Vinod Khosla

To watch the video version of this episode, visit incumbentsandinsurgence.com. This podcast is presented by the Towerbrook Healthcare Institute. incumbents and insurgents in healthcare with Eric Larson. This presentation is for informational purposes only. Please review the full disclaimer at the end of the episode. Ian, good morning. Good morning. What have you become a private equity guy with that vest? I guess, man. It's just like osmosis. Is that a designer vest? Did you just have a logo on it? No logo, but it definitely, you know, at a distance that they're like, yeah, that's, that's B.C. That's P.C. P.E. It's out of it. Yeah. No, it's on the metamorphosis that you've gone through here. That's right. That's right. That's right. What are we doing? We are introducing our dear friend and colleague, Vinod Kozla. And, and look, I got to tell you, man, I have been looking forward to this for so long. I unapologetically love Vinod Kozla. I just think he's such, first of all, I think he's like probably the most, no, I'm not going to qualify it. He is the most prescient technologist of the last 50 years. And, you know, he's, he's, he's one of the titans of Silicon Valley, one of the titans of, you know, if there were a Mount Rushmore of venture capitalists, I think he'd be etched in maybe with the biggest sort of like, you know, sculpture. Because, you know, he's just had this deep clairvoyance about how technology is going to like ricochet across society. And, he's been present and really instrumental in each paradigmatic shift. You know, let's go back to internet, right? I mean, his legendary investment was Juniper Network. Sorry about the fact that he co-founded Sun Microsystems, right? But, he put a $3 million investment into Juniper. And, then, I put it with a $3 billion outcome. So, a $7,000 return, $7,000 full return on his investment. But, he was also the first outside investor into open AI. And, just has this deep intuition. But, then, he's also sort of the techno philosopher of Silicon Valley. And, and has written some really seminal thought pieces that have framed the dialogue. But, above that, and then, you know what, you know what, you know, you've gotten to know him really well over the last few years, too. Like, I just love his insatiable learning. I mean, the guy's 70, 71 years old. He's planning on his next 25 years. He's gonna outlive both you and me. You and I have a lot of stress in our lives. So, I'm super excited about this. What are your thoughts? Look, I agree with everything that you've said. I'm thrilled that you were able to, you guys were able to find the time. Have this conversation, you know, early on in this journey that you've, you've provoked with us, with Towerbrook, with Towerbrook Health Institute. But, I'll tell you, Vinod, I, I mean, just, you know, do or do or do or do or do or do or do or. And, the fact that he has set his sights on healthcare after various industries that he has set his sights on. And, he continues, obviously, to be involved in so many businesses, seen and unseen. But, his fearlessness in terms of attacking in industry like he is attacked healthcare. The persistence from the time he wrote that paper in 2016 on the 80% doctor and the positions doctor. Sorry, 20% doctor, but that 80% of what physicians are gonna be doing, you know, will be automated. And then to write, to have it be 100 pages long and go through almost every specialty. The insatiable appetite to learn, to challenge, to throw down meaningful statements like that, you know, the, the 20% doctor. And like he's done in other areas with that he's attacked in healthcare, behavioral healthcare. Various types, parts of business automation, various parts of primary care. With his son Neal and the, and the, and the business that they have started as well. It's just impressive with the fearlessness with which he's gone about it. One, two, you know, to hear him talk about each one of these businesses. It's not like any venture capitalist. It's almost as if he's talking about his family members and his children. The role of this leaves nature that he has that you can see that he can get into the weeds and understand the details. And then he can fly high and see the farthest. I mean, it's, it's, it's, look, it's what makes him so special and what makes him so, so successful. And then lastly, what I'd say is this, this understanding how fast things are beginning to move now. And his orientation towards the incumbents, towards working with the likes of Towerbrook. To working with the likes of our portfolio companies and co-investing with us. To seeing that accelerating the impact of things that he is working on and funding could actually come through either the combination or co-investment in some larger businesses. I think we're going to see more, more of the mainstay venture firms. Look, they're already beginning to do that. I think we're going to see more of that. I think it's incredibly prescient as, you know, as well. So super, super fortunate for the ability to work and be exposed to his brain and be able to work with him. And his firm and, and super fortunate that you've been able to have this conversation. So I'm looking forward to sharing this with our readership and our listenership and viewership. And we're going to talk about the dislocation on the jobs front, of course. We're going to talk about this notion of a medical superintelligence. And, and this is a topic that, you know, we've had a lot of energetic backs and forths with the note over the last few years because, you know, we've each been sort of conceptualizing what will this look like? And, you know, what are the consequences for the single most respected profession in the United States, which is the practice of medicine. And doctors are at the top of the occupational compensation and societal prestige hierarchy. And yet you're going to have the biggest fundamental reconception of what it means to be a doctor, what it means to be a teacher, what it means to be a consultant, what it means to be a financier. The node is enough of a realist and enough of a pragmatist that he sees the societal benefits that are going to diffuse, that are going to be massively deflationary, that are going to massively improve quality of outcomes. But he's also sees the probability that they're going to be real shifts. I mean, and it's going to be in a really telescope time period. So exploring that together and then almost being a decoder ring between Silicon Valley and the establishment of the 150. You know, you and I always talk about this sort of mutual incomprehension between, you know, the oligopoly and healthcare and how entrenched our institutions are. And then we're spending all this time in Silicon Valley with the insurgency. And I think the node has really, especially in the last few years, you know, turned his attention to bringing the 150 into this sort of like vortex of the technology and help them be an arbiter and a judge and bring their wisdom to what is applicable and what is not. And we know that just the outright resistance isn't going to work in the face of a technology of this potency and really quite frankly, in inevitability. So I'm excited to share this conversation. I learn every time I hang out with