Latest / Two PhDs on a Pod / The Future of Medicine is AI
Transcript
- Dr. Guru Banavar: Hello, everyone. Hey, Momo. Hello, Guru. ⁓ okay, so today we are joined by a super, super exciting guest. He's one of the most influential entrepreneurs and venture capitalists of our time. From co-founding Sun Microsystems back in the 1980s to backing groundbreaking companies with venture capital in AI, healthcare, clean energy, and biotechnology. Vinod Kosla has spent more than four decades helping turn bold ideas into technologies that improve millions of lives. I'm really excited about this one. Yeah, Momo, you know, growing up and going to school in the in the computing world, I have heard about Vinod Kosla's many different ⁓ accomplishments, adventures, investments, and all that stuff. And I've always kind of looked at him as a you know model for many people to follow on entrepreneurship-wise, at least, right? And the the other thing I really love about Vinod, I'm I met him ⁓ around you know the same time that you you met him, which is you know, when we were talking about ⁓ investing in Vyome back in 2016, 2017. What I was really impressed with with ⁓ Vinod when I first met him was that he is sort of in equal parts a big Picture visionary. And at the same time, he's very familiar with all the minutiae of the technologies, of the new developments. And he's able to engage you with very detailed information about many things, both on the tech side as well as the medicine side. And so that makes him very unique. That you know he's got that breadth and the depth that he's able to kind of make the right decisions and everything. So I'm super excited about talking to him. Agreed, agreed. Just a very unique individual with a very rich background. So we're gonna get into it soon. So I am Momo, a biochemist. And I'm Guru, a data and AI expert. We're two PhDs on a pod. ⁓ what's on your radar, Momo? Well, currently, summer is absolutely on my mind every day. Here in the Pacific Northwest, we have Such long day daytime. You know, the sun rises in around five and sets around nine for like two or three months. And the weather is just perfect. It's like the highs are in the 60s and 70s, the lows are in the 50s, both on the coast and in the mountains. And it's just gorgeous out here. So next episode, I'm going to show some pictures ⁓ that I've taken because I'm a very, very intense amateur photographer. ⁓ but yeah, the summer here is just phenomenal. And then the other ex thing that I'm always excited about is cooking. Wait, before you go to the second one, I wanna I just want to mention my brother just moved back to Seattle. So I will be visiting probably more often or staying a little bit longer. I mean I already visit Seattle quite quite often because you're located there and our our biome headquarters is over there and so on. ⁓ but I will be there even more frequently I believe and maybe we'll have a chance to go explore some ⁓ some of the Pacific Northwest Wilderness. Love that. Love that. Okay, great. Yes. Good. ⁓ so that's on my radar. And then of course, cooking is always on my radar because ⁓ we live to eat and cook and so on. So recently, what I'm really excited about is that I've been doing a lot of literature, a lot of searching ⁓ the ancient literature about mole recipes. So mole is a traditional Central American sauce. And ⁓ now you can go to a Mexican restaurant and many of them will have moles. So they'll have a green mole, red mole, black mole, typically those three. and these were recipes that were actually developed by Native Americans prior to 1492, but they were pretty much reserved for royalty. They were not necessarily ⁓ well, the cocoa one was reserved for royalty and chile one was reserved for royalty. There were others that were used in normal households. But, anyways, I was able to research the history of mole. And there are pre-Columbian and post-Columbian recipes, and we decided that we should make a recipe from 1500s that the Spaniards actually recorded. So this is somewhere in the 1550s. First recipes of mole were documented, and we picked a black mole because I love cocoa and it's just so nutritious and so delicious. So yeah, we made a black mole with sort of like an authentic recipe that combines both New World and Old World foods. And it's just super rich and tasty. So very unique flavor that I don't think any other cuisine has. Nice, nice, nice. Is it is it hundred percent vegetarian? Yeah, it's all vegan. Absolutely. You can put the sauce you the the sauce is hundred percent vegan, and then you can put it onto any kind of ⁓ protein source you'd like. ⁓ the Central Americans obviously use turkey, that was their major source of meat. ⁓ but they had other sources of meat as well. And then today we can of course choose tofu or Anything you'd like. That sounds great. Momo, ⁓ it would be nice if you can put the recipe on the show notes because I want to try this. Yeah, I'll put the recipe in the show notes and I'll show pictures of all the ingredients in the next episode. So make sure you tune into the next episode. All right, so that was my radar. What's on your radar? All right, so I'll be short because I want to get to the discussion with Vinod as quickly as possible. So first, over the last couple of weeks, I have definitely had the FIFA World Cup. Football. I still call it football, although people in America call it soccer. I love watching those those matches, especially the ones with I call of them like artists, all the strong stars of ⁓ football, like Messi and ⁓ you know, Harry Kane from England. ⁓ and my favorite is this new guy who's shown up on the scene called Erling Holland from ⁓ Norway. He's like he's like a Viking dude. ⁓ big guy, very Powerful looking, but also very agile and very fast. And he's made all these goals and just amazing stuff. And today, actually, there's another match between ⁓ Norway and England. So I'm you know excited about watching that. So that's one thing. ⁓ one thing on my radar. The other thing that on my radar for the last couple of weeks has been a bunch of acquisitions that Vyome has been making and evaluating ⁓ for. Expanding our portfolio of solutions for good health and longevity, right? So the the idea, of course, of VIOM has always been that we want to understand biology