Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety

2026-06-03 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Cybersecurity & Data Privacy · Depth: Expert, quick

Summary

A new standards-derived rubric addresses the "evidence-type gap" in autonomous driving safety, where popular XAI methods like SHAP fail to provide the directed cause-and-effect evidence mandated by standards such as AMLAS, ISO 26262, ISO21448, and ISO/PAS 8800. Researchers derived 19 testable evidentiary criteria across 7 lifecycle stages from these standards and structurally scored six XAI method classes. Causal XAI emerged as structurally required for hazard identification (+62% rubric gap), incident investigation (+50%), and data management (+50%), while correlational or language-based methods suffice for the remaining four stages. A single-VLA proof of concept, using 1,996 real-world driving clips (79,840 rows, ten splits), consistently matched observed output types with rubric predictions. This work highlights that XAI method selection for Autonomous Driving Systems (ADS) safety assurance should prioritize lifecycle-stage evidence demand over method popularity, noting that validating output fidelity remains an open assurance challenge.

Key takeaway

For Robotics Engineers developing autonomous driving systems, your XAI method selection must move beyond popularity. If you are working on hazard identification, incident investigation, or data management, you should prioritize Causal XAI techniques to meet regulatory evidence demands for directed cause-and-effect chains. Relying solely on correlational methods like SHAP for these critical stages will likely create an evidence-type gap, hindering compliance and assurance efforts.

Key insights

Autonomous driving safety standards demand specific evidence types that popular XAI methods often cannot provide, necessitating a shift to causal XAI for critical stages.

Principles

• XAI method selection must align with evidence demand.
• Causal XAI is critical for specific safety lifecycle stages.
• Structural admissibility is necessary, not sufficient, for compliance.

Method

Derive 19 evidentiary criteria from safety standards (AMLAS, ISO 26262, ISO21448, ISO/PAS 8800) across 7 lifecycle stages. Score XAI method classes structurally against these criteria to identify required output types.

In practice

• Prioritize Causal XAI for hazard identification.
• Use Causal XAI for incident investigation.
• Apply Causal XAI in data management.

Topics

• Autonomous Driving Systems
• XAI Admissibility
• Safety Standards
• Causal XAI
• ISO 26262
• Evidence-Type Gap

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Robotics Engineer, AI Security Engineer

Related on AIssential

• Output Type Before Quality: A Standards-Derived XAI Admissibility Rubric for Autonomous-Driving Safety — Autonomous driving safety standards demand causal evidence that many popular XAI methods cannot structurally provide.
• Towards Rigorous Explainability by Feature Attribution — Existing SHAP scores for ML explainability are flawed and misleading; rigorous alternatives are necessary.
• People need to understand why systems make decisions, not merely whether the systems appear to perform well. A map of the Explainable AI (xAI) landscape. — Explainability is now a prerequisite for trustworthy AI, demanding tailored methods for diverse contexts.
• Risk Assessment of Autonomous Driving: Integrating Technical Failures, Ethical Dilemmas, and Policy Frameworks — Autonomous driving risks require integrated governance addressing technical failures, ethical dilemmas, and regulatory inconsistencies.
• I Built an RSI for My RSI — Recursive self-improvement systems can encounter specification gaming, requiring robust validation methods.
• Physical AI in the Digital Omnibus – Update - Taylor Wessing — The EU's Digital Omnibus streamlines AI regulation for physical machinery, shifting focus to sector-specific product safety.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.