Explanations as Dialogues: Toward Human-Centered Conversational Explainable AI

Explanations work differently in conversation than in static formats. The paper argues that how AI explanations are delivered through dialogue—including timing, tone, and conversational context—fundamentally changes how well people understand them, yet most AI research treats explanations as fixed documents rather than interactive exchanges.

Conversational elements matter as much as content. Beyond just what an AI explains, the how of explanation (persona, conversational history, and tone) are "critical constituents" of effectiveness, suggesting that responsible AI systems need to be designed with dialogue dynamics in mind, not just accurate information.

Current AI explanation research has a significant gap. The paper identifies that while AI systems are increasingly conversational (like chatbots), the study of explanations hasn't caught up—researchers still primarily focus on static explanations rather than how people actually experience them through back-and-forth conversation.

Human-centered design requires rethinking explainability. The authors propose "Human-Centered Conversational XAI" as a framework, implying that truly responsible AI needs to move beyond one-way explanations and instead design for interactive dialogue where users can ask follow-up questions and clarify misunderstandings in real time.