## Agent Hospital: A Simulacrum of Hospital with Evolvable Medical Agents
- article: https://pub.towardsai.net/worlds-first-ai-agent-hospital-42-ai-doctors-4-nurses-0-humans-1cd051a397b0
- arxiv: https://arxiv.org/pdf/2405.02957
This source describes the creation and evaluation of **Agent Hospital**, a simulation populated by **Large Language Model (LLM)**-powered agents acting as **patients, doctors, and nurses** to replicate a real hospital environment. The primary goal is to enable **doctor agents to learn and improve their medical skills** through interaction within this virtual world, treating automatically generated patient agents and learning from both successful and unsuccessful cases without manual data labeling. By simulating the entire patient treatment process and allowing doctor agents to accumulate experience, the research demonstrates that these **evolved doctor agents can achieve high diagnostic accuracy** in the virtual environment and even outperform existing methods on real-world medical benchmarks. This approach, termed **Simulacrum-based Evolutionary Agent Learning (SEAL)**, offers a novel paradigm for training AI agents in complex, task-specific domains by building realistic simulations for data generation and agent evolution.