Towards a Grounded Dialog Model for Explainable Artificial Intelligence

Explanations require both cognitive and social elements: The paper emphasizes that effective AI explanations aren't just about conveying technical information—they involve social interaction processes, meaning AI systems need to be designed to engage in meaningful dialogue rather than simply delivering facts.

A structured dialog model can improve AI transparency: By analyzing 398 real explanation conversations using grounded theory, the researchers identified key components and patterns in how humans naturally explain things. This bottom-up approach resulted in a generalized state model that can guide how AI systems should structure explanations to users.

Meaningful interaction between explainer and explainee matters: The paper highlights that successful explanation depends on the quality of back-and-forth interaction—not one-way communication. This suggests responsible AI systems should be designed to allow users to ask follow-up questions and engage in dialogue rather than receiving static explanations.

The research provides a practical framework for XAI design: By comparing their derived model with existing conceptual frameworks, the authors offer designers of explainable AI systems a grounded, evidence-based approach to structuring how AI communicates its decisions to users, potentially making AI systems more trustworthy and understandable.