this dude and, you know, we're the fact that we get to invest with him and think with him is just a great privilege. Looking forward, looking forward. Nice. We'll see you. The node, it is nice to see you my friend. How are you? Oh, I'm doing great. But all this innovation, it's exciting. It is exciting. Well, I have been looking forward to this for a long time. I get the privilege of talking with you regularly and and, you know, I'm such an evangelist, the node for your mental model, your insight. You know, kind of think about you as the most prescient technologist of the last 50 years. And I'm going to separate the fact that you're in the pantheon of all venture capitalists. But I just think of you as almost a techno anthropologist in somebody who has this really like deep intuition about how technology is going to ramify across society and be civilization shaping. But, you know, the domain that you and I always talk about is my, you know, my, not just vocation, but my consecration, which is to the industry of healthcare. And I know you've got the same passion for this and the same sort of like evangelical, you know, sort of energy around it. But I'm going to try to start us off with a little bit of a head nod to the title of this podcast. And, you know, the title is incumbents and insurgents. And one of the granule themes that you and I've talked about over the years is this notion that insurgents and just about every other sector of the US economy have misplaced the incumbents. You know, they sort of graduated to ascendancy. And it wasn't curts or avis that innovated, you know, the ride sharing world. It was Uber and Lyft. It wasn't Lockheed Martin and Boeing that innovated space exploration. It was SpaceX. And healthcare is a really troubling exception to that. In fact, I'm hard pressed come with a single insurgent that's graduated to the ranks of incumbents, maybe with the possible debatable exception of Epic, a narrative 60 years in the making. And we'll see how that plays out. But I want to ask you because you've been out cross-alatizing the one I refer to them as the 150, you know, the establishment and healthcare. And I think you've done, you know, a dozen board retreats in probably the last few months. I want to start by asking you, what are you telling the establishment, the incumbents in healthcare today that they need to know about this moment in technology and this moment in history? Well, the thing to keep in mind, incumbents do just fine if none of the other incumbents are innovating rapidly. So they only have to keep up with other incumbents. If there's a startup that's super disruptive, that everybody needs to respond. A good example in a different industry is nobody in the automotive business want the largest in the world. 100 million vehicles are more per year while doing electric vehicles. Then Tesla came along, bet the company on just doing electric vehicles, and that then very rapidly forced change at every other automaker. And they isn't an automaker that isn't doing electric vehicles. And the fastest growing automaker today in the world. But Tesla got more market cap than the next 10 traditional incumbent auto companies combined. And then new ones emerged like B by the other China is probably the largest exporter of electric cars. Why point is when an entrepreneur comes along and drives change, then everybody has to respond comparatively. In healthcare, you have to keep in mind the chat GPT moment was less than three years ago, less than three years ago. Nobody in most parts of the world had heard of open AI or chat GPT three years ago. And it ran from zero to a hundred million users in in something like 90 days. Today, it's at 800 million users about the 10th of the global population that's using it on a weekly base weekly. When an incumbent come along in healthcare and they are slowed down by regulation, that will happen. And my back for the 150, the bigger opportunity is to say, instead of letting a new incumbent come in, we will lead the innovation way, I frame it. Certainly, we are at a point in the next couple of years where all expertise is free. And theologist doesn't cost you any more than a health coach or no practitioner doesn't, it costs the same as a gastroenterologist or a cardiologist. So, leaving aside interventional medicine or in hospital care for a moment, we can come back and talk about that. Almost all expertise should be free. All systems are designed today to date expert, the most expensive expertise you get through a patient has to go to many hoops to get to a neurologist. But I would invert that in a world. And if an incumbent decided we will use almost free expertise to reduce expense easier for pairs to do and we can talk about pairs versus providers. It's a massive opportunity for us to have say a 30 to 50% reduction in their cost of expertise compared to every other competitor, if somebody makes that move, everybody else will have to follow and the leader in this, the one driving or instigating this change, instigating is the right word I like to use. It's going to have a dramatic increase in shifts, particularly through in risk bearing or Medicare advantage. And we can come back and talk to my recommendations on the other side of Medicare advantage, the government. Yeah, I want to decompose that a bit of a node because there's that's so information dense and it has such implications for the industry and maybe we can go systematically through some of the incumbents and and almost like a typology of the incumbents. Because, you know, something that you and I've done over the last two to three years is bring cohorts of the 150. So I've been bringing these groups of CEOs from life sciences companies and payers and providers to come meet with you to meet with Sam to meet with Dario to get evangelized is what's happening with the exponentiality of the tech. And the theme that you're articulating about, you know, there isn't asymmetric first mover advantage. And if all the incumbents in a cartel or in some sort of racketeering agreement, you know, sort of complicitly decide not to compete, then you can have some sort of equilibrium. And later in our conversations, we're going to talk about some of the geopolitical implications of this because we've got some exogenous forces like in China and the GCC that are propelling this forward. So I think about incumbency for a minute, let's just take health systems, right? Health systems are the single largest vertical in the US economy, $1.6 trillion. And it's super oligopolistic, right? The top 100 health systems represent 936 billion out of the 1.6 trillion dollar sector. So the beautiful irony is if you can persuade just a couple of these, it's sort of cascades change. But my mental model, the node for how Jenny, I is going to ricochet through health care is in four buckets, first is an administrative simplification, second is in care, augmentation and a medical super intelligence, which I hope we'll talk about. Third is in synthetic and computational biology and fourth is in consumer empowerment. Hospitals, my intuition is that this is about employment, right? US health care employees, 23.8 million Americans. US health care is the only industrial vertical to see negative productivity growth over