AI in Healthcare: Full Breakdown & Transcript

[00:00] Welcome back to the Stanford Healthcare AI Podcast.

[00:15] We are thrilled to be joined by a friend, Eric Larson, who is president at Towerbrook Advisors, venture partner at Thrive Capital, venture partner at SignalFire.

[00:27] He is an investor in us at Qualified Health as well, and really just a healthcare visionary working for 25 years and being president of the advisory board. And being known by my name is Whisperer to what is coming next in healthcare.

[00:42] Welcome, Eric. Justin, it's an honor to be here with you and Matt. I obviously think the world of you both. And proud investor in Qualified, Rocketship, you guys are doing amazing stuff. So it's an honor to be part of the pod.

[00:55] Appreciate it. Well, there are way too many topics to discuss, and even as our pre-conversation recording live, there's too much to go on. But we have to start with maybe the most publicized recent paper on autonomous AI exceeding positions

[01:16] that Zeke Emanuel, Neil Kozlov, and Ed Kozlov wrote just a few weeks back. I'll pull up the headlines here, but Matt, as a practicing physician amongst the three of us,

[01:28] what's your take? Let's get us started. I mean, so, okay, so just for the audience, unless you're under a rock, you've definitely heard about this paper, or you've at least heard a take on it. And this is just sort of the kind of culmination of some,

[01:40] I would say maybe cherry-picked to be slightly provocative, but in terms of the evidence that they kind of brought to bear for this, you know, largely opinion piece, but grounded in some reality. And I think we've seen some of these trends and talked about some of them, and they include things like when you have these AI versus physician kind of papers, and then you do AI plus physician, and you look at these different tasks that they ask, you know, folks who do these academic papers, we keep finding that AI plus physician is actually worse than just the AI alone, which often beats, you know, physician.

[02:12] And so the conclusion that they're kind of taking this logically towards is, you know, should we just let AI autonomously deliver care? And, you know, which, again, is very provocative.

[02:25] It's caused tons of hot takes and reactions. I think of all the different reactions I've read, I align probably closest with Bob Walker's response, and he wrote a really nice substack on this. And he kind of called it a doorman problem.

[02:38] But the point is that if you narrow things down to a task, and this has happened to us in radiology since the beginning, if you're deep down looking for pneumonia, that's like 1% of what I do, right? And I think it's the comprehensive collection of tasks that we're still not putting our finger on in terms of a benchmark.

[02:55] Now, the better question I think to ask is not do we just throw AI out there and have it take care of all patients. It's more how do we need to reinvent the practice of medicine so that we get the benefits of what this can deliver in terms of access to care, better knowledge, better, you know, prescriptions, better eye care, you know, surgical decisions, whatever those things are, but then also allow us to cover the rest of the tasks that we do.

[03:21] And I would also argue that when you follow someone around, I think Graham had a really nice talk about this. If you follow someone around on a given day, within five minutes in the clinic, you're going to be like, oh, wait, okay, we're going to need a lot more advancements, a lot more capability to do the 19 things I just saw you do in the last five minutes.

[03:43] And so, like, it's a balance to me. And I don't know the answer. We've been talking a lot about administrative tasks being the place to really focus and freeing up physicians. But I'm not sure that's necessarily the right approach either.

[03:56] I'm curious to hear what you guys think. But to me, it's like there is a, yes, I would argue that any benchmark you can create, you can hill climb and you can certainly beat humans on. The question is, are these benchmarks reflective of reality to the extent that they need to be to draw these conclusions?

[04:12] And I just don't know yet. I'm just not convinced yet on the benchmarks. well maybe just an old pile on i mean i'm not a clinician so i proceed with a ton of humility here

[04:26] on the topic but i do this article this opinion piece really resonated with me and obviously we're you know we we think the world of the node and neil and zeke we do a lot with them i will say

[04:40] that Vinod has been probably the most clairvoyant technologist for the last 75 years, maybe with the exception of Ray Kurzweil, in terms of his predictions, him being super vindicated in his predictions.

[04:53] And in 20, I think it was 2016, he wrote Doctor Algorithm and the 20% Doctor. And, you know, and that was a really inflammatory piece, even though I think it's going to prove to be totally prophetic.

[05:10] But everybody sort of rose up in arms. And I've noticed the node over the last few years get a lot more sort of like diplomatic and calibrated. The doctor algorithm was pretty incendiary, pretty flamethrowing.

[05:25] Now, you know, I think this is a really measured piece that is a meta-analysis of a lot of other industries too, So basically suggesting that, you know, in these five cognitive domains for a clinician, you know, the AI is demonstrably superior and more accurate and has more fidelity to ground truth.

[05:46] But the bigger intuition is that it's kind of this, you know, Garry Kasparov, 1997, Deep Blue, getting defeated. And that was sort of this big epistemic kind of like shockwave.

[05:59] And then, you know, you have this period, I think the last time a human beat one of the real premier chess models was in whatever, the mid-2000s.

[06:13] Then you have this period from 2005 to 2012 where this centaur idea, Gary Kasparov actually wrote a book about human-machine sort of optimization. And from 2005 to 2012, human-plush machine was ascendant.

[06:28] But then after 2012, the machines beat everybody. And what Vinod and Neil and Zeke are saying in this is that superimposing human judgment on top of a functionally verifiable domain actually degrades the accuracy of the AI.

[06:46] And everybody's freaking out, and they're finding all sorts of methodological flaws in the study. And Matt, you're commentary on it is super principled and measured, but a lot of the online stuff has been pretty ad hominem and kind of stuff.

[07:03] And I think there's a real meta intuition in this, which I subscribe to, which is if you look at the trajectory of the cognitive ability of the AI in every epistemic domain,

[07:16] starting with areas of functional verifiability then going to codifiability then things you can decontextualize you know one of our refuges used to be tribal knowledge or embodied knowledge or

[07:28] tacit knowledge but it turns out Eric Brignolson just wrote this amazing study your colleague at Stanford that a lot of the tacit knowledge is now discoverable

[07:40] or learnable he uses autonomous driving as an example so I guess where I land is I mean, go ahead and find methodological flaws in this perspective piece. Go ahead and, like, rage against the dying of the light.

[07:54] But the fact is, I think we're going to see the prestige professions systematically dislocated by this ever-increasing synthetic Jupiter brain.

[08:07] and the implications for medicine are huge because our entire our entire organization of our six trillion dollar industry has been predicated on scarcity scarcity of cognition scarcity of it

[08:21] takes 12 years to train radiologists and therefore you can't you know there's no elasticities of supply i think what the node and zeke and neil are pointing to is that humanity get ready like

[08:33] you are not the apex predator. And I think the implications for medicine are very good if we can soar this transition and, you know, and compensate these self-sacrificial physicians

[08:46] who spent 12 years of their life and incurred $250,000 in academic debt. So anyway, that's my take on this. I love the piece. I thought it was really profound.

[08:58] Well, just a really quick reaction before Justin, you jump in. I will say, like, I think that what I'm talking about is we're falling into this, we're talking about one thing in an exponential. And any discussion around an exponential, to me, is hard, right?

[09:12] And I think it's much more looking at, okay, the trajectory, the slope is much more important than the snapshot, right? And I think, again, I was talking about this yesterday, I'm a little worried we're not thinking delusional enough.

[09:26] Like, even this piece, to me, isn't delusional enough. And the reason I say that, it feels to me like I did a public health degree, and, you know, one of the things we learned about exponential spread, the COVID being such a beautiful example in some ways,

[09:38] it feels like we're in that stage where, remember when we were like, hey, if we don't let that cruise ship of COVID people dock, we'll stop COVID from spreading in the U.S. So let's not let them dock, or let's ground some flights.

[09:50] But we already knew the exponential, it was already gone. Like, we already lost them. They are not. Right. So like to me it like I feel like we having some of these debates and again they useful to have everywhere every corner of medicine but at the same time time it the slope that I missing and you looking at that as saying guys it already decided almost

[10:15] So we have to start reorganizing society, how we think about every institution, but certainly health care is right smack in the center of that because it's tomorrow, then the next day, then the next day,

[10:29] we're progressing at a rate that the current discussion isn't going to capture. You're saying something really important, and Justin, sorry, Matt and I are monologuing. I'm going to say something really quickly and then shut up.

[10:41] You know, L.T. Morris, who was this MIT sort of, you know, luminary professor, wrote this amazing piece in 1955 called Man, Machines, and Modern Times. It felt a little anachronistic in the title, but Man, Machines, and Modern Times.

[10:56] And what he said was that when a society is confronted with a new radical technology, it reacts in a pretty predictable, formulaic way. First, it ignores it.

