Debug Information

Processing Details

• VTT File: HYMANDIGITALLLC2059275436.vtt
• Processing Time: September 11, 2025 at 02:02 PM
• Total Chunks: 2
• Transcript Length: 81,425 characters
• Caption Count: 630 captions

Prompts Used

Prompt 1: Context Setup

You are an expert data extractor tasked with analyzing a podcast transcript. 
I will provide you with part 1 of 2 from a podcast transcript. 
I will then ask you to extract different types of information from this content in subsequent messages. Please confirm you have received and understood the transcript content.

Transcript section:
[00:00:00.320 --> 00:00:02.000] Coming up on this episode of the Dr.
[00:00:02.000 --> 00:00:02.880] Hyman Show.
[00:00:02.880 --> 00:00:11.840] The thing about like replacing doctors, the line that I really like, I think it's Eric Philpel's, which is AI won't replace doctors, but doctors who use AI will replace doctors who don't.
[00:00:11.840 --> 00:00:16.640] And I think that is a really good way to put it because it is a tool.
[00:00:18.560 --> 00:00:25.520] If you're living on caffeine, constantly tired, and feel like your stress is stuck on high, there's a good chance you're low in magnesium.
[00:00:25.520 --> 00:00:31.200] Magnesium is critical for turning food into energy, calming your nervous system, helping you sleep, and regulating your mood.
[00:00:31.200 --> 00:00:34.320] And here's the kicker: up to 80% of us aren't getting enough.
[00:00:34.320 --> 00:00:42.240] That kind of chronic deficiency may contribute to feelings of burnout, occasional tension, irritability, and the kind of fatigue that coffee can't fix.
[00:00:42.240 --> 00:00:44.240] That's why I recommend Magnesium Breakthrough.
[00:00:44.240 --> 00:00:45.360] Why Bioptimizers?
[00:00:45.360 --> 00:00:52.400] It's the only formula that combines all seven essential forms of magnesium, including glycinate for calm and citrate for digestion.
[00:00:52.400 --> 00:00:55.520] Most magnesium supplements only offer one or two forms.
[00:00:55.520 --> 00:01:01.600] Magnesium Breakthrough covers them all for deeper sleep, better focus, less tension, and real support during stressful times.
[00:01:01.600 --> 00:01:04.720] If your body's feeling off, this might be the missing link.
[00:01:04.720 --> 00:01:10.480] Try it risk-free at bioptimizers.com/slash hymen and get 15% off your order.
[00:01:10.480 --> 00:01:15.120] Before we jump into today's episode, I want to share a few ways you can go deeper on your health journey.
[00:01:15.120 --> 00:01:18.960] While I wish I could work with everyone one-on-one, there just isn't enough time in the day.
[00:01:18.960 --> 00:01:21.760] So I built several tools to help you take control of your health.
[00:01:21.760 --> 00:01:30.080] If you're looking for guidance, education, and community, check out my private membership, The Hyman Hive, for live QAs, exclusive content, and direct connection.
[00:01:30.080 --> 00:01:34.400] For real-time lab testing and personalized insights into your biology, visit Function Health.
[00:01:34.400 --> 00:01:39.760] You can also explore my curated doctor-trusted supplements and health products at drhyman.com.
[00:01:39.760 --> 00:01:46.080] And if you prefer to listen without any breaks, don't forget you can enjoy every episode of this podcast ad-free with Hyman Plus.
[00:01:46.080 --> 00:01:50.800] Just open Apple Podcasts and tap try-free to start your seven-day free trial.
[00:01:51.040 --> 00:01:59.760] I think the interesting thing about the AI scene is it really didn't get real until, let's say, seven, eight years ago.
[00:02:01.480 --> 00:02:09.080] And it really, for our space of medicine, it was confined to medical images, scans.
[00:02:09.720 --> 00:02:12.440] And that was the deep learning phase of AI.
[00:02:12.440 --> 00:02:14.440] And it really has been formidable.
[00:02:14.440 --> 00:02:46.840] That is, just about every type of scan you can imagine, but path slides, electrocardiograms, the retina, as you mentioned, skin lesions, they could be interpreted as well or better by machines that were trained with so-called supervised learning, meaning that, of course, you had to have thousands, tens of thousands, hundreds of thousands of images that were annotated by expert physicians, and then you could train a model to do better than humans.
[00:02:46.840 --> 00:02:49.080] So that was really great.
[00:02:49.080 --> 00:02:56.520] And, you know, back in 2019, when I wrote Deep Medicine, it was about that phase of deep learning.
[00:02:57.000 --> 00:02:58.440] That's like ancient history now, right?
[00:02:58.440 --> 00:02:59.000] 2019.
[00:02:59.080 --> 00:03:00.040] Yeah, I know.
[00:03:00.760 --> 00:03:03.400] It's amazing how quickly that has gone.
[00:03:03.640 --> 00:03:05.080] Yeah, really, Mark.
[00:03:05.080 --> 00:03:17.320] But what's interesting is, you know, I wrote in the book that what we need is a new model because we didn't have one that could take all the layers of what makes us unique.
[00:03:17.880 --> 00:03:31.800] You know, you've alluded to that, not just the electronic health record, but our genome, you know, our gut microbiome, our sensors, our environment, our immunome, the works, right?
[00:03:31.800 --> 00:03:39.400] And the fact that that those data changes over time, and the fact that we could get the corpus of medical knowledge into that as well.
[00:03:39.400 --> 00:04:02.400] So that's where we are now with this transformer model, also known as large language model phase, which is, of course, got major jump in a year ago with Chat GPT, and now, of course, the GPT-4, Gemini, and future models, GPT-5, sometime next year, in fact.
[00:04:02.400 --> 00:04:22.160] And that, of course, is getting us to that state where we could take all that data for a given patient or individual and be able to not only define what is so critical about predicting a condition, better treatment, better prevention.
[00:04:22.720 --> 00:04:27.040] So we're on the cusp, but we haven't done it yet, to be honest.
[00:04:27.040 --> 00:04:35.840] So, you know, no one has actually done multiple layers.
[00:04:35.840 --> 00:04:42.400] They've done electronic health records and a genome, electronic health records, and a scan.
[00:04:42.400 --> 00:04:50.640] But to take multiple layers, including sensors, that's an analytical AI challenge that has yet to be solved.
[00:04:50.640 --> 00:04:54.080] It will be imminently, and that's exciting.
[00:04:54.080 --> 00:05:01.920] Yeah, I mean, you know, when you talk about it, you wrote an article that I thought was just so prescient, and it was such a good description in a short amount of time.
[00:05:01.920 --> 00:05:08.400] And I encourage people to read it called As Artificial Intelligence Goes Multimodal, Medical Applications Multiply.
[00:05:08.400 --> 00:05:39.880] And you talked about how we're going to be getting high-dimensional data that underlie the uniqueness of all of us and how it can be captured from all these different sources that you mentioned, including all the biomarkers we have through biosensors, wearables, implantables, our genome, our microbiome, our metabolome, our immunome, the transcriptome, proteome, epigenome, it goes on and on, and then our electronic health records, our lab tests, our family history, unstructured text from our medical records, and also things that are air pollution sensors we could be wearing.
[00:05:39.880 --> 00:05:44.920] I just got one of those that someone sent me to try to wear it on my air pollution, environmental stressors.
[00:05:45.400 --> 00:05:54.600] All these things are going to be then informed by the whole Medline, a National Library of Medicine database of peer-reviewed data.
[00:05:54.600 --> 00:05:58.040] And it's going to create so much information.
[00:05:58.040 --> 00:06:05.480] And it seems to me there's an intersection of a number of trends right now, which are going to transform medicine in a way that we can barely imagine.
[00:06:05.480 --> 00:06:18.360] And it's going to happen very soon, which is the omics revolution, the systems biology, and medicine revolution, the biosensors and wearable revolution, and then the AI, machine learning, and big data analytic capacity that we have.
[00:06:18.360 --> 00:06:30.840] And so, those five basic trends are all converging in a way that I think is within even four or five years, we're going to see medicine be profoundly different because the acceleration of this is happening so fast.
[00:06:30.840 --> 00:06:56.280] And I'm excited about it because I feel like I've been trying to, with my little brain, put my head around all these immense complexity of human biology, which we've managed to navigate through this reductionist model of medicine and science into siloed specialties where you're super sub-sub-sub-sub-specialist on X, Y, or Z topic, but you don't understand how it all connects and interacts.
[00:06:56.280 --> 00:06:58.920] And so, the first time with AI, it seems like we're gonna be able to do that.
[00:06:58.920 --> 00:07:01.640] So, how do you see this unfolding?
[00:07:01.640 --> 00:07:03.400] And how is this kind of happening?
[00:07:03.400 --> 00:07:04.360] And where are we going?
[00:07:04.360 --> 00:07:07.560] Because I feel like I'm sitting on the edge of my seat.
[00:07:07.560 --> 00:07:17.760] And right now, I feel like we're about to kind of get out of our little dark ages and enter into an era where we're going to be able to make a real transformation in people's health.
[00:07:14.840 --> 00:07:18.880] Well, I think you're right.
[00:07:19.440 --> 00:07:23.600] It's extraordinary, this convergence that you're getting at.
[00:07:24.240 --> 00:07:26.480] And it's going to happen in phases.
[00:07:26.480 --> 00:07:31.920] So the first one is more the practical, which is, you know, I've been calling keyboard liberation.
[00:07:31.920 --> 00:07:32.240] Yeah.
[00:07:32.240 --> 00:07:32.960] Thank God.
[00:07:32.960 --> 00:07:34.000] I heard that you say that.
[00:07:34.000 --> 00:07:35.600] I'm like, hallelujah.
[00:07:35.600 --> 00:07:39.760] Because every doctor is stuck on their keyboard looking at the computer instead of looking at the patient.
[00:07:39.760 --> 00:07:42.000] And so being free of that is so huge.
[00:07:42.160 --> 00:07:45.680] So it's hated mutually by doctors and nurses and patients.
[00:07:45.680 --> 00:07:53.600] I mean, it's everything that people love to hate because it's destroyed that bond, that human-human bond.
[00:07:53.600 --> 00:08:09.280] And that's going to be basically history of data clerk function because we're already seeing now in many health systems around the country that you can do all this through the conversation.
[00:08:09.280 --> 00:08:15.360] The only adjustment you have to make, Mark, is to articulate the physical exam findings with the patient.
[00:08:15.360 --> 00:08:22.240] But other than that, the notes are far superior than the ones that are pecked along.
[00:08:22.240 --> 00:08:30.160] And what's great is once you have that note digitized and it's got all the juice in it, two big things happen.
[00:08:30.160 --> 00:08:42.800] One is that, of course, you could put it in any format conducive for the patient, you know, in terms of educational level or language or, you know, whatever cultural bent.
[00:08:42.800 --> 00:08:51.840] You could also, that patient has the audio file, so if they don't understand something in that note, they can link it right to the auto file, listen to it again.
[00:08:51.840 --> 00:08:56.800] And you know how many patients that you see where they're confused or they don't remember things.
[00:08:56.800 --> 00:09:12.600] But the other big thing is on the clinician side, because instead of having to peck through all this stuff, the orders for new tests and labs and return appointments, prescriptions, billing, pre-authorization, it's all done.
[00:09:12.600 --> 00:09:13.720] It's all done.
[00:09:13.720 --> 00:09:21.000] And the nudges to the patient subsequent about the things that were discussed, like blood pressure, did you check?
[00:09:21.000 --> 00:09:22.200] What were the results?
[00:09:22.200 --> 00:09:27.000] You know, the AI picks that up, gets it back to the physician.
[00:09:27.000 --> 00:09:29.720] You know, all these things are now automated.
[00:09:29.720 --> 00:09:32.920] So that will in itself be welcome.
[00:09:32.920 --> 00:09:40.040] You know, instead of the things that all clinicians want to hate, this is, I think, something that will be widely embraced.
[00:09:40.040 --> 00:09:48.600] And there's no, you know, as you know very well, Mark, there's a lot of concerns about confabulation, hallucination, but that doesn't apply here.
[00:09:48.600 --> 00:09:54.360] I mean, this is, this is not that the AI is not going to be making things up about this kind of thing.
[00:09:54.680 --> 00:09:56.200] Do you have that in your office yet?
[00:09:56.200 --> 00:09:57.400] Do you have that in your office?
[00:09:57.800 --> 00:10:05.080] I've used it at Scripps Health, where I have cardiology practice.
[00:10:05.320 --> 00:10:11.160] They haven't used what I consider the best of these, but they have done a pilot.
[00:10:11.880 --> 00:10:24.360] The largest one is the Microsoft Nuance, but the company that I've advised is Abridge Health, which is derived from University of Pittsburgh and Carnegie Mellon.
[00:10:24.360 --> 00:10:25.640] But there's been several.
[00:10:25.640 --> 00:10:28.680] I mean, there's about 20 of these out there in various testing.
[00:10:29.240 --> 00:10:31.480] I want to get one right away for my practice.
[00:10:32.120 --> 00:10:43.080] Yeah, I mean, I don't, I think this is an inevitability because this is finally the payback for all these bad years of having to become data clerks.
[00:10:43.080 --> 00:10:44.400] But it's just the beginning.
[00:10:44.200 --> 00:10:48.000] You know, it's just one thing that's going to be remarkably different.
[00:10:48.480 --> 00:10:52.560] And that helps us to care better, but it doesn't change what we're doing.
[00:10:52.560 --> 00:11:02.480] In other words, you know, we're going to be able to read x-rays better and MRI imaging better and pathology reports better and EKGs better and retinal imaging that tells us so much about a patient's health.
[00:11:02.480 --> 00:11:10.160] And these are incredible advances that are going to create much more refinement and understanding of how to be precise in our diagnosis of patients.
[00:11:10.160 --> 00:11:12.560] And that's going to up-level medicine for sure.
[00:11:12.560 --> 00:11:16.160] But let me, can I just say one thing?
[00:11:16.160 --> 00:11:16.560] Yeah.
[00:11:16.560 --> 00:11:21.440] Because the retinal image is something that is extraordinary.
[00:11:21.760 --> 00:11:23.440] So before we just pass over that.
[00:11:23.680 --> 00:11:24.240] Yeah, yeah, I know.
[00:11:24.640 --> 00:11:34.720] You know, I just want to point out that, you know, the original task was to see if the AI could interpret the image as well as a clinician.
[00:11:34.720 --> 00:11:42.000] But what wasn't envisioned is that the AI could see things that humans will never see.
[00:11:42.000 --> 00:12:16.640] So with the retina, as you touched on, the ability to predict Alzheimer's disease, Parkinson's five to seven years before there's any symptoms, the issue of, of course, the hepatobiliary tract, kidney disease, cardiac risk, risk of, you know, across all systems, diabetes control, blood pressure control, someday we will be taking pictures of our own retina and get as a checkup with an AI.
[00:12:16.640 --> 00:12:18.160] So it's pretty amazing.
[00:12:18.160 --> 00:12:21.920] And of course, that extends to cardiograms and chest x-rays.
[00:12:21.920 --> 00:12:31.400] Each of them, there's all this stuff that the AI can see, if you will, that humans will never see it.
[00:12:31.560 --> 00:12:33.560] So it's even better, better than humans.
[00:12:33.640 --> 00:12:34.120] Right.
[00:12:29.760 --> 00:12:34.600] Yeah, yeah.
[00:12:34.760 --> 00:12:49.160] I mean, this is why, you know, when I interviewed Jeff Hinton recently for the podcast I do, Ground Truths, he said, you know, he's worried about AI because it's getting advanced so quickly, but not for medicine.
[00:12:49.160 --> 00:12:51.000] He thinks this is the sweet spot.
[00:12:51.000 --> 00:12:54.920] This is really where the good is extraordinary.
[00:12:54.920 --> 00:12:55.400] I agree.
[00:12:55.400 --> 00:13:00.040] I mean, you know, I remember in medical school, you had the ophthalmoscope and you had to look in someone's eye.
[00:13:00.040 --> 00:13:05.000] And you know, okay, you learn about AV nicking and high blood pressure and diabetic retinopathy and macular generation.
[00:13:05.000 --> 00:13:06.360] You could see all that stuff.
[00:13:06.360 --> 00:13:08.600] But there wasn't a whole lot else you could kind of figure out.
[00:13:08.600 --> 00:13:13.000] You know, and if you're an ophthalmologist, you might have a few more refinements in your ability to see things.
[00:13:13.000 --> 00:13:15.800] But what you're saying is you can see things like Alzheimer's.
[00:13:15.800 --> 00:13:17.400] So how does it pick that up?
[00:13:17.400 --> 00:13:21.720] What is it actually seeing and looking at, for example, for Alzheimer's?
[00:13:22.040 --> 00:13:35.400] Well, you know, this goes back to when the realization was made, and that was when you showed the retina picture to ophthalmologists and you say, is this retina from a man or a woman?
[00:13:35.400 --> 00:13:38.280] They got it right 50% of the time.
[00:13:38.280 --> 00:13:41.880] And the AI got it right 97% of the time.
[00:13:41.880 --> 00:13:44.920] And the answer is, we don't really know.
[00:13:44.920 --> 00:13:45.560] Okay.
[00:13:45.560 --> 00:13:56.360] That is, there's explainability work to, you know, define these so-called saliency maps to try to deconvolute the model.
[00:13:56.360 --> 00:14:04.440] But as far as what is it picking up to see the risk of Alzheimer's or Parkinson's or a pattern billiaries, it isn't clear.
[00:14:04.440 --> 00:14:19.280] I mean, there's some aspects that have been determined, but basically, because these models are so extraordinary in terms of what they've learned, and this is all from deep learning.
[00:14:14.840 --> 00:14:22.160] This isn't even from this transformer model era.
[00:14:22.320 --> 00:14:24.320] So can you just stop here for a sec?
[00:14:24.640 --> 00:14:26.880] You're talking about deep learning, transformer model.
[00:14:26.880 --> 00:14:30.320] Can you just explain the sort of shift in what you're thinking?
[00:14:30.320 --> 00:14:32.640] And because I don't think most people understand what that is.
[00:14:32.960 --> 00:14:33.440] Right.
[00:14:33.440 --> 00:14:38.720] So what was the phase of AI that lit up the world?
[00:14:38.960 --> 00:14:53.120] Jeff Hinton and his colleagues like Jan Lacun and many others, they basically found that there was this ability to input data that was supervised.
[00:14:53.120 --> 00:15:01.440] That is, for our purposes, it was labeled by experts, so-called ground truths.
[00:15:01.440 --> 00:15:13.760] And so they put it what they knew was the actual image interpretation and train with tens of hundreds of thousands of these images so that the machine could see stuff.
[00:15:14.000 --> 00:15:17.520] So this is a knowledge-based or expert-informed AI, right?
[00:15:17.520 --> 00:15:18.240] Yeah, yeah.
[00:15:18.240 --> 00:15:21.760] So that really was, you know, deep neural networks.
[00:15:21.760 --> 00:15:22.800] That was the story.
[00:15:22.800 --> 00:15:26.480] It required a single task, unimodal.
[00:15:27.040 --> 00:15:36.480] And then what happened, a Google team in 2017 discovered what they call transformer models.
[00:15:36.720 --> 00:15:40.480] The title of the preprint, Attention is All You Need.
[00:15:40.480 --> 00:16:05.720] And basically it changed the attention from a single bit of information, like a word in a sentence, to basically the context of the entire sentence, or of course, much broader than that, what turned out to be unsupervised, putting in the entire internet, Wikipedia, 100,000 books, 200,000 books.