the past generation. From 1970 to day, the number of doctors in the country increased 150% over that time period, same time period, the number of health care administrators increased 4,300%. And so for me, this is about labor reduction. And it's about contracting the number of employees that we have, if the typical health system spends 60 cents of every dollar on labor. And what we're talking about the node, about substituting technology for labor and that a first mover who does that is going to take these anemic sub 1% operating margins, turn into something more robust and grow market share, contiguously, non contiguously. Am I thinking about that the wrong way? I think you're thinking about it the right way when it comes to administrative labor. Yes, absolutely that can be cut and I am not sure why it can't be cut drastically. If I were running a health system now on the provider side, I would personally target to cut the number of administrative people by more than half in the next five years. I would guess that's a conservative goal, not an aggressive goal. I would plan on providing patients on the physician expertise side, three to five times more contact, face time with the physicians. Then anything else, one of our companies tortoise did a study in the UK, physicians using their system spent 25% additional face time with the patients, 25% that's like increasing physician capacity by 25%. That's a pretty big thing. So I think on the care provision side, you can provide five times amount of care. The average patient can see their PCP or a physician in the ironic thing is that no different reaction, the cost of a no sort of gastroenterologist, five times more often per year than they do. That's an opportunity which will reduce other costs or increase participation satisfaction, hopefully increase employment. I see no reason on the provider side. First, AI doesn't do all the patient intake before they see the physician when they're booking an appointment. At the time they come in and to book an appointment, all of the intake happens, probably a diagnosis happens that's not given to the patient, recommended treatments happen, dosages, casts, all that gets recommended by the AI. That's entirely possible today. Only thing you can't do is hand it over to the patient, then you can decide what to do with it. If it's low equity, you can pass it to a nurse or do column medicine. If it's low equity or low sensitivity, if they need to see a patient or if it's a high value patient, economically, you bump somebody else to move them into your schedule. For example, so there's that side effect, intake should be done this way. We're already doing a lot of scribing and that's been a big help. No physician in this day and age should ever touch epic ever. A voice interface into it is incredibly possible. Hospitals or other provider groups haven't been bold enough to say we don't want to touch epic. It still remains as the system of record, but the AI runs epic and the physician just talks to an AI is the vision. And then from that comes all the other stuff like prior art or claims or other things which can which are all administrative and AI can do really well at all that. In patient care, there's no reason today. Every patient visit shouldn't be followed up with a call two days later. Did you get your medication? Are you complying? Is it sitting well with you? Do we need to change the dosage? Yeah, I can do all that. Those are not clinical functions. Every visit can have one, two or three follow up depending upon the patient in the security, see how they're doing. That might cost a dollar or two to follow up in the context of patient satisfaction and avoided phone calls or customer support calls. That's very, very possible today. I would love to build both the intake and the follow up systems that basically done by many of our companies, the strivings done, the prior art has done. So it's just nobody's bold enough and they had too many committees and IT and security and legal get in the way. None of it is that well critical, all solvable problems, fully hypercompliant, all that kind of stuff. Yeah, even though that totally resonates in, you know, one of my favorite clothes is trim up, Denson Claire, and he said it's impossible to get a person to understand the opposite of what his salary tells them to understand. And if I just define this fully salary compliant, you know, it's absolutely time intake gets them higher value patients on their schedule and maybe bumps some less smaller to the patients, which, by the way, is good for the patient, they don't have to come to the doctor's office. And they can do a television or a chat visit follow up is good for the patient, and absolutely for physician income, what you're describing is sort of ambidextrous works in a delegated risk environment. It works in a fee for service environment. It's augmentative of panel sizes, right. So if the typical primary care panel is 1 to 2,200. I don't see why we couldn't have 1 to 10,000 1 to 20,000 panels. And what you're describing really resonates and I want to unpack it a little bit because, you know, I would categorize what you're describing the note is the deburacridization of the clinician's practice. And if the typical primary care doctor only spends 30% of her time laying hands on patients and 70% is spent in administrative and fighting with the EMR and arguing adversarial with payers, it's a total suboptimal time distribution. And it feeds directly into these really scary kind of apocalyptic predictions of physician shortage, which I don't believe, right. Like the American Association of specialty colleges is predicting, you know, 137,000 scarcity by 2033. To me, that's a really self-serving, you know, sky is falling projection and doesn't account for what you're describing. But I want to ask you a question about the next oral area to this, right. Physicians, there are 950,000 physicians in the country, predominantly their W2s, right. 54% of US physicians are employed by hospitals. And then another 26% are employed by private equity players, vertically integrating payers like United, so their W2s, right. So it's about only 18% of patients are physicians are independent. See, of 950,000 doctors, they're the most respected profession in the country, followed by nurses, followed by military veterans. They're the highest compensated occupation in the country, 9 of the top 10 and 20 of the top 25 highest compensated occupations in the country are medical. And yet we reimburse our clinicians on a fee for service, our BRVS model, unless they're capitated, which is a minority. And so my question to you, and you've talked at length about this, about this demonetization of expertise, lawyers, consultants, financiers, doctors, where the cost of expertise asymptotically moves toward three. And if you talk about almost a million doctors, making an average of $350,000 per year, that's a $300 billion wealth transfer every year. How does what you're describing the node, even just from a deep rockerization point of view, we haven't even touched on the medical superintelligence dimension of this, which I want to turn to in a second. How do we reimburse for these doctors, when they graduate with medical school at the quarter million, if