a fundamental rethink of how we measure if an explanation is actually faithful to the math.