A Science study reported that Microsoft’s MAI-DxO diagnosed 83 percent of hard simulated cases, against 36 percent for generalist physicians working alone. Specialists were not in the comparison, and much of the gain is attributed to the underlying models.
What the comparison leaves out
The cases were simulated, drawn from complex NEJM-style problems, not a clinic where a doctor can examine a patient.
TechCrunch noted that much of the improvement may come from the models underneath the system, including GPT-5, rather than from Microsoft’s diagnostic wrapper alone.
A reader who sees “AI beat doctors” and needs the missing line: generalists, alone, on a simulation. It is not a license to skip a clinician.