Making Things Explainable vs Explaining: Requirements and Challenges under the GDPR

One-size-fits-all explanations don't meet legal and user needs: Current explainable AI (XAI) approaches produce generic explanations that fail to address the GDPR's requirement for user-centered explanations of automated decision-making systems, meaning the same explanation is given to everyone regardless of their specific needs or background.

Making something explainable is not the same as explaining it: The paper distinguishes between "explainability" (the technical capacity to explain a system) and "explanations" (actual communication tailored to specific users), arguing that the XAI field has conflated these concepts and focused too much on the former while neglecting the latter.

Interactive, narrative-based explanations are needed: Rather than static technical explanations, the authors propose "explanatory AI" (YAI) that presents information as interactive narratives or "explanatory discourses," allowing users to explore different explanation paths and find what works best for their individual circumstances and understanding.

GDPR compliance requires rethinking AI transparency: The EU's regulatory framework through the GDPR and AI-HLEG guidance demands that AI systems be explainable to affected individuals, which cannot be achieved through current one-size-fits-all technical approaches and requires fundamentally redesigning how explanations are generated and delivered.

Users need agency in understanding decisions affecting them: The paper emphasizes that people affected by automated decisions should be able to interactively explore explanations suited to their needs, rather than passively receiving predetermined technical explanations.