[11:08] Then it seeks to rationally rebut it. If that fails, it mocks it and uses ad hominem to make fun of the champions. and then only when a leader of sufficient moral authority and energy comes in and rides in over the top does the adaptation happen.

[11:26] And I think you're going to see this with law, consulting, finance, and medicine, where the guild, which is at the absolute top of the hierarchy, both in competition and societal prestige,

[11:40] you know, is seeing it's kind of threatened. And it's going to react in this very formulaic way. And I just saw the full spectrum of this. I saw people, some people ignoring it, although we're past that.

[11:56] I saw the rational rebuttal, sometimes really intellectually rigorous, but mostly just like, you know, oh, they cited the wrong study, or they didn't include a methodology section. I mean, give me a break.

[12:08] Sorry, we love to curse in your pod. And the third is mocking them. Oh, this is a multi-billionaire, and his son has a company, Curai.

[12:22] They're doing the ad hominems. You will see the guilds respond with vigor. Now, it was really smart to bring Zeke into this, because Zeke has such credibility,

[12:34] and the combination of Vinod and Zeke and Nero is pretty formidable. I just think, look, I'm glad this has incited this level of conversation, but to me, the irreducible big thing is we have deified our intellect, and that is the top of

[12:54] the hierarchy. That's the thing that we kind of worship in a secular society. Suddenly, we've speciated something that not just rivals, but in some emerging domains, and Vinod and others

[13:06] point to the five cognitive functions, we're not as good. What do we do in this world? What do we do economically? What do we do sociologically? What do we do theologically? I mean, if you remove scarcity, what happens to the profession?

[13:22] Well, it demonetizes. The more you produce something, the cheaper it becomes. Well, the more you produce cognition and expertise, the cheaper it becomes. And that's great. And you guys have been kind and listening to some of my pods.

[13:35] you know, when I did the pod with Dario on a day, I asked him, I'm like, look, there's 23.8 million Americans employed in healthcare, and you're talking about 50% entry-level job dislocation and 10 to 20% unemployment. What is your resolution here? And he's like, I don't know. I'm thinking about the patient.

[13:52] Great answer. So actually, I think my job actually on this show with this audience is to just provoke you both and get out of the way. whereas again it's to find a group that is both knows health care and respects and understands

[14:11] the industry with you both and are both more ai killed than myself uh and so let me push you both let me push you both on a few of these which is uh one i think and to respond quickly

[14:27] i totally agree i'm very happy we're having this discussion now and it was great to have and I've had this discussion actually at a conference almost a year ago was pushing Zeke and we were trying to come up with a bet of,

[14:39] no, no, no, like, look, this AI thing is real. Patients getting empowered in this way. And he was actually pushing me on, well, okay, Justin, what is the bet we could make if you wanted to bet a chocolate bar or something like that? So high stakes for what we would measure to show that it is better,

[14:55] to show that patient outcomes are better, to show that cops in town, to show something. And I struggled with that to be candid a year ago. It seems like he has evolved quite a bit in his thinking in the power of these tools and where we're coming from this.

[15:09] And so it's useful to have the discussion now to push all of ourselves, push ourselves on the employment discussions, push ourselves on the profession, push ourselves on the compensation and costs. But what are we not yet pushing ourselves on enough yet?

[15:21] Matt and Eric, to you both. I'll bring up Eric. You wrote this piece on, you know, I'm going to make it almost outdated at the moment. What I'm pushing you on is like, what, like two months ago, you know, we were talking about everything coming out on this piece.

[15:36] And it's just the pace of how fast things are moving. And I know you're already, you know, writing something else. But you wrote about this idea of health care is often a high-level moment. What do we need to educate people on? Where is this going?

[15:48] You know, Matt, I know you're working on something big at the moment to talk about, you know, what does the future of the academic medical center look like, these pieces? Like, what do we need to push ourselves down? And it's fun.

[16:00] My job now is to push you both to be unfiltered. Be unfiltered with where you think things are going. And I'll say what's so hard about this is every day one of my jobs as I'm running Qualified Health

[16:12] is to have discussions with health system leaders and to describe where we're going. But one of the things I struggle with is if I go too far, then I lose people. If I go too far into the future for what's happening, then I lose people.

[16:26] on, maybe the person I'm speaking with will be ready there and be ready to move, but they won't be able to convince the organization that this is the time to make the investment to run, to move at this pace.

[16:38] And so with that as background and context, can you go, what are the things where you're really pushing forward? Let's really go out there, because again, you both are some of the people I respect most, knowing healthcare and seeing where things are going, but let's push each other.

