Towards autonomous medical artificial intelligence agents.
- Open access
- 29 cites
MIRA, an autonomous AI agent in a sandboxed EHR, surpasses physicians in diagnostic accuracy and makes safe, guideline-concordant clinical decisions across multiple cases.
- Why it matters: Integrating AI into clinical workflows is crucial for effective decision support, but it remains unproven whether such systems can perform at physician level within EHR environments, addressing a significant gap in AI-enabled healthcare.
- What they did: The team developed MIRA, an AI system that navigates a large clinical action space, using real patient case simulations to obtain histories, order tests, generate diagnoses, and formulate treatment plans, including medication and admission decisions.
- The result: MIRA outperformed physicians in diagnostic accuracy and safety, demonstrating the potential for EHR-integrated AI to convert clinical intent into actionable operations, paving the way for more effective decision-support tools.