EP 427: The Role of GenAI in Modern Healthcare - Challenges and Opportunities

This is the Everyday AI show, the Everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life.

One industry that I think is ripe for disruption via artificial intelligence is healthcare. Especially here in the U.S. right. Sometimes I'm still scratching my head, like, why does it seem like the, you know, health care and in medical fields in the US at least are 10 years behind? Right. I understand there's so many privacy concerns, but I think the role of generative AI in modern healthcare is quickly changing. So today we're going to be talking about both some of the challenges and opportunities and see, well, hey, when we, as we roll into 2025, is the landscape going to be shifting or are we still going to be stuck in this kind of AI healthcare, sitting on the fence sort of thing that we've been in the last year or two? All right, I'm excited for today's conversation. Hope you are, too. Welcome to Everyday AI. What's going on, y'? All? My name is Jordan Wilson. I'm the host of Everyday AI. This thing is for you. This is a daily live stream podcast, free daily newsletter helping us all learn and leverage generative AI to grow our companies and our careers. So even if you aren't in the healthcare medical fields, this is something that obviously impacts us all. So I'm excited for today's conversation. I'm also excited for you to go to your everydayai.com we will be recapping the highlights and even giving deeper insights from today's interview in our free daily newsletter. So make sure you go check that out on our website and also go check out more than 430 back episodes. You can go listen to them all, watch them all, read about them all on our website, sorted by category. It is probably the best source of unbiased information on generative AI on the entire Internet. So make sure you go check that out. All right, before we get into today's conversation, let's first start off with the AI news. And there's a ton. So Google Search is apparently going to change and start rolling out a dedicated AI mode. So reports indicate that the new AI mode in Google Search will closely resemble Google's Gemini AI Chatbot, which has been operating separately from the search engine. So the introduction of this mode is expected to expand Gemini's audience as billions of users via Google Search will now have access to an AI enhanced search functionality. So early tests of the AI mode inside Google Search have been spotted in the Google App and on Android devices, suggesting that a rollout across the wider landscape may be imminent.

A new shortcut button for AI mode.

Has been identified in a recent APK teardown, allowing users to quickly switch to this feature and refine their searches with follow up questions. Interesting here. So it looks like Google may be following in the path of of chat, GPT search and perplexity, so we'll see how that one rolls out. Speaking of Google, Google's been crushing it the past two weeks. So Google also unveiled Gemini 2.0 flash thinking. All right, so we saw Google Gemini 2.0 flash, but now we have Flash Thinking essentially their answer or their version of OpenAI's Zero1 model that does this. More chain of thought or reasoning under the hood. So a different kind of, you know, quote unquot AI chatbot. So Google Gemini's 2.0 flash thinking boasts advanced reasoning abilities, enabling it to solve complex problems rapidly while revealing its internal planning steps. That's the big thing. So we have a little bit more transparency under the hood. Pretty interesting. Responded to Andre Karpathy on Twitter and about his thoughts on it, so I'll share that in the newsletter if you want to see. So the model supports multimodal inputs and outputs, allowing users to interact with images, videos and audio, which could enhance user engagement and creativity in various applications. So Gemini 2.0 is being positioned as a competitor to OpenAI's Zero1 model, which has received positive feedback for its powerful reasoning capabilities. So it is available right now for free in Google's AI studio. Just know if you're using Google's AI studio, you can't really opt out of training. So just keep that in mind before you throw, you know, sensitive documents at this new 2.0 flash. All right, last piece of AI news for today. It is the 12th day of Open AI's 12 Days of Ship Miss. All right, so according to reports, OpenAI is poised to potentially reveal a new AI model called O3. Okay, interesting. So according to the information, the new model O3 is expected to replace O1, which was just fully released like a couple of weeks ago. So reports suggest that the decision to to skip O2 is in part due to a UK telecom company's existing use of the same name in OpenAI. CEO Sam Altman had a cryptic tweet after he said ho ho ho. He said, should have said oh oh oh oh three. That's what people are pointing to. And Experts speculate that O3 may have the ability to tackle evaluation tests designed to assess artificial general intelligence so, yeah, are we actually going to get a model today that is AGI? I don't know. We'll see. I'm guessing we're gonna get a live stream blog post and wait list. All right, so if you want to know more, make sure to go to your everydayai.com Sign up for the free daily newsletter. We'll be recapping those stories and a whole lot more. All right, but you probably tuned in today to hear or to listen or to ask questions about the role of generative AI in healthcare. All right, so I'm excited to bring on to the show our guest for today. Please, live stream audience, help me in welcoming William Horton, staff machine learning engineer at Included Health. William, thank you so much for joining the Everyday AI show.