[16:53] Where is this really going? What are the messages you really want to kick it out to the readers for what's coming? Aaron? Matt, you want to start? Well, I mean, listen, I think there's a lot of impact from what you said, Josh,

[17:06] because I think that going back to what we had just talked about, there's like the reality. Again, there's some percentage of the forecasting that is going to sound original because that's what you have to do in this.

[17:18] And that's kind of been my thesis here. But at the same time, right now, as it stands today, are our systems able to truly take advantage of the capabilities or not? And, like, I think you've seen some of this in the broader market, in the less regulated, the less high-risk industries,

[17:33] where, you know, a small company with seven people who are adept at leveraging the tools to achieve business outcomes are running circles around the large companies.

[17:45] I mean, the best example that we can point to recently is the cursor story, right? Like, Microsoft in 2024 had the opening IIP, they had GitHub, and they have the leading IDE in the world, like, globally.

[18:02] You would be insane to try to go up against that combination of things and say, I'm going to build a coding harness for a model that someone else owns the IP of that's competing with me, and I'm going to still win in the market.

[18:15] Like, it's insane to think about, but yet that's not how it played out, right? Like, their team was focused on their use case. They were leveraging the technology at every step of the way. They didn't have any legacy things to bring along with it,

[18:27] and they were able to execute in a way that's been obviously now shown to be quite prescient. And so, like, how does that relate to healthcare? To me, we have to construct a system, to Eric's point, around the idea that knowledge is scarce, expertise is scarce.

[18:43] and the entire academy was developed on the way we teach people, the way we get evidence, the way we do peer review. And now we're in this environment where by the time the paper comes out, all of the technology they were writing about is out of date.

[18:56] By the time I'm trying to implement something for a use case, there's a better version of the technology. That is really hard to deal with in a somewhat ossified, structured system. That's a great point. What do you do now?

[19:08] Like, do you need to do it? Instead of doing a code-based shift, do you just start fresh? Yeah, and I think there's some people really doing some interesting things in that space. And, Eric, I'd love to hear your thoughts on this, because it really is, I think we're going to have to build in place, build this next to, almost, as opposed to trying to pour intelligence into systems that exist today.

[19:30] Look, I really resonate with what you said. I mean, I think about it as, look, this is, you know, you talk about the exponentials, right? This is the most exponential technology in the history of the world, right? and the obsolescence of, like, a framework or an idea.

[19:44] And, you know, there's a lot in Oppenheimer that was only released on June 15th that I repudiate. And we should talk about that. I mean, you know, and I'm writing my new piece right now. I thought about, you know, writing, like, or amending Oppenheimer,

[19:59] and I'm like, no, forget it. Like, this is just going too fast. Now, I do think, like, Kurtz is a great example and a $60 billion outcome, and, you know, and I think that illustrative of the insurgency right incumbent versus insurgents and there no more you could use the word ossified there no more ossified like you know establishment than in healthcare

[20:22] I mean, I think about, like, we have these sort of buildings in granite, you know, with the stone inscriptions above, you know, above the entrance. And, you know, the AMA has existed for a century.

[20:34] And the AHA is similarly, like, you know, deeply established. And, you know, I find it interesting that our most prestigious institutions, like the Cleveland Clinic and the Mayo Clinic, are also centenarians, right?

[20:50] That's an unnatural act in this sort of epic, right, where you've got this Darwinian, you know, sort of evolutionary, like, competition.

[21:02] And so what does the most established granite sector do when confronted with the most exponential high-velocity tech in history?

[21:17] Well, I think you're starting to see it. I think you're seeing, you know, this sort of opposition. You're seeing this, you know, this resistance. and, you know, kind of this Ancien Regime clinging to its prerogatives.

[21:33] And I don't mean to sound dismissive or disrespectful. I mean, I feel I'm going to get the same disintermediation as everybody else, right? Like, this is not externalizing.

[21:46] But to me, the question, Justin, that you're asking really is the right one, which is, if the exponentials are true, and if the trajectory continues unabated, even if it slows down, Like, how do we prepare? What if medical expertise really does become free or moves asymptotically toward the cost of compute, right? Like, if you think about, you know, what happens when the scarce resource becomes super abundant?