[00:16:05.720 --> 00:16:10.920] So that's what the transformer model, large language model, generative AI era that we're in now.
[00:16:10.920 --> 00:16:18.760] It didn't start when ChatGPT was released last year, but it actually was in incubation.
[00:16:19.160 --> 00:16:24.520] It was being pursued about six years now, but it's now blossomed.
[00:16:24.520 --> 00:16:33.320] And that we basically have two big types of AI now: the old, if you will, the old and the new.
[00:16:33.320 --> 00:16:43.400] Yeah, I mean, it just seems it's going to accelerate the pace of medical discovery because, you know, if a simple retinol scan can pick up things that we didn't even know we were missing, you know, we didn't even know we didn't know.
[00:16:43.400 --> 00:16:46.200] They were unknown unknowns, as Donald Russell said.
[00:16:46.520 --> 00:16:46.760] Exactly.
[00:16:47.080 --> 00:16:49.880] And that's just the back of the eye.
[00:16:49.880 --> 00:16:59.000] Imagine when we put in all these things that we just mentioned: the whole omics field, the biosensors, your pictures of what you're eating, your movement pattern.
[00:16:59.000 --> 00:17:08.680] I mean, it's just an enormous amount of data that's going to pick up patterns in that data that we've never seen before and that are going to inform what's happening on a biological level.
[00:17:08.680 --> 00:17:18.840] That I think is going to redefine medicine, just as we sort of redefine physics from a Newtonian or a world is flat view to a quantum view to even beyond that.
[00:17:18.840 --> 00:17:27.960] It's like we're kind of in that era of biology where we basically have a profound revolution that's going to upend medicine.
[00:17:27.960 --> 00:17:38.600] And I'd love to hear your perspective on as we sort of enter that era and we start learning these things and understand the body as a network, understand the body as a system instead of these siloed specialties.
[00:17:38.600 --> 00:17:44.520] How do you see that shifting medicine, medical education, medical practice, reimbursement?
[00:17:44.640 --> 00:17:47.520] I mean, this is a massive shift.
[00:17:47.840 --> 00:17:49.520] Well, it is seismic.
[00:17:50.000 --> 00:17:56.560] It's going to be a challenge because medicine, as you know, doesn't change easily.
[00:17:56.560 --> 00:18:07.280] And then you got, you know, throw in all these other practical matters like, you know, reimbursement and education, regulatory, trust, implementation.
[00:18:07.280 --> 00:18:10.960] I mean, there's a long list here of challenges.
[00:18:10.960 --> 00:18:18.320] So, you know, this isn't going to be easy, but it's going to be, you know, the biggest shakeup in the history of medicine.
[00:18:18.320 --> 00:18:36.720] The question is how we adapt, how we, you know, our problem at the moment, outside of a practical thing like we discussed with the keyboard thing, is to get things implemented, we've got to have compelling evidence.
[00:18:36.720 --> 00:18:51.280] And there's a dearth of that because, you know, just like you can't get thousands of doctors to annotate images, and that's why this new form, transformer model, doesn't require supervised learning.
[00:18:51.280 --> 00:18:52.800] It's self-supervised.
[00:18:52.800 --> 00:18:57.200] So it basically is the bypass to what was holding back medicine.
[00:18:57.200 --> 00:19:08.000] But just like that problem, you know, we have the problem of lack of dedication to do prospective trials, whether they're randomized or not.
[00:19:08.000 --> 00:19:20.480] But getting the compelling evidence, which basically says to everyone in the medical community, this is it, you know, that this is going to lead to better patient outcomes, better, you know, better everything.
[00:19:20.800 --> 00:19:27.040] And there's always going to be some risk, of course, when there's never going to be, you know, total positive side of the story.
[00:19:27.040 --> 00:19:45.880] But we, except for the gastroenterologists who have done 33 randomized trials of colonoscopy with machine vision and a few other randomized trials and radiology that have been quite impressive, particularly mammography, there hasn't been much compelling evidence so far.
[00:19:45.880 --> 00:19:46.840] Yeah, it's true.
[00:19:46.840 --> 00:19:47.240] It's true.
[00:19:47.240 --> 00:20:00.680] But, you know, on the other hand, you look at the amount of deaths that are caused by medical practice, probably a third or fourth leading cause of death or complications or reactions to drugs or medical errors.
[00:20:00.680 --> 00:20:01.480] It's huge.
[00:20:01.480 --> 00:20:06.600] And I was listening to Elon Musk talk about cars and AI and self-driven cars.
[00:20:06.600 --> 00:20:10.760] And he says, you know, about 40,000 people in America die from car accidents every year.
[00:20:10.760 --> 00:20:12.760] You know, what if that was reduced to 10,000?
[00:20:12.760 --> 00:20:18.040] But, you know, that's a dramatic drop, but still, you're going to have some people dying from a self-driving car.
[00:20:18.040 --> 00:20:19.720] And are we willing to accept that?
[00:20:19.720 --> 00:20:36.920] You know, so I think that's really a point where we have to kind of understand the value proposition and understand that there is some risk, but the upside in terms of reducing our healthcare costs, the burden on our healthcare system is going to be profound.
[00:20:39.800 --> 00:20:42.440] You don't need more caffeine or another stress supplement.
[00:20:42.440 --> 00:20:44.040] You might just need more magnesium.
[00:20:44.040 --> 00:20:48.040] Magnesium supports sleep, mood, energy, and focus, but most of us are missing it.
[00:20:48.040 --> 00:20:51.400] That's why I recommend Magnesium Breakthrough by Bioptimizers.
[00:20:51.400 --> 00:20:56.040] It combines seven forms of magnesium for results you can feel in your mood, sleep, focus, and more.
[00:20:56.040 --> 00:21:01.800] Try it now for 15% off at bioptimizers.com/slash hyman and feel the difference.
[00:21:07.240 --> 00:21:12.680] One of the things that we've done recently is to get into digital twins.
[00:21:12.680 --> 00:21:18.400] And so a digital twin is a representation of your body's physiology.
[00:21:18.400 --> 00:21:21.120] And we've done this first for brain health.
[00:21:21.120 --> 00:21:30.080] And so what we can actually do in this case is, and we're going to release a test on this, you know, a product based on this next year.
[00:21:30.080 --> 00:21:39.440] But basically, what you can do is you can monitor for a number of these blood measures, your genetics, cognitive assessments, and so forth.
[00:21:39.440 --> 00:21:43.840] And you can then run a simulation based on your particular biology.
[00:21:43.840 --> 00:21:51.040] And it's based around understanding from a physiologic and molecular level what's driving brain health.
[00:21:51.040 --> 00:22:00.000] And you can actually forecast the likely amount of time that you have with a healthy brain given your current state.
[00:22:00.000 --> 00:22:11.280] More importantly, you can go to personalized recommendations for different kinds of things that people can do, some of which are exercise to keep your oxygenation in your brain high.
[00:22:11.280 --> 00:22:14.480] You can get into things like phosphatidylcholine.
[00:22:14.480 --> 00:22:17.600] Turns out that that becomes rate limiting under low oxygen conditions.
[00:22:17.760 --> 00:22:21.200] Latest people are developing dementia, hugely important.
[00:22:21.520 --> 00:22:23.200] Vitamin D, very simple one.
[00:22:23.200 --> 00:22:24.800] We could talk a lot more about that one.
[00:22:24.880 --> 00:22:26.320] Turns out to be very important.
[00:22:26.320 --> 00:22:28.000] There's many, many of these.
[00:22:28.000 --> 00:22:42.640] But the point is that what you can actually do with the digital twins is you can get a representation of a person's individual risk profile and then tailor the precise recommendations.
[00:22:42.640 --> 00:22:44.800] These recommendations are very different person to person.
[00:22:44.800 --> 00:22:51.280] Once you get to four recommendations, only 1% of people actually benefit from what's the best thing, the best for in the population.
[00:22:51.280 --> 00:22:55.200] We just did those simulations, you know, so it's very interesting when you do that.
[00:22:55.200 --> 00:23:04.280] And so, so you get this intense personalization, and you can get into the physiology and you can start to make sense of this because you have to take the complexity of all these measures.
[00:23:04.360 --> 00:23:11.880] You can't place that on the person, you have to put that into the algorithms and deliver back simple, actionable information.
[00:23:11.880 --> 00:23:22.760] And then, the other side of the coin, which I'll just mention here briefly, is the ChatGPT and all these things that we've, you know, that have shocked the world over the last year.
[00:23:22.760 --> 00:23:39.880] The ability now to deliver personalized insights that give you a lot of context and that you can have a back and forth with, and you can get access to a dialogue, even with what your digital twin is saying, or what you're learning about your body.
[00:23:41.800 --> 00:23:49.480] The capability for us to develop personalization on that front is just radically better than any of us thought it was going to be a couple years ago.
[00:23:49.480 --> 00:23:59.800] And so, those things together are really pushing us into this new world of where we're going to be able to harness so much more of this complexity than we could have even thought about before.
[00:23:59.800 --> 00:24:01.400] I mean, is Chat GPT there?
[00:24:01.400 --> 00:24:08.440] Like now, for example, if I put in all my symptoms, I enter in all my lab data, and I hit, you know, tell me what's wrong and what to do about it.
[00:24:08.440 --> 00:24:11.720] Would it give me anything useful at this point, or is it still far off?
[00:24:11.720 --> 00:24:14.360] So, I've played with this a lot, so maybe I'll jump in on that.
[00:24:14.360 --> 00:24:17.880] But it's pretty much what I do in my free time.
[00:24:17.880 --> 00:24:20.520] I don't do anything else.
[00:24:20.840 --> 00:24:21.800] You're a hypochondriac.
[00:24:22.360 --> 00:24:24.120] You put it on your symptoms.
[00:24:24.120 --> 00:24:25.080] My stomach nerds.
[00:24:25.080 --> 00:24:25.880] I got a head pig.
[00:24:26.040 --> 00:24:26.360] Yeah.
[00:24:26.360 --> 00:24:28.360] So it's partially there.
[00:24:28.360 --> 00:24:34.360] If you use like earlier versions, like the GPT 3.5, for example, you'll get lots of hallucinations.
[00:24:34.360 --> 00:24:36.440] It's sometimes useful, sometimes not.
[00:24:36.440 --> 00:24:41.080] GPT-4 is pretty good, except anyway, there's this weird trend.
[00:24:41.080 --> 00:24:42.760] It's not as good as it used to be.
[00:24:42.760 --> 00:24:44.520] And there's a lot of chatter around that on it.
[00:24:44.520 --> 00:24:46.400] It doesn't let you go as deep as it used to.
[00:24:46.400 --> 00:24:50.800] I don't know if it's legal or they're not really put guardrails on it.
[00:24:44.840 --> 00:24:50.880] Yeah.
[00:24:51.120 --> 00:24:54.240] They put guardrails and various kinds on it and so forth.
[00:24:54.240 --> 00:25:02.160] But as long as your question is reasonably well dealt with in available text that it's generating from, it can be quite good.
[00:25:02.160 --> 00:25:09.280] And I've had, and I've used it, you know, not just on medical issues, but you know, explain statistical analysis of this kind of data or something like that.
[00:25:09.280 --> 00:25:14.000] And it actually gives back really reasonable kinds of information.
[00:25:14.000 --> 00:25:15.920] Now, it's not fully to where it wants to.
[00:25:16.000 --> 00:25:17.600] Oh, and I did see a survey.
[00:25:17.600 --> 00:25:19.120] Maybe you saw this as well.
[00:25:19.120 --> 00:25:26.880] They pulled doctors, and apparently, 60% of doctors are using GPT today, right now, in the background on things that they do.
[00:25:26.880 --> 00:25:29.280] So I saw if you saw that survey.
[00:25:29.840 --> 00:25:34.320] But I was actually not totally ready for prime time, but just to say that.
[00:25:34.320 --> 00:25:35.120] Yeah, go ahead.
[00:25:35.120 --> 00:25:46.320] Well, no, I was at this big medical conference in Lake Nona, and they had this guy from Microsoft with, I think, Prometheus, which was kind of a new version of like Chat GPT that was like, you know, for doctors.
[00:25:46.320 --> 00:25:51.760] And they had a case report that they were sharing and they were entering in this case study.
[00:25:51.760 --> 00:25:54.720] And it got it totally wrong.
[00:25:54.720 --> 00:25:56.480] And I guessed it immediately.
[00:25:56.480 --> 00:25:57.600] Like, I wouldn't guess it.
[00:25:57.600 --> 00:25:59.920] I just knew what it was because I listened to the story.
[00:25:59.920 --> 00:26:19.360] But, you know, it was basically a patient who had frequent urination, fever, chills, you know, had had, I think maybe had had a history of rheumatoid strep long ago or something like that, or had a murmur, or maybe had a murmur as a sort of part of the exam.
[00:26:19.320 --> 00:26:20.480] And it was just a murmur.
[00:26:20.480 --> 00:26:22.480] And I'm like, oh, this guy has endocarditis.
[00:26:22.480 --> 00:26:24.240] This guy has bacterial endocarditis.
[00:26:24.240 --> 00:26:28.480] And the chat, the Prometheus thing said, oh, he's got a kidney infection.
[00:26:28.640 --> 00:26:30.520] And I'm like, no, he's not having a kidney infection.
[00:26:30.520 --> 00:26:31.320] And it was wrong.
[00:26:31.320 --> 00:26:34.040] And it was like in front of like 500 people.
[00:26:29.920 --> 00:26:36.120] So, you know, I kind of wonder.
[00:26:36.280 --> 00:26:40.760] But I do think that things are changing.
[00:26:40.760 --> 00:26:57.560] So as you've gotten into sort of looking at these sort of enormous amounts of data through the phenome typing of people, when that goes into these machine learning AI models, like, you know, where is the next step in this in medicine?
[00:26:58.120 --> 00:27:00.200] Are we all kind of moving towards this?
[00:27:00.200 --> 00:27:02.520] Are doctors going to become in some ways obsolete?
[00:27:02.520 --> 00:27:08.600] Or are they just going to be helping to kind of implement some of the decision support that these tools give?
[00:27:08.600 --> 00:27:23.080] Because personally, I would love to be able to put all the data for my patients in, and instead of spending hours and hours modeling over it and thinking about it, trying to remember every study I ever read and what to do in my medical school training, like this is going to give me kind of a roadmap to start with and then implement it.
[00:27:23.480 --> 00:27:25.320] How far are we away from that?
[00:27:25.320 --> 00:27:27.960] Well, I'll make a couple of comments.
[00:27:28.200 --> 00:27:40.840] I think a really important thing about these large language models, which is what GPT and the other things we've talked about are, is that they have to be educated properly.
[00:27:40.840 --> 00:27:57.000] So if you take a large language model and you expose it to the internet, you expose it to the conspiracy theories and the lying and all of those other things, you have an enormous susceptibility in that device.
[00:27:57.000 --> 00:28:08.200] And my argument is for health, we ought to have a GPT that has only been educated with biomedical data.
[00:28:08.200 --> 00:28:13.160] And we're actually collaborating with a group that has one of those.
[00:28:13.480 --> 00:28:24.480] And what our hope is, is we'll, and part of the education has been to put PubMed into the device, which gives you an enormous amount of data.
[00:28:24.480 --> 00:28:29.040] Now, some is right and some is wrong, and you'll still have to make judgments.
[00:28:29.040 --> 00:28:37.840] But what we plan to do is we have access, for example, to Google's knowledge graph.
[00:28:37.840 --> 00:28:46.400] And this is a graph that connected roughly 50 different features from the literature.
[00:28:46.400 --> 00:28:58.400] So it's assembled from the PubMed literature, all of the relationships between genes and proteins and diseases and drugs, and on and on and on.
[00:28:58.400 --> 00:29:03.440] PubMed, for those listening, is just the entire body of peer-reviewed, published medical.
[00:29:03.680 --> 00:29:05.200] Biological information.
[00:29:05.200 --> 00:29:08.800] Yeah, it's a lot, it's millions and millions of studies.
[00:29:08.800 --> 00:29:23.040] Well, this knowledge graph has 50 million nodes and 850 million edges, which means an enormous number of relationships.
[00:29:23.040 --> 00:29:31.600] So we're going to put this knowledge graph in this medically educated GTP.
[00:29:31.600 --> 00:29:37.440] And we're going to put in, we're building now a knowledge graph for the kidney.
[00:29:37.440 --> 00:29:41.520] We'd certainly like to put in the knowledge graph for brain health.
[00:29:41.520 --> 00:29:50.320] All of the knowledge graphs and digital twins that we have should go into educating this thing.
[00:29:50.320 --> 00:30:28.600] And then my hope is the following: we'll be able to take the data, genome, and phenome from each individual, enormously more complicated than what we did in Airvail, maybe 10 times as much data as we had initially, and put it in there and ask it to generate from tens of thousands of actionable possibilities, the ordered priority of actionable possibilities that you as an individual can use to optimize your health or avoid disease or whatever.
[00:30:28.920 --> 00:30:38.680] And what the AI will actually do is send this information to a doctor, and there'll be two things the information will have to do.
[00:30:38.680 --> 00:30:46.040] One, clearly explain the actionable possibility and what the doctor and the patient will be expected to do.
[00:30:46.040 --> 00:30:57.800] But two, it's to give the physician the medical evidence for this actionable possibility to assure him or her it's bona fide.
[00:30:58.120 --> 00:31:12.200] And the dramatic result of this is you will be able to take a family practitioner and make him a domain expert in virtually every field of medicine.
[00:31:12.200 --> 00:31:20.600] It gives you this global reach that you were talking about and the capacity to handle virtually anything.
[00:31:20.600 --> 00:31:26.680] And that democratizes medicine in an incredible way.
[00:31:26.680 --> 00:31:39.480] And I'll argue, we'll never ever get rid of the physician because they're, in the end, still an integrative factor that we're a long ways from being able to replicate and so forth.
[00:31:39.800 --> 00:31:47.760] But he will have the tools to become a world expert in every field of medicine.
[00:31:44.600 --> 00:31:51.680] Really, quite a remarkable promise for the future.
[00:31:52.000 --> 00:32:04.720] And what it promises for patients, that is the optimization of this wellness and prevention, Nathan and I have talked about, I think is really dramatic.
[00:32:04.720 --> 00:32:07.120] So, how far away from this are we?
[00:32:07.440 --> 00:32:16.640] So, I think we'll begin to see the effects of this within the next year or so as these things get.
[00:32:16.640 --> 00:32:23.840] I mean, we won't have him in the full glory for, you know, who knows?
[00:32:23.840 --> 00:32:32.080] Maybe 10 years is way too long to say, because look, what, I mean, that 60% of the doctors would use a tool like this.
[00:32:32.080 --> 00:32:39.600] I would have said there's no way in the world that that conservative group of people would ever go into AI like this.
[00:32:39.600 --> 00:32:40.000] And yet.
[00:32:40.160 --> 00:32:43.120] So they're putting their patients' history in there and saying, hey, what's wrong?
[00:32:43.120 --> 00:32:44.160] Is that what they're doing?
[00:32:44.480 --> 00:32:44.960] Yeah.