deeply, prevent, reverse chronic conditions, cancers, and aging, right? So we want to provide all kinds of solutions, something that's super integrated into your lifestyle, like eating, you know, nutrition, supplements. And now we are also considering other kinds of interventions that can help you get to better health and longer life's ⁓ health span, I should say. So ⁓ in the not too distant future, I think we will announce these ⁓ acquisitions and ⁓ so you're keeping them secret for now. Yeah, I think we should keep them a little bit secret until we actually publicly announce it. But you know what? The way we'll do that is to maybe bring some of those people into our show and have them discuss some of the things that ⁓ ⁓ that they they have brought into our portfolio. So that's that's a radar. That's awesome. If we go back to the future, for those who watch this like a year from now, they can go back to the future and figure out, find out what acquisitions we made and ⁓ and he and learn about ⁓ So yeah, that's pretty exciting. Cool. Awesome. Let's get into it. Let's bring our s vis our guest. All right, let's do that. Okay. Well, welcome to another episode of two PhDs on a pod. And we have a super, super extraordinary human today. as a guest. So this is ⁓ Vinod Kosla and ⁓ he is going to talk to us about the future of medicine, integration of AI and ⁓ investments into startups and some big companies. And so I'm going to introduce him. I've known Vinod since 2016 and I'm just super happy to have followed him and to be ⁓ in discussions with him all the time. So first of all I think of Vinod as a great human being. That's to me the most important thing. He's super nice, super knowledgeable, patient, and reasonable. So that's super important. But let's talk a little bit about his background because it's pretty staggering. ⁓ he has an undergraduate degree in electrical engineering from the Indian Institute of Technology in New Delhi. And then he came to the United States and had a master's in biomedical engineering. So he knows quite a bit of biology, ⁓ which is very unusual for this kind of a profile. And then he has an MBA from Stanford. And then he co-founded Sun Microsystems. So that's one phase of his life where he co-founded Sun Microsystems in 1984 and really shaped modern internet and enterprise computing. And then he moved over to Kleiner Perkins, which is an investment company and backed many pioneering in internet infrastructure companies such as Juniper Networks. And then in 2004, he started Kostla Ventures, which he has been running now for 22 years. And he's a great investor, one of the best and won many awards and accolades and so on. And Vinod is really visionary in in understanding what are the companies that are going to change the world. He's not thinking small, small projects and how to exit quickly and make some money. He's really thinking how can we use venture capitalism to back companies that are going to transform the world in many, many different areas. And so ⁓ that's really I think a foundational difference between Vinod and all. most other investors. So he has consistently championed technologies that are aimed at improving health and creating a better future for the world. I'll just list a few companies that he has backed very early, like one of the earliest investors in open AI in block or square that ⁓ it it includes, you know, that allows small merchants to to be a part of the financial system. And then DoorDash and Rocket Lab and Garden Health and also Viome. He's ⁓ series A major investor in Viome and has been such a big supporter of IOM. So we're just super grateful. So Vinod, thank you so much for so many things for as a member of the society and as a co-founder of IOM and so on. Welcome. It's an honor for us that you joined us. No, happy to happy to. Super. Absolutely. Yeah. So let's focus on. ⁓ ⁓ the future of medicine, focusing obviously on preventive medicine and integration of AI. So I'll just start off with okay, so we're shifting towards preventive medicine. We now have the tools and knowledge to get more involved because so far medicine has been to a very large degree very reactionary, waiting until symptoms set in. So tell us please about your sort of grand vision of ⁓ true preventive medicine and how AI fits in that model. Well, the first thing to realize. Is AI has made enough progress. Modulo a few things that need to be fixed, and we can come back and talk about it. The expertise of any human doctor or researcher can be built into NI quite easily today. So the idea that we should gate or limit The amount of physician time is a silly idea now. It made sense when humans had to do the job and you went to a nurse first and then a doctor, primary care doctor, and then a specialist. So a the more skilled a professional, the more gated access to them was. You went through four doctors before you saw a neurologist or a brain surgeon. Because they were more expensive. That notion is collapsing very quickly because the cost of all expertise is a dollar or two an hour, and we shouldn't be getting it. That comes back to your question on preventive medicine. We don't use experts in preventive medicine because they're too expensive. But if they're near free, it's the cost computed. then we should use them abundantly at as first line therapy. So preventive medicine becomes much more feasible in the world of AI. We can also make it very available to the consumer and we should talk about that. So I think the big i issue in healthcare is engagement. Yeah. People don't think about their health because they don't deal with it very often. Exactly. In fact, the most frequent thing people deal ⁓ with their health on is supplements or things like that. ⁓ There was a recent study on Alexa if you just reminded people of a chronic disease like hypertension or diabetes by simply not anything fancy, just simply talking to them in the first thing in the morning with two or three questions, which is a reminder that you have the disease really more than anything else. There was a dramatic reduction in blood pressure and blood sugar. Wow. It says just reminding people is sufficient and they're a little more cautious that day. ⁓ so in other parts of social media, we've learned how to engage users. This is one little example. But In