Title: OpenHospital: A Self-Contained Environment for the Development and Assessment of LLM-Driven Collective Intelligence

Abstract: Collective Intelligence (CI) powered by Large Language Models (LLMs) offers a compelling solution to the data wall challenge, enabling the continuous improvement of LLM agents. Despite this potential, the field currently lacks a specialized platform designed specifically for the evolution and benchmarking of LLM-based CI. To bridge this void, we present OpenHospital, an interactive simulation environment where physician agents can develop CI through dynamic interactions with patient agents. Utilizing a "data-in-agent-self" paradigm, the platform rapidly upgrades agent capabilities while delivering rigorous evaluation metrics to assess both medical competence and system performance. Our experimental results confirm that OpenHospital is effective in both cultivating and measuring Collective Intelligence.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC