Recontextualizing NLP in Healthcare: A Survey on LLM-Based Multi-Agent AI Hospitals

ACL ARR 2025 May Submission680 Authors

Abstract: Large Language Models (LLMs) are increasingly being used as autonomous agents in high-stakes domains, yet their behavior in complex, real-world environments remains underexplored. This survey introduces the concept of AI hospitals—LLM-driven multi-agent ecosystems that simulate clinical workflows and support a wide range of medical applications. We review 70+ recent studies and propose a taxonomy covering core components and application areas. By analyzing how these systems integrate language, knowledge, and interaction in dynamic settings, we highlight AI hospitals as a powerful testbed for evaluating LLMs beyond static benchmarks. We also outline open challenges in aligning LLM behavior with clinical reasoning, safety, and patient-centered goals, offering a roadmap for the future at the intersection of NLP and healthcare.

Paper Type: Long

Research Area: Special Theme (conference specific)

Research Area Keywords: Interdisciplinary NLP, Multi-Agent Systems, Simulation-Based Evaluation, NLP for Healthcare, Recontextualized LLM Applications

Contribution Types: Surveys

Languages Studied: english

Keywords: Interdisciplinary NLP, Multi-Agent Systems, Simulation-Based Evaluation, NLP for Healthcare, Recontextualized LLM Applications

Submission Number: 680