Why Did Apple Fall: Evaluating Curiosity in Large Language Models
Summary: arXiv:2510.20635v2 Announce Type: replace-cross
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 the 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.
Introduction
Curiosity is a fundamental aspect of human cognition, driving exploration and the pursuit of knowledge. In the realm of artificial intelligence, particularly with the rise of large language models, understanding and evaluating curiosity becomes crucial. This article evaluates how curiosity is manifested in LLMs and its implications for future advancements in AI.
Evaluation Framework
To assess curiosity in LLMs, we developed a comprehensive evaluation framework that incorporates various dimensions:
- Information Seeking: This dimension evaluates how effectively LLMs seek out new information to enrich their understanding.
- Thrill Seeking: This aspect examines the model’s willingness to engage with novel and risky information, which can lead to unexpected learning outcomes.
- Social Curiosity: Here, we analyze how LLMs interact with social contexts and their ability to learn from social cues.
Findings
Our research revealed several key findings:
- LLMs demonstrated a remarkable capacity for knowledge acquisition, often surpassing human curiosity in their quest for information.
- Despite their thirst for knowledge, LLMs exhibited a tendency to make conservative choices when faced with uncertain environments, prioritizing safer options over riskier ones.
- The interplay between curiosity and reasoning was significant; increased curiosity led to enhanced reasoning and active learning capabilities in LLMs.
Implications for Future Research
The implications of our findings are profound. By understanding how curiosity operates within LLMs, researchers can work towards developing models that are not only more intelligent but also more adaptable and innovative. Future research may focus on:
- Enhancing curiosity-driven learning mechanisms in LLMs.
- Exploring the ethical implications of AI curiosity and its impact on human-AI interaction.
- Investigating the potential for curiosity to foster creativity in AI systems.
Conclusion
As we delve deeper into the capabilities of large language models, understanding the nature of curiosity is essential. Our research provides foundational insights into how LLMs can exhibit curiosity similar to humans, paving the way for enhanced learning capabilities and innovative research in artificial intelligence.