the world of Facebook and TikTok and Instagram, we've learned how shades of color on the ad behind you dramatically change engagement. If we could do something similar for healthcare in engaged people in some sort of More engaging thing. Like, do you know the most recent signs relevant to your condition? ⁓ and I don't know exactly what that is, but we've done it in so many other spaces that we should be able to do it in healthcare. If we do, we've made huge progress. Okay. Then downstream of that, we can direct a lot of behavior and take advantage of the fact ⁓ we were talking about earlier, which is expertise is all free. Give them every specialist because they all cost a dollar or two an hour. Yeah. ⁓ go ahead, Guru. No, no, I was I was gonna say, you know, every time I make a decision in my in my daily life, right, wha what to what to eat. You know, I eat ⁓ three times a day or and snack and all of that stuff. And every time I go out and you know to a restaurant or to buy groceries or you know Do things with other people that have to do with my health, which is, for example, exercise or, you know, play play a game or something like this, right? Every time something that is relevant to health happens in my life, I think your point is that if you kind of somehow engage them and say, hey, this is what it means to your health, that keeps it front of mind and maybe it'll it'll change their behavior for a day. That makes a huge difference. Yeah, no, it will make a huge difference. So ⁓ you you've seen a lot of talk about AI that sees what you see, hears what you hear, there's glasses, ⁓ that may be looking at your food and looking at the conversations. It uses that, yeah, that ambient listening and watching to trigger behavior, say, Maybe you should ⁓ have half that dessert. Exactly. Yeah, you don't need four scoops of ice cream. ⁓ I love this idea, Vinod, and I think that in the since you're a big ⁓ investor, of course, I think this can potentially s inspire many potential entrepreneurs here to think about how can they create a social media platform that both entertains people but also provides this sort of this sort of ⁓ visual or audio guidance during the day. I you know what? I I would call it ambient health, Mama. It's like you know, it's ambient health. It's like it's around spontaneous health, daily health, everyday health, something like that. So entrepreneurs, think about how you can leverage existing technologies such as ⁓ you know, video glasses and ⁓ audio listening devices to basically remind people. And and like Vinod said, maybe it's not something so intrusive that it's like every minute of every day, but even just a gentle reminder just in the morning or maybe a couple of times a day is gonna make a big difference. So I think that ⁓ basically a company that will engage people in preventive health, that's the key. And and and ⁓ you know, come up with a business idea and and pitch it to Vinod and maybe ⁓ maybe there'll be a big successful company there. But yes, somebody will get it right and then we will be on the path. Because you can't have clinical interventions till first you have engagement. Exactly. And by the time you have clinical engagement, it may already be too late, right? You know, because you know, these chronic diseases take so long to develop that by the time you realize that you need to go talk to a doctor, it's already probably quite a bit progressed. Super important point, yes. Okay, so we have basically a part of the the demographic that is not yet engaged in preventive medicine, but we do have a part of demographic that actually want actively to pursue preventive medicine. And those guys are watching YouTube videos, they're listening to podcasts, they're reading scientific papers. They're really into this. They're testing different supplements. They're testing different diets. ⁓ how can AI today, and and what's it going to look like two, three, four, four years down the road, enable those people who are engaged to actually practice preventive medicine ⁓ without having to go through four layers of specialists and and doing a lot of Testing that's guided by different doctors. Tell us about that, please. Well, again, AI can help a lot. You know, in the past, ⁓ first it was expensive to take a blood test that had, say, 100 variables every quarter. It's sort of what I do. I measure 100 blood parameters every quarter. ⁓ first it's expensive, but then you have to have somebody go. Through the results and say a result of this test, you should do X. If your vitamin B12 is low, that's a pretty simple interpretation. Right. But as you know, many of these relationships are complex and multivariate. Yep. ⁓ and this is why I believe most of the studies that say vitamins and supplements don't help. are wrong because they didn't measure anything in the blood and then recommend a supplement. They just put together everybody taking the supplement and said, what was the result? If you don't need a supplement, then it's gonna have have no effect. And if 90% of the people don't need that particular supplement, ⁓ then the results will be washed out by the ninety percent instead of the ten percent who benefited. Right. So I think the very nature of these studies have been pretty flawed. I haven't seen a good study. Well, I'd like to add, you know, that also they're short term. We have to think about our health as decades pr decades long process. And so giving someone a multivitamin pill for three months and then looking for clinical outcomes. You may not observe any changes in three months. It's just too short of a period. Absolutely. Absolutely. So, you know, and most diseases, most of serious, you know, used to be we were in the world of infectious diseases. We are now in the world of mostly chronic diseases as being the bulk of healthcare in the West. Yep. ⁓ Chronic diseases start 30 years before they manifest. I remember doing a calcium scoring test when I was 30 40 years ago. Wow. And I was in the 90th presentile, not unusual for Indians who are very at risk of cardiac disease. And I said, it'd be nice to do. I talked to my doctor, and he says, ⁓ don't worry about it. You're 30. Yeah. I I wish I had worried about it. So the establishment is not very good. Now AI can get very, very personalized and help define strategies