Thank you so much for having me.

Oh, man. So much AI news going on today took took a minute. William was just waiting patiently there in the waiting room. But can you tell us a little bit about what you do in your role at Included Health?

Yes. So I work on our machine learning platform team, and in the last year and a half, a lot of that has been around building a platform for generative AI. That's kind of the big topic now. So what I've been working on with my team is building tools that let people do different things with large language models. And that's, you know, access the latest and greatest, evaluate the outputs, learn more about how they can prompt effectively, as well as serving models internally. So we're kind of building a whole set of tools to let people really use these things effectively.

Yeah. And, you know, before we get too deep into today's topic, could you even just tell us all in case those, those that aren't aware what is Included Health, what do you all do?

Yeah, that's. That's a good question. So Included Health, kind of our motto is all Included Care. And so what we offer is a combination of health benefits, navigation, as well as telemedicine to work with a patient through their entire journey. So our members, we call them members, they can come to us with questions about what do I pick during open enrollment. That's a very popular one. Or what would I pay to go see a primary care physician? But we can take that all the way to saying, okay, if you need actual health care, we have doctors on staff. We can do virtual urgent care, behavioral health, primary care for you through telemedicine. So we're kind of selling this all in one journey for the patient through Included Health.

All right, so I know this is going to be the Most open ended and vague question I could possibly ask William, but can you give us an overview of where we're at right now, at least here in the US with AI and healthcare and the medical field? Because you know, we've, we've had some great guests on the show before, but it seems like, at least to me, that this, you know, because of privacy concerns. Hipaa, Right. So many things. It seems like to me that the field's not going as quickly as other sectors, probably for reasons that make a ton of sense. But can you just give us like a super zoomed out view of like, where the heck are we at?

Yeah, I'll give you my zoomed out view of healthcare. I think that, you know, there's obviously, like you said, security and privacy concerns when it comes to this and also, you know, considering the, the risk that goes into making these decisions, I think, you know, Genai has made a lot of progress in certain areas, I think mostly in kind of back office or administrative tasks. So you see a ton of companies now in the medical scribing business as a very popular thing or otherwise trying to streamline operations. I think that's a very rich area that's already seen a lot of progress. But at the same time, you know, people are starting to push the frontier of bringing it into actual patient care and patient questions. So I think the first kind of stage of that, if you look at it broadly, is companies that, you know, let people ask questions or try to answer, you know, medical things, kind of like WebMD, but smarter. I'm sure there's a lot going on in that space. And then I think the next level, which you don't see yet, but the research is getting there, is how do we use medicine as a diagnostic or, sorry, how do we use the models as a diagnostic tool in medicine? So there's papers already that are saying how can we get them to help doctors figure out complicated cases? And I don't think you see that widespread right now due to the risks involved. But that's kind of the next step. I would see it looking at the kind of global landscape.

Yeah. And maybe, you know, I just assume everyone understands the privacy concerns. But you know, William, from someone on your side, can you explain what those are? Right. Like is it, you know, because a lot of people say, well, you know, hey, what's the difference? Right? What's the difference if our, you know, healthcare organization, you know, uses, you know, Google's cloud, You know, like what's the difference if we're actually then using or you know, tapping into, you know, one of these large language models on the back end. So can you just give us a little bit, you know, of the look on the privacy side and you know why it's important with medical records and health information.

