Model curiosity questions boost deeper user exploration and insight
What Makes a Question Curious? Measuring Model Curiosity in Human-LLM Interactions
Computation and LanguageArtificial Intelligence
Summary
People increasingly use large language models (LLMs) not just to get answers but as conversation partners who ask questions too. The authors studied a specific kind of model question called curiosity questions, which aim to explore unknowns or assumptions rather than just clarify. They found these curious questions are rare but lead to richer and more exploratory user responses. Their study also showed these questions are helpful especially when users seek advice, creative input, or research help.
What this means in practice
- •For virtual assistant developers: Incorporate curiosity-driven questions in assistants to prompt users for more insightful and exploratory conversations in advice and creative tasks.
- •For customer support teams: Design support chatbots to ask targeted curiosity questions to uncover user needs and assumptions for better problem resolution.
Authors
Angana Borah, Rachana Ramchandar, Inderjeet Nair, Rada Mihalcea
Abstract
Large language models increasingly act as partners in everyday communication. In these interactions, useful assistance may require not only answering, but also asking the right questions that surface knowledge gaps, assumptions, and alternatives. In this study, we refer to such epistemic question-asking as model curiosity, and introduce a theory-grounded taxonomy that distinguishes curiosity-oriented questions from generic model questioning. Across four conversation datasets, we find that curiosity-oriented questions remain rare (17-28% of all questions), but are associated with more exploratory and epistemically rich subsequent user responses, with varying patterns across datasets. We further conduct a human study to evaluate the usefulness of model curiosity, finding that curiosity can encourage exploration, especially in advice, creative, and research contexts. These findings have implications for building LLMs that ask more selectively to support deeper exploration and human-AI collaboration.