[22:15] the organizations that were built to shepherd and steward and arbitrate and officiate this scarcity what happens to them when you think about a hospital as an example let alone a prestigious

[22:29] academic you know the hospitals are a 1.7 trillion dollar sector they have over a trillion dollars in real estate the idea is you you get this geographically co-located big you know sprawling

[22:44] you know in insulation and you aggregate multiple specialists and you got the gastroenterologist the cardiometabolic specialist in there and the neuroscientist you need them all co-located because our reaction to complexity is hyper sub-specialization you know we sort of look at

[23:02] the body through a straw because our beautiful biological brain can only encompass so much you know and this is stylized and and will be ridiculed methodologically but in 1950 took 50 years for the sum total of human medical knowledge to double.

[23:16] And by some estimates, by 2019, it was 73 days. You know, our reaction to cognitive overload is to draw smaller and smaller circles around what we know. Experts know more and more about less and less.

[23:28] So we create this multi-trillion dollar bureaucracy and these edifices to bring the specialists together. Well, what happens, at least in the cognitive domain, we're not talking about optimists

[23:41] doing surgery or physical AI. yet, but when that cognitive scarcity becomes cognitive abundance, do you need this multi-trillion dollar sort of industrial complex?

[23:56] And especially, I mean, there's something really intrinsic about LLMs in the first instantiation, right? They sort of, well, they democratize expertise. And the purpose of a corporation is to take individual experts and aggregate them and

[24:12] coordinate them. But the LLMs are arming the rebels. They're arming the individual. I become a polymath. With my trusting superintelligence, I become a lawyer and a doctor and a consultant,

[24:25] and the experts can bemoan that. And it's a jagged frontier of intelligence and a jagged frontier of capability. So Terry Tao isn't getting disintermediated by this for math,

[24:37] even though now all the Erdos problems are falling by the week. I guess my point is that But Justin, the thing I want to think about is we spend 18% of our GDP on U.S. healthcare. We spend $6 trillion.

[24:50] It's super inequitable. Our health system is great if you're rich, white, and urban. It's not great if you're not in one of those privileged classes. What happens when medical expertise becomes free and democratized?

[25:04] Do the institutions that were built to guard and meter out and arbitrate the scarcity, do they continue to exist in their future, in their current forms? And I would say the answer is no.

[25:16] And how do we steward the transition in a thoughtful, compassionate way? That's what I'm thinking about. Well, because you both brought this up at the end. I'll push once more, and then there's a couple other papers and topics we could cover.

[25:30] But you talked about, Eric just now, how do the institutions make the transition? Matt, you talked about actually we have to build in parallel and I guess like

[25:43] Eric, I want to push you on that can the current institutions make the transitions and if so, what do they really have to do versus do you believe it's going to be the insurgents, the cursors

[25:55] of the world who are going to make this I love it, and that's right of course, and the cursor Microsoft example is very analogous right, I mean I thought Satya was the CEO of his generation until January of 2025 when he was at Davos.

[26:14] You can point the moment that he lost a trillion dollars in market cap. He's at Davos and Sam and Larry Ellison and Masa are in the Roosevelt Room at the White House announcing Stargate.

[26:26] and somebody at Davos interviews Satya about this exploding CapEx and Satya's like all I know is I'm good for my 80 billion Microsoft lost the race

[26:39] in that nanosecond so Cursor went against the empire and now is the most formidable entrepreneur in the history of the world

[26:51] with Elon the question is can an incumbent self-disrupt and an incumbent move at the speed of an insurgent. When the insurgent is unencumbered with legacy systems, with a lot of employment, with having

[27:04] to do the change management versus just build like unencumbered, you know, healthcare is the most regulated, litigious, you know, sort of incrementalist sector we have.

[27:17] And actually, as an investor, I think about healthcare as that affords you a little bit of a head start to figure it out. but you know incumbency is not some invariant law of nature incumbency is a head start what do you do with it

[27:30] and what do you do with it in this like sacred intimate space of healthcare where if you move too fast people die and that sounds dramatic but it's actually quite true

[27:43] I think if this weren't a global dynamic the counterfactual would be what health system or payer or life sciences company

[27:56] or medtech company moves faster than its competitors that's not the counterfactual counterfactual is what do the authoritarian and monarchical regimes do what are you going to see in Saudi Arabia

[28:08] and China I mean Xinhua University opened its AI hospital I'm actually going to give one of the keynotes to the Dell Med opening and you know I'm so intrigued by what we can do if you sort of start from first principles and ask the question, what does AI permit us to do?

