Can Agentic AI Fix Healthcare?

AI assistants like transcription co-pilots and chat-bots are yesterday’s news. Now, we’re entering the new era of agentic AI. The new tools are powerful but a little unsettling. AI giving advice is one thing, but are we ready for AI to take action on our behalf?

In this episode of CareTalk, hosts John Driscoll and David Williams explore the dawn of agentic AI, systems that act autonomously to perform complex tasks, including in healthcare.

🎙️⚕️ABOUT CARETALK
CareTalk is a weekly podcast that provides an incisive, no B.S. view of the US healthcare industry. Join co-hosts John Driscoll (President U.S. Healthcare and EVP, Walgreens Boots Alliance) and David Williams (President, Health Business Group) as they debate the latest in US healthcare news, business and policy.

GET IN TOUCH
Follow CareTalk on LinkedIn
Become a CareTalk sponsor
Guest appearance requests
Visit us on the web
Subscribe to the CareTalk Newsletter 

Transcript

David: 0:00

AI assistance like transcription, copilots, and chatbots are yesterday's news. Now we're entering a new era of agentic ai. The new tools are powerful, but a little unsettling. AI giving advice is one thing, but are we ready for AI to take action on our behalf? Welcome to Care Talk, America's home for incisive debate about healthcare, business, and policy. I'm David Williams, president of

John: 0:30

Health Business Group. And I'm John Driscoll, chairman of Yukon Health. David. David. David, do you think the a AI is and the, the artificial intelligent robots are about to take over our lives? What are you, what are you getting at here? Well, John, we've,

David: 0:46

we've seen what kind of, uh, impact generative AI can make, but I understand that we are, you know, moving quickly into the area of agent ai, which goes beyond generative

John: 0:56

ai. Well, just, you know, you, you consultants, you know, you have all this confusing, so, okay. SmartyAnts, what is ai? Let's start with ai, gen ai and then agen AI so that we at least understand the wacky terms you're using. What is ai?

David: 1:12

The part I could probably answer is I thought you'd say, what's gen? Like, what's Gen Z? I could probably talk about that. I mean, so AI is, is artificial intelligence and it's basically, you know, like a alien intelligence is another way to describe it. It's just some intelligence that's being, that's, you know, intelligent things that are being done by machines. And it used to be thought of as, okay, something that's kind of. Stimulating a human, but it isn't necessarily John. It's like an intelligence that's taking place outside of the person. You're caught up

John: 1:39

in your sci-fi fantasy. Artificial intelligence is that intelligence that's basically comparable to big compute, large computers, crunching lots of information, giving us information that is modeled historically, like the way computers answer and ask questions, but at a speed. And a reach using lots of data that just generally unheard of. It's, it's the original version of AI is can we create a computer that can think faster, smarter, better than a human? That comes from the late fifties when it was earlier set up, and it, and has really been beyond reach and in the last few decades, the, with the advent of more computing power. More powerful chips, more access to, to data service and frankly, more standardized data because the machines need standardi, historically standardized data in order to learn or create models that a AI or the promise of big compute has started to be delivered. Now, it, it, we've very quickly moved from a period when the ma, the, the machine could beat. IBM machine be big. Blue could beat the great chess master and Google's Alpha Go. Another AI model could beat the world's greatest, um, uh, player of Go, which is a highly complicated, uh, multilayered intellectually game. Strategy, but it's again, a board game with novel approaches. And the interesting thing was not just crunching the, the, the, the, the movies faster, which was true and smarter, which is true of chess, but actually creatively that we've now moved to something called, uh. Uh, generative ai. Now, what exactly is generative ai? What's it generating? The way I think

David: 3:34

about gener generative ai, it, it's actually producing content in response to a prompt. That's what it's generating. It's generating content. I think about it like a creative writer. Or a tool where you give it a prompt and it can give a response. And whereas Ag agentic ai, which is where we're going, is ai, where it's an agent that's acting autonomously and it's like a project manager that's gonna orchestrate a lot of different tasks that are going, that are going on, it's gonna do that, uh, for you. But you're right, John, to start with a regular. Ai, which is, which is impressive. And it is, in fact, I'll go back and say, as an alien intelligence,

