cs.CLOct 24, 2026
SaveWhat Makes a Question Curious? Measuring Model Curiosity in Human-LLM Interactions
Organizations: University of Michigan - Ann Arbor, USA · PES University, Bangalore, India
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.
Figures & tables
Figure 1: Overview of questions in human-model interaction. Given the same user prompt, we define a taxonomy to distinguish between the two types: (1) non-curiosity questions support task flow, clarification, or engagement, and (2) curiosity-oriented questions invite evidence-seeking, reflection, and exploration instead of directly answering with assumptions. Note that each category includes additional subtypes, we show three here for illustration.
| Question type | Theoretical grounding | Example |
| Curiosity: Information-gap | Information-gap theory; epistemic curiosity | “What problem does it solve?” |
| Curiosity: Causal-explanatory | Explanation-seeking; epistemic curiosity; explanations as understanding | “Why did this happen?” |
| Curiosity: Exploratory broadening | Epistemic exploration; diversive and specific curiosity | “What else might explain this?” |
| Curiosity: Reflective-metacognitive | Metacognition; epistemic reflection; uncertainty monitoring | “What assumptions are you making?” |
| Curiosity: Perspective-taking | Perspective-taking; role-taking theory; empathy | “How might they interpret this?” |
| Curiosity: Counterfactual-hypothetical | Counterfactual reasoning; hypothetical exploration; epistemic simulation | “What if the context changed?” |
Table 1: Taxonomy of assistant-to-user questions. Curiosity-oriented questions are grounded in theories of information gaps, explanation-seeking, epistemic exploration, metacognition, perspective-taking, and counterfactual reasoning.
Figure 2: Labels across datasets. Curiosity-oriented questions appear less frequently, with WildChat showing the highest share.
Figure 3: Curiosity across datasets
Figure 4: Topic/context analysis of human questions that lead to LLM curiosity.
Figure 5: Matched Analysis. Curiosity questions lead to more engagement and curiosity overall, however it varies by dataset. Estimates are observational and do not establish causal effects.
| Data | AnyQ | ExplQ | Epist. | Words | Judge |
|---|---|---|---|---|---|
| WildChat | +2.0 ∗∗ | +0.053 + | +0.165 + | +11.31 + | +12.1 + |
| OASST1 | 1.9 | 0.008 | +0.085 ∗ | -0.16 | +1.5 |
| LMSYS | -3.5 ∗ | +0.035 + | +0.072 + | +5.90 + | +9.5 + |
| ShareGPT | -1.8 | +0.014 ∗ | +0.084 + | +3.28 + | +13.7 + |
Table 2: Regression and validation estimates for downstream user engagement. AnyQ is the adjusted change in next-turn question asking; ExplQ , Epist. , and Words are adjusted coeff. for exploratory questions, epistemic markers, and response length; and Judge is the matched LLM-judge curiosity estimate. Significance markers denote two-sided tests: , , .
Figure 6: Strict LLM-judged human-curiosity rates in matched next-user turns following curiosity-oriented model questions and clean controls.
Figure 7: Human Study: Participants often prefer directness, and it is low for curiosity questions. However, that differs by subtype.
Figure 8: Human-Study: Some model curiosity subtypes are preferred (reflective, exploratory) more than others (causal, counterfactual).
Figure 9: Human-Study: Question-Asking Preferences vary by the query context. Advice/Reasoning/Creative contexts prefer more model questions.
| Model | Acc. | Macro-F1 | AUC |
|---|---|---|---|
| Majority baseline | 0.685 | 0.406 | 0.500 |
| Affordance only | 0.652 | 0.625 | 0.663 |
| Subtype only | 0.562 | 0.559 | 0.605 |
| User + question TF-IDF | 0.668 | 0.618 | 0.670 |
Table 3: Proof-of-concept prediction of item-level validated useful curiosity. Accuracy is affected by class imbalance, macro-F1 better reflects recovery of useful-curiosity cases.
Appendix figures & tables21 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 10: Prompt used for LLM-judge annotation of human-LLM conversations.
Figure 11: Human annotation instructions for validating assistant-question extraction and curiosity labels.
| Question type | Theoretical grounding | Operational definition | Example |
| Curiosity: Information-gap | Information-gap theory; epistemic curiosity ( Loewenstein, 1994 ; Litman, 2005 ) | Asks about missing knowledge that could deepen understanding or surface an epistemic gap, rather than merely filling a task slot. | “What problem does it solve?” |
| Curiosity: Causal-explanatory | Explanation-seeking; explanations as understanding ( Chi et al., 1994 ; Lombrozo, 2006 ) | Asks for causes, mechanisms, reasons, or why/how something happens. | “Why did this happen?” |
| Curiosity: Exploratory broadening | Epistemic exploration; diversive and specific curiosity ( Berlyne, 1966 ; Litman et al., 2010 ) | Broadens inquiry toward alternatives, implications, extensions, or new angles. | “What else might explain this?” |
| Curiosity: Reflective-metacognitive | Metacognition; epistemic reflection; uncertainty monitoring ( Flavell, 1979 ; Nelson, 1990 ) | Prompts reflection on assumptions, evidence, confidence, uncertainty, values, or reasoning. | “What assumptions are you making?” |
| Curiosity: Perspective-taking | Perspective-taking; role-taking theory; empathy ( Davis, 1983 ; Galinsky and Moskowitz, 2000 ) | Invites the user to consider another person, stakeholder, viewpoint, intention, or interpretive frame. | “How might they interpret this?” |
| Curiosity: Counterfactual-hypothetical | Counterfactual reasoning; hypothetical exploration; epistemic simulation ( Kahneman and Miller, 1986 ; Roese, 1997 ) | Asks the user to reason about changed conditions, alternative possibilities, or what would happen if something were different. | “What if the context changed?” |
Table 4: Taxonomy of assistant-to-user questions. Curiosity-oriented questions are grounded in theories of information gaps, explanation-seeking, epistemic exploration, metacognition, perspective-taking, and counterfactual reasoning. Non-curiosity questions capture task, clarification, engagement, and other interactional functions. Generated artifacts are excluded before classification.