[28:29] And if you don't have to sort of rehabilitate or reconstruct an existing organization, you can build from scratch. Obviously, you've got to work within the regulatory regime. I do think Matt is right, and you've got to build in parallel.

[28:44] And that's why you're starting to see a stratification. you know Justin you and I often you know we talk about this oligopoly of of leaders right and 150 CEOs that have this real sort of deterministic effect on how things play out I am seeing a

[28:59] stratification you are seeing a stratification you're embedded with some of the real pioneering health systems you're starting to see some of these players move ahead of their peers but too infrequently do we have a conversation about where where are we going to be in the next

[29:15] several years if the exponentials continue. And that's why I'm so energized by the conversation we're having right now. My comment is, like, as I think about, okay, well, listen, and to your point, there's plenty of, like, asterisks on, like, just, you know,

[29:30] the reality of the regulations. You mentioned that there's high risk, low risk. There's all kinds of different ways we could slice this. But there is a hybrid, you know, analogy to the cursor Microsoft, which I think is probably the sales force.

[29:42] and very recently. So again, do I believe that that's the path? Maybe not, right? But this is like a middle ground where it's like, okay, I can accept the fact that most of the things that use software

[29:54] in the very near future are going to be agentic. Right? So we're already seeing that with the internet. So as a software incumbent who wants to at least maintain relevance in the exponential, I become headless and I basically invite,

[30:07] you know, the barbarians at the gate all right on in, right? And it sort of like can I ride the exponential with my incumbency with my system of record And so like there is a story here where you know because I guess what you would say to this parallel idea okay there going to be an upstart EHR that going to be a first principles health system

[30:26] that's going to be AI native and it's going to serve healthcare autonomously as often as it can and whatever it can that is a story that you could you could play out but I think the other version this is does an epic or a miracle or

[30:38] existing incumbent sort of come around to the concept of no longer are my users humans clicking buttons, my users are going to be agentics. So what do I have to do to my stack to maintain the system of record

[30:52] incumbency while also delivering the value for the change to come? And I don't know if any system of record is currently capable of that, but that is another future for this, right, where, again,

[31:04] it's accepting the fact that the intelligence layer is going to be primarily operating with each other, meaning agent to agent and agent to software, to deliver X value. And so what does that look like for healthcare?

[31:17] I don't know. And the regulations boggle the mind. Yeah. But the regulators are trying. They're having the discussions. They're inviting people to the table.

[31:30] They typically, right, regulators work with the incumbents. The regulators are reaching out to the insurgents, to Eric's analogies that he's using on his podcast, on his show. And so the regulators are moving, and so it is a very interesting dynamic.

[31:47] Eric, the one point I wanted to maybe disagree with slightly, your point is, we need to have the exponential continue. Oh, great. That's where I guess I'd like to just say, like, we are so far.

[32:01] Great point. We are so, we could have a decade of implementing the current technology to its potential without anything improving whatsoever. A year ago, we had more than enough.

[32:13] We have way more than enough right now without any kind of artificial intelligence. Yes, yes. And the place to look to is just how a leading organization,

[32:27] and I'll skip bringing up the slides, but a leading organization like an open AI, anthropic or a cursor-type startup is using tools versus the median versus healthcare. they are already orders of magnitude ahead on the current tools and growing

[32:42] and so we don't need because I think the conversation happens with readers though we'll just wait they'll get better or we'll see how much it gets better then I'll figure this out it's like no no no no we already have more than enough

[32:54] today to be orders of magnitude ahead of where we are but maybe Eric the last point I know we've talked about a few different things and you know we have some positivity and you know maybe negativity of what's happening in the future.

[33:08] Eric, you brought up this point as we were getting ready on the cost. Yeah. Yeah. Yeah. Yeah. Yeah. Yeah. Are we actually already doing better? Well, I'll pause here.

[33:20] Keep jumping for whatever you want to highlight. Yeah. This paper, everybody should read. Cutler and Clarnett. It's a Harvard paper. I think it's a pre-peer review. It's still in archives. This thing melted my brain.

[33:33] And the idea here is that, you know, I've been sort of promulgating this idea that we can take 500 to 700 basis points off the percentage of US GDP allocated to healthcare in just the next few years.

[33:45] And I got, like, villainized for that, you know. And, you know, I'm sort of making myself a little bit unpopular in some circles because I'm sort of saying the quiet thing out loud, which is we're a $6 trillion sector, of which $3.2 trillion is labor.

