Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety
A new rubric for Explainable AI (XAI) in autonomous driving safety has been proposed, highlighting a significant gap between current XAI methods and the evidence required by safety standards. The proposed rubric, derived from automotive safety standards like ISO 26262, identifies that causal XAI methods are structurally necessary for critical stages such as hazard identification and incident investigation. The research suggests that XAI method selection should prioritize the evidence demands of specific lifecycle stages rather than relying on method popularity. AI
IMPACT Highlights the need for causal XAI methods to meet stringent safety standards in autonomous driving, potentially guiding future development and validation.