John: 4:08

oh dev, to get off with your star trekky sci-fi a October Halloween surprise nonsense. Let's, let's, let's, let's, so we've gone from the power of computers growing. Nearly exponentially in certain areas. The ability to, you know, and, and, and, and gen AI being things like chat, GPT, clawed, Gemini, the big tools, um, Misra in Europe that people are using in a prompt is simply a. Targeted attuned question and context that then allows the machine to come back with content. Uh, a lot of people, grok would be another example where, and there's a massive battle among the big tech titans to build the biggest and the most, the most expensive models. And these valuations are going through the roof and Nvidia creates the. The chips and all that. We don't need to get into how it happens, but Gen AI is something that has grown faster than any other software tool or perhaps consumer tool ever in American history. It's growing 400 times, I think, as fast as search grew. So it's a big deal because even in, in some cases it's being integrated into business, it it's everywhere being integrated into white collar work, and now we're, and that is a. To your point, you create this targeted series of questions and context called a prompt. And now we're moving to agentic ai. So this is where the robots take your job, David. I mean, what is, what is agent? What kind of agents are we talking about?

David: 5:45

Yeah. They could take my, my volunteer job of, uh, you know, trying to arrange, arrange all sorts of nonsense, uh, in the world. Uh, it's where they act, um, you know, sort of on, on my behalf. So they're gonna go out and instead of say, you know, me going to a website and setting something up or searching for something, uh, some, something to shop for and then, uh, buy it and approve it and decide where it's gonna go, it could theoretically do all of that for me. So you're seeing applications in places like shopping. And I think, yeah, you could see how that could work. And there's earlier versions of that. Uh, you know, you can imagine picking out your airline and your hotel and all that. But, but Dave,

John: 6:22

Dave, what's different? I mean, people have had shopping bots that are working either for shoppers when you're on a retail site or the retail site is using it to respond to giving you the next most likely thing to buy. Suggesting the next model, uh, next movie for you to watch. What's different about agentic AI and, and effectively the kinds of things that Netflix and Amazon is already doing, which has some form of, of large compute and, and intelligence associated with it.

David: 6:54

So like a lot of these things, the, uh, the dividing lines aren't so crystal clear, but in general, the boss that you have now. Are more like the, the chat bots, so like chat, GPT, similar idea. I can ask it. They can tell me and could suggest this and that. This is more, I'm gonna say, you know, John, would you mind going out and just like booking this trip for me? Like think about like a travel agent. I don't say to the travel agent, Hey, what's this? What's that? I say, I, you know, I need to go here. You know, my preferences in general, go and book it and send me the itinerary. It's more like that.

John: 7:28

Well, and I, and I would go even further, David, now let, let's, let's shift into healthcare. You know, I'm, um. On the board of an agentic AI company and the, the, the, the agents, if you will, have the ability, the computer generated agents that that can interact directly with providers, doctors, n nurses, et et cetera, and patients to answer complex benefit. Coverage and design questions, how conditions about, um, that that can go with certain drugs, the, the drugs and fulfillment. And so effectively we're taking some of the roles that would've been handled by call centers, if you will, and hubs for pharmacy and really. Turning them into agentic AI engines that are 7 24 available that have complete information that can be completely updated. And in many cases, the bots appear to be as sympathetic or more than the people who otherwise would be handling and, and, and parsing through all the bureaucratic. Nonsense that, that, that, that really sits in the middle of people, people needing care and people getting care. And so I, I think this ag agentic ai, when you can actually set up a, an interactive computer interface that's got some intelligence associated with it, we may actually be able to use these things if we're careful because it's healthcare, we're talking about to simplify and. And accelerate people's access to care. But that's the positive case. David, this is healthcare. What could possibly go wrong?

David: 9:06

Well, you know, in, in your previous question about whether it's gonna take my job, and I know it's not that you were joking, you were just looking for something to stick me with, which is fair. That's part of the rules of engagement here. What AI is doing at this stage is actually taking the jobs of some of the people in the outsourced call centers, let's say in the Philippines or in India, that can handle the basic questions. Now, as you, as we've been discussing before, I'm in the process of helping my, uh, parents who are in their eighties to make a move. And there's a bunch of things that have to be done with, uh, with healthcare and my experience of being on the phone. You'll talk to a person, say an agent. Then they invariably have to check with their supervisor the next level of support you get handed off in a lot of places. So what I'm now thinking about is, you know, the Syngen ai, maybe it's that first level before they say, Hey, I have to check with my supervisor because you know, you aren't authorized to talk about your father unless we do that, where they're, oh, they're also moving and not just changing plans. The main processes can't seem to handle that today without a lot of heavy human interaction, and sometimes has errors. So I think what's gonna happen is it's gonna, we're gonna start with a bot. It's gonna be almost more like dealing with that entry level piece. Now, maybe you could flip it around and say, actually no, the entry level's gonna stay as the person just, but the, the agentic part is gonna be dealing with all those supervisory requirements. Maybe it goes that direction. I'm not exactly sure.