more, if not more in debt, like what's the economic model to accommodate this level of transformation? Yeah, so here's what I would say you talked on a lot of different issues. The first thing I would say is there is no reason every physician shouldn't double their panel size in the next three to five years, double their panel size. The more they go at risk, the more income they make, but even in fee for service, you can double the panel size, cut the visits a time in half, but more importantly, give the patient five times as many visits with an AI intern who works under the supervision of the doctor. Now that's how you cut panel sizes and AI intern, the patient sees 500% more engagement with the provider, most of it with an AI, all supervised by the physician and we can go into the details of how this modern intern model works. It would be the same if you hired a fresh graduate from Harvard medical school, you'd almost use them as an intern, not let them go wild with patients, so that's a good model works much better in complicated environments, but it works in fee for service environments here. Let me mention something funny, so most of the 150 will have a hard time believing what I'm saying. So I was recently at the conference and most of them probably know E. Camano, who is the vice provost for global health initiatives and at the end director of the health transformation institute and advisor to president Obama on health. Him and I are co-editing an article that he has titled, "Bither the Physician in an AI World." He came up to me recently and said, "I was very wrong 10 years ago when we talked about it, but you were wrong too, because he said it would take till 2040, it's only going to take in the 2020s." So this hopefully will be published sometime soon, which is more credible than me saying it, inputting his weight behind it, that's credible, because he's knowledgeable, more knowledgeable than I am about this. I do think that's the way to think about it. For now, just set the simple goal of cutting administrative costs in less than half, panel sizes, double, physician income, either doesn't change or goes out, and risk bearing becomes a huge comparative advantage. You and I are totally aligned with a delegated risk model and the deployment of not just the administrative simplification applications of GNI, but the clinical augmentation aspects. I almost, you know, before we turn to the next topic, and we've got an infinity of topics I want to go through in limited time, so I'm going to have to ration. I want to ask you almost a philosophical question about this, because the thing about interns, and I really, your unstruck resonates, because it's in the nomenclature of medicine, it's apprenticeship-based, we have interns that are five to seven years out of medical school, it's a mental model that kind of is intuitive. But the thing about interns is they grow up and they get smart. And if we know anything about the exponentiality of this tech is that it is going toward omniscience, pretty fast. And so I want to ask you a philosophical question, because, you know, we've really enshrined physician judgment and almost subjectivity as an almost theological tenet of the practice. And you have talked over the years of the practice of medicine, evolving into the science of medicine. But when these interns grow up and these multi-model, multi-model foundation models that are superhuman from a differential diagnostic and a care protocol and treatment point of view, what is the role of a human physician in a post-AI maturity and medical superintelligence world? Again, you covered a lot of different topics, so let me comment on them, seriously. First, the notion of primary care, which is primarily a gay keeper and a router to expensive care should go away. I now call it not primary care, but multi-specialty primary care. In some of our companies, a primary care physician, even acting as a supervisor over an AI, can handle 50, 60, 70% of specialty care because the AI is knowledgeable and the human is really keeping providing endocrinology advice. Because the AI is providing that advice, the humans saying it's appropriate to give to the patient and they have enough judgment to do that or can do a quick check with the AI or what publication might support something like that. We've seen things like open-air evidence and curi and others do the same thing. So the notion of primary care should be changed to multi-specialty primary care, so multi-specialty visits get cut in half. I think that's a very feasible goal for 2030 because the AI is assisting the human primary care physician under current regulation who has to fully oversee any patient interaction. By the way, not history taking, that's happening today, that's legal. You could take all the history in direct conversations with the patient. So that's one thing to keep in mind as to physician judgment. Let me give you a following study. Multi-center study on complex disease diagnosis run by Onnie Milstein, which probably won the pre-eminent names in quality of care in the country. Out of Sanford, multiple centers, I believe they use clinical centers with high quality and remote locations where the physicians were not researchers. Here's the data. Physicians had 73% accuracy in complex disease diagnosis. That means 27% of the patients got the wrong diagnosis or a suboptimal diagnosis. AI alone was 88% accurate, so substantially better in complex disease diagnosis, and this study is published by Onnie Milstein, so anybody can access it? Here's the funny part. They gave the AI to physicians who improved from 73% to 76% and I think I have the numbers roughly right, but it degraded the AI from 88%. So this thing you call human judgment is applying a lot of human biases, like recent sea bias. I recently saw a patient with COVID, so I'll diagnose you with COVID. If a disease like ADHD is mentioned in the New York Times, diagnosis of the disease in the following month goes up substantially because physicians have this recent sea bias about reading it about it in New York Times, you know all this. So I would be cautious about human judgment, but I do think human judgment between now and 2030 is very, very valuable in putting the AI on a leash and monitoring it, interacting it, I believe the humans will get better at their job. And then slowly deciding how much to let the leash out and frankly regulations don't allow today and they should, but they don't. The idea that AI talks does a diagnosis or prescription for a patient or gives medical advice. So I think within that current regulatory framework, all this is possible, much better for the patient, much better for the physician, whether they are fee for service or capricated, I think all that is very, very exciting. Now, you also asked the question, what happens when these interns go wrong? And I think that's a real question to face. You know, if e-command you can write with me about whether the physician in the word of AI, I think we have to acknowledge this is happening and we can try and block it or use it to our advantage. If I were in the healthcare system and planning my business for 2035, I would be moving as quickly as possible to the application of every source. You know, we have a company