Position: Explainable AI is Causality in Disguise

The XAI field lacks consensus because it's missing a fundamental anchor: The paper argues that the persistent confusion in Explainable AI—including conflicting metrics, failed sanity checks, and unresolved debates—stems from the field trying to define "correct explanations" without grounding them in causal models of how systems actually work.

Causality is the missing ground truth for AI explanations: Rather than explanations being inherently subjective, the authors prove that causal models are both necessary and sufficient for XAI. This means the "correct" explanation for an AI decision is fundamentally a causal one—understanding what actually caused the model to make that decision.

Current XAI methods are incomplete without causal reasoning: Many popular explanation techniques (like feature importance scores or attention visualizations) can appear to work but lack causal grounding, making them potentially misleading. Without understanding the causal relationships in a system, explanations remain "unmoored" and unreliable.

The field should prioritize causal discovery methods: The authors recommend the AI community shift focus toward "advanced concept and causal discovery" as a way to resolve the fragmentation and uncertainty currently plaguing XAI research and practice.

This reframing has practical implications for responsible AI: If explanations must be causal to be valid, then organizations deploying AI systems need to invest in understanding the causal mechanisms behind their models' decisions—not just surface-level correlations—to ensure truly trustworthy and accountable AI.