[00:32:46.080 --> 00:32:49.280] Well, I don't want, we should probably not over.
[00:32:49.440 --> 00:32:51.440] It means they use it to some degree.
[00:32:51.440 --> 00:33:03.680] Because the thing about replacing doctors, the line that I really like, I think it's Eric Topel's, which is, you know, AI won't replace doctors, but doctors who use AI will replace doctors who don't.
[00:33:03.680 --> 00:33:09.200] And I think that is a really good way to put it because it is a tool.
[00:33:09.200 --> 00:33:12.480] And I think it's like today, it's already a super useful tool.
[00:33:12.480 --> 00:33:25.440] Like, if you're trying to remember something or if you want to delve into the literature, it's so, you know, you can, and especially with these particular GPTs that are based around PubMed and things like that, they're already an assist, right?
[00:33:25.440 --> 00:33:30.000] So it's just already a function of how strongly that assist can be made.
[00:33:30.760 --> 00:33:40.840] And I think the doctor's still going to be the quarterback, but your ability to block and tackle and just solve lots of issues with the AIs is incredible.
[00:33:40.840 --> 00:33:41.960] And it's not just the LLMs.
[00:33:41.960 --> 00:33:50.040] I mean, one of the really biggest uses that's straightforward right off the bat is getting rid of as many medical errors as possible, right?
[00:33:50.040 --> 00:33:56.200] Because a doctor who's tired, it's easy to, you got a long, complicated name, and there's two of them that look almost exactly the same.
[00:33:56.200 --> 00:33:59.960] It's pretty easy to accidentally check the wrong box.
[00:33:59.960 --> 00:34:08.280] But if the AI actually knows, well, you said your patient has diabetes and that's a drug, did you actually mean this drug for multiple sclerosis?
[00:34:08.280 --> 00:34:08.840] Right?
[00:34:08.840 --> 00:34:10.520] And that's already happening today, right?
[00:34:10.520 --> 00:34:22.040] Hospital systems have saved millions of lives already by just implementing some of those really simple things, the kind of mistake that's easy to make as a human and a computer won't make.
[00:34:22.040 --> 00:34:29.640] Now, vice versa, computers will make the kind of, you know, and AIs will make errors that a human never would because they don't understand causality.
[00:34:29.640 --> 00:34:31.000] They don't understand the context.
[00:34:31.000 --> 00:34:37.960] They don't, you know, there's all kinds of stuff, like the case study that you got right that the AI didn't, like there's things that it doesn't know.
[00:34:37.960 --> 00:34:52.520] So a hybrid or what we call centaur AI in the book, a hybrid approach really makes a lot of sense so you can cover your bases because those two kinds of intelligence, human intelligence and AI, actually operate quite differently.
[00:34:52.520 --> 00:34:54.440] And the kind of errors you make are very different.
[00:34:54.440 --> 00:34:56.680] So combining them is powerful.
[00:34:56.760 --> 00:35:10.280] What you're talking about is definitely going to help transform the expertise of physicians and allow them to practice medicine that's more up-to-date, that reflects the scientific literature, that is based on understanding a wide network of biological factors that they haven't been able to consider before.
[00:35:10.280 --> 00:35:11.880] And that's going to be fantastic.
[00:35:11.880 --> 00:35:17.120] But the truth is that wellness, health, does not happen in a doctor's office, right?
[00:35:17.120 --> 00:35:28.960] And so 80 to 90% of the things that determine your health actually don't require a doctor and are things that you can learn about yourself and fix without a doctor's help.
[00:35:28.960 --> 00:35:37.120] And so, in a way, this is also going to help, I think, disintermediate people from the healthcare system and from doctors because we don't really have a healthcare system.
[00:35:37.120 --> 00:35:38.400] We have a sick care system.
[00:35:38.400 --> 00:35:58.000] And so, what you're talking about is actually a new kind of healthcare system where people are going to be empowered with their own health data, guided by these big, dense data clouds of their own biological information from all their omics to their blood panels, to things we don't even measure now that we're going to measure to their wearables and biometrics.
[00:35:58.000 --> 00:35:59.360] I mean, I have a Garmin watch.
[00:35:59.360 --> 00:36:06.880] I mean, I know everything about myself: my pulse ox, my heart rate ability, how much I slept, how much deep sleep, how much light sleep, about my training printing this is what you know, like how much time I need to recover.
[00:36:06.880 --> 00:36:11.840] I mean, it's pretty impressive, and all that is just sitting out there ready to be kind of harvested and used.
[00:36:11.840 --> 00:36:25.360] And so, individuals, I think, are in this moment where they can become more empowered to be actors in determining their own degree of wellness and health, and then know when to go to the doctor.
[00:36:25.360 --> 00:36:28.240] Like, oh, well, gee, you know, your creatinine's like five.
[00:36:28.240 --> 00:36:30.720] You better get your ass over to the nephrologist tomorrow.
[00:36:30.720 --> 00:36:32.800] So, that's going to for sure be still there.
[00:36:32.800 --> 00:36:37.280] But a lot of the stuff that actually requires a physician isn't really needed.
[00:36:37.280 --> 00:36:44.320] It's really diet, lifestyle, you know, behavioral changes, supplements, and other practices that they have access to.
[00:36:44.320 --> 00:36:53.440] So, how do you see this kind of being a tool that the individuals and patients and consumers can use in a way that is really going to disrupt healthcare?
[00:36:53.440 --> 00:37:01.800] You know, Mark, I think you made a really excellent point, and that is the importance of education for the consumer, if you will.
[00:36:59.840 --> 00:37:05.160] And we're doing a number of things in that regard.
[00:37:05.480 --> 00:37:40.440] For example, this past year, an educational team at the Institute for Systems Biology that I initiated 20 years ago to deal with K through 12 science education problems has put together a four-module one-year course based on two chapters several of us wrote in the systems biology and systems medicine book, one on systems medicine, one on P4 healthcare.
[00:37:40.760 --> 00:37:59.160] And the essence of this module is to give them the picture that is portrayed in our book of what healthcare is going to be in the future and to clearly explain the responsibilities they'll have for their own education.
[00:37:59.160 --> 00:38:13.880] And it makes very strongly the point: the core of your health is going to be diet, exercise, sleep, stress, et cetera.
[00:38:13.880 --> 00:38:20.520] And these are things you can do about it, and these are tools and devices you can use to measure it.
[00:38:20.520 --> 00:38:32.040] And oh, by the way, there is this more sophisticated medicine of assaying your blood and your gut microbiome that can tell us.
[00:38:32.040 --> 00:38:46.720] And by the time students will get done with that year course, I'll guarantee they'll know more about what I think, what we think the future of medicine is than 95% of the physicians out there.
[00:38:46.720 --> 00:39:03.680] I mean, this revolution in transforming healthcare from a disease orientation to an orientation of wellness and prevention, I can't stress how important that's going to be in doing two things.
[00:39:03.680 --> 00:39:12.080] One, improving the quality of health for every single individual that practices, even partially.
[00:39:12.560 --> 00:39:26.480] And two, it's going to lead to enormous cost savings in the healthcare system by avoiding what costs 86% of our health care dollars today, namely chronic diseases.
[00:39:26.480 --> 00:39:35.600] And Mark, I'd love to kind of weigh on that question as well that you asked, because I think it's such an important thing because you're exactly right.
[00:39:35.600 --> 00:39:40.960] Because more and more of what we can call, you know, put under healthcare, especially if we start talking about wellness care, right?
[00:39:40.960 --> 00:39:45.360] We like to say scientific wellness should be the front door of the healthcare system.
[00:39:45.360 --> 00:39:58.160] Most of that effort should really be on this maintenance of health, and then you get referred back into the disease care system when hopefully early enough to really make a difference, but with some advanced warning.
[00:39:58.160 --> 00:40:08.880] But the ability for us to deliver this really efficiently and low cost, I totally agree with you, is pushing this more and more to the home remotely, making it easier.
[00:40:08.880 --> 00:40:18.480] So some of the things that we've done, for example, we've spent the last few years developing an essentially, painless, you know, at-home blood collection device.
[00:40:18.880 --> 00:40:21.680] It used to be called the OneDraw, now called the NanoDrop.
[00:40:22.000 --> 00:40:23.760] But that's like one feature of it.
[00:40:24.160 --> 00:40:27.280] You're not going to go to jail like Elizabeth Holmes with this, are you?
[00:40:27.600 --> 00:40:28.400] Not at all.
[00:40:28.400 --> 00:40:29.800] Yes, exactly.
[00:40:29.120 --> 00:40:32.680] That was my objection to the name change, obviously.
[00:40:29.280 --> 00:40:34.600] Sounds like very familiar.
[00:40:36.040 --> 00:40:38.360] I have watched, yeah, I have gotten into her story.
[00:40:38.360 --> 00:40:39.960] The Mano Tainer.
[00:40:41.800 --> 00:40:43.480] Yeah, I read the book.
[00:40:43.480 --> 00:40:45.800] I watched the documentary like 12 times.
[00:40:45.800 --> 00:40:48.840] I watched the dramatization one they did of it.
[00:40:49.160 --> 00:40:52.680] It's a fascinating story in many ways.
[00:40:52.680 --> 00:40:53.960] But you can move to home, right?
[00:40:53.960 --> 00:40:55.080] Microbiome testing, right?
[00:40:55.080 --> 00:40:56.120] You can do that in your home.
[00:40:56.120 --> 00:40:58.920] You can get access to this with AIs.
[00:40:59.560 --> 00:41:04.760] We developed something called the microbiome wipe to make that as easy as possible for people and so forth.
[00:41:04.760 --> 00:41:27.400] But the whole idea is that we should be able to deliver health information to people in ways that are much more efficient, much more user-friendly, not nearly as expensive, and that people can have a real control over that kind of health and be informed by really deep data.
[00:41:27.720 --> 00:41:30.040] I think that's really the key.
[00:41:30.040 --> 00:41:36.760] Oh, and on the you know, coming back to some of these, you know, like small measurements, you know, you brought up Elizabeth Holmes and so forth.
[00:41:36.760 --> 00:41:44.120] One of the things that's important is that a lot of people have failed in trying to take traditional measures and miniaturizing them.
[00:41:44.440 --> 00:41:49.000] You know, at you know, at least doing a lot of them at the same time.
[00:41:49.000 --> 00:42:02.240] But the kind of things that we're talking about in terms of omics, like a metabolome where you can make thousands of measures, which we're going to do on this device, a protein proteome that you can do, right, again, you know, thousands of measurements.
[00:42:02.240 --> 00:42:05.000] Those are only ever done on small amounts of blood.
[00:42:05.000 --> 00:42:12.680] So, if you know, if Lee and I are running something on that in our lab or any of the top labs in the world, you only ever run those things on time.
[00:42:13.320 --> 00:42:18.080] If you gave them a huge bat of blood, all they would do is take a tiny amount out of it and run it on the mass spec.
[00:42:18.080 --> 00:42:20.640] There's no such thing as running this through it.
[00:42:20.640 --> 00:42:24.160] So, you're talking about technologies that are miniaturized already.
[00:42:24.160 --> 00:42:26.720] That's the only way, that's the way that they work.
[00:42:26.720 --> 00:42:34.880] And so, there isn't actually a technological breakthrough of any kind that's needed to use this small amount of blood to get those many measurements.
[00:42:34.880 --> 00:42:39.760] The breakthrough is you have to understand how to read the information.
[00:42:39.760 --> 00:42:53.200] But in the modern world, I'd much rather have an information challenge than a technology challenge because the information challenge can actually be overcome by getting access to samples, the AIs, the long term.
[00:42:53.440 --> 00:42:55.680] And I'll give one interesting example.
[00:42:55.680 --> 00:42:58.560] So, think about what happened in genomics.
[00:42:58.560 --> 00:43:03.840] So, in the genome, initially, one of the traits that we couldn't predict from the genome was height.
[00:43:03.840 --> 00:43:05.360] Now, we all know height is heritable, right?
[00:43:05.360 --> 00:43:07.040] If you have tall parents, you have tall kids.
[00:43:07.040 --> 00:43:16.960] If you have short, you know, if you're short, it depends part on what you're there's some other factors, but by and large, it's fairly heritable, right?
[00:43:17.600 --> 00:43:22.720] So, in the early days, there's no gene for height, and there's no small set of genes for height.
[00:43:22.720 --> 00:43:28.960] But you fast forward to now, and height is now the number one trait that we can predict with the highest accuracy.
[00:43:28.960 --> 00:43:42.400] You can capture over 60% of the variance in height by a genome prediction, but that genome prediction requires over 180,000 genetic variants.
[00:43:42.400 --> 00:43:45.440] So, it's distributed across this long tail.
[00:43:45.440 --> 00:43:49.440] So, one of the things that we don't know yet is how you mean SNPs.
[00:43:49.440 --> 00:43:50.960] You mean you're talking about SNPs?
[00:43:51.160 --> 00:43:51.400] Yeah.
[00:43:51.440 --> 00:43:52.000] SNPs, yeah.
[00:43:52.000 --> 00:44:01.160] Which is like one single nucleotide polymorphism, which in English means you substitute out one nucleotide in that gene sequence that changes the function of the gene.
[00:44:01.160 --> 00:44:06.520] So, you need 180,000 of these slight little spelling variations in order to actually predict what's going on.
[00:43:59.840 --> 00:44:07.240] That's impressive.
[00:44:07.560 --> 00:44:08.280] Pretty high.
[00:44:08.280 --> 00:44:13.480] But you could see that there was a really interesting paper, and one of the people they included was Sean Bradley.
[00:44:13.480 --> 00:44:16.440] If you remember him, he was a basketball player.
[00:44:16.440 --> 00:44:18.440] He was 7'6, huge outlier.
[00:44:18.440 --> 00:44:23.080] And you look at this, and you get a distribution, and he's a massive outlier.
[00:44:23.080 --> 00:44:28.440] Like, if you looked at his genome at birth, you could have predicted that he was going to be crazy tall.
[00:44:28.440 --> 00:44:32.600] And so, you can do this in the NBA, you can do it in all these different groups.
[00:44:32.600 --> 00:44:47.880] And so, coming back to the blood, the thing that we don't know yet is it might be possible once we're able to make, say, tens of thousands of measurements out of the blood instead of the handful that we do in medicine, we might find that there's a lot of information in that long tail.
[00:44:47.880 --> 00:44:52.280] It's a little harder because it's not as digital as the genome, but it might be there.
[00:44:52.280 --> 00:45:06.440] And so, it's an open question, but these are some of the things that are really fascinating as we go forward because there might be a ton of signal that will let us optimize health in many ways and look for early warning signs or clear them and so forth.
[00:45:06.440 --> 00:45:11.480] And there is just an incredible amount of data you can pull out of blood that we haven't harnessed yet.
[00:45:11.480 --> 00:45:23.720] One of the things that is, I think, a major force right now, and we saw it with COVID in many ways, is that people are taking charge of their own health care and that they're actually very hungry to do so.
[00:45:23.720 --> 00:45:27.560] And the means that they're looking for today isn't working.
[00:45:27.560 --> 00:45:33.080] And this is coming at the same time where there's actually now all these tools that do miraculous things.
[00:45:33.080 --> 00:45:40.040] You see what you can do with GLP-1s, you can see what you can do with CGMs, you know, these glucose monitors.
[00:45:40.040 --> 00:45:43.000] Metabolic health is such an exciting area.
[00:45:43.000 --> 00:45:54.320] There's numerous areas in health that are being driven by patients and patients as consumers, not as products of the healthcare system, but as real active drivers of it.
[00:45:54.320 --> 00:45:56.880] And that's one of the key areas that we've been interested in.
[00:45:56.880 --> 00:46:00.880] And Daisy and I have been working on that space together.
[00:46:00.880 --> 00:46:11.520] And, you know, we're seeing that basically, I think what's growing is a movement of like-minded companies, like-minded founders that there's an opportunity to really transform healthcare in this way.
[00:46:11.920 --> 00:46:17.840] There's many aspects to healthcare, so this is one part of it, but this part actually I think is really ripe for disruption.
[00:46:17.840 --> 00:46:31.200] And by enabling people to understand their health, whether we're talking about diet, fitness, primary care, and beyond, I think these are areas that are actually something that people are building in today.
[00:46:31.200 --> 00:46:32.720] Yeah, I couldn't agree more.
[00:46:32.720 --> 00:46:42.400] I think we talk a lot about in healthcare, you know, problems of cost and access, but what we don't talk about is how broken the consumer experience is.
[00:46:42.400 --> 00:46:47.680] And it's broken because consumers are not seen as the end customer in healthcare.
[00:46:47.680 --> 00:46:57.120] You know, providers and hospital systems see the insurance company who pays them as their end customer and therefore don't optimize around consumer experience.
[00:46:57.120 --> 00:47:11.520] And what results from that is like, even if you are a highly motivated patient who wants to take control of your health, it's really hard to make appointments and get tests and understand those tests and understand what you can be doing.
[00:47:12.000 --> 00:47:17.200] And then we have problems of behavior change and everyone's like, oh, that's a cultural issue.
[00:47:17.200 --> 00:47:21.480] But I think what we ignore is that the best companies fundamentally change consumer behavior.
[00:47:21.480 --> 00:47:24.480] And we see that all the time in other industries.
[00:47:26.320 --> 00:47:31.080] And so I think we're really ripe for consumer disruption in healthcare.
[00:47:31.720 --> 00:47:34.920] And function is at the forefront of that.
[00:47:29.680 --> 00:47:36.120] Yeah, it's exciting.
[00:47:36.760 --> 00:47:39.000] You think about what healthcare looks like today.
[00:47:39.000 --> 00:47:47.720] And we were just talking about this earlier, but my healthcare records are across a bunch of different doctors' offices and different states.
[00:47:48.200 --> 00:47:52.680] And it's really hard to understand what's happening in my body and how it's changing.
[00:47:52.680 --> 00:47:57.400] And with function, you get, you know, your data is tracked.
[00:47:57.400 --> 00:48:00.680] Every three or six months, you have all these comprehensive tests.
[00:48:00.680 --> 00:48:02.760] You can see how your biomarkers are moving.
[00:48:02.760 --> 00:48:11.960] It plugs, you know, it's going to plug into EHRs and have all the data that happens at a doctor's visit, all the data from your wearable devices.
[00:48:12.280 --> 00:48:18.840] And it's going to be, you know, everything that's happening in every person's body, you know, in one database for them.
[00:48:18.840 --> 00:48:21.720] You know, I think that's an incredible vision.