and maybe even give you real Trade-offs. If you wanna put in this much effort, this is what you should do. Okay. You know, like give you a menu of choices or remind you periodically what the trade-offs are. So you can get extremely personalized, hyperpersonalized in with AI. Labor you can't afford when you're using a physician making two hundred and fifty thousand dollars a year or a nutritionist or something. So ⁓ hyper personalized attention, ⁓ attending to these details, deciding which is the right trade offs. I'll give you my favorite study. It's it's it's almost is unbelievable for patients visiting the NHS service in the UK. Just from their medical records. They're not visiting for cancer or oncology or any specific condition. You screen them, you can tell with fairly high sensitivity and specificity whether they have cancer from just the data in their medical record. And NHS records tend to be more complete than records here in the US. Wow. Way more sensitive. Then a t a test we think of as early cancer detection like Grail. In fact, I would say they've detected a hundred times more cancers in the UK than GRAIL has detected worldwide. And ⁓ and this is done by AI or some algorithm that are that are computer algorithms. ⁓ and so it's it's a ⁓ dramatic enrichment of the population that should be tested for cancer. Right. When you have a generic screening test like the Grail test, one percent of the people may have cancer, but the 99% false positives will ⁓ will make ⁓ swamp the results of the 1%. Right. And so s These statistics have to be understood well. You can't run screening tests unless there's some other criteria for screening. Otherwise, they're not cost effective and they lead to a lot of heartburn from wrong information. But if you can enrich a population, you have a 60-70% probability of having cancer, then it's worth looking. Then finding it six months earlier or 12 months earlier. Which is what we found. ⁓ it's pretty damn good. Okay. That's worth spending the money because you'll save the money. And that's I don't know whether you call it preventive or post preventive, but pre these whatever you call it. Early detection. Screening. That's when it becomes really cost effective. Yeah. So so this is now a tool that is integrated into NHS or not yet integrated? Well, they're doing it by patient record. For primary care practices that want to screen their patients for it. Okay. And is is this at the order of half a million people ⁓ have been tested, and fifteen percent of the people had an anomaly of this sort. Now these are patients seeking help, so they're not a random population survey. Right. But still pretty impressive. Pretty impressive. So ⁓ so let's talk about when we talk about AI, let's get a little more specific. AI is sort of this ⁓ system-wide capability, but let's talk about a little more specific. So let's say there is a 50-year-old person and they want they they have not paid attention to their health, but now they're they woke up and they're like, Wow, I really want to intentionally every day, I want to do that. They're inspired to actually make daily changes in their nutrition and lifestyle. How can AI help them today? Do you recommend something like ChatGPT or do you recommend what is your recommendation today? You bring up an important point. I don't recommend things like Anthropic or ChatGPT straight out of the box. Okay. Why? We all know these systems hallucinate. Okay. So we just did a study using GPT-5, and the same is true of Google's AMI product. So And said, how good are they at triage? If you look at triage, they generally sound pretty damn good, but you look at the triage error rate, it's a north of 30%. Because of hallucination, because of other things like missed questions, questions the AI failed to ask in doing its differential diagnosis. What we did is my son runs an AI primary care company called Cura. We have a whole harness around GPT-5. So they're using GPT-5, always using the best model. But the harness adds a whole bunch of other systems to complete the diagnosis, to add safety checks, do the appropriate triage when you're uncertain to a human being for now. All these systems, the error rate. went from 30% to 0%. Well it's a formal study they did. Okay. So these systems with the harness ⁓ reduce error rates. And that's what all the coding startups have also done, added harnesses that put constraints in safety check systems and ⁓ and that's the right way to do it. Okay, so these are specialized medical. Let me ask a question about ⁓ about the harnesses, ⁓ you know, because I think this is an our audience here is both medical as well as AI oriented. So so let's just get into a little bit more detail on the harness part. So does the harness use ground truth of any type that is that is curated well, or is it is it ⁓ you know, human in the loop type of training for a you know significant period of time? What are the techniques people use for developing these harnesses? So there are many ways. There's literature, there's old systems. We've used every possible system. We've used medical records. ⁓ none of them are particularly effective, but a safety system we've built of multiple experts, sort of ⁓ to put it more technically, mixtures are mixture of experts approaches. with each system having a different slight bias, one for accuracy, one for imagination, one for exploration, you start to get the right answers. And so it takes years of iterative development beyond just GPT five. But you want to do it in a way when GPT six comes out, you use GPT six. You don't set back to GPT four where you develop the system. So the harness has to be independent of the AI system you're using. ⁓ I saw this ⁓ publication in Nature called The Fragile Nature of GPT-5 of four or five in medicine. I looked at it, I said, it's a complete abomination of how bad the researchers were in defining the problem. Wasn't GPT five. They just used it the wrong way. Because when we used it the right way, none of that showed up. Now the flip side of it, which is equally important to talk about, is what I'll be comparing against. Yep. So Annie Milstein, who's probably the best person on quality of medical care in this country, very traditional MB ⁓ sits on most of the quality boards and things, did a study of complex