John: 10:24

But I think what you're pointing out is that there are a lot of breakdowns in the car cust, the current customer facing applications to access choice visits, bills follow up. Changes that at least theoretically the agent AI should be able to replace completely because it's got much better information. And it is, it is. Um. It, it's it, you know, it, it can perform with incredible consistency if it's programmed correctly, to constantly be upgraded and updated to get the patients the access to the care they need. And, and, and by the way, when they're available, it's not held to any human schedule. Yeah. But again, I go back, Dave, to what could go wrong.

David: 11:15

Right. Well. I know I was actually gonna go the other way, John, and say what could go right. What would you like to see? Because I, I think we can, I was just thinking we can very quickly get into why this isn't gonna work. What are some things we would like to see happen? And one of the things is that when you have an interaction with a healthcare system, a lot of times because you're sick or somebody you're helping is sick, you're under stress, you're, you know, you don't wanna deal with a lot of new administrative issues. So what are some things we might like? Well, one issue is that after you're coming out of the hospital from a discharge, make sure you get to the right doctors. Make sure that your prescriptions are filled, make sure someone's gonna come and see you. I'd like to see those sort of things done. That would be one use case that I think would be good. Another one we talk about sometimes is, you know, gaps in care that get flagged, which means basically, hey, somebody's not getting what they, they need and as a result, they're probably gonna end up in the emergency room. We wanna actually identify those things and intervene earlier. Those are a couple places that I would love to see. Uh. These agents actually take over for me and I know what I want. I wanna come outta the hospital, I wanna get my appointments, I wanna get my medications, I wanna get my follow up, and I wanna deal with the, with the insurance. So, I mean, those are some things that I would, I would like. Now I will answer your question, what could go wrong? So here's one thing that could go wrong, John. I saw a headline about this the other day. Um, my, my AI agent just got phished. Right. So, you know, uh, not, it wasn't, it was a headline, it wasn't an email, uh, to me. But, you know, there's like, there's things like that could go wrong in healthcare or not.

John: 12:40

Well, I, I, I think Dave, the, the reality is anything can go, but, but we have a system right now that's kind of lousy for consumers to access. It's hard for doctors to kind of do their work and there's a lot of just friction in the process, and I still think we're dealing with. The long tail of, you know, of the, of people working incredibly hard during the, during the, the, the COVID crisis and not getting a break. I mean, we, we are, we, we will be hundreds of thousands of clinicians short over the next decade or so in terms of hospitals and nurses. We hopefully won't continue to tighten, uh, the access for foreign medical. Professionals to come to the United States, but that's a real challenge. One of the few areas of I think, uh, white Space and hope is that we can leverage tools like Ag Agentic AI to replace humans. And I honestly make doctors and nurses, social workers and psychiatrists and psychologists jobs easier as well as the administrators of all these hospitals and doctors' offices. I mean, I, I'm actually quite, um, you know, quite. Quite optimistic that the power of having seen, um, that, you know, through our own deployment of AgTech AI with complex call centers, I think we could do that for call centers in health plans. We could do it for doctor's offices and just the explosion in ambient listening where the, the, it's called scribes. Where, uh, everyone from Epic to some early stage startups is built and, and, and, and, and Athe and open AI are building tools where the doctor no longer has to be pit pattering in entering lots of data into the Epic EMR, they can actually make eye contact and listen to the patient and then the. The, the ambient scribe, the agent scribe that's listening to that patient interview and patient interaction can code that in ways that would identify the next most important action and, and also autonomously code it into what should be more likely to be an accurate and complete bill, you know, which is, uh, to to, to sort of shift the responsibility of the doctor and the nurse from generating the perfect bill to perhaps generating the perfect care.

David: 14:59

John, a lot of the things we've been talking about here are, you know, administrative in nature. Either it's things that are being done in, in their essence, like in enrollment, uh, with nonclinical staff, administrative staff, or a bunch of administrative things that clinical people have to do. And I see opportunities there. Uh, you know, my concern is that the processes aren't robust enough that then if the AI goes in there, it may cause new problems. But I see, I do see a lot of opportunity if that can be done well. What about on the clinical side? So actually getting, you know, better care, um, or getting it faster.

