Revolutionary AI: GPT4's Impact on Healthcare Revealed - Successfully Diagnoses Rare Conditions

0:00Since the launch of the original ChatGPT and it's now ChatGPT4 has come out, a lot of people have been talking about how this revolutionized pretty much every industry. And one of the massive areas that people are talking about is medical. Today on the podcast, we're going to talk about how ChatGPT is going to revolutionize the medical industry. And that is because there is a new book that is recently being published. We're going to dive into that by a prominent healthcare worker and computer scientist. And it's being co-authored with the VP, one of the VPs over at Microsoft, who obviously has a lot to do with OpenAI and everything going on over there.

0:37So let's jump into it. So a lot of this is being talked about in the news right now. There's an article in Insider that says the newest version of ChatGPT passed the U.S. medical licensing exam with flying colors and diagnosed a 1 in 100 ,000 condition in seconds. And so today we want to talk about that. A lot of this is coming from Dr. Isaac Kohan, who he's both a computer scientist and he's at Harvard and he's also a physician. And so he was teaming up to kind of test drive GPT-4 on its, you know, before it originally launched. And his goal was to see how this was going to be able to work with doctors in kind of a medical setting.

1:24He said, I'm stunned to say better than many doctors I've observed. So he's literally saying that this thing is functioning better than a lot of doctors, which is kind of scary. And it's not too surprising. I mean, a lot of this test taking kind of stuff you could say, because this thing was in the 90th percentile when it took the medical exam or the medical, yeah, the main medical exam. So this thing is incredibly powerful. Obviously, the medical licensing exam. I've said a couple other benefits or I guess features that this is able to do. GPT-4, it's not just a good test taker and fact finder, but it's also a really good translator.

2:04So this makes it very useful in the medical field because it's able to translate discharge information for a patient who speaks Portuguese. And it can also go and it's really good at kind of distilling wonky technical jargon into something that sixth graders could read. So this thing is obviously incredibly powerful, and a lot of people are saying that just the language translation alone is going to help it to make medical discoveries in different locations more easily accessible all around the world. So he has a couple kind of vivid examples, and he says that GPT-4 can also give doctors really helpful suggestions about bedside manner.

2:44it can offer tips on how to talk to patients about their conditions and a compassionate but clear language right because you can ask chat gpt you know say there's a patient with these problems how would i address them in a compassionate clear manner blah blah you can put anything you want there and it can um and it can obviously spit out a really good response now this has been criticized um not this exact use but there there was recently a school shooting and the school that notified the parents about the school shooting used ChatGPT to write the memo they sent out. And so a lot of people were quite up in arms, you know, they're saying, with something so atrocious as this, you know, surely a real person could have written it.

3:25And it's kind of hard because I guess playing the, talking from the other side, I think the people that originally wrote it, they themselves, you know, there was school faculty and they themselves had a lot of trauma and didn't really want to sit down and write something so painful, so they used this tool to kind of help them. So, I mean, there's two sides of that story and controversy, and I know people have a lot of different opinions on that, but all I'm saying is it was a controversial topic that people are discussing, right? Should ChatGPT write sensitive or highly emotionally charged or, you know, really sad news like this?

4:01So, in any case, it is something that a lot of doctors struggle with, right? Maybe they don't know the right words to say, and this is something that will help them. And I personally can, I guess I have some empathy for those doctors in that situation, as I myself would not, you know, know necessarily how to comfort someone in such a super hard situations. So that's one area that it will be used. Another that they discuss in this book is that it can be read in, it can read really lengthy reports and studies, and it can summarize them in like seconds, right? Like super fast. And this is really useful because the people that are, you know, obviously doctors that are diagnosing diseases, if they have the latest studies, which are always new ones are always coming out, but if they're able to summarize some of the latest studies coming out in a particular field, maybe relating to a particular disease that they specialize in, this could be really powerful for staying up to date on the latest information in your field.