| Dataset | Candidate Qs | Kept real Qs | Dropped/artifact | Curiosity Qs |
|---|---|---|---|---|
| WildChat | 75,543 | 50,989 (67.5%) | 24,554 (32.5%) | 19,755 (26.2%) |
| OASST1 | 3,704 | 3,116 (84.1%) | 588 (15.9%) | 876 (23.7%) |
| LMSYS | 202,552 | 159,918 (79.0%) | 42,634 (21.0%) | 33,938 (16.8%) |
| ShareGPT | 28,110 | 21,462 (76.4%) | 6,648 (23.6%) | 6,190 (22.0%) |
Table 5: Question-event preprocessing statistics. Candidate Qs are assistant-question spans extracted by rule- based preprocessing. Kept real Qs are retained by the LLM judge as real assistant-to-user questions. Curiosity Qs are retained questions classified as curiosity-oriented; percentages are relative to candidate questions.
Figure 12: Examples of extracted assistant questions and their assigned labels across datasets.
Figure 13: Topics and contexts that lead to LLM curiosity across datasets.
| Model | Dataset(s) | Curiosity rate |
|---|---|---|
| gpt-4 | WildChat, LMSYS | 54.18% |
| claude-2 | LMSYS | 47.90% |
| claude-1 | LMSYS | 46.22% |
| claude-instant-1 | LMSYS | 42.52% |
| gpt-3.5-turbo | WildChat, LMSYS | 41.43% |
| vicuna-33b | LMSYS | 34.22% |
Table 6: Curiosity rate by explicit model name. Rates are computed over real, classified assistant questions only. Dataset-level placeholders such as oasst1 and sharegpt are excluded.
Figure 14: Matched Effects Analysis - Stage 1.
Figure 15: Curiosity Rates - Stage 2.
Figure 16: Curiosity Types by User - Stage 2.
| Covariate | WildChat | OASST1 | LMSYS | ShareGPT | All |
|---|---|---|---|---|---|
| Prior user words | 0.278 | 0.516 | 0.507 | 0.187 | 0.338 |
| Prior questions | 0.140 | 0.107 | 0.197 | 0.166 | 0.138 |
| Prior epistemic markers | 0.274 | 0.481 | 0.316 | 0.193 | 0.254 |
| Prior exploratory questions | 0.237 | 0.234 | 0.186 | 0.154 | 0.180 |
| Assistant words | 0.809 | 0.718 | -0.023 | 0.640 | -0.008 |
| Assistant turn index | 0.154 | -0.002 | 0.024 | -0.080 | 0.036 |
Table 7: Standardized mean differences between matched curiosity and non-curiosity events on pre-treatment covariates. Positive values indicate higher covariate values for curiosity-oriented treatment turns; negative values indicate higher values for matched controls.
Figure 17: Subtype-specific differences in mean next-user epistemic-marker count relative to clean controls. Error bars show 95% conversation-clustered intervals.
Figure 18: Pooled treatment-control differences in judged human-curiosity presence across five later user turns
Figure 19: Subtype-control differences in judged human-curiosity presence across five later user turns. Points show rate differences and bars show 95% confidence intervals.
Figure 20: Model-question composition within the nine named preceding-user contexts in the clean causal-analysis event table.
Figure 21: ATT point estimates for next-turn epistemic markers by prior-user-engagement tercile and dataset.
Figure 22: Annotation guidelines for the Prolific human study on useful model curiosity.
Figure 23: Example item from the Prolific human study comparing a direct answer, a non-curiosity question, and a curiosity-oriented question.
Figure 24: Gender-Based Differences in Prolific Study for Q3
Figure 25: Age-Based Differences in Prolific Study for Q3
| Prompt | Any Q | Curious Q | Other Q | Generic Q |
|---|---|---|---|---|
| Direct | 7.5 | 1.2 | 6.2 | 7.5 |
| Ask-if-needed | 96.2 | 52.5 | 43.8 | 41.2 |
| Curiosity | 97.5 | 83.8 | 11.2 | 11.2 |
Table 8: Percentage of generated responses classified by question type in the 80-item prompting pilot. Results are preliminary and do not measure downstream user outcomes.