-
Posts
-
By ad47uk · Posted
I don;t think I would trust it, seeing the mistakes all these AI are making.l -
By CHUNWEI · Posted
15 years is just yesterday ... -
By Martog · Posted
I'm stumped people think DEI is just a quota and nothing else. -
By +thexfile · Posted
Microsoft's AI is highly effective for identifying targets on the battlefield or for firing a lot of workers. 🫤 -
By zikalify · Posted
Microsoft AI diagnoses complex medical cases four times better than human doctors by Paul Hill Microsoft AI has revealed that its AI Diagnostic Orchestrator (MAI-DxO) is able to correctly diagnose 85% of complex cases from the New England Journal of Medicine (NEJM) case proceedings. What makes this more impressive is that the cases published in NEJM are very diagnostically complex and intellectually demanding. They typically require multiple specialists and diagnostic tests to reach a definitive diagnosis, so the fact AI can get it right most of the time is a big deal. The MAI-DxO turns language models into a virtual panel of clinicians that is able to ask follow-up questions, order tests, or deliver diagnoses. MAI-DxO boosted the diagnostic performance of every model that was tested by Microsoft, with the best results when it was paired with OpenAI’s o3. When MAI-DxO was using o3, it was able to correctly solve 85.5% of NEJM benchmark cases. Microsoft compared this to humans; it took 21 practicing physicians from the US and UK with 5-20 years of clinical experience, and on the same tasks, they achieved a mean accuracy of just 20% across completed cases. Microsoft believes that these tools can significantly shake up healthcare by empowering patients to self-manage routine aspects of care and equipping clinicians with advanced decision support for complex cases. To find out how AI would perform on the NEJM cases, Microsoft had to create the Sequential Diagnosis Benchmark (SD Bench) which transforms 304 recent NEJM cases into stepwise diagnostic encounters. Models can then iteratively ask questions and order tests. As new information becomes available, the model will update its reasoning and gradually move toward a final diagnosis which can be compared to what was published in the NEJM. As mentioned before, the MAI-DxO system emulates a virtual panel of physicians that is able to ask follow-up questions, order tests, or deliver diagnoses. Aside from this, it can operate within defined cost constraints to help prevent excessive testing. While Microsoft’s experiment is showing very promising results, this research is just a first step. Before generative AI can be used safely in healthcare scenarios, more evidence needs to be gathered from real clinical environments. There also need to be appropriate governance and regulatory frameworks to make sure models are reliable and safe. To accomplish all of this, Microsoft is partnering with health organizations to test and validate its approaches before any broader rollout.
-
-
Recent Achievements
-
CHUNWEI earned a badge
One Month Later
-
TIGOSS earned a badge
Week One Done
-
henryj earned a badge
First Post
-
CarolynHelen earned a badge
First Post
-
henryj earned a badge
Reacting Well
-
-
Popular Contributors
-
Tell a friend
Recommended Posts