4:58Something else that Chatshubut can do is it can explain its reasoning. It talks about making a diagnosis or solving a problem, and a lot of people are saying that that looks a lot like, you know, human-styled intelligence if you ask it to explain its reasoning behind something. But if you ask ChatGPT how it does all this, it's likely just going to tell you that its intelligence is still limited to patterns in the data, that it doesn't really involve any true understanding or intentionality. So, you know, know, that's what GPT told a lot of different people when they've asked about that. But even with these types of limitations, GPT-4 can mimic how doctors diagnose conditions with a really stunning success rate, although it is not perfect, which is, I think, a really important point to bring up here.

5:45So a lot of doctors that are kind of looking at this right now and taking it, are looking at this. There's one in particular, he took it through kind of a clinical thought experiment and he based it off of a real life case that involved a newborn baby he treated several years earlier and so he gave it a few different details about the baby he gathered from its physical exam as well as some information from the ultrasound and hormone levels that he collected and GPT-4 was actually able to correctly diagnose a one in a hundred thousand condition which is called congenial adrenal hyperplasty and anyways this doctor said just as he would have with all of his years of study and experience, GPT-4 was able to do this in a couple seconds.

6:29And so, you know, he was both impressed and kind of horrified. And if you think about it, it's like on the one hand, it's, you know, obviously solving this really sophisticated problem and giving this really sophisticated diagnosis, which is really impressive and amazing. But on the other hand, it's kind of mind-blowing to think that, you know, so many people are going to have access to this, and it's hard to guarantee or, you know, certify that GPT-4 or any other AI models advice is actually going to be safe and effective, right? Like, you would hate for something like this to not be perfect and to have any sort of issues.

7:07So it's important to know that GPT-4 isn't always right, and some people also say they note, you know, GPT-4 doesn't have an ethical compass. I think that, you know, when every doctor in the U.S. takes the, when they go and take the medical licensing exam, they do something which is called taking the Hippocratic Oath, which I think is just like saying, you know, they'll never harm another human. Obviously, GPT-4 has not taken the Hippocratic Oath, and so people are concerned that using it for doctor's practices, maybe, you know, it will do something that will harm people. And obviously, we've seen a lot of instances where GPT-4 does what people call hallucinates, just comes up with a random answer or it doesn't do something correctly.

7:50When talking about a lot of these different issues that it has, there are ways around it. I mean, obviously you're going to need a real human doctor to give a second opinion and look over these results. And also just on the sense of getting GPT-4 to give better results in these areas, you can also go, I think one trick a lot of people do is you start a fresh tab on chat GPT and give it all the data and ask it if there's any mistakes in there and get it to look at it with a fresh set of eyes similar to having a person look at something with a fresh set of eyes or having you know getting the giving the opinion of another trusted advisor you can also get it to explain I think this is a really powerful thing that doctors will want to be using is get it to explain its reasoning process behind a particular diagnosis or something because as it goes through and it kind of outlines why it said something, you can see if there's any flaws in the logic that brought it to its final conclusion.

8:47So I think this is going to be all in all a really incredible tool. I think it's going to free up a lot of time for different healthcare providers and clinicians to be able to be a lot more present with their patients instead of just on the computer screen all the time. But we also have to, I think, you know, because doctors right now are saying, like, hey, this thing's not perfect, it makes mistakes, blah, blah, blah. And I think that's totally valid and true. And we definitely don't want to rely on this. But I do think it's important for us to be thinking about the fact that we're continually going to be living in a world where this technology gets better and better, these results get better and better.

9:22And eventually, it will be right, like right now, if it's in the 90th percentile, eventually, it will be in the 99th percentile. And what does that mean? That means that this thing is better than 99 % of any doctors taking this. And if it's really that much better, like people are going to solely rely on this thing. People are going to build different tools and bots, whether it's regulated or unregulated to get their medical advice from these kinds of sources. And so I think it's going to be, it's an interesting question that we have to start talking about. And we have to kind of force ourselves to imagine what this world will look like with these machines that are essentially smarter than us.

10:01And we're going to have to decide, you know, how do we want that world to work? What do we want that to look like? It's going to be a lot of crazy stuff, a lot of big questions. And it's going to be something that obviously is going to require a lot of very smart people focusing on it. So it's going to be incredibly interesting to see how that evolves in the future.