Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Why Did Apple Fall To The Ground: Evaluating Curiosity In Large Language Model

Haoyu Wang, Sihang Jiang, Yuyan Chen, Yitong Wang, Yanghua Xiao

TL;DR

The paper investigates whether large language models (LLMs) exhibit human-like curiosity by adapting the Five-Dimensional Curiosity Scale Revised ($5DCR$) to prompts and behavior. It combines questionnaire-based assessments, three behavioral experiments (Information Seeking, Thrill Seeking, Social Curiosity), and curiosity-driven learning (CoQ vs CoT, with SFT and GRPO) to study the relationship between curiosity and learning. Findings indicate LLMs show stronger Information Seeking than humans but are more cautious in uncertain contexts, with social curiosity broadly comparable; intriguingly, curiosity-driven questioning improves active learning and reasoning, supporting curiosity as a pathway to autonomous learning in LLMs. The work provides a systematic, multi-method framework for evaluating and leveraging intrinsic curiosity in LLMs and highlights implications for future curiosity-directed AI research and applications.

Abstract

Curiosity serves as a pivotal conduit for human beings to discover and learn new knowledge. Recent advancements of large language models (LLMs) in natural language processing have sparked discussions regarding whether these models possess capability of curiosity-driven learning akin to humans. In this paper, starting from the human curiosity assessment questionnaire Five-Dimensional Curiosity scale Revised (5DCR), we design a comprehensive evaluation framework that covers dimensions such as Information Seeking, Thrill Seeking, and Social Curiosity to assess the extent of curiosity exhibited by LLMs. The results demonstrate that LLMs exhibit a stronger thirst for knowledge than humans but still tend to make conservative choices when faced with uncertain environments. We further investigated the relationship between curiosity and thinking of LLMs, confirming that curious behaviors can enhance the model's reasoning and active learning abilities. These findings suggest that LLMs have the potential to exhibit curiosity similar to that of humans, providing experimental support for the future development of learning capabilities and innovative research in LLMs.

Why Did Apple Fall To The Ground: Evaluating Curiosity In Large Language Model

TL;DR

The paper investigates whether large language models (LLMs) exhibit human-like curiosity by adapting the Five-Dimensional Curiosity Scale Revised () to prompts and behavior. It combines questionnaire-based assessments, three behavioral experiments (Information Seeking, Thrill Seeking, Social Curiosity), and curiosity-driven learning (CoQ vs CoT, with SFT and GRPO) to study the relationship between curiosity and learning. Findings indicate LLMs show stronger Information Seeking than humans but are more cautious in uncertain contexts, with social curiosity broadly comparable; intriguingly, curiosity-driven questioning improves active learning and reasoning, supporting curiosity as a pathway to autonomous learning in LLMs. The work provides a systematic, multi-method framework for evaluating and leveraging intrinsic curiosity in LLMs and highlights implications for future curiosity-directed AI research and applications.

Abstract

Curiosity serves as a pivotal conduit for human beings to discover and learn new knowledge. Recent advancements of large language models (LLMs) in natural language processing have sparked discussions regarding whether these models possess capability of curiosity-driven learning akin to humans. In this paper, starting from the human curiosity assessment questionnaire Five-Dimensional Curiosity scale Revised (5DCR), we design a comprehensive evaluation framework that covers dimensions such as Information Seeking, Thrill Seeking, and Social Curiosity to assess the extent of curiosity exhibited by LLMs. The results demonstrate that LLMs exhibit a stronger thirst for knowledge than humans but still tend to make conservative choices when faced with uncertain environments. We further investigated the relationship between curiosity and thinking of LLMs, confirming that curious behaviors can enhance the model's reasoning and active learning abilities. These findings suggest that LLMs have the potential to exhibit curiosity similar to that of humans, providing experimental support for the future development of learning capabilities and innovative research in LLMs.
Paper Structure (24 sections, 16 equations, 5 figures, 15 tables)

This paper contains 24 sections, 16 equations, 5 figures, 15 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Evaluation of Curiosity
  4. Curiosity in Language Model
  5. Method
  6. Curiosity Questionnaire Study
  7. Curious Behavior Study
  8. Curiosity-Driven Learning
  9. Experiments
  10. Experimental Setup
  11. Results of RQ1
  12. Results of RQ2
  13. Results of RQ3
  14. Conclusion
  15. Appendix
  16. ...and 9 more sections

Figures (5)

  • Figure 1: Overview of our evaluation. A) Three dimensions of human curiosity. B) Large Language Models (LLMs) are prompted to self-assess using the Five-Dimensional Curiosity Rating (5DCR) scale. C) We investigate three types of curious behaviors exhibited by LLMs. D) We design a curiosity-driven questioning and thinking pipeline for LLMs to investigate the relationship between curiosity and learning in LLMs.
  • Figure 2: Examples of Curiosity-Driven Information Seeking, Thrill Seeking and Social Curiosity studies.
  • Figure 3: Comparison of the model’s results with human sample across six dimensions of curiosity.
  • Figure 4: Results of curiosity-driven Information Seeking, Thrill Seeking and Social Curiosity studies.
  • Figure 5: Trial selection paths for the underwater game jirout2012children. The value represents the uncertainty corresponding to the two Windows. Pink, red and blue lines correspond to the choice of the model in the previous round of the game is "confirm to choose", "medium confirm" and "not sure".