Microsoft’s AI Has Now Diagnosed Patients More Accurately Than Actual Doctors

Microsoft has unveiled a groundbreaking artificial intelligence system, the Microsoft AI Diagnostic Orchestrator (MAI-DxO), which demonstrates unprecedented accuracy in diagnosing complex medical cases.

According to a recent study, MAI-DxO achieved an 85.5% diagnostic accuracy rate on challenging cases from the New England Journal of Medicine (NEJM), outperforming a panel of experienced physicians who averaged 20% accuracy.

This development, described by Microsoft AI CEO Mustafa Suleyman as “a genuine step toward medical superintelligence,” signals a transformative potential for healthcare diagnostics, promising greater accuracy and cost efficiency.

The cornerstone of Microsoft’s research is the Sequential Diagnosis Benchmark (SDBench), a novel testing framework designed to mirror real-world clinical decision-making.

Unlike traditional AI evaluations that rely on multiple-choice questions, SDBench uses 304 NEJM case studies—known for their diagnostic complexity—to simulate the iterative process of diagnosing patients.

The system requires AI models or human physicians to ask questions, order tests, and refine hypotheses step-by-step, with each action incurring a virtual cost to reflect real-world healthcare expenditures.

MAI-DxO integrates multiple leading AI models, including OpenAI’s GPT, Google’s Gemini, Anthropic’s Claude, Meta’s Llama, xAI’s Grok, and DeepSeek, to emulate a virtual panel of physicians.

This “chain-of-debate” approach allows the system to combine diverse diagnostic perspectives, enhancing its ability to handle intricate cases.

When paired with OpenAI’s o3 model, MAI-DxO achieved its peak performance, correctly diagnosing 85.5% of cases while reducing diagnostic costs by approximately 20% compared to human physicians.

Outperforming Human Experts

The study compared MAI-DxO’s performance against 21 U.S. and U.K. physicians with 5 to 20 years of experience.

These doctors, restricted from using external resources like colleagues, textbooks, or AI tools, achieved a mean accuracy of 20% on the same NEJM cases.

Microsoft noted that this limitation ensured a fair comparison of “raw human performance” against the AI system.

However, some experts, including MIT scientist David Sontag, caution that real-world physicians often use digital tools, which could narrow the performance gap if included.

Dominic King, a vice president at Microsoft involved in the project, emphasized the system’s cost-effectiveness, stating, “Our model performs incredibly well, both getting to the diagnosis and getting to that diagnosis very cost-effectively.”

By prioritizing less expensive tests, MAI-DxO reduced diagnostic costs to $2,396 per case in its budget-conscious mode, compared to $2,963 for human physicians.

This efficiency could address the estimated 25% waste in U.S. healthcare spending, which nears 20% of GDP.

Microsoft defines medical superintelligence as a system that surpasses the best human clinicians across a broad range of medical expertise.

MAI-DxO’s ability to integrate multiple AI models and process vast medical data positions it as a significant step toward this goal.

“This orchestration mechanism—multiple agents that work together in this chain-of-debate style—that’s what’s going to drive us closer to medical superintelligence,” Suleyman told WIRED.

The system’s design mimics a team of specialists, with AI agents assigned roles such as “Dr. Hypothesis” (ranking differential diagnoses), “Dr. Test-Chooser” (recommending tests), and “Dr. Stewardship” (optimizing costs).

This structured approach allows MAI-DxO to navigate complex cases with a level of reasoning that surpasses individual human capabilities, particularly in rare or multifaceted conditions.

Implications for Healthcare

With over 50 million health-related searches conducted daily across Microsoft’s consumer products like Bing and Copilot, the demand for reliable digital health tools is evident.

MAI-DxO’s potential applications include empowering patients to self-manage routine care and providing clinicians with advanced decision support for complex cases.

However, Microsoft emphasizes that the system is not intended to replace physicians but to complement their expertise.

“While AI is becoming a powerful tool in healthcare, clinical roles are much broader than simply making a diagnosis,” the company stated, highlighting the importance of human empathy and trust-building in patient care.

Experts like Eric Topol of the Scripps Research Institute praise the study’s rigor, noting its focus on complex cases and cost optimization.

However, both Topol and Sontag stress the need for clinical trials to validate MAI-DxO’s performance in real-world settings before it can be deployed in hospitals or integrated into platforms like Bing.

Despite its promising results, MAI-DxO is not yet approved for clinical use.

Microsoft acknowledges that further testing is needed, particularly for common ailments, to ensure safety and efficacy.

Regulatory approval and real-world clinical trials will be critical to confirm the system’s reliability and address concerns about over-reliance on AI.

A recent survey by the American Medical Association found that 60% of Americans are uncomfortable with providers relying on AI, and 57% believe it could harm patient-provider relationships.

Microsoft’s research also builds on the work of its AI health unit, led by Suleyman, a former co-founder of Google DeepMind.

The company’s strategic recruitment of top AI talent from Google underscores the intensifying competition in the AI healthcare space.

While Microsoft has not finalized plans for MAI-DxO’s deployment, possibilities include integration into Bing for public symptom analysis or tools to automate routine tasks for physicians.

Microsoft’s advancements signal a shift toward AI-driven healthcare solutions that prioritize accuracy, efficiency, and accessibility.

By addressing diagnostic complexity and cost challenges, MAI-DxO could help alleviate the strain on global healthcare systems, where rising costs and delayed diagnoses remain significant barriers.

As Suleyman told The Guardian, he envisions AI systems becoming “almost error-free” within the next 5-10 years, potentially reshaping the healthcare landscape.

For now, Microsoft’s research represents a proof-of-concept, demonstrating AI’s potential to augment human expertise.

As the company moves toward clinical validation, the path to medical superintelligence may redefine how we approach diagnosis and care, blending cutting-edge technology with the irreplaceable human touch of medicine.

Also Read: Coinbase Is Now Seeking SEC Approval For Tokenized Stock Trading

Visit the Homepage ⬅️ for our extensive library of news, or read news for you below.

---

News For You

---

About The Author