[REVISION HISTORY]

AI development in medical diagnostics

Updated 3 times since CLSTR started tracking revisions of this situation.

What changed

2026-10-23 13:42 UTC → 2026-10-23 20:07 UTC · added ~~removed~~

Recent developments in artificial intelligence show increasing capabilities in clinical diagnostic settings. A study published in The Lancet regarding Google’s AMIE chatbot indicated the feasibility of using conversational AI to help patients prepare for primary care appointments. While clinical evaluators rated conversation quality favorably in most cases, researchers noted instances of hallucination and one interaction rated as “somewhat harmful” by a physician due to patient anxiety caused by a suggested cancer diagnosis. In a separate advancement reported in the journal Science, Microsoft researchers introduced MAI-DxO, a system designed to mimic the step-by-step diagnostic process of physicians. This tool achieved an 83.2% accuracy rate in diagnosing over 300 complex and unusual cases from The New England Journal of Medicine, significantly exceeding the 36.1% accuracy rate of generalist physicians using medical references and the internet. The system utilizes an “orchestrator” to coordinate simulated clinical experts that sequentially determine necessary patient questions and tests. MAI-DxO also demonstrated superior performance in emergency department scenarios, reaching approximately 80.3% accuracy compared to 46.8% for human physicians. Beyond accuracy, research indicates the system may reduce diagnostic costs through more efficient testing paths. Microsoft officials, including AI lead Mustafa Suleyman and Health unit VP Dominic King, emphasized that the system has only been tested on historical data and requires further clinical validation. Microsoft plans to conduct real-world validation studies in collaboration with Beth Israel Deaconess Medical Center. Recent data suggests the system’s performance was notably high when integrated with OpenAI’s GPT-5 and GPT-5.5 models.

Versions

Only revisions since CLSTR began indexing content versions appear here. Select a version to see what changed compared to the one before it.