diagnosis. at a mul ⁓ it's a multicentre study with multiple academic multicent ⁓ medical centers. A human accuracy in doing complex diagnosis had seventy three percent accuracy. Which means twenty-seven percent of the patients got the wrong or suboptimal diagnosis. Or no diagnosis. Or no diagnosis. Yep. That's like if you're driving to work in the morning, only once a week will you have an accident. Right. Worse than that. ⁓ it's it's so we have to compare to what is the best we have available, which is good doctors. Yeah. And I I'm not critiquing doctors here because medicine has improved every decade for the last hundred years. It's gotten better and better and better. But it's not good enough if we have a better alternative. When we had no alternative, we had to use that. ⁓ but when we do, let's use this. AI with the right kind of safety and triage and other ⁓ accompaniments is so much better than humans. We should always use that first. Okay. All right. And we should we should get to this issue of what should what what will it take to do that? ⁓ there's a fundamental problem in the FTS system, but if that's available, let me finish my story. 73% accuracy. Then they used an AI system, it had 88% accuracy. Then they gave the AI to the doctors to use. And guess what? The doctors improved, but degraded the AI. So they improved from 73% to 76%, but the AI degraded from 88 to 76. Yes. So doctors made the AI worse because they have pre-built biases. Yep. You have recency bias. Recently I saw a patient like this, or I just read about ADHD in the New York Times. So maybe you have an ADHD diagnosis. It's well established. The New York Times publishes article on ADHD. The diagnosis goes up the following few weeks. ⁓ huh. Because there's this recency bias. Humans have about twenty different types of biases. ⁓ I first met Arnie Millsteen, because many years ago he was characterizing the kinds of biases that show up in medicine in human-driven medicine long before AI was even a thing. ⁓ And so these biases do show up. Yeah. And so we have to say the right AI systems can do much better than humans, and every diagnosis should start there. My son's company, almost all the dialogue is handled by the AI directly with the patient today. Almost all of it. The AI might do 20 questions back and forth. Because of regulatory requirements, when the AI is done, it makes a diagnosis, maybe any testing recommendations, maybe prescription recommendations, including dosage and all that. But they can't give it to the patient. So a physician comes on qualified in that state and asks a few more questions. May I add three or four more questions? And then they can issue a prescription. Which is fine for now because it helps ⁓ verification of the system. but in the end, the right system should be able to work. So that leads to the question: what's the right way to say, hey, GPT 5 is not good enough, but QRI system is better, or Enthropic isn't good enough, some other system is better. We have a process for that, which is called FDA approval. ⁓ If we have FDA approval of software as a medical device, which is a known category, ⁓ then we should be able to do it. Only one problem. You you have to do it de novo or you can have a predicate device. I've argued with the FDA. Humans should be a predicate device to compare any new software device performance against. Very interesting. ⁓ Congress never anticipated that, so it's not in the FDA mandate formally, either included or excluded. So Congress just has to c clarify language, saying, Okay to use humans as a predicate device, in which case ⁓ you would have a mechanism to approve a new doctor say it performs better than humans. That makes sense. Wow. ⁓ so that's what's needed, and it doesn't matter whether it's oncology or radiology or primary care or gastroontology, it's the same mechanism. ⁓ nice thing in AI is the primary care physician is also a great gastroontologist because it's the same AI, or the ⁓ or the best ⁓ endocrinologist because it's the same AI. ⁓ people haven't grabbed this notion. So we keep putting AI back in the old context instead of imagining how the system should look in this world with this. So that's probably the starting point. There's a couple of other things that we don't talk a lot about in medicine. A big thing in medicine in the US is medical fraud. Every AI encounter will be audited, auditable. In fact, you could audit 100% of them offline. And so you could ensure nothing got coded that shouldn't have been coded. Because you have the exact Conversation between the AI and the patient, and another AI can monitor for fraud. That's what we do in financial systems for fraud. We have systems monitoring the transactions and fraud and history taking and all that stuff. So we could eliminate fraud, and that would really remove a lot of the constraints. Half the constraints in medicine are because we are afraid of fraud. Wow. So there's a lot of consequences downstream for this system. Okay. I have some questions we know. So going back to that fifty year old who wants to get into like basically longevity medicine, th that that person wants to delay the disease onset for as long as possible. So there's this cure AI, which is your son's company. ⁓ is that covered by insurance? Can can anyone just go to cure AI and and get healthcare there? Yes, they can. ⁓ and QRI will build their health insurer. Okay. ⁓ by the way, you can access QRI on the Walmart website too. Okay, great. Okay. So so that's basically the democratization of AI access and healthcare. That's basically the But there's still a problem. Okay. The American Medical Association is the one that approves CPT codes. They approve got what gets paid and what doesn't and how much. It's based on physician time. Physician time in AI is approaching zero, right? And every time Cure I does a visit, some doctor doesn't get an in-clinic visit, so they don't get paid. The AMA has generally been opposed to the use of AI because it reduces their billings. And most insurers. against their own interest are paying less for an AI encounter than for a in clinic encounter, which is silly. You know, ⁓ in clinic encounter might cost $150. Your patient goes to a doctor, that visit is a bill $150, depending upon where and who and all that. If a patient does telemedicine, they