[00:48:21.720 --> 00:48:28.280] And one of the things that I'm curious about your perspective on is the types of innovations that are happening.
[00:48:28.280 --> 00:48:37.160] Because when I was at Cleveland Clinic, Toby Cosgrove's one of my heroes, you know, brought the kind of discover, whatever we call it, of Watson.
[00:48:37.160 --> 00:48:40.280] It was IBM's sort of supercomputer.
[00:48:40.280 --> 00:48:50.920] And, you know, the big kind of tagline was Watson goes to medical school and was able to sort of ingest all of medical textbooks and knowledge and pass exams and do all that great.
[00:48:50.920 --> 00:48:57.080] And what really struck me was that it was sort of like rearranging the deck chairs in the Titanic.
[00:48:57.080 --> 00:49:08.920] It was using incredible technology to do the same thing better, not to do something fundamentally different that what I would call scientific wellness or functional medicine or systems medicine or whatever you want to call it doesn't matter.
[00:49:08.920 --> 00:49:10.520] It's just going to be medicine.
[00:49:10.840 --> 00:49:19.920] But this paradigm shift is not, from my perspective, not really emerging from a lot of the new startups, new businesses, new innovations that are happening.
[00:49:20.880 --> 00:49:31.520] And I see just incrementalism in innovation, not a fundamental shift in how we think about health and healthcare and disease and diagnosis and treatment.
[00:49:31.920 --> 00:49:37.680] What are you seeing come across your desk that is different?
[00:49:37.680 --> 00:49:41.120] Or are you just seeing the same kind of thing that I think I'm seeing?
[00:49:41.520 --> 00:49:42.400] Am I wrong?
[00:49:42.720 --> 00:49:46.000] Or this is actually how things are shaping up?
[00:49:46.000 --> 00:49:48.720] I don't think you're wrong in a sense that for two factors.
[00:49:48.720 --> 00:49:58.560] One is that, look, I mean, changing a system as complex as healthcare, 20% of US GDP, that's not something that's easy to do.
[00:49:58.560 --> 00:50:03.360] And in fact, too, you can change, you can improve one part, but it's a complex system.
[00:50:03.360 --> 00:50:05.120] That doesn't mean the whole thing improves.
[00:50:05.120 --> 00:50:07.200] So the task is really hard.
[00:50:07.200 --> 00:50:12.640] And then also, there are probably only going to be a few companies that really make this kind of revolutionary change.
[00:50:12.640 --> 00:50:19.680] You think about the companies that have revolutionized other industries, like Spotify revolutionized music.
[00:50:20.000 --> 00:50:25.520] That's something that it was basically one company that did that or a few companies.
[00:50:25.520 --> 00:50:27.280] It's not like hundreds of companies.
[00:50:27.280 --> 00:50:33.920] You can go through Lyft and Uber for transportation or Airbnb for hotels.
[00:50:33.920 --> 00:50:35.680] These are only going to be a few companies.
[00:50:35.680 --> 00:50:38.480] There are going to be many that will try in a couple different ways.
[00:50:38.480 --> 00:50:42.880] But I think what will happen in this space is that a few will really stand out.
[00:50:42.880 --> 00:50:45.440] And these are the ones that will be transformative.
[00:50:45.440 --> 00:50:51.440] We review like thousands of companies before we invest and in a year.
[00:50:51.680 --> 00:51:02.440] And so there's many brilliant, hardworking entrepreneurs in this area, but making this type of change is something that only a few people could do and only a few companies will do.
[00:50:59.680 --> 00:51:03.960] And those are the ones that we're looking for.
[00:51:04.600 --> 00:51:12.840] And what do you think, both of you, around your vision for healthcare and what are the big disruptive innovations that are really game changers for us coming up?
[00:51:13.160 --> 00:51:25.640] I'd love to hear your perspective because you, like I said, you have these sort of crystal ball looking at the future and seeing what's bubbling up and also understanding the complexity of healthcare and understanding the challenges and looking for ways to really shift.
[00:51:25.640 --> 00:51:28.120] So I'd love to kind of hear your vision for the future.
[00:51:28.120 --> 00:51:31.000] Maybe I'll take one area and Daisy can take another.
[00:51:31.000 --> 00:51:39.720] So and we can list more, but like if I were to pick one that is the one that's been on my mind is AI.
[00:51:39.720 --> 00:51:44.200] And when you think about healthcare, what are the big issues in healthcare right now?
[00:51:44.200 --> 00:51:50.120] I think if I were to name the top three, I would call them cost, quality, and access.
[00:51:50.440 --> 00:51:52.840] And AI has a hope to address each one of those.
[00:51:52.840 --> 00:51:54.440] What about outcomes?
[00:51:54.760 --> 00:51:56.680] That's the one I care about as a doctor.
[00:51:56.840 --> 00:51:59.480] I put that in terms of quality, like the quality of outcomes.
[00:51:59.480 --> 00:51:59.880] Yeah.
[00:52:00.200 --> 00:52:11.160] You know, in terms of cost, I think one thing that we're already seeing is that AI is a pilot for doc co-pilot for doctors today and may take on more and more tasks.
[00:52:11.160 --> 00:52:20.760] That's something that can actually, what's exciting about it is that when it can be trained from the very best doctors, it can give access effectively of the very best doctors to everyone.
[00:52:20.760 --> 00:52:22.520] And that's something that we just don't have today.
[00:52:22.520 --> 00:52:26.440] And that democratization of medicine, I think, would be very exciting.
[00:52:26.440 --> 00:52:28.680] So that would be cost and access.
[00:52:28.680 --> 00:52:39.880] And then in terms of quality, you know, when we saw a similar arc in other areas, like in, let's say, on Wall Street 20 years ago, people were talking about using computers to do trading.
[00:52:39.880 --> 00:52:42.200] And the reaction was like, that's ridiculous.
[00:52:42.200 --> 00:52:45.000] Being an expert trader takes like decades and decades, right?
[00:52:45.520 --> 00:52:48.880] And there's no way a computer is going to beat a human being.
[00:52:48.880 --> 00:52:50.640] You know, like, there's no way.
[00:52:50.640 --> 00:52:53.280] And then 20 years later, it's like, well, that's ridiculous.
[00:52:53.280 --> 00:52:56.240] There's no way a human being is going to beat a computer.
[00:52:56.240 --> 00:52:58.080] You know, and we saw this in chess.
[00:52:58.080 --> 00:53:00.480] We saw this in so many different areas.
[00:53:00.480 --> 00:53:10.160] And I think it's the flip that we're in the middle of now is that it feels like hard for some to imagine that, you know, a computer and AI couldn't do what a human being can do.
[00:53:10.160 --> 00:53:13.120] But sometimes you think about what we're asking doctors to do.
[00:53:13.120 --> 00:53:22.640] We're asking them to be machines to grind through all of this information, all this medical data about me and about the world, and instantaneously come up with the answer.
[00:53:22.880 --> 00:53:25.120] That's a lot to put on somebody's shoulders.
[00:53:25.360 --> 00:53:30.560] But I think the hope was that AI working with doctors will be the best of both worlds.
[00:53:30.560 --> 00:53:37.440] And the future of, in terms of cost quality and access, would be dramatically improved.
[00:53:37.440 --> 00:53:41.440] Yeah, I think that's a beautiful vision because I think those are three elements.
[00:53:41.440 --> 00:53:48.400] On the quality bucket, I would put the paradigm shift that's happening too in medicine because, you know, we can do the same things better, right?
[00:53:48.400 --> 00:53:49.360] Which needs to happen.
[00:53:49.360 --> 00:54:09.360] And often when I hear about quality-based care, value-based care, it really to me is often about improving things around the margin, like improving medical efficiencies, reducing errors, care coordination, better EMRs, better tracking of data, maybe better preventive screening, but it's still diagnosing the same diseases, prescribing the same drugs.
[00:54:09.680 --> 00:54:15.040] How do you think AI can play a role in really disrupting the medical paradigm itself, the scientific paradigm?
[00:54:15.040 --> 00:54:22.720] Not just the practice of medicine and getting people access and democratizing it, decentralizing, and bringing down costs and improving all of that.
[00:54:22.720 --> 00:54:25.840] But how did it really change the scientific paradigm?
[00:54:25.840 --> 00:54:28.320] Yeah, I think we talked about the data analysis part.
[00:54:28.320 --> 00:54:29.520] I think that's part of it.
[00:54:29.520 --> 00:54:38.680] But then I think part, and you would know better than I, but like, I think the part of making medicine successful is giving the right care at the right time at the right place.
[00:54:39.160 --> 00:54:44.840] And AI helping doctors and helping medical systems make sure that happens.
[00:54:44.840 --> 00:54:47.240] And this is a win for providers.
[00:54:47.240 --> 00:54:54.840] You know, doctors want to make healthcare better, but it's also a win for payers in that if we can do that, we can keep people healthier.
[00:54:54.840 --> 00:54:59.880] And healthier patients are obviously less expensive, which is the win-win.
[00:54:59.880 --> 00:55:05.160] We think about what healthcare will look like in 20, 30, 40 years, and then we work backwards from that.
[00:55:05.160 --> 00:55:13.720] And we have invested in a lot of companies who are taking on pieces of that puzzle to build us, you know, toward a better tomorrow.
[00:55:13.720 --> 00:55:21.400] But I think, you know, 30 years from now, we probably 90% of healthcare is delivered via your phone.
[00:55:21.400 --> 00:55:35.560] So we're going to have amazing wearable devices, both, you know, in terms of watches, rings, et cetera, but also subcutaneous that are monitoring all sorts of molecules and things happening in our bloodstream in real time.
[00:55:35.800 --> 00:55:37.480] We're going to all be doing functions.
[00:55:38.040 --> 00:55:41.320] We're going to have at-home, you know, blood collection by then.
[00:55:41.320 --> 00:55:43.080] We probably won't meet up with Otomist.
[00:55:43.080 --> 00:55:44.520] We'll have a device to do it.
[00:55:44.520 --> 00:55:48.200] And so we'll have a real monitoring of our health.
[00:55:48.200 --> 00:55:56.440] And you were describing this earlier, but we're going to have all of our health data in this one place and you're going to be able to chat with your phone and say, I have a stomachache.
[00:55:56.440 --> 00:55:57.000] What's going on?
[00:55:57.000 --> 00:55:59.400] Does anything seem weird in my body right now?
[00:55:59.960 --> 00:56:01.560] We'll ask you questions, right?
[00:56:02.200 --> 00:56:02.840] Yes.
[00:56:02.840 --> 00:56:08.600] And we're all going to have access to like the world's best AI and human doctors through our smartphone.
[00:56:08.600 --> 00:56:14.120] And then probably 10% of healthcare will be you know going to the hospital for procedures.
[00:56:14.120 --> 00:57:45.480] But more and more every year is going to be something that's you know you can do at home um with you know and then we'll have you know drug delivery into the home so i i think it's going to look very different um you know 10 20 30 years from now and i hope it happens faster rather than it seems like the costs will then come way down i mean it seems like the the the costs in healthcare are just kind of crazy and i don't i wonder if you're seeing any technology companies that are creating transparency because you know i can send a patient of mine i did this not too long ago who before function uh who wanted to get some lab work done i wanted to sort of check a bunch of things and i i did kind of an abbreviated panel of what's in function and she she her insurance didn't cover it and she sent me said mark like i don't know what to do like the bill is like ten thousand dollars and i'm like oh i'm sorry um let me call the company and so i i call the lab like hey you know like this is not our pricing like you give us a different pricing and so there's such variability in elasticity in the marketplace you can go to one hospital and get a scan for my knee for four hundred dollars another scan is another hospital it's twenty five hundred dollars for the same scan the same machine and and the consumer doesn't know any of this and they're completely confused i went to go get a a knee uh knee exam and i need a knee brace for something like messed up my knee and i get a call from the hospital today they said oh uh just let you know your insurance didn't cover that knee brace and it's a thousand dollars i'm like a thousand dollars for a knee brace i gotta got a new knee you know and and so the the elasticity and pricing is is and the lack of transparency in pricing you know leaves the healthcare so padded with costs.
[00:57:45.480 --> 00:57:54.200] You know we we spend twice as much as any other developed nation and get much worse healthcare outcomes uh you know, we're like the bottom of the pile of developed nations.
[00:57:54.200 --> 00:58:04.280] So, how do you see kind of this evolving and us actually using technology and ai to help create transparency and kind of more democratize healthcare?
[00:58:04.280 --> 00:58:06.120] Because it's so messed up right now.
[00:58:06.280 --> 00:58:07.880] Yeah, it's funny, Mark.
[00:58:08.120 --> 00:58:13.800] We all work in healthcare, and I think none of us understand how the pricing works or what we're going to get.
[00:58:14.440 --> 00:58:15.960] You know, what kind of bill we'll get in the mail.
[00:58:15.960 --> 00:58:19.080] I was actually trying to figure out if I'd hit a deductible today.
[00:58:19.080 --> 00:58:22.680] And it is purposely very confusing.
[00:58:22.680 --> 00:58:26.200] But I think there's a lot of promising changes on the horizon.
[00:58:26.200 --> 00:58:29.560] We're getting some regulatory changes around price transparency.
[00:58:29.560 --> 00:58:37.640] We're investors at a company called Turquoise that's helping consumers and other entities in healthcare understand what everyone's pricing is.
[00:58:37.640 --> 00:58:52.280] And so I do think we're starting to see, and you have a lot of people moving on to high-deductible health plans, which is probably not a great trend in healthcare where you have to, you know, you have to pay out of pocket for the first $5,000, $10,000, $20,000 before your health insurance kicks in.
[00:58:52.520 --> 00:59:00.680] But the silver lining of that is I do think it enables some more free market dynamics where people are going to start shopping for their care and comparing prices.
[00:59:00.680 --> 00:59:04.360] And we are, we're definitely seeing some of that in consumer behavior today.
[00:59:04.360 --> 00:59:06.760] And we actually saw it in relation to function.
[00:59:06.760 --> 00:59:09.080] I think we saw, you know, $500 a year.
[00:59:09.080 --> 00:59:12.280] Is that something that most Americans are going to want to pay?
[00:59:12.280 --> 00:59:21.320] And what really struck us when we were going through all of the customer surveys is how many people were like, this is amazing value.
[00:59:22.360 --> 00:59:23.320] Something's wrong with my health.
[00:59:23.320 --> 00:59:26.840] I'm bouncing around the healthcare system trying to figure out what's going on.
[00:59:26.840 --> 00:59:30.680] And I know these tests would cost me $10,000 elsewhere.
[00:59:31.320 --> 00:59:35.640] And so you guys are obviously doing amazing things for cost in healthcare.
[00:59:36.040 --> 00:59:49.840] But I think to the question about AI, we also obviously, it's funny, BJ and I have talked about this a lot, but AI has way worse margins and it's way more expensive than traditional software, but it is way cheaper than human services.
[00:59:50.160 --> 00:59:52.880] And healthcare is a $4 trillion industry.
[00:59:52.880 --> 00:59:58.240] That's like 90% human services and a lot of expensive human services in doctors.
[00:59:58.240 --> 01:00:01.280] And so I think we're going to see a lot of cost reduction from that.
[01:00:01.280 --> 01:00:20.960] Yeah, I mean, it is striking to me how the value we're getting is so low in terms of the diseases going up, people getting sicker and sicker, you know, rising costs, rising hospital burdens, rising disease burdens, and we're spending more and more than any other nation and getting less and less.
[01:00:20.960 --> 01:00:22.960] And that can't stick.
[01:00:22.960 --> 01:00:30.080] And, you know, I meet with senators and congressmen, and I work in Washington on food policy and healthcare policy.
[01:00:30.080 --> 01:00:32.880] And I don't think any of them even have a clear v.
[01:00:33.200 --> 01:00:43.680] I said to one of the other night, I said, you know that $1.8 trillion of the entire federal budget is spent, which is about a third of the entire federal budget is spent just on healthcare.
[01:00:43.680 --> 01:00:49.920] And not just through Medicare, but Medicaid, the Department of Defense, the Indian Health Services, VA, I mean, you name it, put it all together.
[01:00:49.920 --> 01:00:53.520] It's a ton of dough, and they're not even managing it.
[01:00:53.520 --> 01:00:55.920] They're not even thinking about it as one problem.
[01:00:56.240 --> 01:01:08.960] And so, and the reason I love function is that to me, it's kind of like this little rascal on the outside of healthcare that's trying to give people what they want and bypassing all the red tape, all the confusion, all the lack of transparency.
[01:01:08.960 --> 01:01:15.600] I mean, like I said, I could literally get more than two function memberships for the price of one knee brace.
[01:01:15.600 --> 01:01:18.000] You know, it's like, that's nuts.
[01:01:18.640 --> 01:01:27.120] The other thing that I think anyone who's gotten sick has seen or has loved ones that got sick is that you kind of have to be the one managing that process, right?
[01:01:27.120 --> 01:01:33.320] You kind of like your house is a body and you have to be the general contractor for all the people coming to help fix it.
[01:01:29.760 --> 01:01:35.080] And that's really hard to do.
[01:01:35.400 --> 01:01:42.680] But if you realize that's what's going to happen if you get sick, I think you start having this mindset shift that maybe I can do that while I'm healthy.
[01:01:42.920 --> 01:01:46.280] I don't have to wait till I'm sick to sort of be the general contractor there.
[01:01:46.280 --> 01:01:48.120] I should be thinking about my health.
[01:01:48.120 --> 01:01:49.800] I should be on top of this.
[01:01:49.800 --> 01:01:56.040] And we see more and more people thinking that way with, you know, for all these different reasons, they come to it that healthcare is top of mind.
[01:01:56.040 --> 01:01:59.000] And then they start looking and they start looking for alternatives.
[01:01:59.000 --> 01:02:00.680] And I think that's the opportunity.
[01:02:00.680 --> 01:02:03.160] That's the market opportunity to present those alternatives.
[01:02:03.160 --> 01:02:07.240] If you love this podcast, please share it with someone else you think would also enjoy it.
[01:02:07.240 --> 01:02:09.560] You can find me on all social media channels at Dr.
[01:02:09.560 --> 01:02:10.280] Mark Hyman.
[01:02:10.280 --> 01:02:10.840] Please reach out.
[01:02:10.840 --> 01:02:12.840] I'd love to hear your comments and questions.
[01:02:12.840 --> 01:02:15.160] Don't forget to rate, review, and subscribe to the Dr.
[01:02:15.160 --> 01:02:17.320] Hyman Show wherever you get your podcasts.
[01:02:17.320 --> 01:02:19.560] And don't forget to check out my YouTube channel at Dr.
[01:02:19.560 --> 01:02:22.760] MarkHyman for video versions of this podcast and more.
[01:02:22.760 --> 01:02:24.680] Thank you so much again for tuning in.
[01:02:24.680 --> 01:02:26.040] We'll see you next time on the Dr.
[01:02:26.040 --> 01:02:27.000] Hyman Show.
[01:02:27.000 --> 01:02:34.040] This podcast is separate from my clinical practice at the Ultra Wellness Center, my work at Cleveland Clinic, and F

