Bridging the Transparency Gap: What Can Explainable AI Learn From the AI Act?

Transparency means different things to regulators and AI researchers: The EU AI Act views transparency as a tool to support accountability and human rights, while the explainable AI (XAI) field focuses narrowly on explaining how algorithms work. This "transparency gap" means XAI solutions may technically comply with regulations without actually addressing the broader societal concerns regulators intended to protect.

Current XAI approaches ignore socio-technical context: The paper finds that XAI typically focuses on algorithmic properties in isolation, without considering the real-world social and technical systems where AI operates. This narrow focus risks leaving important transparency issues unaddressed that regulators and affected communities actually care about.

Four practical areas need development to bridge the gap: The authors identify specific areas where progress is needed: (1) defining what scope of transparency is actually required, (2) clarifying the legal status of XAI tools under regulations like the AI Act and GDPR, (3) improving how conformity with transparency requirements is assessed, and (4) developing explainability methods that address datasets, not just algorithms.

Regulatory requirements can guide XAI research priorities: Rather than viewing regulation as a constraint, the paper suggests the EU AI Act and GDPR provide valuable guidance about what transparency actually needs to accomplish in practice—moving XAI beyond technical explanations toward supporting accountability and protecting human rights.