Yeah. So, you know, the good thing for the US consumer is that the government has strong protections for your data. And that comes in the form of hipaa, which is a law that I think a lot of people have heard of nowadays. And so HIPAA has requirements. And one of the main things that I've run into in my work in trying to set up this platform is to work with other people, you need to have what's called a business associates Agreement, a baa, and that is the other party agreeing that they're going to treat your data according to all of these regulations. And starting out with our platform, that was one of the challenges because there weren't many, if any, providers that would actually sign a BAA to use these large language models through their API. But the good news for I think everybody is that increasingly the major cloud providers have their own APIs. And so nowadays you could get a BAA to use Amazon, Bedrock, Google, Gemini through Vertex AI. And I heard you mention, yeah, Gemini Studio still hasn't, which is tough because we can't get the latest stuff that comes out there. But Vertex AI, you can get a BAA as well as Azure, OpenAI services and even OpenAI now will has a process to get a BAA. So I think that's the good news for, I mean both people building companies and for consumers is that, you know, the providers of these models are seeing that it's important to set up the security and privacy infrastructure to be able to make these guarantees so that companies in regulated industries like ours can actually work with them.

And for, for our livestream audience, now's a great time. If you do have a question for William, please get it in. So let's talk a little bit here about the opportunity. Right. So you know, obviously this depends, right, Because I know that there's some very forward facing, you know, healthcare organizations that are really and have been for many, many decades, right. Been using traditional artificial intelligence. But you know, when it comes to this generative AI wave, large language models, William, where would you say is the biggest opportunity for healthcare organizations?

Yeah, I mean, I'll pick the biggest. I think there's a couple, but I think one that I'm cognizant of is like the physician burnout crisis. And this is something that we're doing a lot of work on and included health to try to reduce the administrative burden of being a doctor because people, people didn't become a doctor to fill out a chart like on an ehr and it's kind of a difficult task. So I think that's one of the biggest opportunities is, you know, we have a shortage of physicians in the country and if we could free up more of their time to be able to actually work with patients, which is what they got into this to do. I think that's probably the biggest opportunity with the tech that we have now is to say like the doctor can be face to face working with you instead of behind a screen.

Yeah. And it's, it's. I find it still like extremely frustrating for me personally. Right. Like, if I'm going to see a doctor, first of all, it takes forever. Right? But then, you know, the doctor's there sitting and you know, oh, doctor's running 45 minutes behind and you know, come in and ask you questions and you know, the doctor is over there just pounding on his keyboard, but very slowly. And I'm like, yeah, why can't we use some of this like, you know, voice dictation, right? So, you know, what are, can you help us all understand, you know, so these healthcare organizations that maybe aren't, you know, really yet using AI, like, why do you know, like, like I know that's a huge question, but like, why?

Yeah, I think one thing I can say from the work me and my team has been doing is integration is very challenging in healthcare. So like, I could build a demo using the OpenAI API and the tools that I have. But then, you know, we have several different software systems to then integrate with. Right. So for, for us, like we use Salesforce Health cloud and then we also have our own internal, like software services and maybe we have like Athena and so how to get that all to work together? I mean, that's not necessarily even an AI problem, that's just an engineering problem. But I think that's part of the challenge is modern healthcare. You have so many different software services that you use and the doctors run into this just as much as like software engineers. So I think that's part of the challenge is how do we get all these systems to talk to each other? You know, you can't do anything useful with the data if you don't have all the data together and available.

What are, you know, what are some of the challenges of, of that exact same thing, Right. Like being able to grab data from, from different systems that are maybe right And I, I could be wrong here, but I feel what a lot of, you know, healthcare organizations are using somewhat antiquated, you know, systems. You know, what, what are some of the challenges of, of making, of being able to grab all that data, you know, having the data be able to talk with each other and then using it actually, you know, for a large language model.

Yeah, I think part of it is output formats. So the industry has standard called fhir, but. And it's, it's starting to be more adopted, but it's still not like super, you know, you're not going to get everything in FHIR format from all of your services. And yeah, I think then it's just like, I guess one other challenge is how they represent a patient or a person. Right. So how do you connect the different identifiers used in different systems? That's a challenge we run into at included. Health as well, is like if I know that you are Jordan Wilson in my say ehr and I know you're Jordan Wilson in my proprietary app, how do I connect those Jordan Wilsons to make sure that I have all of that together? No.

