Causality-Inspired Taxonomy for Explainable Artificial Intelligence

Causality is essential for complete AI explainability: The paper argues that explainable AI (xAI) cannot be fully understood without incorporating causal reasoning—moving beyond simply identifying which features influence a decision to understanding why those features matter and how they causally affect outcomes.

A new framework organizes xAI approaches: The authors propose an "xAI Ladder" that categorizes different explanation methods, providing a structured taxonomy that helps researchers and practitioners understand where different explainability techniques fit and what level of explanation they provide.

Biometric systems reveal xAI gaps: Analysis of 81 papers on biometric technologies (facial recognition, fingerprinting, etc.) shows that many current systems lack causal explanations, highlighting a practical area where better explainability is urgently needed for responsible deployment.

Current xAI methods are incomplete: The paper suggests that many existing explanation techniques only partially address what users need to understand—they may show correlations but not causal relationships, which is critical for building trust and accountability in AI systems used in high-stakes decisions.

Causality framework enables better AI governance: By connecting causality to explainability, this work provides a foundation for developing AI systems that can be more thoroughly audited and understood, supporting efforts to ensure AI systems are fair, transparent, and accountable.