Prompt 2: Key Takeaways

Now please extract the key takeaways from the transcript content I provided.

Extract the most important key takeaways from this part of the conversation. Use a single sentence statement (the key takeaway) rather than milquetoast descriptions like "the hosts discuss...". 

Limit the key takeaways to a maximum of 3. The key takeaways should be insightful and knowledge-additive. 

IMPORTANT: Return ONLY valid JSON, no explanations or markdown. Ensure:
- All strings are properly quoted and escaped
- No trailing commas
- All braces and brackets are balanced
Format: {"key_takeaways": ["takeaway 1", "takeaway 2"]}

Prompt 3: Segments

Now identify 2-4 distinct topical segments from this part of the conversation.

For each segment, identify:
- Descriptive title (3-6 words)  
- START timestamp when this topic begins (HH:MM:SS format)
- Double check that the timestamp is accurate - a timestamp will NEVER be greater than the total length of the audio
- Most important Key takeaway from that segment. Key takeaway must be specific and knowledge-additive.
- Brief summary of the discussion

IMPORTANT: The timestamp should mark when the topic/segment STARTS, not a range. Look for topic transitions and conversation shifts.

Return ONLY valid JSON. Ensure all strings are properly quoted, no trailing commas:
{
  "segments": [
    {
      "segment_title": "Topic Discussion", 
      "timestamp": "01:15:30",
      "key_takeaway": "main point from this segment",
      "segment_summary": "brief description of what was discussed"
    }
  ]
}

Timestamp format: HH:MM:SS (e.g., 00:05:30, 01:22:45) marking the START of each segment.

Prompt 4: Media Mentions

Now scan the transcript content I provided for ACTUAL mentions of specific media titles:

Find explicit mentions of:
- Books (with specific titles)
- Movies (with specific titles)  
- TV Shows (with specific titles)
- Music/Songs (with specific titles)

DO NOT include:
- Websites, URLs, or web services
- Other podcasts or podcast names

IMPORTANT: 
- Only include items explicitly mentioned by name. Do not invent titles.
- Valid categories are: "Book", "Movie", "TV Show", "Music"
- Include the exact phrase where each item was mentioned
- Find the nearest proximate timestamp where it appears in the conversation
- THE TIMESTAMP OF THE MEDIA MENTION IS IMPORTANT - DO NOT INVENT TIMESTAMPS AND DO NOT MISATTRIBUTE TIMESTAMPS
- Double check that the timestamp is accurate - a timestamp will NEVER be greater than the total length of the audio
- Timestamps are given as ranges, e.g. 01:13:42.520 --> 01:13:46.720. Use the EARLIER of the 2 timestamps in the range.

Return ONLY valid JSON. Ensure all strings are properly quoted and escaped, no trailing commas:
{
  "media_mentions": [
    {
      "title": "Exact Title as Mentioned",
      "category": "Book",
      "author_artist": "N/A",
      "context": "Brief context of why it was mentioned",
      "context_phrase": "The exact sentence or phrase where it was mentioned",
      "timestamp": "estimated time like 01:15:30"
    }
  ]
}

If no media is mentioned, return: {"media_mentions": []}

Prompt 5: Context Setup

You are an expert data extractor tasked with analyzing a podcast transcript. 
I will provide you with part 2 of 2 from a podcast transcript. 
I will then ask you to extract different types of information from this content in subsequent messages. Please confirm you have received and understood the transcript content.

Transcript section:
e coming to help fix it.
[01:01:29.760 --> 01:01:35.080] And that's really hard to do.
[01:01:35.400 --> 01:01:42.680] But if you realize that's what's going to happen if you get sick, I think you start having this mindset shift that maybe I can do that while I'm healthy.
[01:01:42.920 --> 01:01:46.280] I don't have to wait till I'm sick to sort of be the general contractor there.
[01:01:46.280 --> 01:01:48.120] I should be thinking about my health.
[01:01:48.120 --> 01:01:49.800] I should be on top of this.
[01:01:49.800 --> 01:01:56.040] And we see more and more people thinking that way with, you know, for all these different reasons, they come to it that healthcare is top of mind.
[01:01:56.040 --> 01:01:59.000] And then they start looking and they start looking for alternatives.
[01:01:59.000 --> 01:02:00.680] And I think that's the opportunity.
[01:02:00.680 --> 01:02:03.160] That's the market opportunity to present those alternatives.
[01:02:03.160 --> 01:02:07.240] If you love this podcast, please share it with someone else you think would also enjoy it.
[01:02:07.240 --> 01:02:09.560] You can find me on all social media channels at Dr.
[01:02:09.560 --> 01:02:10.280] Mark Hyman.
[01:02:10.280 --> 01:02:10.840] Please reach out.
[01:02:10.840 --> 01:02:12.840] I'd love to hear your comments and questions.
[01:02:12.840 --> 01:02:15.160] Don't forget to rate, review, and subscribe to the Dr.
[01:02:15.160 --> 01:02:17.320] Hyman Show wherever you get your podcasts.
[01:02:17.320 --> 01:02:19.560] And don't forget to check out my YouTube channel at Dr.
[01:02:19.560 --> 01:02:22.760] MarkHyman for video versions of this podcast and more.
[01:02:22.760 --> 01:02:24.680] Thank you so much again for tuning in.
[01:02:24.680 --> 01:02:26.040] We'll see you next time on the Dr.
[01:02:26.040 --> 01:02:27.000] Hyman Show.
[01:02:27.000 --> 01:02:34.040] This podcast is separate from my clinical practice at the Ultra Wellness Center, my work at Cleveland Clinic, and Function Health, where I am chief medical officer.
[01:02:34.040 --> 01:02:36.920] This podcast represents my opinions and my guests' opinions.
[01:02:36.920 --> 01:02:40.760] Neither myself nor the podcast endorses the views or statements of my guests.
[01:02:40.760 --> 01:02:47.800] This podcast is for educational purposes only and is not a substitute for professional care by a doctor or other qualified medical professional.
[01:02:47.800 --> 01:02:54.040] This podcast is provided with the understanding that it does not constitute medical or other professional advice or services.
[01:02:54.040 --> 01:02:58.440] If you're looking for help in your journey, please seek out a qualified medical practitioner.
[01:02:58.440 --> 01:03:06.760] And if you're looking for a functional medicine practitioner, visit my clinic, the ultrawellnesscenter at ultrawellnesscenter.com, and request to become a patient.
[01:03:06.760 --> 01:03:14.720] It's important to have someone in your corner who is a trained, licensed healthcare practitioner and can help you make changes, especially when it comes to your health.
[01:03:14.120 --> 01:03:19.280] This podcast is free as part of my mission to bring practical ways of improving health to the public.
[01:03:19.440 --> 01:03:23.840] So I'd like to express gratitude to sponsors that made today's podcast possible.
[01:03:23.840 --> 01:03:25.840] Thanks so much again for listening.

Prompt 6: Key Takeaways

Now please extract the key takeaways from the transcript content I provided.

Extract the most important key takeaways from this part of the conversation. Use a single sentence statement (the key takeaway) rather than milquetoast descriptions like "the hosts discuss...". 

Limit the key takeaways to a maximum of 3. The key takeaways should be insightful and knowledge-additive. 

IMPORTANT: Return ONLY valid JSON, no explanations or markdown. Ensure:
- All strings are properly quoted and escaped
- No trailing commas
- All braces and brackets are balanced
Format: {"key_takeaways": ["takeaway 1", "takeaway 2"]}

Prompt 7: Segments

Now identify 2-4 distinct topical segments from this part of the conversation.

For each segment, identify:
- Descriptive title (3-6 words)  
- START timestamp when this topic begins (HH:MM:SS format)
- Double check that the timestamp is accurate - a timestamp will NEVER be greater than the total length of the audio
- Most important Key takeaway from that segment. Key takeaway must be specific and knowledge-additive.
- Brief summary of the discussion

IMPORTANT: The timestamp should mark when the topic/segment STARTS, not a range. Look for topic transitions and conversation shifts.

Return ONLY valid JSON. Ensure all strings are properly quoted, no trailing commas:
{
  "segments": [
    {
      "segment_title": "Topic Discussion", 
      "timestamp": "01:15:30",
      "key_takeaway": "main point from this segment",
      "segment_summary": "brief description of what was discussed"
    }
  ]
}

Timestamp format: HH:MM:SS (e.g., 00:05:30, 01:22:45) marking the START of each segment.

Prompt 8: Media Mentions

Now scan the transcript content I provided for ACTUAL mentions of specific media titles:

Find explicit mentions of:
- Books (with specific titles)
- Movies (with specific titles)  
- TV Shows (with specific titles)
- Music/Songs (with specific titles)

DO NOT include:
- Websites, URLs, or web services
- Other podcasts or podcast names

IMPORTANT: 
- Only include items explicitly mentioned by name. Do not invent titles.
- Valid categories are: "Book", "Movie", "TV Show", "Music"
- Include the exact phrase where each item was mentioned
- Find the nearest proximate timestamp where it appears in the conversation
- THE TIMESTAMP OF THE MEDIA MENTION IS IMPORTANT - DO NOT INVENT TIMESTAMPS AND DO NOT MISATTRIBUTE TIMESTAMPS
- Double check that the timestamp is accurate - a timestamp will NEVER be greater than the total length of the audio
- Timestamps are given as ranges, e.g. 01:13:42.520 --> 01:13:46.720. Use the EARLIER of the 2 timestamps in the range.

Return ONLY valid JSON. Ensure all strings are properly quoted and escaped, no trailing commas:
{
  "media_mentions": [
    {
      "title": "Exact Title as Mentioned",
      "category": "Book",
      "author_artist": "N/A",
      "context": "Brief context of why it was mentioned",
      "context_phrase": "The exact sentence or phrase where it was mentioned",
      "timestamp": "estimated time like 01:15:30"
    }
  ]
}

