Overview
- The MAI Diagnostic Orchestrator (MAI-DxO) uses a multi-agent framework that queries leading AI models—GPT, Gemini, Claude, Llama and Grok—to mimic a collaborative panel of physicians
- In Sequential Diagnosis Benchmark testing, MAI-DxO achieved between 80 and 85.5 percent diagnostic accuracy versus about 20 percent for a panel of 21 generalist physicians
- By selecting more cost-effective tests and procedures, the AI system reduced estimated diagnostic costs by around 20 percent compared with human doctors
- Microsoft has secured partnerships with health systems and is preparing real-world clinical trials and regulatory reviews before any patient-care deployment
- Researchers and external experts caution that the prototype must be validated across diverse patient populations, integrated into actual workflows and cleared by regulators before clinical use