Yeah, that. So many challenges. Right. And I guess another, you know, upcoming potential challenge. Right. Read, read these stories all the time about, you know, potential shortages. Right. Someone in our, our live stream here was, was saying this, Kofi saying, you know, shortage of positions, aging population that's living longer. Right. You know, nursing shortages. We've been reading about these since COVID What are like, how realistic are these shortages and if they are true, right. Like all these studies, you know, that are saying, you know, hey, by this year, you know, we're going to be this many nurses short, this many doctors short. How real are those and are they a huge concern for the healthcare system here in the U.S. yeah, the, the.

Physician shortage problem is real and we definitely should be concerned about it. I think both on an overall level and then one thing we see it included in health is also what you would call like health deserts. So even within the US there's geographic locations that don't have as good access to high quality physicians. And so I think there's a number of ways you can address that. We partly try to do that through telemedicine. So that helps because, you know, you don't have to drive two hours to see the nearest doctor. But I do think AI has a role to play moving forward. I think some things there, or at least one thing would be like virtual triage. Right. So if you could describe your symptoms to an intelligent bottle and Then it could tell you, like, just like, is this actually really. Should you go to the ER or not? Right. Or like, is this really urgent matter? I think some of that is in the works in the industry. I'm sure people are already working on that. But I think that could really help because, you know, you want the doctors to be working on the difficult cases, and maybe some of the other ones can be, you know, the bot tells you to go get some Tylenol because it's not a. It's not a huge deal. I think that's probably something that's coming in the future.

Yeah. And I mean, what's coming in the future? I think for me personally, it's inevitable. Right. Like, I think, you know, AI and using large language models for daily healthcare. It's inevitable. Right. So between the nursing doctor shortage, we just talked about, but some recent studies. So we shared this in our newsletter. But there was a recent study published in jama, I think that's what it's called. Right. So it showed, and this was an actual test, there was researchers from Stanford and other universities, and it showed that doctors, essentially there was doctors that went, you know, 50 doctors tested on six challenging medical cases. So doctors scored an average score of 76% if they use ChatGPT. If they did not, they scored 74. Right. So using CHAT GPT provided a slight bump for Doctors going from a 74 to a 76%. But then chat GPT without the doctors scored a 90 accuracy rate. Right. So there's all this information out there. You know, everyone's like, oh, I could never talk to a, you know, an AI, like, I need a doctor like William. How can we as, as humans that have always sought care from a human doctor, how can we begin to accept maybe this reality that, okay, maybe these large language models might be better suited to make some of these decisions? Is that crazy to think?

Are you still running in circles trying to figure out how to actually grow your business with AI? Maybe your company has been tinkering with large language models for a year or more, but can't really get traction to find ROI on Gen AI. Hey, this is Jordan Wilson, host of this very podcast. Companies like Adobe, Microsoft and Nvidia have partnered with us because they trust our expertise in educating the masses around generative AI to get ahead. And some of the most innovative companies in the country hire us to help with their AI strategy and to train hundreds of their employees on how to use Gen AI. So whether you're looking for ChatGPT training for thousands or Just need help building your front end AI strategy? Or you can partner with us too.

Just like some of the biggest companies in the world do.

Go to your everyday AI.com partner to get in contact with our team or you can just click on the partner section of our website. We'll help you stop running in those AI circles and help get your team ahead and build a straight path to ROI on gen AI.

Yeah, I think it's difficult. I mean it's something that everyone is wrestling with. I think it's something we're discussing. But to me it's, it's getting used to like a different way of looking at it psychologically where, you know, part of what a doctor provides is warmth. Part of what a doctor provides is when they tell you some bad news, they can give that to you in, in a way that makes you maybe feel not so bad. And I think, I think in the immediate future, say the next couple years it's probably going to be some kind of hybrid system. Right. Like the AI can be really good at diagnosis maybe, but you still maybe want a human to interpret that and deliver you the results. I mean, I mean AI has voice now, so maybe they could also tell you. And there's developments there. But I think at least in the short term, in terms of the next couple years, like people are not going to be necessarily comfortable with like total AI driven medicine. I think actually a good analogy is to self driving. Right. Like it could be that self driving cars don't crash as much. I think like there's data that shows that they're doing a pretty good job with safety but sometimes it's still uncomfortable to get in a car with no driver. And I think it's a similar thing in medicine. If you're talking about AI for diagnosis is like maybe in this case like the, the model diagnoses 90% and the doctors only get 76% but you don't necessarily want to get into that car right now. And I think that's a good point.

