Paging Dr. Algorithm: Microsoft’s AI Diagnoses Like House, Bills Like Costco

Key Points

1. MAI-DxO, a model-agnostic orchestrator, simulates a panel of virtual physicians and achieved 85.5% diagnostic accuracy in testing against 304 complex cases from the New England Journal of Medicine.

2. A group of 21 experienced physicians achieved only a 20% success rate on the same cases, highlighting the AI's superior performance.

3. MAI-DxO operates through the Sequential Diagnosis Benchmark, balancing diagnostic accuracy with cost efficiency, achieving 80% accuracy at $2,397 per case and 85.5% accuracy at $7,184 per case.

4. The system is designed to work with AI models from various companies, boosting performance by an average of 11% across models from OpenAI, Google, Anthropic, Meta, and others.

5. Microsoft emphasizes that AI will augment rather than replace doctors, with plans to submit the research for peer review and collaborate on clinical validation and regulatory reviews.