John: 15:32

I think if you think about the fact that you've got this brain, this clinical brain that is pulling in all of this information, you know, it typically takes about 15 to 17 years between the time that. Doctors, the academies, the, the most, uh, uh, important authorities in medicine agree on an advancement of care or a novel thera therapy or therapeutic to the time it's actually deployed consistently with the in, in the places where it should be. That should go to almost zero. Once you've got agentic AI partnering with the human in the loop, the doctors to actually upgrade that care. I think you could develop, you could you, we should be able to develop a therapeutic intelligence layer that could really make care coordination and care enhancement. Um, and what the, the. More consistent, um, and, uh, and more and more intelligent and, and, and hopefully lead to better care faster. Let's talk

David: 16:34

about maybe, um, from a primary care physician, how this might change their, their world and what they do. So now, uh, patient comes in. And they, you know, discuss things with their physician. A physician may be able to handle it themselves, or they may make a referral. If they make a referral. They do it based on who they know or who's in network and things like that. But maybe if you think about it differently and you consider it empowering primary care with an ai, uh, agent that they have a problem and they could go out and say, find me the best specialist that could actually, uh, work on this. And actually don't just refer the patient to the specialist. Let them go through the friction of getting the. You know, the authorization to see the patient, to see that, uh, that physician and then getting the appointment and so on, but more physician to physician interaction with an agent. We take it the other way too and say those specialists that have a lot of knowledge, some of them may have written textbooks or written articles, could actually create even avatars of themselves that start with what they know and then the avatars may be able to interact. That, those may be agents actually, and they may be interact.

John: 17:37

You don't have to go to your, you know, again, your sci-fi movies to say, look, there's gonna be intelligence at your fingertips. That and a effectively an intelligent companion that's gonna help support the decisions. I mean, healthcare is super complicated and people are super fragile. And then you get to billing reimbursement formularies protocols. So you've got three real levels of complexity. If the doctor can just primarily focus on the patient and the not the billing. And be as intelligent as they possibly can be. Also nurses and social workers and psychiatrists, everybody. I think that's a big boon for the patients. Now, I, I still think that, that, to your point, these are, these, the, the, the every technology is fragile and we need to harden them to keep them, protect them from cyber ev and when you're doing a novel technology, it can, it can definitely, um. You know, the, the technical world is literally hallucinate, and believe me, they do. The other day we had, we were testing, playing around with chat, GPT, which is one of the larger, uh, uh, agentic models, uh, perhaps the most advanced one in, in terms of compute. And it was arguing with us as to whether the East Wing had actually been demolished and whether there was a ballroom being built. So it can, it can really, um, these models are built on vast amounts largely of internet. You know, informed information and sometimes the signals can cross as they do with people. But, but, but, but e but, but remarkably, factually wrong. So I still think we are at the level of anything in healthcare has to be held to a higher standard. But gosh, if you were to pick a, an area of, of. Work where you need more intelligence, healthcare, which is full of administrative costs that need to go away. Healthcare and where it is really where, where it's, where we don't have enough labor to cover the people. I think that it's, uh, I think this, I think healthcare is prone to be one of the best areas. Now Dave, you, you haven't really been willing to take the bait on what can go wrong?

David: 19:57

Okay, I'll try. You know, you know, I say take debate, John, that like, it's, it's not a criticism of the first. Okay. Lemme, I

John: 20:05

thought you, I thought you'd done all the research, but I, I read a really interesting piece of research. Yeah. That, that, you know, one of the things that, one of the places

David: 20:14

Oh, I was just about to take, I was just about to take debate. But you go ahead. Okay. Go, go for it. No, go for it. Here's what I'm worried about, John. So healthcare you, you ticked off all the reasons why it's a really good candidate for agent ai and I agree. Now, here's the issue though, as you talk about coming outta the pandemic, one of the things that's very important for healthcare and that's been lost to a great degree is trust. People don't trust the system. They don't even trust their own doctor anymore, or the pharmacist and nevermind politicians or people running healthcare systems. So to me, it seems that if we could actually. And it seems to me all else being equal, that introduction of AI is gonna reduce trust. So if we could flip it around somehow and find a way to have it increase trust, I think that would be very powerful. And really the only way it's gonna be successful in healthcare.

