00:00 Speaker a
As technologies like Chat GPT become part of our everyday lexicons, the use of artificial intelligence is wiping out the world, and Microsoft believes that its AI could end up in your doctor's office. The Microsoft AI team is investigating how AI can help diagnose medical cases. A new study from today's team shows that the company's diagnostic orchestrators using open AI technology can correctly diagnose up to 85% of case procedures in the New England Journal of Medicine. This is more than four times higher than the group of experienced physicians, and according to the survey results it is more cost-effective. Joining me right now is Mustafa Suleman, CEO of Microsoft AI. I'm happy to meet you, Mustafa. A long follower of your work. So let's start with today's news. How will this breakthrough help the health, um, the health care system go on?
01:39 Mustafa Suleman
Well, the incredible thing is that we at Microsoft AI already have 50 million health-related questions per day through Copilot and Bing. And our goal is to provide the highest quality, most accurate, and ultimately the cheapest health advice and support to all consumers and more people in the future. So this is early research, but as we put this into production, we hope that, um, very high quality health information will become accessible to everyone.
02:26 Speaker a
The first thing I thought, Mustafa, when I saw this, why do I need my doctor? And because of these advances in AI, will doctors disappear over time?
02:47 Mustafa Suleman
No, you definitely need your doctor. Your doctor will make the right decision call at the right time, plan treatments that will come soon after diagnosis, and are basically there to oversee, organize and retain accountability to the AI itself. Essentially, what these models do is look at a huge array of existing medical information, synthesize it, and present it to humans at the right time. This is a great interactive experience. AIS actually asks questions about the data itself, recommends which tests to get, and then improves the accuracy of the diagnosis in order as they learn more information. And by doing that, if you work with your doctor and do it, uh, you know, obviously, in itself, it's the way this is unfolding.
04:05 Speaker a
How are doctors trained with AI?
04:11 Mustafa Suleman
Well, I, today, I know, you know, you need to pass the USMLE, medical license exam, and you know, you need to get around 60% with multiple choice questions. Well, these models get more than 95% accuracy for these questions. Well, what's really unique about this model is that it doesn't just use one AI. In reality, we use all four major AI developers and are essentially passed on to the most accurate ones at any time. That's what an orchestrator does. Decide when to use a larger or smaller model, when to use a lot of thought time, and when to use it. You can also try your best to get a more accurate answer.
05:34 Speaker a
Well, in many cases health care decisions deal with life and death, Mustafa. How do you build trust in this technology?
05:45 Mustafa Suleman
Now, what's cool about this method is that you can actually see why AI is reaching a certain conclusion. Additionally, adding more data will update the best estimate of what the diagnosis will look like. And that gives you a lot of transparency about how it thinks, what data it takes into account, the tests it does. So this is actually an unprecedented amount of transparency. Of course, you can trust it because you can actually go back and see the whole conversation that took place between many different agents coordinating with each other.