If no media is mentioned, return: {"media_mentions": []}

Full Transcript

[00:00:00.320 --> 00:00:02.000] Coming up on this episode of the Dr.
[00:00:02.000 --> 00:00:02.880] Hyman Show.
[00:00:02.880 --> 00:00:11.840] The thing about like replacing doctors, the line that I really like, I think it's Eric Philpel's, which is AI won't replace doctors, but doctors who use AI will replace doctors who don't.
[00:00:11.840 --> 00:00:16.640] And I think that is a really good way to put it because it is a tool.
[00:00:18.560 --> 00:00:25.520] If you're living on caffeine, constantly tired, and feel like your stress is stuck on high, there's a good chance you're low in magnesium.
[00:00:25.520 --> 00:00:31.200] Magnesium is critical for turning food into energy, calming your nervous system, helping you sleep, and regulating your mood.
[00:00:31.200 --> 00:00:34.320] And here's the kicker: up to 80% of us aren't getting enough.
[00:00:34.320 --> 00:00:42.240] That kind of chronic deficiency may contribute to feelings of burnout, occasional tension, irritability, and the kind of fatigue that coffee can't fix.
[00:00:42.240 --> 00:00:44.240] That's why I recommend Magnesium Breakthrough.
[00:00:44.240 --> 00:00:45.360] Why Bioptimizers?
[00:00:45.360 --> 00:00:52.400] It's the only formula that combines all seven essential forms of magnesium, including glycinate for calm and citrate for digestion.
[00:00:52.400 --> 00:00:55.520] Most magnesium supplements only offer one or two forms.
[00:00:55.520 --> 00:01:01.600] Magnesium Breakthrough covers them all for deeper sleep, better focus, less tension, and real support during stressful times.
[00:01:01.600 --> 00:01:04.720] If your body's feeling off, this might be the missing link.
[00:01:04.720 --> 00:01:10.480] Try it risk-free at bioptimizers.com/slash hymen and get 15% off your order.
[00:01:10.480 --> 00:01:15.120] Before we jump into today's episode, I want to share a few ways you can go deeper on your health journey.
[00:01:15.120 --> 00:01:18.960] While I wish I could work with everyone one-on-one, there just isn't enough time in the day.
[00:01:18.960 --> 00:01:21.760] So I built several tools to help you take control of your health.
[00:01:21.760 --> 00:01:30.080] If you're looking for guidance, education, and community, check out my private membership, The Hyman Hive, for live QAs, exclusive content, and direct connection.
[00:01:30.080 --> 00:01:34.400] For real-time lab testing and personalized insights into your biology, visit Function Health.
[00:01:34.400 --> 00:01:39.760] You can also explore my curated doctor-trusted supplements and health products at drhyman.com.
[00:01:39.760 --> 00:01:46.080] And if you prefer to listen without any breaks, don't forget you can enjoy every episode of this podcast ad-free with Hyman Plus.
[00:01:46.080 --> 00:01:50.800] Just open Apple Podcasts and tap try-free to start your seven-day free trial.
[00:01:51.040 --> 00:01:59.760] I think the interesting thing about the AI scene is it really didn't get real until, let's say, seven, eight years ago.
[00:02:01.480 --> 00:02:09.080] And it really, for our space of medicine, it was confined to medical images, scans.
[00:02:09.720 --> 00:02:12.440] And that was the deep learning phase of AI.
[00:02:12.440 --> 00:02:14.440] And it really has been formidable.
[00:02:14.440 --> 00:02:46.840] That is, just about every type of scan you can imagine, but path slides, electrocardiograms, the retina, as you mentioned, skin lesions, they could be interpreted as well or better by machines that were trained with so-called supervised learning, meaning that, of course, you had to have thousands, tens of thousands, hundreds of thousands of images that were annotated by expert physicians, and then you could train a model to do better than humans.
[00:02:46.840 --> 00:02:49.080] So that was really great.
[00:02:49.080 --> 00:02:56.520] And, you know, back in 2019, when I wrote Deep Medicine, it was about that phase of deep learning.
[00:02:57.000 --> 00:02:58.440] That's like ancient history now, right?
[00:02:58.440 --> 00:02:59.000] 2019.
[00:02:59.080 --> 00:03:00.040] Yeah, I know.
[00:03:00.760 --> 00:03:03.400] It's amazing how quickly that has gone.
[00:03:03.640 --> 00:03:05.080] Yeah, really, Mark.
[00:03:05.080 --> 00:03:17.320] But what's interesting is, you know, I wrote in the book that what we need is a new model because we didn't have one that could take all the layers of what makes us unique.
[00:03:17.880 --> 00:03:31.800] You know, you've alluded to that, not just the electronic health record, but our genome, you know, our gut microbiome, our sensors, our environment, our immunome, the works, right?
[00:03:31.800 --> 00:03:39.400] And the fact that that those data changes over time, and the fact that we could get the corpus of medical knowledge into that as well.
[00:03:39.400 --> 00:04:02.400] So that's where we are now with this transformer model, also known as large language model phase, which is, of course, got major jump in a year ago with Chat GPT, and now, of course, the GPT-4, Gemini, and future models, GPT-5, sometime next year, in fact.
[00:04:02.400 --> 00:04:22.160] And that, of course, is getting us to that state where we could take all that data for a given patient or individual and be able to not only define what is so critical about predicting a condition, better treatment, better prevention.
[00:04:22.720 --> 00:04:27.040] So we're on the cusp, but we haven't done it yet, to be honest.
[00:04:27.040 --> 00:04:35.840] So, you know, no one has actually done multiple layers.
[00:04:35.840 --> 00:04:42.400] They've done electronic health records and a genome, electronic health records, and a scan.
[00:04:42.400 --> 00:04:50.640] But to take multiple layers, including sensors, that's an analytical AI challenge that has yet to be solved.
[00:04:50.640 --> 00:04:54.080] It will be imminently, and that's exciting.
[00:04:54.080 --> 00:05:01.920] Yeah, I mean, you know, when you talk about it, you wrote an article that I thought was just so prescient, and it was such a good description in a short amount of time.
[00:05:01.920 --> 00:05:08.400] And I encourage people to read it called As Artificial Intelligence Goes Multimodal, Medical Applications Multiply.
[00:05:08.400 --> 00:05:39.880] And you talked about how we're going to be getting high-dimensional data that underlie the uniqueness of all of us and how it can be captured from all these different sources that you mentioned, including all the biomarkers we have through biosensors, wearables, implantables, our genome, our microbiome, our metabolome, our immunome, the transcriptome, proteome, epigenome, it goes on and on, and then our electronic health records, our lab tests, our family history, unstructured text from our medical records, and also things that are air pollution sensors we could be wearing.
[00:05:39.880 --> 00:05:44.920] I just got one of those that someone sent me to try to wear it on my air pollution, environmental stressors.
[00:05:45.400 --> 00:05:54.600] All these things are going to be then informed by the whole Medline, a National Library of Medicine database of peer-reviewed data.
[00:05:54.600 --> 00:05:58.040] And it's going to create so much information.
[00:05:58.040 --> 00:06:05.480] And it seems to me there's an intersection of a number of trends right now, which are going to transform medicine in a way that we can barely imagine.
[00:06:05.480 --> 00:06:18.360] And it's going to happen very soon, which is the omics revolution, the systems biology, and medicine revolution, the biosensors and wearable revolution, and then the AI, machine learning, and big data analytic capacity that we have.
[00:06:18.360 --> 00:06:30.840] And so, those five basic trends are all converging in a way that I think is within even four or five years, we're going to see medicine be profoundly different because the acceleration of this is happening so fast.
[00:06:30.840 --> 00:06:56.280] And I'm excited about it because I feel like I've been trying to, with my little brain, put my head around all these immense complexity of human biology, which we've managed to navigate through this reductionist model of medicine and science into siloed specialties where you're super sub-sub-sub-sub-specialist on X, Y, or Z topic, but you don't understand how it all connects and interacts.
[00:06:56.280 --> 00:06:58.920] And so, the first time with AI, it seems like we're gonna be able to do that.
[00:06:58.920 --> 00:07:01.640] So, how do you see this unfolding?
[00:07:01.640 --> 00:07:03.400] And how is this kind of happening?
[00:07:03.400 --> 00:07:04.360] And where are we going?
[00:07:04.360 --> 00:07:07.560] Because I feel like I'm sitting on the edge of my seat.
[00:07:07.560 --> 00:07:17.760] And right now, I feel like we're about to kind of get out of our little dark ages and enter into an era where we're going to be able to make a real transformation in people's health.
[00:07:14.840 --> 00:07:18.880] Well, I think you're right.
[00:07:19.440 --> 00:07:23.600] It's extraordinary, this convergence that you're getting at.
[00:07:24.240 --> 00:07:26.480] And it's going to happen in phases.
[00:07:26.480 --> 00:07:31.920] So the first one is more the practical, which is, you know, I've been calling keyboard liberation.
[00:07:31.920 --> 00:07:32.240] Yeah.
[00:07:32.240 --> 00:07:32.960] Thank God.
[00:07:32.960 --> 00:07:34.000] I heard that you say that.
[00:07:34.000 --> 00:07:35.600] I'm like, hallelujah.
[00:07:35.600 --> 00:07:39.760] Because every doctor is stuck on their keyboard looking at the computer instead of looking at the patient.
[00:07:39.760 --> 00:07:42.000] And so being free of that is so huge.
[00:07:42.160 --> 00:07:45.680] So it's hated mutually by doctors and nurses and patients.
[00:07:45.680 --> 00:07:53.600] I mean, it's everything that people love to hate because it's destroyed that bond, that human-human bond.
[00:07:53.600 --> 00:08:09.280] And that's going to be basically history of data clerk function because we're already seeing now in many health systems around the country that you can do all this through the conversation.
[00:08:09.280 --> 00:08:15.360] The only adjustment you have to make, Mark, is to articulate the physical exam findings with the patient.
[00:08:15.360 --> 00:08:22.240] But other than that, the notes are far superior than the ones that are pecked along.
[00:08:22.240 --> 00:08:30.160] And what's great is once you have that note digitized and it's got all the juice in it, two big things happen.
[00:08:30.160 --> 00:08:42.800] One is that, of course, you could put it in any format conducive for the patient, you know, in terms of educational level or language or, you know, whatever cultural bent.
[00:08:42.800 --> 00:08:51.840] You could also, that patient has the audio file, so if they don't understand something in that note, they can link it right to the auto file, listen to it again.
[00:08:51.840 --> 00:08:56.800] And you know how many patients that you see where they're confused or they don't remember things.
[00:08:56.800 --> 00:09:12.600] But the other big thing is on the clinician side, because instead of having to peck through all this stuff, the orders for new tests and labs and return appointments, prescriptions, billing, pre-authorization, it's all done.
[00:09:12.600 --> 00:09:13.720] It's all done.
[00:09:13.720 --> 00:09:21.000] And the nudges to the patient subsequent about the things that were discussed, like blood pressure, did you check?
[00:09:21.000 --> 00:09:22.200] What were the results?
[00:09:22.200 --> 00:09:27.000] You know, the AI picks that up, gets it back to the physician.
[00:09:27.000 --> 00:09:29.720] You know, all these things are now automated.
[00:09:29.720 --> 00:09:32.920] So that will in itself be welcome.
[00:09:32.920 --> 00:09:40.040] You know, instead of the things that all clinicians want to hate, this is, I think, something that will be widely embraced.
[00:09:40.040 --> 00:09:48.600] And there's no, you know, as you know very well, Mark, there's a lot of concerns about confabulation, hallucination, but that doesn't apply here.
[00:09:48.600 --> 00:09:54.360] I mean, this is, this is not that the AI is not going to be making things up about this kind of thing.
[00:09:54.680 --> 00:09:56.200] Do you have that in your office yet?
[00:09:56.200 --> 00:09:57.400] Do you have that in your office?
[00:09:57.800 --> 00:10:05.080] I've used it at Scripps Health, where I have cardiology practice.
[00:10:05.320 --> 00:10:11.160] They haven't used what I consider the best of these, but they have done a pilot.
[00:10:11.880 --> 00:10:24.360] The largest one is the Microsoft Nuance, but the company that I've advised is Abridge Health, which is derived from University of Pittsburgh and Carnegie Mellon.
[00:10:24.360 --> 00:10:25.640] But there's been several.
[00:10:25.640 --> 00:10:28.680] I mean, there's about 20 of these out there in various testing.
[00:10:29.240 --> 00:10:31.480] I want to get one right away for my practice.
[00:10:32.120 --> 00:10:43.080] Yeah, I mean, I don't, I think this is an inevitability because this is finally the payback for all these bad years of having to become data clerks.
[00:10:43.080 --> 00:10:44.400] But it's just the beginning.
[00:10:44.200 --> 00:10:48.000] You know, it's just one thing that's going to be remarkably different.
[00:10:48.480 --> 00:10:52.560] And that helps us to care better, but it doesn't change what we're doing.
[00:10:52.560 --> 00:11:02.480] In other words, you know, we're going to be able to read x-rays better and MRI imaging better and pathology reports better and EKGs better and retinal imaging that tells us so much about a patient's health.
[00:11:02.480 --> 00:11:10.160] And these are incredible advances that are going to create much more refinement and understanding of how to be precise in our diagnosis of patients.
[00:11:10.160 --> 00:11:12.560] And that's going to up-level medicine for sure.
[00:11:12.560 --> 00:11:16.160] But let me, can I just say one thing?
[00:11:16.160 --> 00:11:16.560] Yeah.
[00:11:16.560 --> 00:11:21.440] Because the retinal image is something that is extraordinary.
[00:11:21.760 --> 00:11:23.440] So before we just pass over that.
[00:11:23.680 --> 00:11:24.240] Yeah, yeah, I know.
[00:11:24.640 --> 00:11:34.720] You know, I just want to point out that, you know, the original task was to see if the AI could interpret the image as well as a clinician.
[00:11:34.720 --> 00:11:42.000] But what wasn't envisioned is that the AI could see things that humans will never see.
[00:11:42.000 --> 00:12:16.640] So with the retina, as you touched on, the ability to predict Alzheimer's disease, Parkinson's five to seven years before there's any symptoms, the issue of, of course, the hepatobiliary tract, kidney disease, cardiac risk, risk of, you know, across all systems, diabetes control, blood pressure control, someday we will be taking pictures of our own retina and get as a checkup with an AI.
[00:12:16.640 --> 00:12:18.160] So it's pretty amazing.
[00:12:18.160 --> 00:12:21.920] And of course, that extends to cardiograms and chest x-rays.
[00:12:21.920 --> 00:12:31.400] Each of them, there's all this stuff that the AI can see, if you will, that humans will never see it.
[00:12:31.560 --> 00:12:33.560] So it's even better, better than humans.
[00:12:33.640 --> 00:12:34.120] Right.
[00:12:29.760 --> 00:12:34.600] Yeah, yeah.
[00:12:34.760 --> 00:12:49.160] I mean, this is why, you know, when I interviewed Jeff Hinton recently for the podcast I do, Ground Truths, he said, you know, he's worried about AI because it's getting advanced so quickly, but not for medicine.
[00:12:49.160 --> 00:12:51.000] He thinks this is the sweet spot.
[00:12:51.000 --> 00:12:54.920] This is really where the good is extraordinary.
[00:12:54.920 --> 00:12:55.400] I agree.
[00:12:55.400 --> 00:13:00.040] I mean, you know, I remember in medical school, you had the ophthalmoscope and you had to look in someone's eye.
[00:13:00.040 --> 00:13:05.000] And you know, okay, you learn about AV nicking and high blood pressure and diabetic retinopathy and macular generation.
[00:13:05.000 --> 00:13:06.360] You could see all that stuff.
[00:13:06.360 --> 00:13:08.600] But there wasn't a whole lot else you could kind of figure out.
[00:13:08.600 --> 00:13:13.000] You know, and if you're an ophthalmologist, you might have a few more refinements in your ability to see things.
[00:13:13.000 --> 00:13:15.800] But what you're saying is you can see things like Alzheimer's.
[00:13:15.800 --> 00:13:17.400] So how does it pick that up?
[00:13:17.400 --> 00:13:21.720] What is it actually seeing and looking at, for example, for Alzheimer's?
[00:13:22.040 --> 00:13:35.400] Well, you know, this goes back to when the realization was made, and that was when you showed the retina picture to ophthalmologists and you say, is this retina from a man or a woman?
[00:13:35.400 --> 00:13:38.280] They got it right 50% of the time.
[00:13:38.280 --> 00:13:41.880] And the AI got it right 97% of the time.
[00:13:41.880 --> 00:13:44.920] And the answer is, we don't really know.
[00:13:44.920 --> 00:13:45.560] Okay.
[00:13:45.560 --> 00:13:56.360] That is, there's explainability work to, you know, define these so-called saliency maps to try to deconvolute the model.
[00:13:56.360 --> 00:14:04.440] But as far as what is it picking up to see the risk of Alzheimer's or Parkinson's or a pattern billiaries, it isn't clear.
[00:14:04.440 --> 00:14:19.280] I mean, there's some aspects that have been determined, but basically, because these models are so extraordinary in terms of what they've learned, and this is all from deep learning.
[00:14:14.840 --> 00:14:22.160] This isn't even from this transformer model era.
[00:14:22.320 --> 00:14:24.320] So can you just stop here for a sec?
[00:14:24.640 --> 00:14:26.880] You're talking about deep learning, transformer model.
[00:14:26.880 --> 00:14:30.320] Can you just explain the sort of shift in what you're thinking?
[00:14:30.320 --> 00:14:32.640] And because I don't think most people understand what that is.
[00:14:32.960 --> 00:14:33.440] Right.
[00:14:33.440 --> 00:14:38.720] So what was the phase of AI that lit up the world?
[00:14:38.960 --> 00:14:53.120] Jeff Hinton and his colleagues like Jan Lacun and many others, they basically found that there was this ability to input data that was supervised.
[00:14:53.120 --> 00:15:01.440] That is, for our purposes, it was labeled by experts, so-called ground truths.
[00:15:01.440 --> 00:15:13.760] And so they put it what they knew was the actual image interpretation and train with tens of hundreds of thousands of these images so that the machine could see stuff.
[00:15:14.000 --> 00:15:17.520] So this is a knowledge-based or expert-informed AI, right?
[00:15:17.520 --> 00:15:18.240] Yeah, yeah.
[00:15:18.240 --> 00:15:21.760] So that really was, you know, deep neural networks.
[00:15:21.760 --> 00:15:22.800] That was the story.
[00:15:22.800 --> 00:15:26.480] It required a single task, unimodal.
[00:15:27.040 --> 00:15:36.480] And then what happened, a Google team in 2017 discovered what they call transformer models.
[00:15:36.720 --> 00:15:40.480] The title of the preprint, Attention is All You Need.
[00:15:40.480 --> 00:16:05.720] And basically it changed the attention from a single bit of information, like a word in a sentence, to basically the context of the entire sentence, or of course, much broader than that, what turned out to be unsupervised, putting in the entire internet, Wikipedia, 100,000 books, 200,000 books.
[00:16:05.720 --> 00:16:10.920] So that's what the transformer model, large language model, generative AI era that we're in now.
[00:16:10.920 --> 00:16:18.760] It didn't start when ChatGPT was released last year, but it actually was in incubation.
[00:16:19.160 --> 00:16:24.520] It was being pursued about six years now, but it's now blossomed.
[00:16:24.520 --> 00:16:33.320] And that we basically have two big types of AI now: the old, if you will, the old and the new.
[00:16:33.320 --> 00:16:43.400] Yeah, I mean, it just seems it's going to accelerate the pace of medical discovery because, you know, if a simple retinol scan can pick up things that we didn't even know we were missing, you know, we didn't even know we didn't know.
[00:16:43.400 --> 00:16:46.200] They were unknown unknowns, as Donald Russell said.
[00:16:46.520 --> 00:16:46.760] Exactly.
[00:16:47.080 --> 00:16:49.880] And that's just the back of the eye.
[00:16:49.880 --> 00:16:59.000] Imagine when we put in all these things that we just mentioned: the whole omics field, the biosensors, your pictures of what you're eating, your movement pattern.
