Benchmarked against real-world case records published each week in the New England Journal of Medicine, we show that the Microsoft AI Diagnostic Orchestrator (MAI-DxO) correctly diagnoses up to 85% of NEJM case proceedings, a rate more than four times higher than a group of experienced physicians. MAI-DxO also gets to the correct diagnosis more cost-effectively than physicians.

The Path to Medical Superintelligence, Dominic King & Harsha Nori

I got hit by a van one time while I was riding my bike. I ended up going to urgent care and getting an x-ray of my foot, which was obviously hurt. The doctor currently at the urgent care took at look at the x-ray and was like: hard to say but looks fine to me. The clinic called back the next day, as another doctor took a look at the x-ray, and was like: actually your foot is broken.

That was nice of them to call back. I definitely want doctors that know what they are doing in the world, let’s keep training doctors.

But also x-rays seem like perfect fodder for AI. Shouldn’t we have like 20 million x-rays of feet along with diagnosis sitting around somewhere? I want my foot x-ray in there. Feels like there should be medical imaging “models” that are pretty damn accurate. Shouldn’t there be 20 million cat scans? 20 million ultrasound videos?

We don’t have to, nor should we, put 100% of our trust in the machines, but it seems like it would be an awfully useful tool in medicine.

And likewise, anything with lots of similar video that you can attach to measured outcomes. 20 million videos of swim meets. 20 million security videos of blackjack tables. 20 million birdfeeders. Tell us about the patterns. Tell us about the anomolies. Tell us things that computers can see that we can’t.