called sword help that does physical therapy with AI. They're results are so stunning. They get five to seven engagements per week on physical therapy with an AI therapist because people don't have to book three weeks in advance to make sure they show up. They do it at night. When I fractured my hip, whenever I landed in that city at 11 p.m. I'd go to my hotel and do my physical therapy. Last time we were together, you told me you did it on Christmas day too. Last Christmas for four surgical physical therapy, 50% of the patients did their physical therapy on Christmas day. I get to, I fractured my hip. I was pretty serious about getting back because my mortality rate was, I think, 25% for males over 70. If you fracture a hip, that's the mortality rate. It's, I think, something closer to 30 or 35% next five years risk. So most acute surgery, physical therapy, 50% of patients complied on Christmas day will never happen in a physical therapy world. And that's what's so exciting. The, in fact, their engagement is so good. They added your money back guarantee to employers and pairs where the patient wasn't satisfied after their 12 week soft therapy. Think about it. That's amazing. That's amazing. That's a no-brainer. That's what AI enables in physical therapy. I'll give you another example. We have a small company, a warehouse working workforce, so people making $15 an hour or less. This employer provided curi primary care to their patients. 38% of patients who got the primary care said they avoided an emergency room visit when they would have gone to emergency room had they not had this text based primary care access. Think about it. And by the way, same thing, we're doing AI oncologists, AI mental health therapists. We should definitely talk about that. Let's turn, let's turn to that now, because Vinod, what you're saying is so compelling in, you know, I'm writing a new paper, apologies in advance to my readership. But the theme is, how do you take 700 basis points off the percentage of US GDP allocated to healthcare? And when you start to stack these various interventions on the administrative simplification and the care augmentation sides, the numbers start to get really massive, right? Because out of our $4.9 trillion industry, 2.7 trillion is in labor. And if we can either substitute or augment that labor or see more patients synchronously and asynchronously, not only is MLR going to reduce, but FGNA is going to reduce dramatically too. And the math pencils. And so I'm trying to take pains to show that this could be, this could be a watershed, you know, because we sort of been fighting against this for a long time, but you mentioned something Vinod, I want to double, double down on. Because to me, one of the most exciting dimensions of the emergent capabilities of Gen AI are in that behavioral domain. And if Sam has 800 million weekly active users, the number one use case is therapy and companionship. And what we're seeing is that, you know, big, big news flash humans lied to other humans to avoid stigmatization, judgment, discrimination, embarrassment, but they'll tell the truth to a chatbot. And as the chatbots increase in conversational ability and down to 230 millisecond latency and having sentiment analysis and psycho fancy and the best in worst senses of that word and that matches the personalities like 24/7 availability, no judgment, you know, you backed two companies and perhaps others that I've been totally like fascinated by one is replica. Yeah, you genie acute is company and I think she's amazing as a founder 40 million people signed up with replica and I just use the beta and it's utterly captivating right. And then the other that I actually personally invested in is limbic Dr. Ross Harper in the NHS, you know, from the UK and creating this multimodal behavioral list, right. Like, tell me about how behavioral and gen ai is going to transform health care. Let me actually go broader than that. When I said multi specialty primary care, that included mental health. Here's what you can get from curi in new service. They'll be launching in January or so. They've already launched it in beta. It's everything health. So it doesn't matter whether it's clinical or non clinical because the most important thing is to engage the patient. There's no health plan program instance where patients engage five or more times a month. That's our goal. We will offer that and full clinical care, including mental health care, mental health, primary care, much of specialty care for $30 a month. Unlimited use of clinicians, clinical diagnosis, prescriptions, everything at $30 a month, including mental health. More important than just the price point, which nobody else can meet. And we expect like two or three clinical episodes where a physician is involved per month in that price point. But we will have simplified the physician's job whether he's doing a primary diagnosis and thinks you need a supplement or you need prescription medicine or you need those titration you can do all of that for $30 a month. That's pretty stunning. But it doesn't matter. Most Americans engage with the health on supplements or well less, more than clinical, all of that included. So you want advice on meditation or which supplement to take or this iron is upsetting my stomach, what form should I take or should I do surgery or should I do physical therapy. All those range of question should be in this broad AI agent. And that's possible. And I want to next cover some cautions. There was a recent op-ed in I think I believe it was in nature. And it was titled the fragile intelligence of GPD 5 in medicine. And it was a peer reviewed publication. My problem is I'm and I'm writing a response to it. It was so poorly done by physicians. They used the wrong system. They used a car without seat belts or seat bags or steering wails and open GPD and said, oh, it doesn't work. So the question I want to give is out of the box and many of them LLM companies are pitching they can do medicine, but without safety systems, without triage systems, without all kinds of cross checks, without sophisticated prompting on how you get the right answers and how you avoid hallucinations of all these LLMs. And I'll go back and come back to hallucinations, build a system, something like your eye works really, really well. And this new health coach AI health coach on everything. I can go hard to exercise how to look at the diagnosis, sacrifinia or build muscle mass or all the things you should be cautious of, what knowledge about vaccines, all that. So that's primary care broadly, but I'm even more excited about mental health specifically. So there again, you've seen lawsuits, you've seen press reports of chat parts going while leading people to suicide, again, because these guardrails safety rails, this grounding in clinical medicine is missing. Kira has grounding in clinical medicine, limbic for mental health has grounding in clinical psychiatry so they don't just chat with the patient. They are doing a set of clinical findings that help with the diagnosis and their diagnostic accuracy for mental health conditions for the top 10 CBD indications is 93%. There's no human that reaches that level of diagnostic accuracy and this was a study of 100,000 patients in