[00:16:59.000 --> 00:17:08.680] I mean, it's just an enormous amount of data that's going to pick up patterns in that data that we've never seen before and that are going to inform what's happening on a biological level.
[00:17:08.680 --> 00:17:18.840] That I think is going to redefine medicine, just as we sort of redefine physics from a Newtonian or a world is flat view to a quantum view to even beyond that.
[00:17:18.840 --> 00:17:27.960] It's like we're kind of in that era of biology where we basically have a profound revolution that's going to upend medicine.
[00:17:27.960 --> 00:17:38.600] And I'd love to hear your perspective on as we sort of enter that era and we start learning these things and understand the body as a network, understand the body as a system instead of these siloed specialties.
[00:17:38.600 --> 00:17:44.520] How do you see that shifting medicine, medical education, medical practice, reimbursement?
[00:17:44.640 --> 00:17:47.520] I mean, this is a massive shift.
[00:17:47.840 --> 00:17:49.520] Well, it is seismic.
[00:17:50.000 --> 00:17:56.560] It's going to be a challenge because medicine, as you know, doesn't change easily.
[00:17:56.560 --> 00:18:07.280] And then you got, you know, throw in all these other practical matters like, you know, reimbursement and education, regulatory, trust, implementation.
[00:18:07.280 --> 00:18:10.960] I mean, there's a long list here of challenges.
[00:18:10.960 --> 00:18:18.320] So, you know, this isn't going to be easy, but it's going to be, you know, the biggest shakeup in the history of medicine.
[00:18:18.320 --> 00:18:36.720] The question is how we adapt, how we, you know, our problem at the moment, outside of a practical thing like we discussed with the keyboard thing, is to get things implemented, we've got to have compelling evidence.
[00:18:36.720 --> 00:18:51.280] And there's a dearth of that because, you know, just like you can't get thousands of doctors to annotate images, and that's why this new form, transformer model, doesn't require supervised learning.
[00:18:51.280 --> 00:18:52.800] It's self-supervised.
[00:18:52.800 --> 00:18:57.200] So it basically is the bypass to what was holding back medicine.
[00:18:57.200 --> 00:19:08.000] But just like that problem, you know, we have the problem of lack of dedication to do prospective trials, whether they're randomized or not.
[00:19:08.000 --> 00:19:20.480] But getting the compelling evidence, which basically says to everyone in the medical community, this is it, you know, that this is going to lead to better patient outcomes, better, you know, better everything.
[00:19:20.800 --> 00:19:27.040] And there's always going to be some risk, of course, when there's never going to be, you know, total positive side of the story.
[00:19:27.040 --> 00:19:45.880] But we, except for the gastroenterologists who have done 33 randomized trials of colonoscopy with machine vision and a few other randomized trials and radiology that have been quite impressive, particularly mammography, there hasn't been much compelling evidence so far.
[00:19:45.880 --> 00:19:46.840] Yeah, it's true.
[00:19:46.840 --> 00:19:47.240] It's true.
[00:19:47.240 --> 00:20:00.680] But, you know, on the other hand, you look at the amount of deaths that are caused by medical practice, probably a third or fourth leading cause of death or complications or reactions to drugs or medical errors.
[00:20:00.680 --> 00:20:01.480] It's huge.
[00:20:01.480 --> 00:20:06.600] And I was listening to Elon Musk talk about cars and AI and self-driven cars.
[00:20:06.600 --> 00:20:10.760] And he says, you know, about 40,000 people in America die from car accidents every year.
[00:20:10.760 --> 00:20:12.760] You know, what if that was reduced to 10,000?
[00:20:12.760 --> 00:20:18.040] But, you know, that's a dramatic drop, but still, you're going to have some people dying from a self-driving car.
[00:20:18.040 --> 00:20:19.720] And are we willing to accept that?
[00:20:19.720 --> 00:20:36.920] You know, so I think that's really a point where we have to kind of understand the value proposition and understand that there is some risk, but the upside in terms of reducing our healthcare costs, the burden on our healthcare system is going to be profound.
[00:20:39.800 --> 00:20:42.440] You don't need more caffeine or another stress supplement.
[00:20:42.440 --> 00:20:44.040] You might just need more magnesium.
[00:20:44.040 --> 00:20:48.040] Magnesium supports sleep, mood, energy, and focus, but most of us are missing it.
[00:20:48.040 --> 00:20:51.400] That's why I recommend Magnesium Breakthrough by Bioptimizers.
[00:20:51.400 --> 00:20:56.040] It combines seven forms of magnesium for results you can feel in your mood, sleep, focus, and more.
[00:20:56.040 --> 00:21:01.800] Try it now for 15% off at bioptimizers.com/slash hyman and feel the difference.
[00:21:07.240 --> 00:21:12.680] One of the things that we've done recently is to get into digital twins.
[00:21:12.680 --> 00:21:18.400] And so a digital twin is a representation of your body's physiology.
[00:21:18.400 --> 00:21:21.120] And we've done this first for brain health.
[00:21:21.120 --> 00:21:30.080] And so what we can actually do in this case is, and we're going to release a test on this, you know, a product based on this next year.
[00:21:30.080 --> 00:21:39.440] But basically, what you can do is you can monitor for a number of these blood measures, your genetics, cognitive assessments, and so forth.
[00:21:39.440 --> 00:21:43.840] And you can then run a simulation based on your particular biology.
[00:21:43.840 --> 00:21:51.040] And it's based around understanding from a physiologic and molecular level what's driving brain health.
[00:21:51.040 --> 00:22:00.000] And you can actually forecast the likely amount of time that you have with a healthy brain given your current state.
[00:22:00.000 --> 00:22:11.280] More importantly, you can go to personalized recommendations for different kinds of things that people can do, some of which are exercise to keep your oxygenation in your brain high.
[00:22:11.280 --> 00:22:14.480] You can get into things like phosphatidylcholine.
[00:22:14.480 --> 00:22:17.600] Turns out that that becomes rate limiting under low oxygen conditions.
[00:22:17.760 --> 00:22:21.200] Latest people are developing dementia, hugely important.
[00:22:21.520 --> 00:22:23.200] Vitamin D, very simple one.
[00:22:23.200 --> 00:22:24.800] We could talk a lot more about that one.
[00:22:24.880 --> 00:22:26.320] Turns out to be very important.
[00:22:26.320 --> 00:22:28.000] There's many, many of these.
[00:22:28.000 --> 00:22:42.640] But the point is that what you can actually do with the digital twins is you can get a representation of a person's individual risk profile and then tailor the precise recommendations.
[00:22:42.640 --> 00:22:44.800] These recommendations are very different person to person.
[00:22:44.800 --> 00:22:51.280] Once you get to four recommendations, only 1% of people actually benefit from what's the best thing, the best for in the population.
[00:22:51.280 --> 00:22:55.200] We just did those simulations, you know, so it's very interesting when you do that.
[00:22:55.200 --> 00:23:04.280] And so, so you get this intense personalization, and you can get into the physiology and you can start to make sense of this because you have to take the complexity of all these measures.
[00:23:04.360 --> 00:23:11.880] You can't place that on the person, you have to put that into the algorithms and deliver back simple, actionable information.
[00:23:11.880 --> 00:23:22.760] And then, the other side of the coin, which I'll just mention here briefly, is the ChatGPT and all these things that we've, you know, that have shocked the world over the last year.
[00:23:22.760 --> 00:23:39.880] The ability now to deliver personalized insights that give you a lot of context and that you can have a back and forth with, and you can get access to a dialogue, even with what your digital twin is saying, or what you're learning about your body.
[00:23:41.800 --> 00:23:49.480] The capability for us to develop personalization on that front is just radically better than any of us thought it was going to be a couple years ago.
[00:23:49.480 --> 00:23:59.800] And so, those things together are really pushing us into this new world of where we're going to be able to harness so much more of this complexity than we could have even thought about before.
[00:23:59.800 --> 00:24:01.400] I mean, is Chat GPT there?
[00:24:01.400 --> 00:24:08.440] Like now, for example, if I put in all my symptoms, I enter in all my lab data, and I hit, you know, tell me what's wrong and what to do about it.
[00:24:08.440 --> 00:24:11.720] Would it give me anything useful at this point, or is it still far off?
[00:24:11.720 --> 00:24:14.360] So, I've played with this a lot, so maybe I'll jump in on that.
[00:24:14.360 --> 00:24:17.880] But it's pretty much what I do in my free time.
[00:24:17.880 --> 00:24:20.520] I don't do anything else.
[00:24:20.840 --> 00:24:21.800] You're a hypochondriac.
[00:24:22.360 --> 00:24:24.120] You put it on your symptoms.
[00:24:24.120 --> 00:24:25.080] My stomach nerds.
[00:24:25.080 --> 00:24:25.880] I got a head pig.
[00:24:26.040 --> 00:24:26.360] Yeah.
[00:24:26.360 --> 00:24:28.360] So it's partially there.
[00:24:28.360 --> 00:24:34.360] If you use like earlier versions, like the GPT 3.5, for example, you'll get lots of hallucinations.
[00:24:34.360 --> 00:24:36.440] It's sometimes useful, sometimes not.
[00:24:36.440 --> 00:24:41.080] GPT-4 is pretty good, except anyway, there's this weird trend.
[00:24:41.080 --> 00:24:42.760] It's not as good as it used to be.
[00:24:42.760 --> 00:24:44.520] And there's a lot of chatter around that on it.
[00:24:44.520 --> 00:24:46.400] It doesn't let you go as deep as it used to.
[00:24:46.400 --> 00:24:50.800] I don't know if it's legal or they're not really put guardrails on it.
[00:24:44.840 --> 00:24:50.880] Yeah.
[00:24:51.120 --> 00:24:54.240] They put guardrails and various kinds on it and so forth.
[00:24:54.240 --> 00:25:02.160] But as long as your question is reasonably well dealt with in available text that it's generating from, it can be quite good.
[00:25:02.160 --> 00:25:09.280] And I've had, and I've used it, you know, not just on medical issues, but you know, explain statistical analysis of this kind of data or something like that.
[00:25:09.280 --> 00:25:14.000] And it actually gives back really reasonable kinds of information.
[00:25:14.000 --> 00:25:15.920] Now, it's not fully to where it wants to.
[00:25:16.000 --> 00:25:17.600] Oh, and I did see a survey.
[00:25:17.600 --> 00:25:19.120] Maybe you saw this as well.
[00:25:19.120 --> 00:25:26.880] They pulled doctors, and apparently, 60% of doctors are using GPT today, right now, in the background on things that they do.
[00:25:26.880 --> 00:25:29.280] So I saw if you saw that survey.
[00:25:29.840 --> 00:25:34.320] But I was actually not totally ready for prime time, but just to say that.
[00:25:34.320 --> 00:25:35.120] Yeah, go ahead.
[00:25:35.120 --> 00:25:46.320] Well, no, I was at this big medical conference in Lake Nona, and they had this guy from Microsoft with, I think, Prometheus, which was kind of a new version of like Chat GPT that was like, you know, for doctors.
[00:25:46.320 --> 00:25:51.760] And they had a case report that they were sharing and they were entering in this case study.
[00:25:51.760 --> 00:25:54.720] And it got it totally wrong.
[00:25:54.720 --> 00:25:56.480] And I guessed it immediately.
[00:25:56.480 --> 00:25:57.600] Like, I wouldn't guess it.
[00:25:57.600 --> 00:25:59.920] I just knew what it was because I listened to the story.
[00:25:59.920 --> 00:26:19.360] But, you know, it was basically a patient who had frequent urination, fever, chills, you know, had had, I think maybe had had a history of rheumatoid strep long ago or something like that, or had a murmur, or maybe had a murmur as a sort of part of the exam.
[00:26:19.320 --> 00:26:20.480] And it was just a murmur.
[00:26:20.480 --> 00:26:22.480] And I'm like, oh, this guy has endocarditis.
[00:26:22.480 --> 00:26:24.240] This guy has bacterial endocarditis.
[00:26:24.240 --> 00:26:28.480] And the chat, the Prometheus thing said, oh, he's got a kidney infection.
[00:26:28.640 --> 00:26:30.520] And I'm like, no, he's not having a kidney infection.
[00:26:30.520 --> 00:26:31.320] And it was wrong.
[00:26:31.320 --> 00:26:34.040] And it was like in front of like 500 people.
[00:26:29.920 --> 00:26:36.120] So, you know, I kind of wonder.
[00:26:36.280 --> 00:26:40.760] But I do think that things are changing.
[00:26:40.760 --> 00:26:57.560] So as you've gotten into sort of looking at these sort of enormous amounts of data through the phenome typing of people, when that goes into these machine learning AI models, like, you know, where is the next step in this in medicine?
[00:26:58.120 --> 00:27:00.200] Are we all kind of moving towards this?
[00:27:00.200 --> 00:27:02.520] Are doctors going to become in some ways obsolete?
[00:27:02.520 --> 00:27:08.600] Or are they just going to be helping to kind of implement some of the decision support that these tools give?
[00:27:08.600 --> 00:27:23.080] Because personally, I would love to be able to put all the data for my patients in, and instead of spending hours and hours modeling over it and thinking about it, trying to remember every study I ever read and what to do in my medical school training, like this is going to give me kind of a roadmap to start with and then implement it.
[00:27:23.480 --> 00:27:25.320] How far are we away from that?
[00:27:25.320 --> 00:27:27.960] Well, I'll make a couple of comments.
[00:27:28.200 --> 00:27:40.840] I think a really important thing about these large language models, which is what GPT and the other things we've talked about are, is that they have to be educated properly.
[00:27:40.840 --> 00:27:57.000] So if you take a large language model and you expose it to the internet, you expose it to the conspiracy theories and the lying and all of those other things, you have an enormous susceptibility in that device.
[00:27:57.000 --> 00:28:08.200] And my argument is for health, we ought to have a GPT that has only been educated with biomedical data.
[00:28:08.200 --> 00:28:13.160] And we're actually collaborating with a group that has one of those.
[00:28:13.480 --> 00:28:24.480] And what our hope is, is we'll, and part of the education has been to put PubMed into the device, which gives you an enormous amount of data.
[00:28:24.480 --> 00:28:29.040] Now, some is right and some is wrong, and you'll still have to make judgments.
[00:28:29.040 --> 00:28:37.840] But what we plan to do is we have access, for example, to Google's knowledge graph.
[00:28:37.840 --> 00:28:46.400] And this is a graph that connected roughly 50 different features from the literature.
[00:28:46.400 --> 00:28:58.400] So it's assembled from the PubMed literature, all of the relationships between genes and proteins and diseases and drugs, and on and on and on.
[00:28:58.400 --> 00:29:03.440] PubMed, for those listening, is just the entire body of peer-reviewed, published medical.
[00:29:03.680 --> 00:29:05.200] Biological information.
[00:29:05.200 --> 00:29:08.800] Yeah, it's a lot, it's millions and millions of studies.
[00:29:08.800 --> 00:29:23.040] Well, this knowledge graph has 50 million nodes and 850 million edges, which means an enormous number of relationships.
[00:29:23.040 --> 00:29:31.600] So we're going to put this knowledge graph in this medically educated GTP.
[00:29:31.600 --> 00:29:37.440] And we're going to put in, we're building now a knowledge graph for the kidney.
[00:29:37.440 --> 00:29:41.520] We'd certainly like to put in the knowledge graph for brain health.
[00:29:41.520 --> 00:29:50.320] All of the knowledge graphs and digital twins that we have should go into educating this thing.
[00:29:50.320 --> 00:30:28.600] And then my hope is the following: we'll be able to take the data, genome, and phenome from each individual, enormously more complicated than what we did in Airvail, maybe 10 times as much data as we had initially, and put it in there and ask it to generate from tens of thousands of actionable possibilities, the ordered priority of actionable possibilities that you as an individual can use to optimize your health or avoid disease or whatever.
[00:30:28.920 --> 00:30:38.680] And what the AI will actually do is send this information to a doctor, and there'll be two things the information will have to do.
[00:30:38.680 --> 00:30:46.040] One, clearly explain the actionable possibility and what the doctor and the patient will be expected to do.
[00:30:46.040 --> 00:30:57.800] But two, it's to give the physician the medical evidence for this actionable possibility to assure him or her it's bona fide.
[00:30:58.120 --> 00:31:12.200] And the dramatic result of this is you will be able to take a family practitioner and make him a domain expert in virtually every field of medicine.
[00:31:12.200 --> 00:31:20.600] It gives you this global reach that you were talking about and the capacity to handle virtually anything.
[00:31:20.600 --> 00:31:26.680] And that democratizes medicine in an incredible way.
[00:31:26.680 --> 00:31:39.480] And I'll argue, we'll never ever get rid of the physician because they're, in the end, still an integrative factor that we're a long ways from being able to replicate and so forth.
[00:31:39.800 --> 00:31:47.760] But he will have the tools to become a world expert in every field of medicine.
[00:31:44.600 --> 00:31:51.680] Really, quite a remarkable promise for the future.
[00:31:52.000 --> 00:32:04.720] And what it promises for patients, that is the optimization of this wellness and prevention, Nathan and I have talked about, I think is really dramatic.
[00:32:04.720 --> 00:32:07.120] So, how far away from this are we?
[00:32:07.440 --> 00:32:16.640] So, I think we'll begin to see the effects of this within the next year or so as these things get.
[00:32:16.640 --> 00:32:23.840] I mean, we won't have him in the full glory for, you know, who knows?
[00:32:23.840 --> 00:32:32.080] Maybe 10 years is way too long to say, because look, what, I mean, that 60% of the doctors would use a tool like this.
[00:32:32.080 --> 00:32:39.600] I would have said there's no way in the world that that conservative group of people would ever go into AI like this.
[00:32:39.600 --> 00:32:40.000] And yet.
[00:32:40.160 --> 00:32:43.120] So they're putting their patients' history in there and saying, hey, what's wrong?
[00:32:43.120 --> 00:32:44.160] Is that what they're doing?
[00:32:44.480 --> 00:32:44.960] Yeah.
[00:32:46.080 --> 00:32:49.280] Well, I don't want, we should probably not over.
[00:32:49.440 --> 00:32:51.440] It means they use it to some degree.
[00:32:51.440 --> 00:33:03.680] Because the thing about replacing doctors, the line that I really like, I think it's Eric Topel's, which is, you know, AI won't replace doctors, but doctors who use AI will replace doctors who don't.
[00:33:03.680 --> 00:33:09.200] And I think that is a really good way to put it because it is a tool.
[00:33:09.200 --> 00:33:12.480] And I think it's like today, it's already a super useful tool.
[00:33:12.480 --> 00:33:25.440] Like, if you're trying to remember something or if you want to delve into the literature, it's so, you know, you can, and especially with these particular GPTs that are based around PubMed and things like that, they're already an assist, right?
[00:33:25.440 --> 00:33:30.000] So it's just already a function of how strongly that assist can be made.
[00:33:30.760 --> 00:33:40.840] And I think the doctor's still going to be the quarterback, but your ability to block and tackle and just solve lots of issues with the AIs is incredible.
[00:33:40.840 --> 00:33:41.960] And it's not just the LLMs.
[00:33:41.960 --> 00:33:50.040] I mean, one of the really biggest uses that's straightforward right off the bat is getting rid of as many medical errors as possible, right?
[00:33:50.040 --> 00:33:56.200] Because a doctor who's tired, it's easy to, you got a long, complicated name, and there's two of them that look almost exactly the same.
[00:33:56.200 --> 00:33:59.960] It's pretty easy to accidentally check the wrong box.
[00:33:59.960 --> 00:34:08.280] But if the AI actually knows, well, you said your patient has diabetes and that's a drug, did you actually mean this drug for multiple sclerosis?
[00:34:08.280 --> 00:34:08.840] Right?
[00:34:08.840 --> 00:34:10.520] And that's already happening today, right?
[00:34:10.520 --> 00:34:22.040] Hospital systems have saved millions of lives already by just implementing some of those really simple things, the kind of mistake that's easy to make as a human and a computer won't make.
