A three-Level Framework for LLM-Enhanced eXplainable AI: From technical explanations to natural language

Different audiences need different explanations: The paper demonstrates that developers, domain experts, end-users, and the general public have distinct needs when understanding AI systems. A one-size-fits-all explanation approach fails to build trust across these diverse groups, requiring instead a three-level framework that tailors explanations to each stakeholder's specific concerns and expertise.

Large Language Models can make AI more understandable: The research shows that LLMs serve as effective "translators" that convert complex technical AI explanations into natural language narratives that non-experts can grasp. This capability enables interactive, conversational explanations rather than static technical documentation, making AI systems more accessible to broader audiences.

Explainability is essential for trust in high-stakes applications: The paper emphasizes that as AI systems are increasingly deployed in sensitive domains (healthcare, criminal justice, finance), accuracy alone is insufficient—systems must also be explainable and trustworthy. The framework positions explanation design as a core component of responsible AI implementation, not an afterthought.

Explanations must address ethical and social concerns, not just technical ones: Beyond explaining how an AI system works, the framework includes a "social explainability" layer that addresses broader questions about fairness, accountability, and societal impact. This reflects the reality that stakeholders care about whether AI decisions are ethical and just, not merely how they were computed.

Interactive dialogue improves AI transparency: The paper's case studies demonstrate that conversational, dynamic explanations—enabled by LLM capabilities—enhance both user engagement and understanding compared to static explanations, creating opportunities for stakeholders to ask follow-up questions and clarify concerns in real time.