Framework Explores Human-AI Curiosity Ecosystems.

This paper introduces a conceptual "toy framework" for analyzing curiosity within both single and multi-agent human-AI ecosystems. The framework first examines an individual agent's inquiry policy, which dictates when, how, and why an agent asks questions. This policy is modeled as a function of several factors: the immediate value of reducing uncertainty, the costs associated with inquiry, the potential for delayed returns, and the strategic value of keeping a question unresolved. A key insight is that the weighting of these decision-related terms can dynamically change based on an agent's past experiences. For instance, a period of low-cost, quickly answered questions might alter an agent's perception of inquiry costs and influence the types of questions it pursues long-term. The framework then extends these principles to scenarios involving multiple agents exploring a shared knowledge landscape. In this multi-agent context, the framework tracks various metrics such as the overall volume of inquiries, the diversity of topics being explored, the extent of frontier-directed inquiry (seeking new knowledge), redundancy in questioning, and the accumulation of reusable knowledge. The ultimate goal of this conceptual framework is to provide a structured approach for studying the ecology of curiosity and to inform the future design of multi-agent AI systems geared towards discovery. It serves as a foundational piece for deeper exploration into how curiosity can be fostered and managed in complex human-AI interactions.