Yeah, yeah, it's like, it's like, oh, I'm curious by the self driving cars or the waymos. But yeah, do I really want to get in there yet? I don't know. Right. Part of it's like curiosity but then part of it's like yeah, it's probably, I would feel personally safer driving the actual car. Right. So William, earlier you mentioned obviously some high level use cases of generative AI. Right. And the opportunities relieving administrative tasks. Right. Helping address a potential upcoming shortfall of doctors and nurses, etc. Right. But as our AI systems become more and more advanced. Right. And less and less technical. Right. The ability for live video, live video that can infer in motion, live video that can see and we can interact with what are maybe some of the next, I guess, aspects of healthcare that we should look at because you know, like I said, you know, transcribing and dictation and note taking and organization, all the administrative stuff, it's like, okay, yes, but maybe where should we be looking next? Or where are you excited as someone working in this field looking at.

Yeah, I think that I'd call it a couple of things. I think you mentioned kind of multimodal models and I think that's like a, an area that is going to expand. So they're getting much better at doing things like reading X rays and, and other things, maybe even video like you can imagine, like a video like physical therapists like you, you walk and it starts to tell you how you could change your gait or to recover from some injury. I mean, that's exciting stuff. I think the other thing which you touched on a little bit earlier is just increasing use with your personal data. Right. So it's like, I mean, I wear a Fitbit, I gather data on myself and I kind of want to get some insights to that data. And right now I can see graphs, right. But it doesn't necessarily capture it. So you know, what if I could take my Fitbit data and like blood testing data and get the AI to interpret that and explain it to me in like layman's terms. I think that that's another area you're going to start to see is people using AI. And again, like, you want to be careful with this, but I think you'll start to see people using AI to, you know, get these answers from data that they're collecting on themselves because people are naturally curious.

It's, it's a hot topic. Right. And there's been a lot of kind of, you know, posts online that have gone viral in the last couple of weeks. You know, people literally doing just that, right. Collecting all their, all their healthcare records, you know, dumping it into, you know, chat, GPT or Claude or Gemini as these, you know, computer vision models get, get better and more accurate. And they're essentially just trying to figure out long standing medical problems that they haven't been able to get answers to. In a lot of cases it's working. What are the dangers? Right. In, in that. Right. Or you know, kind of like what you said, these wearable Devices are collecting more and more information. Right. Should we be exporting all the information from our smart devices and our results that we get online and, and you know, using AI to try to figure out, you know, things that have been nagging us from a healthcare perspective, is that good or bad?

I think, yeah, I think the dangers of kind of do it yourself and, and I will say like I, I've totally done this. I've gone to chap GPT, put in symptoms or put in the results from a test and I've been like, what does this mean? But I, I think the dangers are like it's not really validated to do that. Right? Like it, it can do that, it even can do that well, but you really going in, don't know like how well because that's not really something that it, it gets benchmarked on. And I think that's an opportunity for healthcare companies even like ours where we can say, okay, we are using say the same models, but we've actually done the work to validate like okay, if we ask it health questions, is it going to give you like answers that make sense? And so I think that that's where there's the opportunity to say and also on the privacy perspective to say like, okay, well like we're subject to hipaa. If you send us like all your data, it's not going to get used in some nefarious way. So, so that's, those are the, I guess the risks of like just using it from a consumer app is like you don't know what you're going to get. I've gotten good things. I've also gotten things that I needed to kind of take a second look at. So I would definitely caution people it can be useful, but I think that there's going to be an increasing space for, I mean like I said like WebMD plus LLMs, right? Like has it been somewhat validated by clinicians and, and people who went through and said okay, like can we be more sure that this is doing the right thing versus like a vanilla chatgpt?

Great, great comment here from Samuel that I want to get your take on William. So he's saying, I think patients will have justified concerns with the privacy of their conversations with their doctors if an AI is listening in. Right? Yeah, that's, that's the big right. One of the biggest hold ups because this technology that simply just, you know, uses AI and transcribes conversations, I think I had the president of the American Medical association on just under a year ago and I Think the stat at the time was like, only 30% of, of their members were even using this information. Right. So as, as someone, William, that's building, you know, AI technology and healthcare on the technical side, how can these things be addressed? Right. Some of these seemingly, you know, simple or simpler, you know, integrations of large language models into healthcare. How can these things be addressed?

