Name: Can AI Fix Health Care?
Uploaded: 2024-09-23
Duration: 5 min 43 s
Description: As AI reshapes industries worldwide, Amazon’s Neil Lindsay and Harvard professor Isaac Kohane explore the transformative impact AI can have on health care. They address concerns around AI bias and hallucinations, how AI can reduce costs, and how it can improve both the patient and provider experience.

As AI reshapes industries worldwide, Amazon’s Neil Lindsay and Harvard professor Isaac Kohane explore the transformative impact AI can have on health care. They address concerns around AI bias and hallucinations, how AI can reduce costs, and how it can improve both the patient and provider experience.

Can AI Fix Health Care?

Health

As AI reshapes industries worldwide, Amazon’s Neil Lindsay and Harvard professor Isaac Kohane explore the transformative impact AI can have on health care. 

There are two mega trends that are going to define medicine in the next five years and then going onwards. One is that a is capabilities, keep on improving literally month to month. The large language models that I used a year ago are nowhere near as good as the ones that I I'm using today. At the same time, primary care is falling apart. I cannot find primary care doctors to refer my friends or faculty to. And that provides an opportunity on a patient centered focus where a lot of the missing functions that the primary care doctors used to provide, we can provide and augment maybe with nurse practitioners, maybe with physician assistants or maybe just with the social network of patients in a consumer focused way. A I has been around a long time. It has become the buzz because I think the end user is now seeing a lot more of it directly. But there's so much opportunity for A I and technology, machine learning in general in health care. And I think about it, especially in three particular areas. I think about the ways in which we can take cost out of the system. There's still a lot of phone calls and faxes and humans involved in work that doesn't really add that much value that machines can do and should do. Because then we can invest that savings in the experience that actually helps patients get healthier, you know, making more providers available. Improving that experience. The second area is the provider experience. So providers often have to go through pages and pages, not just dozens, sometimes hundreds of pages of history to try to find insights so they can focus with the patient on the topics that are going to matter. Most A I can help extract a lot of insights from that history. Um A I can also help a provider take notes without having to spend their time facing the computer so they can be facing the patient and spend the energy with the patient. And then of course, I think there is an opportunity for customers and patients to find out more about their choices so that they can be more informed in their own health and actually perhaps be navigated using technology uh without necessarily taking up the time of a provider to, to do that navigation for them. When GP T four came out, that was a two year sprint basically from GP T three, but it keeps on accelerating. A year later, many of the large language models started having visual and other multimodal capabilities. And so I took a picture that had never been seen by any doctor. It was just out of the journal medicine, a picture of the back of a man, an older man who had developed a lot of itchiness over the past day and felt really bad. And it was a photograph of his back and there were, these looked like scratch marks, lots of lines. I took everything from the puzzler, all the text and the picture I removed the most important clue and I gave it to GP four. I said, what is this and this with a clue. Most doctors didn't get it. And it said it could be something called Bleomycin toxicity reaction to a chemotherapy treatment. The other is shitake mushroom toxicity. I never heard of it. And what I removed from the history of the patient is that they're eating mushrooms the day before. And so we're having this phenomenal fast moving capability that is behaving with all the hall stations with all the problems better on average than most doctors in terms of depth of knowledge. How do we address the problem? I just asserted. Well, it turns out that right out of the box. When you train an A I on all the information you have on the internet, they'll say a lot of weird things, some of which are politically correct. So open A I and Google put a lot of effort into something called alignment. They give examples of questions and how they want it answered and you give that tens of thousands of such examples and it starts behaving in a certain way. It's aligned to certain behavior. So you can actually align A is to maximize certain outcomes. You could, then if you're making the A I, you could align it to the patient, you can align it to the hospital. No, we're talking about this, but it's actually a well known procedure how to align something called fine tuning and uh reinforcement learning with human feedback. These are all technical terms which basically say I'm taking this whole big machine and I'm telling you how I want it to behave. What are the concerns about A I? There are concerns around bias around um hallucinations. But in my opinion, the biggest concern is who is the A I serving? Is it maximizing the interest of the hospital, of the insurer of public health or the patient? And answering that question I think is perhaps the most important concern. And right now, we're not talking a lot about it.