A framework for single and multi-agent human-AI curiosity ecosystems

arXiv cs.AI · 6 hours ago

Researchers developed a framework for understanding curiosity in both single and multi-agent AI systems by modeling how agents decide when and what to ask based on uncertainty reduction, costs, and delayed rewards. A key insight is that an agent's inquiry preferences shift over time as it gains experience with different types of questions, such as preferring cheaper questions after a period of rapid answers. When extended to multiple agents sharing a knowledge landscape, the framework tracks metrics like inquiry volume, topic diversity, and knowledge reusability to characterize collective discovery dynamics.