Leveraging Generative AI for Human Understanding: Meta-Requirements and Design Principles for Explanatory AI as a new Paradigm

Current AI explanations prioritize showing how algorithms work rather than helping practitioners understand why recommendations make sense in their specific context. The paper distinguishes between traditional explainable AI (XAI), which focuses on algorithmic transparency for validation purposes, and Explanatory AI, which integrates domain knowledge and professional frameworks to support actual decision-making by end users.

Explanatory AI uses generative capabilities to create dynamic, narrative-based explanations rather than static technical outputs. Instead of presenting computational processes, this approach generates contextually relevant explanations that practitioners can interact with, similar to having an explanatory partner who understands both the AI's reasoning and the professional domain.

Effective AI explanations must draw from cognitive science, communication research, and education principles—not just computer science. The research synthesizes multidisciplinary perspectives to identify that human understanding requires explanations addressing "Why does this make sense?" rather than purely "How did the algorithm decide?"

Healthcare contexts reveal a critical gap between how AI systems currently explain themselves and how medical professionals actually need to understand recommendations. Empirical evidence from healthcare and interviews with AI experts demonstrate that practitioners require explanations incorporating their professional judgment and contextual reasoning, not just algorithmic validation.

Five key dimensions distinguish Explanatory AI from traditional XAI: purpose (diagnostic vs. interpretive), communication mode (static vs. dynamic), epistemic stance (algorithmic vs. contextual), and others that collectively shift focus from system validation to human decision-making support.