the NHS. Think about it, there are no total studies in the US for mental health that are at the scale. They might be lucky to get a thousand patients, not a hundred thousand patients, 93% accuracy. By the way, diversity and inclusion went through the roof. For marginalized hierarchies, the more marginalized they were, the better it got. For transgender, it was a 150% increase in inclusion rate. By the way, this is regulated, regulatory approved by the MHRF, which is the UK equivalent of the FDA to do both diagnosis and do therapy. And therapy can cost you a dollar. Yeah, at the marginal cost, at the level, $400 in average, it's $2 at the marginal cost of the compute. And I want to add something we know before you go on, because I'm so high conviction on this, even beyond the synergy between primary care and behavioral. We know that in this country, if you have a medical complaint with the behavioral comorbidity, your medical costs are between 2.5 and 6.7x greater. We know that one of every three people in an industrialized country is lonely and one out of 12 has a behavioral comorbidity because of that. And what you said about the DEI element of what limbic has achieved in the National Health Service is really, it's quite revealing because there's no judgment that psycho-fancy where it matches the personality and the cultural background. When it's used in a recommender engine to buy more stuff, it's kind of diabolical. But if you use that to stimulate adherence to a pharmacologic regimen or a next best step in a treatment protocol or closing gaps in care. And I think about American medicine is hyper fragmented, hyper bolcanized, right? You got medicine over here and pharmacology over there and behavioral over there and SDOH over there. And suddenly, you have this unifying intelligence that can synchronize all of these pieces together. And yet, just a couple of months ago, JB Pritzker, the governor of Illinois, VTOT, actually put a moratorium on all therapy bots. And in his rationale, he said, we're going to protect jobs. And to me, that's unconscionable, right? You have an epidemic. I'd like to invite you right there. I'd like to see how many people he will kill from suicide. How many life years really ruin from depression because they can't get treatment? Here's the statistic. Even if you're doing regular human therapists, the number of sessions to reduce your PHQ-9 score to a level below the distance. The level below the disease threshold gets cut in half when you're using AI bots. You can use them in conjunction. Do therapy, but the patient gets better in half the number of sessions. How does he increase the number of therapists? So this is silly politicians trying to run for president, ruining patient care in this country. Having safety guards and all that, really good idea. Not using chat bots and having standards for care is really good idea. I just sent a note to somebody in the White House on AI. They should set the same bar, what improvement in performance above that of human performance. Does what I have to be to provide patient care? Now that's easy thing to put in place. Whether you're a nurse practitioner, a health coach, a physician, a loan broker, doesn't matter. All humans, they should be a standard set and AI should exceed it by a clear margin so that it can practice. I think that's the way it is and there's a deep hypocrisy in how we approach technology. We hold technology to an inhuman standard of perfection, not the standard of human equivalency or even human superiority. And so the risk here is these things are fallible. There are going to be mistakes and some of them may be even catastrophic. But to your point, how many lives did the governor of Illinois cost by proscribing this therapy? Which I would go so far as to say is not just a substitution for the human equivalent. It is superior to the human equivalent. It falls superior. The diagnostic performance is superior. The time to wellness is superior. It's possible to maintain wellness between episodes of depression. Let me give you my favorite statistic. If every human today on the roads in America drove as well as a way more car, this year we would have 30,000 fewer deaths in America. You would save a breast cancer level of deaths in America. Now the cars on the road, but there's a period they still have a superior. They're so superior to human drivers. We'd save 30,000 and that's that now my credit is we saved something similar medicine in mental health. I haven't done the math so something similar in drug abuse or in other areas. The note you're so right and autonomous driving is the sort of canonical example. I was just looking at this. There are 45,000 deaths every year due to traffic fatalities. The number three causes trunk driving, texting and distracted driving. All human caused and all immediately eliminated with autonomous driving. The math is very clear. If you just look at self driving cars like VAMO, 400 million miles, humans have one death. Self driving cars have sort of a third of the deaths. So you'd save 30,000 lives this year. Well, there's nobody better than you to make that to serve and then see that argument. Because I actually think there's something very fundamental about this impossible standard we hold technology to. Some of it's a little disingenuous because we are understandably fearful of the dislocation that AI is going to bring. And you talk about 80% of 80% of jobs within the next five years of all jobs. And it's deeply disconcerting. And I think we're going to see a ton of posturing. We're going to see a lot of litigation and we're going to see a lot of regulation. Some of it well intentioned, but a lot of it is pretty cynical. And I want to ask you about that. You know, my own intuition on this is that we're not going to see these advances in the United States. And you know, in my last paper, I put a chapter that somewhat irreverently said, America innovates, Europe regulates, China appropriates. And this new paper I'm chaptering a titling a chapter on clinical AI. And I'm borrowing Leopold Ashen Brenner's opening line from situational awareness. Would you turn me on to for any of our listeners who haven't read situational awareness? It was the most consequential AI piece of 2024. And actually had Leopold come to a dinner and do a talk and he's mesmerizing. But the opening line of situational awareness is you can see the future first in San Francisco. And I plagiarized his line on my clinical AI piece. And I said, you can see the future first in Beijing and Riyadh. And the reason I say that the note and I love your reaction is because you can do things in an autocracy or a monarchy that you cannot do in a messy western liberal democracy. And in other words, you can only innovate at the speed of the regulatory capture that is dominant in an industry. And our clinical industry is beset with regulatory capture. So I'm worried that we're going to see these advancements in the GCC in China. And then we're going to be reduced to reverse importation of those advances. And even worse, I worry geopolitically that whoever diffuses to the world, free doctors, free teachers is going to propaganda is their own political ideology. And so even healthcare is getting sort of politicized in this. But