Yeah, I think the first thing is definitely like putting the power in the patient's hands and getting explicit consent for some of these things. Right. So if we're going to record your conversation, we want to tell you that and make sure that you are agreeing to that. I think that's a very important part, giving you the opportunity to opt out. And then part of it, I think, is just building trust with your users. Right. So if you're a healthcare technology company, you're getting a lot of sensitive data about your users. Like, if you weren't, then you wouldn't really be able to do anything useful. So part of it is just showing the user that you can be trusted, and I think the other part is showing them that what they're giving you can be used to their benefit. Right? So, like, nobody wants to give you a bunch of data if you're not going to do something for them, but if you can, you know, do something beneficial, then maybe they have a willingness to say, okay, I'll send you this test because I know when you send me back the explanation, like, I get a benefit from that. And so I'm kind of, everyone is making this trade off of how much data do I share for what I'm getting back. And I think the hope for healthcare companies is that we can give people back enough that they're willing to trust us with some of this information in order to, like, do the job.

So I know, included health. One of the things you all do is, you know, personalized virtual care. Is there, is there an opportunity in the future for, you know, a different kind of healthcare system, at least in the U.S. where, you know, I would love this, right? Where it's like, I don't care about my privacy. Yes, Blank healthcare, whatever. Take everything, take it all. I just want to be able to, you know, as an example, talk with a, an AI all the time or talk with a doctor who's using AI and can get me questions or get me answers so much quicker. Is this something where we might just see a complete change in how healthcare works in the future?

I think so. And I think this is a combination of AI plus some of these other trends we've mentioned in terms of, like, personal data gathering. But I do think that's something that's coming where, you know, even with virtual primary care, we're realizing, like, primary care isn't just coming into the office once a year and maybe doing some tests. Right. Like, that isn't really sufficient in a lot of cases, even for a healthy person. And so, you know, can we have you wearing a wearable to. To track some of these things? You know, CGMs are available, there's startups that will send you that to track your blood sugar. And, you know, consumer blood testing is on the rise. So I, I think that. I think it's coming where there's a health care company that can actually integrate both traditional medical data versus this data that people are starting to collect on their own, because I think there's a big trend there. I'm. I'm one of these people. I mean, I love to collect data about myself and my health. So. Yeah, and then just having that all in one place and saying I can talk to an AI about it, that has been validated by clinicians to be giving me, like, useful and good answers and, and that I can escalate to a doctor if I need be. Like, I think something like that is coming in the future, if it's not already here.

All right, William, so we've covered a ton in today's conversation, bouncing all over the place. It's been a fun one for me. But, you know, as we wrap up, what's the one most important thing that you think our audience should know when they're thinking about the role of AI in modern health care?

Yeah, I think the main thing I would say to people is that it's not something to be afraid of. I mean, it's totally valid to be afraid of it. I think there's a lot of reasons to, but I think ultimately we want to use AI to give more power to the patient. And that's kind of my mission. And I think that's something that hopefully we'll see in, in the years to come. But, yeah, it. It can be a very scary tool, but it can also be a tool that gives you abilities that you didn't have before. And I think that's really the. The optimistic side of the view of AI and healthcare.

Power to the patient. We can all get on board with that, right? Yeah. No one, no one's gonna argue with that. All right. This was a great one. William, thank you so much for taking time out of your day to join us. We really appreciate it.

Thank you so much for having me.

All right, as a reminder, y', all, that's not it. There's always more. We're going to be recapping today's conversation and everything else you need to know to keep up with AI. So if you haven't already, please go to your everyday AI.com if you found this helpful, please subscribe. If you're listening on the podcast Apple Spotify, please follow us. Leave us a rating. If you're listening online, tag someone who needs to hear this right? We bring you the experts. You can ask questions, but also so you can be informed and keep your network informed as well. Thank you for tuning in. Again, go to your everydayai.com Sign up for the free daily newsletter. We'll see you back for more Everyday AI. Thanks y'. All.

And that's a wrap for today's edition of Everyday AI. Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. It helps keep us going for a little more AI magic. Visit your everydayai.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.