[34:01] you know we employ 23.8 million people one out of every six working adults is in health care and you know we suffer from balmol's fox disease right productivity you know progress in manufacturing

[34:14] industrials even hospitality things like that like you know that health care has to compete for labor so we don't enjoy the productivity bump but we have to raise our compensation to keep

[34:29] people and as a result we have to raise prices and healthcare has inexorably risen prices and so i've been saying that you know if this is if this is ai is an industrialization of

[34:41] intelligence and you know just like we saved a lot of money going from branded drugs to generics or or inpatient to outpatient side of care right this is taking scarce cognition and making it

[34:56] abundant and when you make something in abundant you demonetize it it becomes cheaper right and so you know a lot of my extrapolations about how you take five seven-hour basis points were

[35:08] intuitions this paper says that we've already done it right and so the paper basically says the cms actuaries i'm going to go by memory um you know in 2010 forecasted that by 2024 the u.s

[35:21] economy would consume, would be, would, healthcare would consume 21.2% of U.S. GDP and represent about $6.3 trillion by 2024.

[35:33] The actual numbers came in $977 billion lower and 320 basis points lower. So instead of 21.2%, we ended at 18%. And instead of $6.2 trillion, we came in at $5.3 trillion, almost a trillion discrepancy.

[35:51] And that over that period, that 14-year period, 2010 to 2024, we saved $6.7 trillion that the CMS actuaries incorporating the projections of the ACA predicted.

[36:07] And in no 14-year period since we started tracking the data in 1960 have we seen this divergence. And relative to other OECD countries, we are actually inflecting the cost curve.

[36:23] And the authors, with a lot of methodological rigor, decompose that number. And they say technology accounted for something like 14% of it. And then site of care shift and a healthier population and the preauthorization and some of the clawbacks from payers, it all sort of agglomerated into this reduction.

[36:45] This is all pre-AI, right? I mean, and so my takeaway, Justin, in that is that we have a roadmap based on the most recent 14-year period for interceding and bending the cost curve.

[37:04] Now, there are some mitigating things. GLP-1s are inflationary in the first instantiation, but they're going to be deflationary. You guys know I'm on the board of EMED. And we're about to publish some data that suggests that in a world of 10% medical inflation,

[37:21] employees that are on GLP-1s that are adherent to it are actually seeing a 5% drop in their total healthcare costs. So most technologies in healthcare, when they were first introduced, are inflationary, right?

[37:34] Because they address something that was previously unaddressed, like a cabbage, as an example. But eventually, you're going to see more cost-effective. I mean, a better example is total hip, total knee, right? So for Medicare, total hip, total knee migration from inpatient to outpatient

[37:50] and other side of care shifts in 2024 accounted for $94 billion in savings out of the $977 billion. And so I look at this study and I think this is a conceptual theoretical roadmap for how AI,

[38:09] so if you went from randed drugs to generics and that deflation and you went from inpatient to outpatient and that deflation I said this earlier what happens with cognition

[38:22] going from scarce to abundant so anyway I would encourage everybody it's an 84 page paper it's kind of impenetrable throw it into Claude throw it into GPT 5-6 or just read the damn thing

[38:35] because it is a seminal paper and I'm really shocked it's not gotten a broader leadership and distribution. If Darren Osmoglu can get a Nobel Prize

[38:47] for basically saying dumb stuff like AI is not going to affect productivity, I'm going to nominate these two for an economics, you know, Nobel Prize.

[39:00] The best part about, I think, this story, though, honestly, is that, like, it's the repercussions of a relatively blunt instrument intervention, right? I mean, some major shifts, some major choices.

[39:14] And then, like, to your point about the GOP, that also was, I mean, let's be honest, 20-year-old, 30-year-old drug, right? So, like, it seems like there's just so much alpha that we haven't even taken advantage of

[39:27] before you get into the discussion about what the technology is capable of doing. It's fascinating to me. And maybe this does leave us on an optimistic trajectory that, again, Justin, to your point, the exponential stops today.

[39:41] Eric, you know, the current interventions are already bending the curve. Like, maybe we do get to a place where some of the major issues about our health system, as you pointed out, the inequities of it, are already on the right path.

[39:54] And so we have reason to leave this conversation super optimistic, right? And I think we're all heading that direction. with that Eric thank you so much for joining us we could go on for hours

[40:06] and we do in other settings so thank you so much for being here and we'll talk more soon thanks gentlemen thank you