May I Ask a Follow-up Question? Understanding the Benefits of Conversations in Neural Network Explainability

Conversations significantly improve how people understand AI explanations: The study found that when users could have free-form conversations with a human expert about static explanations, their comprehension improved substantially compared to reading explanations alone.

Interactive dialogue builds trust in AI systems: Beyond just understanding, participants who engaged in conversations reported higher levels of acceptance and trust in the explanation methods themselves, suggesting that the ability to ask follow-up questions addresses user concerns more effectively than one-time explanations.

One-size-fits-all explanations are insufficient: The research demonstrates that static, standardized explanations cannot adequately serve users with different backgrounds and knowledge levels, highlighting a key limitation of current explainability approaches in responsible AI.

Conversational AI explanations enable better human-AI collaboration: The study measured participants' ability to select the most accurate machine learning model based on explanations, and found that conversations facilitated this collaborative decision-making task more effectively than static explanations alone.

Future AI systems should prioritize interactive explanation formats: The findings suggest that responsible AI development should move beyond providing fixed explanations and instead design systems that allow users to engage in dialogue about how and why AI systems make decisions.