get paid fifty dollars by the same insurer for the same visit. When it could be ten times cheaper if it was AI only, with much more auditability, much more accuracy, but insurers aren't yet paying at parity because they have doctors and the AMA controlling payment mechanisms. The fundamental problem of AI in medicine is not AI's performance in medicine. It's the reimbursement of AI-based medicine and there are established interests who benefit a lot from the current system that don't want it to change. That's a big problem. Yeah. So I wrote a blog in the beginning of 2025 for JPM saying AI in medicine really works well in capitated care or at-risk care or Medicare Advantage. Not When you're paid per visit, because then you're trying to maximize the number of visits, not maximize patient care. Wow. In Medicare Advantage, all encapitated care or IPAs, you are trying to minimize the number of visits, yet meet all the patients' needs and not have their condition escalate, where it becomes even more expensive for you because you're at risk. Yeah, Of course, there's this additional issue. Most patients are paid by an insurer for one or two or three years, but not for 10 years. And so if you do much more preventive medicine, you're not going to get the benefit for four or five years, and the patient may not be ⁓ your patient as a payer. So so ⁓ we know ⁓ along those lines I had ⁓ I had sort of two ⁓ points to to connect to. preventive medicine. One one, when we focus on preventive medicine, you may not have enough encounters with doctors or claims or lab data or anything else. Therefore, you may not have data in the in the in the in in the UK Biobank or in the in the ⁓ NHS L system, right? Because you're still working like you normally work at you know in your day-to-day life and you're trying to do the best you can to prevent ⁓ diseases. So number one, do you think that molecular data, let's say non-invasive molecular data like like what we're trying to do in biome, right? With stool, with saliva, with fingerprint blood blood and so forth, is that a a necessary, essential component for achieving this idea of preventive medicine even before you engage with I'm glad you brought this up. Because I didn't answer Momo's question earlier around supplements. You know, it may be one supplement, but it may be you need three things fixed for the system to behave differently. ⁓ there's a great set of ⁓ science in network medicine by ⁓ by a professor at Northeastern. He has a book called Network Medicine, where it's gene networks that affect The result. So it's not a single point of intervention. Now, taking beyond supplements and all that to molecular medicine, if you're going, if you're knowing what's happening to the transcriptome of the body, it helps a lot to know what genes you're trying to regulate. So today's clinical medicine, humans can Barely do because there's so many, so much science, new science every year. When you add molecular medicine, what do you do with a million transcriptomes? And what do you do with a million transcriptomes for the microbiome and the corresponding body's reaction of the human transcriptome in the blood? Which is the exciting part about wiome. You know, you get what what the microbiome's doing, and you get the body's response, the host's response to the microbiome, which depends on everything you eat. That kind of complex system can only be tackled by an AI. There's no hope a human being can ever do complex systems. So whenever a system is high-dimensionality, humans can't handle it. Yeah. You know, many, many years ago, I think it was 2001, I was a postdoc at the Santa Fe Institute. And I did a little bit of work with this professor who was there. I was a postdoc for four months. ⁓ my geeky side, I took a sabbatical from Venture Capital to be a postdoc at the Santa Fe Institute. I moved there on with the whole family. ⁓ the the human body, just in the blood is A few thousand metabolic pathways interacting through one mixed channel, which is the blood. Each marker doing multiple things. Yeah. You know. So it's not a simple system. And biology didn't design design it to be elegant, just to be functional and efficient. We I can re inverse that and build a model. For human biology. That's real medicine. ⁓ in 2016, I wrote a hundred-page document called the 20% doctor included. I was wrong, it doesn't need to be 20%. I was trying to be polite to doctors. ⁓ but these complex systems are far beyond human. Comprehension. Consumption. You no longer let humans mostly fly planes. It's too complex. Why should we let humans manage a system with three thousand metabolic pathways in one complex system? Doesn't you know, you add other organs like the brain and organoid organ level behavior. You're talking about very, very complex systems only AI can handle. There is no other way except AI. Humans have no role in this kind of molecular medicine. Yeah. And when we do it the right way, with the kinds of stuff, data that ⁓ Wyome is collecting, the variables are collecting, all that interacted together. You're going to get much higher quality ma medicine. At a fraction of the price and much more globally accessible. Yeah. Right? ⁓ in 2016, I wrote about this and I talked about molecular medicine in 2016 and what the AI would do. Yeah. It's on our website, it's on the web somewhere, 20% doctor included. ⁓ I said, very likely, and I today I believe this more than ever. In a village in India, I will get. Better cardiac care than I will get at Stanford. Why? Because the human expert will still be here, and there'll be no human expert in this village in India. So we'll rely on AI and get better care. They l they will leap from the human expert and go directly to AI. Absolutely. Yeah. And so I do absolutely believe this. I wrote a first wrote about all this in 2012, a blog in TechCrunch called Do We Need Doctors? In twenty sixteen I expanded it into a hundred page thesis on medicine. ⁓ and you know, we don't know how bad medicine is today. It's the best it can be, it's the best it's ever been, but it's still pretty bad. And I define where it is today or where it was in twenty sixteen and go on to say where it needs to go. But molecular medicine is absolutely a critical part of this. Thank you. So 2016 to 2026, 10 years ⁓ since you wrote