Microsoft AI beats doctors in complex diagnosis tests

Microsoft’s AI diagnosis tool has beaten general doctors in tests on complex medical cases. The result is striking, but it does not mean the Microsoft AI system can replace doctors. The tests used past case records, not live patients in hospitals.

How Microsoft AI reached its diagnosis

As reported by Reuters, Microsoft’s tool, called MAI-DxO, tries to follow the steps a doctor takes. Instead of reading all the details at once and giving one answer, it works through a case in stages. It first decides what to ask about a patient’s past health and symptoms. It then looks at exam findings and chooses which tests may help. After seeing the results, it can ask for more tests and change its likely diagnosis.

The system uses an “orchestrator” to guide a group of AI agents. These agents act like a panel of medical experts, each helping to assess the case. The aim is to make the process more careful, not just to guess the name of a disease. A study published in the journal Science on October 22 found that the tool was more than twice as accurate as general doctors on more than 300 rare and difficult cases.

What did this study found?

The test used 304 complex cases drawn from The New England Journal of Medicine. The AI reached the right diagnosis in about 83% of cases. General doctors using reference material got it right in about 36% of cases. Doctors working without those tools scored about 22%. The system also did better in tests based on records from emergency room visits. It reached about 80% accuracy, compared with around 47% for the doctors.

The researchers also found that the company AI planned tests at a lower estimated cost than human doctors. It could weigh which checks were useful and avoid some costly steps. The team says this matters because good care is not only about finding the right answer. It is also about choosing sensible tests. Still, the results need care. These were set medical cases, not patients being treated in real time.

What is next for the new Microsoft AI?

Microsoft says the next step is to test the tool in a real hospital business setting. It plans to work with Beth Israel Deaconess Medical Center in Boston to study whether the system can help sort patients in emergency department waiting rooms. The company designed this tool to copy the step by step process of diagnosis. That could help doctors think through hard cases and decide which tests to order.

But a strong score in a study is not proof that an AI tool can safely care for people. It must be checked in real clinics, with doctors overseeing its advice. Researchers will need to see how often it misses a disease, asks for needless tests or gives advice that does not fit a patient. The findings offer a hopeful sign for medical AI. Yet the next test is not a paper or a score. For now, it looks more like a possible aid for doctors than a replacement for them.

Author’s Opinion

Microsoft’s results are promising, but beating doctors in a test is not the same as treating patients. The real value of this tool will depend on how well it works in hospitals. For now, AI should support doctors, not replace them.

ALSO READ: Meta faces $25 billion hit over Cambridge Analytica scandal.