A Grounded Interaction Protocol for Explainable Artificial Intelligence

Explanation requires structured dialogue: The research analyzed 398 explanation dialogues and found that successful AI explanations follow identifiable patterns and components, suggesting that effective XAI systems need to be designed with specific interaction structures rather than providing explanations in ad-hoc ways.

Both cognitive and social processes matter: The paper emphasizes that explaining AI decisions isn't purely about conveying information—it involves social interaction dynamics. This means responsible AI systems must account for how humans actually process and respond to explanations in conversation, not just what information is technically accurate.

A formal protocol improves human-AI interaction: The researchers developed an interaction protocol based on real explanation dialogues and validated it with 101 new dialogues from 14 participants, demonstrating that structured protocols can help AI systems communicate more naturally and effectively with users.

Real dialogue analysis grounds better design: By using grounded theory to analyze actual human explanation conversations (bottom-up approach), rather than imposing predetermined structures, the researchers identified what genuinely works in practice—suggesting that responsible AI design should be informed by how people actually interact, not just theoretical models.

Validation with real users is essential: The study's human-agent interaction evaluation shows the importance of testing XAI systems with actual participants to ensure explanations are meaningful in practice, not just theoretically sound.