that paper, right? Do you see all of your predictions from 2016 happening today? And and and then I want to know what you think will happen in the next 10 years, like 2036. I would say in 2016, I thought it would take 25 years or till 2040 to get there. I was wrong. It's happened now. So In twenty sixteen I was wrong by fifteen years in the first ten years. ⁓ I was too pessimistic. I think Look, the thing that gets in the way is regulation. The thing that gets in the way is human biases and humans in unwillingness to believe that AI can be better. And the systems that are in place already. Yeah. And so we can talk about what it will take to break this system. Yes, please. Let's do that. and extend it all the way into molecular medicine, wearable medicine. You know, we can read facial expressions. There was a great sleep study, ⁓ a foundation model based on sleep studies out of Stanford recently. Detected 120 disease conditions. Now, granted it was retrospective, so it's not a true study, but ⁓ we even with the cadre, they detect it 120 different disease conditions in patients from just their sleep study data. These are patients who went into a clinic, sleep clinic, to get a sleep study done because they had a sleep condition. Nothing to do with the 120 to 119 other conditions they may have, but all detectable in EEG data. That's pretty stunning. Even I couldn't believe it. All those signals. will come to pass and be part of medicine by twenty thirty five if we let it, if the establishment lets it. Imagine cutting the cost of healthcare in half. How many people would object to it? Is there bread and butter? In fact, very likely in 10 years we can cut it by 75%. I just saw an article on the first ⁓ hu ⁓ human droids doing surgery on animals. ⁓ I just saw that this week. ⁓ so everything can be reduced by 75%. Even medicines can be changed. We haven't talked about ⁓ pharmaceuticals. Yeah. ⁓ We are starting a new effort to say, so on our website is a talk, ⁓ Sid Sidbrand. Sid was a CEO of a software company, GitLab. So mo any software person knows GitNab. He was a founder in our portfolio, company done real well, went public, then he got cancer. And the doctors gave him very little time. So he decided he'd be his own medical director, orchestrate his own care. From directing his care to designing his own cancer vaccines with AI, to designing small drugs, everything he could throw at it. Wow. ⁓ five years later, he's still alive, doing real well. And this was before AI was mature. Wow. And now AI is mature. So a software engineer with no knowledge of medicine. outpaced any possible ⁓ Treatment, the medical establishment through ⁓ you know, the care pathways they had established and pri ⁓ could do, he did much, much better with no knowledge of medicine, just he needed to save his own life. Now So we know how do we get past the the establishment? I mean, this this is a this is a big question for the next 10 years, so please help us think through that. So what's different today? Almost certainly. Most cancer patients will be able to design their own care and have it be cheaper than taking Kruda. We have company called Nabla, it can design an antibody to any antigen. So we can figure out the antigens on your cancer cells, design ⁓ antibody. Soon we'll be able to do bispecific or trispecific antibodies for for your cancer cells. So personalized care, what I call in equal to one medicine. Will be possible. Why? Because you the expertise is free, so you don't need somebody to design one drug for a billion people. It's so cheap you can design it for one person. You can bypass regulatory mostly because you can't have a trial if your population size is one. How do you do a trial? So you'll probably have to go through approval of the process, not the drug. Right. ⁓ so lots of interesting things ahead of us, lots of uncertainty, lots of regulatory barriers, but a pretty exciting time. Amazing. Very, very super amazing. We could go on for hours, ⁓ but let's go ahead and switch to investments. ⁓ we had a series of questions for you, but we're just gonna narrow it down to one. What are you excited about in terms of companies pitching to you for investments? What do you think? ⁓ what do you think? What like what are you hoping for in the next let's say twelve months you're gonna get a pitch for that you think is gonna make ⁓ the biggest advancement in this integration of AI into healthcare? Well, very narrowly I would say and equal to one medicine. And whether it's for inflammatory disease, you know, you got tens of billions of dollars of value of single drug that's supposed to build ⁓ serve seven or eight billion and the anti TNF inhibitor. Right. But you could do it for one person. Based on their transcriptome. Yep. ⁓ maybe you have ⁓ CART T therapy for their immune cells that are their autoimmune cells. ⁓ it's starting to happen. And it's starting to be tried in diseases like long COVID and MECFS. Yeah. That's pretty exciting. Okay. But it needs to be scalable as expensive. well with in patient engagement. That's not fancy. That's sort of normal software product management. So there's a wide range. I think most diseases are curable in the next 10 years. Yep. They can get very specific to a patient. Detecting chronic disease early and preventing it for people who want it prevented. I wish I had some tools 40 years ago. I'm 72 now. When I said, ⁓ I got my calcium score as a curiosity, and then when I got it in the 90th percentile, I got scared, but nothing to do. Nobody would do anything in the medical establishment. Right. But by the way, I'm probably the only person who's had cardiac surgery bypass without ever having a symptom. Bypassing every treadmill test. ⁓ I just said I'm looking at the level of LAD occlusion, I'm going to have a bypass and then I'm gonna do preventive care on top. Yep. in And I've been very lo I believe that's saved my life. ⁓ so preventive, yes, preventively. I don't consider myself having diabetes, which I did at age 40. I don't consider myself today as having cardiac disease, though my medical record says I had bypass surgery. I do annual checks on all these things at fine-grained level. And I I design solutions. But this is Not possible for every person, but when it's embodied in an AI that can go to the molecular level, it'll