tell me why I'm wrong because I really don't like this opinion. Well, you are exactly right. I wrote a 25 page paper in August or September of 2024 a year ago. Exactly this issue of what's possible and what will AI lead to in terms of dystopia or utopia and how it might vary by country. You know, people worried about job displacement will slow it down. My bet is certain countries will accelerate it China being the most visible one. But let me say the following with the current regulatory environment. There's a few things that are entirely possible speaking to the 150 influential audience members you're talking about. Without taking under risk with a little bit of courage to internally manage their organizations and that's where courage is needed. You can double panel sizes provide five times the number of contacts per patient per year. Managed chronic disease and by the way, when I talked about $30 a month includes chronic disease management for all diseases, mental health, urgent care, primary care, all of that. That's possible today for the major health plans and pairs in this country within the current regulatory environment. Without violating any rules, maybe upsetting a few people, I think that's possible in will be a major competitive advantage. And so I hope somebody decides to make the move and moves the whole industry to what happened in the automotive industry with zealon and then self-driving cars. I've never been to a gas station in five years. Look, the node that brings us full circle because you started our dialogue kind of reflecting on this sort of mash equilibrium among incumbents. And whoever breaks ranks is going to precipitate the change. And I've been doing my own evangelism to the 150 talking about that dynamic. And I think there's enormous scope that is even outside the regulatory sort of like domain. But I want to, you know, with our, with our waning time here, I've got two more themes quickly for you. First is, you know, I want to talk about, you know, in my view, the two most important words right now in this technology moment in where it meets AI and health care is functional verifiability. And when I think about what that means, the fact that, you know, I look at a stratification among the models, right? You're seeing Google Nudge ahead in multi-modality and world models and science with what Dennis is doing and winning a Nobel Prize for chemistry for alpha fold. I look at what Sam's doing in really creating reasoning and sprinting headed memory and winning 100 to zero on consumerization and productization. I look at what Dario is doing and Dario and I are doing this podcast next week. So we'll go deeper into this. And in, and in coding and autonomy and now clawed 4.5 son at 4.5 can do 30 hours of long horizon and supervised coding and I look at health care and what is automatable. And to me, the areas that are going to be automated first are those that have functional verifiability meaning their subject to the laws of math. There's a right wrong answer. It's correct incorrect. It's provable. You can back propagate those answers into the algorithm, make it better. You can create high quality synthetic data and make the algorithm better. And the area that has that in super abundance is revenue cycle management. And we at Towerbrook and CDNR took the largest revenue cycle management company private R1 last Thanksgiving. And you know, was so honored to know that you participated with us. We invested together behind this. But our conviction was this is going to be the area of fastest automation and agentification. We partnered with Dario and and Thropic in this. We partnered with Alex Carp and Palantir Alex and I are doing this podcast this week in fact. We're going to talk about forward deployed engineers. And so I want to ask you about how do you think about these areas of objective function and besides RCM what's next in health care that is going to lend itself to the fastest automation or agentification. There's so many. It depends on each organization. So if you told any hospital CEO we can avoid your physicians ever touching epic. That'd be like a God's that so they just be chatting as if they're chatting to their MD intern who then touches epic and they never have to like that's the model that MD can be an AI MD. So I think that's just one area. I think it's entirely doable. But imagine if every physician knew every patient got a lot of follow up or they got more face time with patients as happened in the total study in the UK. So there's so many different areas by the way limited by what epic is I came to this country in 1976 and programmed in a language computer programming language called months that epic is still written it. And and so being captive to epic is a really bad idea. Most of this innovation this is what I would say to Dario and others is happening in startups who then take these models and switch seamlessly between open AI model and then topic models some open source model Google's models between them as they develop capability in different areas. But add all the specifics of health care if our compliance security checks safety tasks triage rails guard rails so much needs to be done and how to carry and custom train the model for a health care patient intake conversation or a patient conversation otherwise. All that is what Dario would call RL post training has to be done specifically for health absolutely with always added and I think the models are doing us a disservice by saying you can use them out of the box and I can service repair or provider. I think that's a major shortcoming on how everybody is. I agree and we're going to do that we are doing that with anthropic and I do think you're going to see a real orientation the node toward what you are prescribing because the generalized models are sort of you know they're omnipotent across so many different dimensions but health care. You really need an immersion you need you need you know to RL that that environment with great specificity and and I'm optimistic we're going to get there. Two more questions right the other failure mode I see is the health care organization using a site function to evaluate AI or to pilots honestly they're not qualified to do it if you look what distilled it with elements distilled is a system integrated that understands AI is AI centric. Pretty successful pilot with elements but many organizations that have tried this with their own people have failed at it so there are how to do it as important as what to do so that's one factor. There's another major factor many of these AI systems aren't like buying epic they won't come with a fixed functionality the functionality evolves every month even every week so it's a moving target you don't have a big manual of here's what the system does you have to get used to. This AI intern idea of somebody who gets better and better but the functionality that you buy keeps changing so it's a different mindset than software buying software I would add a critical component of that that I see another major failure mode positions I have seen. Providers a plan say hey give us to a startup give us your software and we'll have our physicians use it but if physicians don't change how they do their work flow they will get inefficiency