John: 20:59

It's a really powerful point, Dave, because without trust you, you're gonna lose patient follow up and follow through. If you don't have that, then honestly, AI is a problem. The other interesting problem that's emerging, and it's a fear of a number of people in medical schools is does the AI itself make us. Stummer because we no longer do, can't do calculations by hand. We don't longer write'cause we let the AI write. Well, there was an interesting study of, uh, doctors, I believe it was gastroenterology, gas, GI doctors. And the, the study was okay, if you have doctors, how well do they do in diagnosis? Test. Okay. They do well. How well do the doctors do with diagnosis? Assisted by chat. GPTI believe chat, GPT could have been another AI agentic AI partner. They were better, but when they took the, after having trained them on how to use ai, it took the AI tools away. The doctor skills were worse. After working with the ai, now it's a small study. It, it, it's, it's not, it's indicative more of a question than a conclusion. But one of the other interesting things is, you know, we, we rely on discernment, judgment. The accumulation of a lot of knowledge for the best and most complex diagnoses that we all want to have access to. We assume that the, the, the, the tools of technology would help that. Well, it definitely helps around innovation and esoteric research, but we really have to somehow find a way to design it that in addition to lowering costs and helping workflow, we actually make sure that the doctors and the diagnosticians and the. Psychiatrists and social workers all become more intelligent, not less when we put these tools in. Because if you think about where trust is gonna fail, is if people start to note that a, uh, an office without the, your bots, your holo crimes, your avatars, and your, your, your David, your ambient David Williams, uh, is not doing as well as the, the, the, the ones with just, uh, you know, doc Dr. Weber or whatever.

David: 23:13

So John let's you know, I know you accused me of being a sci-fi guy, which I'm really not. But let's, let's, I think a very interesting question is whether AI is just like another tool in technology. Or whether that something's fundamentally different. So let's, let's, let's go forward with your discussion there about the physicians and the de-skilling, which I do worry about. So if you talk to these old school physicians, and I'm talking physicians that are like 60 plus old people trained to palpitate, well, you know, they're trained to. Lay hands on, listen to the patient, listen to the chest stethoscope and so on. What they will often say is about their younger colleagues, they don't actually know how to do that.'cause they just, they take the image, right? They just go do the m mri, the ultrasound, whatever it may be. And so they don't have that skill. And so, um, you know, and there's different reasons for that, including incentives in the healthcare system, which is, I can get paid to do an MRI or I'm gonna go outta business if I just listen to your chance. Put that aside and say there's already kind of that trend. Now, is the AI doing a diagnosis, just one more step like that, which we can deal with? Or is it something fundamentally different and in terms of how it's gonna grow and change, you know, what are we looking at in 20, 25 years? What is a physician, you know, is it better actually just to have a lesser trained person, let's say a nurse, social worker or whatever, who's able to interact in person with a patient and then use the intelligence? You know, from the. To do everything that those physicians could do before and maybe, and maybe do it better. Or are we saying somebody who's like a brilliant neurologist or GI now, are they gonna be empowered somehow and be like, really, really good or not? To me, I don't know. I, I, I wear more good in the area. Well,

John: 24:53

I, I think, I think that what, what you really are, and this is probably a good place to wrap 'cause we will be digging into this with practitioners, with people who understand AI, is if you reset. If you say you're, there's sort of a pre AI era, an era where people were dealing with a a with knowledge, training and care in one world, and we've really reached an inflection point that care is now different. We should re-underwrite intellectually and clinically how we're going to actually architect care because otherwise you run the risk of assuming the machine is better or worse. The doctor be is better or worse and the things are gonna work like they did before, when really, if it's really truly novel, everything may have changed. And I think while certain areas like um, the law, which is very rules driven. You can see an absolute replacement call center. You could absolutely see full replacement healthcare. I think we really wanna step back and say, what does all of this mean? And how do we think about the architecture of everything from the training of a doctor and a nurse? To the care of a complex elderly patient, to how we do drug treatment protocol and, and clearing insurance in a way that we actually tech test whether we're getting something better or worse or different and different that we have to then model differently or architect differently.

David: 26:20

That's it for yet another episode of Care Talk. We've been discussing the rise of agent AI and all ai, and how quickly and broadly it will change our world. I'm David Williams, president

John: 26:30

of Health Business Group, and I'm John Driscoll, the chairman of the Yukon Health System. If you liked what you heard or you didn't, we'd love you to.