be possible for every person to do. Amazing. Absolutely amazing. You know, you know, as a as a final question, can I ask one last question before we go? ⁓ so ⁓ you know, there are people who believe that longevity medicine is getting to the point where in the not too distant future, I don't know exactly what the time frame is going to be, but in the not too distant future. Every year that actually passes, you will actually get a little bit younger than a year because because of that. The right way to put it is it's reasonable to say by 2035 or 2040, you'll gain more than one year of life expectancy every year you're alive. There you go. There you go. So so so that'll happen in in in in in your lifetime, our lifetime, right? So so you expect that to happen in 2040. And what do you think are going to be the biggest breakthroughs to get there? Well, it's hard to tell. There's simple stuff to do. I recommend anybody over the age of I used to say anybody over the age of 65, now I say anybody over the age of 55 should take rapamycin. Okay. It's not proven, though I personally finance some clinical studies in things like long COVID and MECFS. And the results are pretty damn good. most people don't know about these studies, but they were done for a disease condition, not for general longevity. But the data is pretty the risk reward is very much in favor of taking rapamycin after age fifty-five. I see. And it's an immune regulator, not an immune suppressor. Which is where most doctors think it's an immune suppression agent. It's an immune regulator. Depends on the concentration, right? One can never be certain. So the risk reward is clear. There are other things you can do that are higher risk and higher reward. And the older you get, the more risk you should take. The younger you are, the fewer the risk, but you have more better chance of fixing the problem. or not letting it progress. So every person has to make their own decision here on longevity. No question. We're starting to make rapid progress on at on point solutions in longevity. At some point it'll evolve into ⁓ a full systemic solution for longevity. ⁓ the good news for the first time ever, And the FDA didn't consider aging as a disease. It gave us aging as an indication in dogs. In one of our companies called Loyal, where you can have a claim of longevity or aging. And and that's a good start. Doing it in dogs first then lets you go to humans. Because age wasn't considered a disease, it was natural. And but people didn't consider the fact that almost all diseases of aging go up dramatically in risk ⁓ as you age. So they are all diseases of aging. Though aging may not be a disease, it should be considered a disease. Yeah. I mean not a pain of living forever either. I'm much more for health span over lifespan. Society would be too ⁓ static too many MAGA people still standing hanging around. Yeah, we need recycling in the society. That's that's absolutely recycling in society. Fresh ideas. ⁓ excellent. Vino, it's been a huge pleasure to host you. We value your advice and your opinions. And ⁓ we will see you again whenever ⁓ we can get together and and host you again to see how your predictions are working out. We look forward to doing that again. Thank you so much. Great. Thank you. Thank you, Vinod. Best wishes. Thanks. Wow, that was great to have in hode and we could have talked for hours. ⁓ he's got just so much experience and knowledge. It's amazing. Yeah, it is. It is. Yeah. All right. So ⁓ just a reminder. You what? Hopefully hopefully, hopefully we'll be able to get him back on the show in in like let's say, you know, six months or a year or something and talk about all the progress that's happening right now. Of course, we can get him back anytime. He's a friend of both us and Viome. So yeah, it should it's gonna be very easy. ⁓ Okay, so the Vio Monthly Challenge, ⁓ just a reminder that for July, we're still just ⁓ looking for good questions posted on our episode. So just ask away and we will pick the best question posted in July, and ⁓ that person will win a full body intelligence kit from Vio. And then go ahead and ⁓ guru, what's in the next episode? Yeah, yeah. So ⁓ our audience may remember that ⁓ we announced we ⁓ have a new product called Gut Health Pro now for providers, gastroenterologists, longevity medicine doctors, functional medicine doctors, and so forth who are interested in gut health, right? So this coming episode, we're gonna have a deep dive into the gut health pro product. We're gonna talk about what the capabilities are, what the utility, the clinical utility of this test is. We're gonna talk about many use cases, ⁓ decision support that could be helpful. For people and so forth, specifically for doctors and patients also to know what would be useful for for them from this test. And we're gonna have a special guest who is a gastroenterologist. ⁓ name is Dr. Mike Bass. ⁓ he's ⁓ definitely a very ⁓ you know valuable member of our team and ⁓ also practices ⁓ gastroenterology every, you know, every day, every week, and ⁓ has a lot of experience in this space with ⁓ both ⁓ entrepreneurial activities as well as his ⁓ his MD-based ⁓ medical practice and so on. So it's going to be tremendous talking to him about the full s spectrum, the breadth and the depth of of the gut health protest. Yeah. A sneak peek there is ⁓ what we talked to Vinod about molecular medicine. This is truly an example of a product that's currently available on the market. Providers can actually order it today and it's really measuring the deepest, deepest corners of human biology, understanding which parts are functioning fine and which parts are not functioning fine, and how to fix those parts that are not functioning fine years or decades before a disease onset. So that's pretty amazing. Absolutely. All right, so I'm Omo, a biochemist. And I'm Guru, a data and AI scientist. We're two PhDs on a far. See everybody ⁓ in a couple of weeks. See you next time. Thank you. See you next time. Bye. HGs on a pod is a Viome Health Science Podcast production. Written and hosted by Viome CTO Guru Banavar and Viome CSO Momo Vuyisic. Produced and edited by Lisa Shomo. Podcast art by Kevin Wu. If you enjoyed today's episode, like and subscribe for new episodes every other week. We'll see you next time.