not efficiency changing the workflow is really really important so my advice to systems start with the startup providing the full service once you benchmark how performances you can let your physicians on board so they have to learn to get the same level that the startup has achieved with its clinicians and and have a program over one two years they transition to internal physicians if they can change their work or how they do things as you know position behavior is very hard to get that resonates and I think when I have you back if you'd be so gracious to do it I really in future want to revisit this question because right now we're still in the frenetic capitalization and the infrastructure building phase and we're very quickly moving into the diffusion and installation phase and companies like distil and we obviously partner with Palantir for R1 I think they're going to have such instrumentality in this because you can't do this off the side of your desk you can't do this extra curricularly you can't ask incumbent health systems to have the competency to deploy the most powerful technology that we've ever seen created and I think the application and the diffusion is very much trailing this sort of multi trillion dollar infrastructure building phase and so I'm excited what distil does I'm excited what you know some of this new generation of forward deployed engineering companies and doing the translational work I'm excited about them and so we'll come back to that in a future conversation video but let me close with a big philosophical question for you ever since 2000 when you were interviewed with the New York Times and you talked about this new reckoning that humanity is going to have to have in a post AI world and then in 2012 you talked about you know you you rather provocatively titled you know a blog do we need teachers and do we need doctors and and you've been very sort of clairvoyant about how technology pays plays out in a piece you and I did together probably three years ago I subtitled one of the chapters from Keynes to Kozla and the reason I said that is because John Maynard Keynes in the year 1930 wrote a seminal article said economic possibilities for our grandchildren and he was looking around the industrialization in the UK and he was extrapolating it forward he's like you know in a hundred years pen impressionally said by 2030 we're not going to need to work we're going to have this post scarcity abundant society and he looked at the indolent British aristocracy he's like well shit that's not great because humans are built for scarcity humans are built for subsistence and he actually prescribed 15 hours of work per week not for sustenance but for spiritual nourishment and you've been saying for a long time and you're timing is looking pretty prophetic that by the year 2030 80% of 80% of jobs are going to get automated across all these different domains and my question to do is this I share your optimism about this post scarcity abundant society or instead of 800 million people out of 8 billion getting the good life we're going to move the 8 billion to that 90th percentile but I worry that you know the most articulate sort of philosophers and anthropologists like you and Dario and others are sort of skipping the little middle part where there's a revolution and what I mean by that just to just kind of complete the plot because I don't want I don't want this to be so grandiose I want to make it real like Mamdani just got elected in New York City when people feel disenfranchisement they are going to vote capitalism out capitalism exists at the pleasure of democracy if Gen AI exacerbates wealth inequality there's this sort of really turbulent transitional period between now and post scarcity what is the how do we navigate the middle two things I would say first I extend extensively address these questions in the blog I referred to earlier called AI dystopia or utopia that I wrote in September of 2024 it's 25 pages of addressing all the dystopic questions first and then addressing the utopic point of view first thing I would say is by 2030 very very likely 80% of 80% of all jobs will be doable by an AI and as Dario he will say I'm being too pessimistic is what he would say but I don't think they'd be widely deployed because AI is capable of doing a job doesn't mean it'll be deployed that will be a political and social and policy question not a technology question and I think different countries will deploy these at different rates in different areas and there'll be a whole social disruption and in this blog I talk a lot about the chaotic intermediate 2030s when deployment starts to be material and by 2035 when I expect we'll have a hugely deflationary economy a hugely deflationary economy because goods and services trend 10 towards free all education will be free all medical expertise will be free legal services will be free I could go on and by the way by then robotics will start to play a lot of a lot bigger role in the physical world between now and 2030 it is really intellectual work and software based stuff not so much the physical stuff but in the next two years it'll start to happen by 2035 robots will be fully deployed probably not in cardiac surgery yet because of the FDA and the regulatory timelines FDA involves but by and large in society we will have chaotic transitions I do think two things will happen one GDP growth will accelerate from 2% to I think well in excess of 5% by 2035 and I think it will be be who was and there'll be policy pressures to take some percentage of that incremental growth above 2% and create some sort of a national sovereign fund or safety nets we have a nalaska oil fund we have a Norwegian oil fund is there a sovereign fund to be had that captures some of the economic benefits of AI and helps the people who are left behind I think the minimum standard of living in this country in almost every area and I address physical things like housing will go up knocked out I love it well I read your stuff religiously and I somehow miss that I'm going to read utopia dystopia but no so grateful to you my friend as always I learn every time we speak and we're just we're we're lucky to have you not just for the industry but for for everything that we're doing across the country thank you very much well thank you it's well fun to work on these hardballs they are exactly thank you my friend thank you incumbents and insurgents in health care is produced by CBTS a tower broke portfolio company this presentation is for informational purposes only and does not constitute and should not be construed as investment advice or as an offer to sell or solicitation of an offer to buy any securities or related financial instruments the statements and opinions expressed are those of the hosts and guests and not necessarily those of tower broke capital partners LP tower broke makes no representations or warranties regarding the accuracy completeness or applicability of the content tower broke expressly disclaims any and all liability or responsibility for any direct, indirect, incidental, special, consequential or other damages or rising out of any individual's use of reference to reliance on or inability to use this presentation or the information presented therein