[00:34:22.040 --> 00:34:29.640] Now, vice versa, computers will make the kind of, you know, and AIs will make errors that a human never would because they don't understand causality.
[00:34:29.640 --> 00:34:31.000] They don't understand the context.
[00:34:31.000 --> 00:34:37.960] They don't, you know, there's all kinds of stuff, like the case study that you got right that the AI didn't, like there's things that it doesn't know.
[00:34:37.960 --> 00:34:52.520] So a hybrid or what we call centaur AI in the book, a hybrid approach really makes a lot of sense so you can cover your bases because those two kinds of intelligence, human intelligence and AI, actually operate quite differently.
[00:34:52.520 --> 00:34:54.440] And the kind of errors you make are very different.
[00:34:54.440 --> 00:34:56.680] So combining them is powerful.
[00:34:56.760 --> 00:35:10.280] What you're talking about is definitely going to help transform the expertise of physicians and allow them to practice medicine that's more up-to-date, that reflects the scientific literature, that is based on understanding a wide network of biological factors that they haven't been able to consider before.
[00:35:10.280 --> 00:35:11.880] And that's going to be fantastic.
[00:35:11.880 --> 00:35:17.120] But the truth is that wellness, health, does not happen in a doctor's office, right?
[00:35:17.120 --> 00:35:28.960] And so 80 to 90% of the things that determine your health actually don't require a doctor and are things that you can learn about yourself and fix without a doctor's help.
[00:35:28.960 --> 00:35:37.120] And so, in a way, this is also going to help, I think, disintermediate people from the healthcare system and from doctors because we don't really have a healthcare system.
[00:35:37.120 --> 00:35:38.400] We have a sick care system.
[00:35:38.400 --> 00:35:58.000] And so, what you're talking about is actually a new kind of healthcare system where people are going to be empowered with their own health data, guided by these big, dense data clouds of their own biological information from all their omics to their blood panels, to things we don't even measure now that we're going to measure to their wearables and biometrics.
[00:35:58.000 --> 00:35:59.360] I mean, I have a Garmin watch.
[00:35:59.360 --> 00:36:06.880] I mean, I know everything about myself: my pulse ox, my heart rate ability, how much I slept, how much deep sleep, how much light sleep, about my training printing this is what you know, like how much time I need to recover.
[00:36:06.880 --> 00:36:11.840] I mean, it's pretty impressive, and all that is just sitting out there ready to be kind of harvested and used.
[00:36:11.840 --> 00:36:25.360] And so, individuals, I think, are in this moment where they can become more empowered to be actors in determining their own degree of wellness and health, and then know when to go to the doctor.
[00:36:25.360 --> 00:36:28.240] Like, oh, well, gee, you know, your creatinine's like five.
[00:36:28.240 --> 00:36:30.720] You better get your ass over to the nephrologist tomorrow.
[00:36:30.720 --> 00:36:32.800] So, that's going to for sure be still there.
[00:36:32.800 --> 00:36:37.280] But a lot of the stuff that actually requires a physician isn't really needed.
[00:36:37.280 --> 00:36:44.320] It's really diet, lifestyle, you know, behavioral changes, supplements, and other practices that they have access to.
[00:36:44.320 --> 00:36:53.440] So, how do you see this kind of being a tool that the individuals and patients and consumers can use in a way that is really going to disrupt healthcare?
[00:36:53.440 --> 00:37:01.800] You know, Mark, I think you made a really excellent point, and that is the importance of education for the consumer, if you will.
[00:36:59.840 --> 00:37:05.160] And we're doing a number of things in that regard.
[00:37:05.480 --> 00:37:40.440] For example, this past year, an educational team at the Institute for Systems Biology that I initiated 20 years ago to deal with K through 12 science education problems has put together a four-module one-year course based on two chapters several of us wrote in the systems biology and systems medicine book, one on systems medicine, one on P4 healthcare.
[00:37:40.760 --> 00:37:59.160] And the essence of this module is to give them the picture that is portrayed in our book of what healthcare is going to be in the future and to clearly explain the responsibilities they'll have for their own education.
[00:37:59.160 --> 00:38:13.880] And it makes very strongly the point: the core of your health is going to be diet, exercise, sleep, stress, et cetera.
[00:38:13.880 --> 00:38:20.520] And these are things you can do about it, and these are tools and devices you can use to measure it.
[00:38:20.520 --> 00:38:32.040] And oh, by the way, there is this more sophisticated medicine of assaying your blood and your gut microbiome that can tell us.
[00:38:32.040 --> 00:38:46.720] And by the time students will get done with that year course, I'll guarantee they'll know more about what I think, what we think the future of medicine is than 95% of the physicians out there.
[00:38:46.720 --> 00:39:03.680] I mean, this revolution in transforming healthcare from a disease orientation to an orientation of wellness and prevention, I can't stress how important that's going to be in doing two things.
[00:39:03.680 --> 00:39:12.080] One, improving the quality of health for every single individual that practices, even partially.
[00:39:12.560 --> 00:39:26.480] And two, it's going to lead to enormous cost savings in the healthcare system by avoiding what costs 86% of our health care dollars today, namely chronic diseases.
[00:39:26.480 --> 00:39:35.600] And Mark, I'd love to kind of weigh on that question as well that you asked, because I think it's such an important thing because you're exactly right.
[00:39:35.600 --> 00:39:40.960] Because more and more of what we can call, you know, put under healthcare, especially if we start talking about wellness care, right?
[00:39:40.960 --> 00:39:45.360] We like to say scientific wellness should be the front door of the healthcare system.
[00:39:45.360 --> 00:39:58.160] Most of that effort should really be on this maintenance of health, and then you get referred back into the disease care system when hopefully early enough to really make a difference, but with some advanced warning.
[00:39:58.160 --> 00:40:08.880] But the ability for us to deliver this really efficiently and low cost, I totally agree with you, is pushing this more and more to the home remotely, making it easier.
[00:40:08.880 --> 00:40:18.480] So some of the things that we've done, for example, we've spent the last few years developing an essentially, painless, you know, at-home blood collection device.
[00:40:18.880 --> 00:40:21.680] It used to be called the OneDraw, now called the NanoDrop.
[00:40:22.000 --> 00:40:23.760] But that's like one feature of it.
[00:40:24.160 --> 00:40:27.280] You're not going to go to jail like Elizabeth Holmes with this, are you?
[00:40:27.600 --> 00:40:28.400] Not at all.
[00:40:28.400 --> 00:40:29.800] Yes, exactly.
[00:40:29.120 --> 00:40:32.680] That was my objection to the name change, obviously.
[00:40:29.280 --> 00:40:34.600] Sounds like very familiar.
[00:40:36.040 --> 00:40:38.360] I have watched, yeah, I have gotten into her story.
[00:40:38.360 --> 00:40:39.960] The Mano Tainer.
[00:40:41.800 --> 00:40:43.480] Yeah, I read the book.
[00:40:43.480 --> 00:40:45.800] I watched the documentary like 12 times.
[00:40:45.800 --> 00:40:48.840] I watched the dramatization one they did of it.
[00:40:49.160 --> 00:40:52.680] It's a fascinating story in many ways.
[00:40:52.680 --> 00:40:53.960] But you can move to home, right?
[00:40:53.960 --> 00:40:55.080] Microbiome testing, right?
[00:40:55.080 --> 00:40:56.120] You can do that in your home.
[00:40:56.120 --> 00:40:58.920] You can get access to this with AIs.
[00:40:59.560 --> 00:41:04.760] We developed something called the microbiome wipe to make that as easy as possible for people and so forth.
[00:41:04.760 --> 00:41:27.400] But the whole idea is that we should be able to deliver health information to people in ways that are much more efficient, much more user-friendly, not nearly as expensive, and that people can have a real control over that kind of health and be informed by really deep data.
[00:41:27.720 --> 00:41:30.040] I think that's really the key.
[00:41:30.040 --> 00:41:36.760] Oh, and on the you know, coming back to some of these, you know, like small measurements, you know, you brought up Elizabeth Holmes and so forth.
[00:41:36.760 --> 00:41:44.120] One of the things that's important is that a lot of people have failed in trying to take traditional measures and miniaturizing them.
[00:41:44.440 --> 00:41:49.000] You know, at you know, at least doing a lot of them at the same time.
[00:41:49.000 --> 00:42:02.240] But the kind of things that we're talking about in terms of omics, like a metabolome where you can make thousands of measures, which we're going to do on this device, a protein proteome that you can do, right, again, you know, thousands of measurements.
[00:42:02.240 --> 00:42:05.000] Those are only ever done on small amounts of blood.
[00:42:05.000 --> 00:42:12.680] So, if you know, if Lee and I are running something on that in our lab or any of the top labs in the world, you only ever run those things on time.
[00:42:13.320 --> 00:42:18.080] If you gave them a huge bat of blood, all they would do is take a tiny amount out of it and run it on the mass spec.
[00:42:18.080 --> 00:42:20.640] There's no such thing as running this through it.
[00:42:20.640 --> 00:42:24.160] So, you're talking about technologies that are miniaturized already.
[00:42:24.160 --> 00:42:26.720] That's the only way, that's the way that they work.
[00:42:26.720 --> 00:42:34.880] And so, there isn't actually a technological breakthrough of any kind that's needed to use this small amount of blood to get those many measurements.
[00:42:34.880 --> 00:42:39.760] The breakthrough is you have to understand how to read the information.
[00:42:39.760 --> 00:42:53.200] But in the modern world, I'd much rather have an information challenge than a technology challenge because the information challenge can actually be overcome by getting access to samples, the AIs, the long term.
[00:42:53.440 --> 00:42:55.680] And I'll give one interesting example.
[00:42:55.680 --> 00:42:58.560] So, think about what happened in genomics.
[00:42:58.560 --> 00:43:03.840] So, in the genome, initially, one of the traits that we couldn't predict from the genome was height.
[00:43:03.840 --> 00:43:05.360] Now, we all know height is heritable, right?
[00:43:05.360 --> 00:43:07.040] If you have tall parents, you have tall kids.
[00:43:07.040 --> 00:43:16.960] If you have short, you know, if you're short, it depends part on what you're there's some other factors, but by and large, it's fairly heritable, right?
[00:43:17.600 --> 00:43:22.720] So, in the early days, there's no gene for height, and there's no small set of genes for height.
[00:43:22.720 --> 00:43:28.960] But you fast forward to now, and height is now the number one trait that we can predict with the highest accuracy.
[00:43:28.960 --> 00:43:42.400] You can capture over 60% of the variance in height by a genome prediction, but that genome prediction requires over 180,000 genetic variants.
[00:43:42.400 --> 00:43:45.440] So, it's distributed across this long tail.
[00:43:45.440 --> 00:43:49.440] So, one of the things that we don't know yet is how you mean SNPs.
[00:43:49.440 --> 00:43:50.960] You mean you're talking about SNPs?
[00:43:51.160 --> 00:43:51.400] Yeah.
[00:43:51.440 --> 00:43:52.000] SNPs, yeah.
[00:43:52.000 --> 00:44:01.160] Which is like one single nucleotide polymorphism, which in English means you substitute out one nucleotide in that gene sequence that changes the function of the gene.
[00:44:01.160 --> 00:44:06.520] So, you need 180,000 of these slight little spelling variations in order to actually predict what's going on.
[00:43:59.840 --> 00:44:07.240] That's impressive.
[00:44:07.560 --> 00:44:08.280] Pretty high.
[00:44:08.280 --> 00:44:13.480] But you could see that there was a really interesting paper, and one of the people they included was Sean Bradley.
[00:44:13.480 --> 00:44:16.440] If you remember him, he was a basketball player.
[00:44:16.440 --> 00:44:18.440] He was 7'6, huge outlier.
[00:44:18.440 --> 00:44:23.080] And you look at this, and you get a distribution, and he's a massive outlier.
[00:44:23.080 --> 00:44:28.440] Like, if you looked at his genome at birth, you could have predicted that he was going to be crazy tall.
[00:44:28.440 --> 00:44:32.600] And so, you can do this in the NBA, you can do it in all these different groups.
[00:44:32.600 --> 00:44:47.880] And so, coming back to the blood, the thing that we don't know yet is it might be possible once we're able to make, say, tens of thousands of measurements out of the blood instead of the handful that we do in medicine, we might find that there's a lot of information in that long tail.
[00:44:47.880 --> 00:44:52.280] It's a little harder because it's not as digital as the genome, but it might be there.
[00:44:52.280 --> 00:45:06.440] And so, it's an open question, but these are some of the things that are really fascinating as we go forward because there might be a ton of signal that will let us optimize health in many ways and look for early warning signs or clear them and so forth.
[00:45:06.440 --> 00:45:11.480] And there is just an incredible amount of data you can pull out of blood that we haven't harnessed yet.
[00:45:11.480 --> 00:45:23.720] One of the things that is, I think, a major force right now, and we saw it with COVID in many ways, is that people are taking charge of their own health care and that they're actually very hungry to do so.
[00:45:23.720 --> 00:45:27.560] And the means that they're looking for today isn't working.
[00:45:27.560 --> 00:45:33.080] And this is coming at the same time where there's actually now all these tools that do miraculous things.
[00:45:33.080 --> 00:45:40.040] You see what you can do with GLP-1s, you can see what you can do with CGMs, you know, these glucose monitors.
[00:45:40.040 --> 00:45:43.000] Metabolic health is such an exciting area.
[00:45:43.000 --> 00:45:54.320] There's numerous areas in health that are being driven by patients and patients as consumers, not as products of the healthcare system, but as real active drivers of it.
[00:45:54.320 --> 00:45:56.880] And that's one of the key areas that we've been interested in.
[00:45:56.880 --> 00:46:00.880] And Daisy and I have been working on that space together.
[00:46:00.880 --> 00:46:11.520] And, you know, we're seeing that basically, I think what's growing is a movement of like-minded companies, like-minded founders that there's an opportunity to really transform healthcare in this way.
[00:46:11.920 --> 00:46:17.840] There's many aspects to healthcare, so this is one part of it, but this part actually I think is really ripe for disruption.
[00:46:17.840 --> 00:46:31.200] And by enabling people to understand their health, whether we're talking about diet, fitness, primary care, and beyond, I think these are areas that are actually something that people are building in today.
[00:46:31.200 --> 00:46:32.720] Yeah, I couldn't agree more.
[00:46:32.720 --> 00:46:42.400] I think we talk a lot about in healthcare, you know, problems of cost and access, but what we don't talk about is how broken the consumer experience is.
[00:46:42.400 --> 00:46:47.680] And it's broken because consumers are not seen as the end customer in healthcare.
[00:46:47.680 --> 00:46:57.120] You know, providers and hospital systems see the insurance company who pays them as their end customer and therefore don't optimize around consumer experience.
[00:46:57.120 --> 00:47:11.520] And what results from that is like, even if you are a highly motivated patient who wants to take control of your health, it's really hard to make appointments and get tests and understand those tests and understand what you can be doing.
[00:47:12.000 --> 00:47:17.200] And then we have problems of behavior change and everyone's like, oh, that's a cultural issue.
[00:47:17.200 --> 00:47:21.480] But I think what we ignore is that the best companies fundamentally change consumer behavior.
[00:47:21.480 --> 00:47:24.480] And we see that all the time in other industries.
[00:47:26.320 --> 00:47:31.080] And so I think we're really ripe for consumer disruption in healthcare.
[00:47:31.720 --> 00:47:34.920] And function is at the forefront of that.
[00:47:29.680 --> 00:47:36.120] Yeah, it's exciting.
[00:47:36.760 --> 00:47:39.000] You think about what healthcare looks like today.
[00:47:39.000 --> 00:47:47.720] And we were just talking about this earlier, but my healthcare records are across a bunch of different doctors' offices and different states.
[00:47:48.200 --> 00:47:52.680] And it's really hard to understand what's happening in my body and how it's changing.
[00:47:52.680 --> 00:47:57.400] And with function, you get, you know, your data is tracked.
[00:47:57.400 --> 00:48:00.680] Every three or six months, you have all these comprehensive tests.
[00:48:00.680 --> 00:48:02.760] You can see how your biomarkers are moving.
[00:48:02.760 --> 00:48:11.960] It plugs, you know, it's going to plug into EHRs and have all the data that happens at a doctor's visit, all the data from your wearable devices.
[00:48:12.280 --> 00:48:18.840] And it's going to be, you know, everything that's happening in every person's body, you know, in one database for them.
[00:48:18.840 --> 00:48:21.720] You know, I think that's an incredible vision.
[00:48:21.720 --> 00:48:28.280] And one of the things that I'm curious about your perspective on is the types of innovations that are happening.
[00:48:28.280 --> 00:48:37.160] Because when I was at Cleveland Clinic, Toby Cosgrove's one of my heroes, you know, brought the kind of discover, whatever we call it, of Watson.
[00:48:37.160 --> 00:48:40.280] It was IBM's sort of supercomputer.
[00:48:40.280 --> 00:48:50.920] And, you know, the big kind of tagline was Watson goes to medical school and was able to sort of ingest all of medical textbooks and knowledge and pass exams and do all that great.
[00:48:50.920 --> 00:48:57.080] And what really struck me was that it was sort of like rearranging the deck chairs in the Titanic.
[00:48:57.080 --> 00:49:08.920] It was using incredible technology to do the same thing better, not to do something fundamentally different that what I would call scientific wellness or functional medicine or systems medicine or whatever you want to call it doesn't matter.
[00:49:08.920 --> 00:49:10.520] It's just going to be medicine.
[00:49:10.840 --> 00:49:19.920] But this paradigm shift is not, from my perspective, not really emerging from a lot of the new startups, new businesses, new innovations that are happening.
[00:49:20.880 --> 00:49:31.520] And I see just incrementalism in innovation, not a fundamental shift in how we think about health and healthcare and disease and diagnosis and treatment.
[00:49:31.920 --> 00:49:37.680] What are you seeing come across your desk that is different?
[00:49:37.680 --> 00:49:41.120] Or are you just seeing the same kind of thing that I think I'm seeing?
[00:49:41.520 --> 00:49:42.400] Am I wrong?
[00:49:42.720 --> 00:49:46.000] Or this is actually how things are shaping up?
[00:49:46.000 --> 00:49:48.720] I don't think you're wrong in a sense that for two factors.
[00:49:48.720 --> 00:49:58.560] One is that, look, I mean, changing a system as complex as healthcare, 20% of US GDP, that's not something that's easy to do.
[00:49:58.560 --> 00:50:03.360] And in fact, too, you can change, you can improve one part, but it's a complex system.
[00:50:03.360 --> 00:50:05.120] That doesn't mean the whole thing improves.
[00:50:05.120 --> 00:50:07.200] So the task is really hard.
[00:50:07.200 --> 00:50:12.640] And then also, there are probably only going to be a few companies that really make this kind of revolutionary change.
[00:50:12.640 --> 00:50:19.680] You think about the companies that have revolutionized other industries, like Spotify revolutionized music.
[00:50:20.000 --> 00:50:25.520] That's something that it was basically one company that did that or a few companies.
[00:50:25.520 --> 00:50:27.280] It's not like hundreds of companies.
[00:50:27.280 --> 00:50:33.920] You can go through Lyft and Uber for transportation or Airbnb for hotels.
[00:50:33.920 --> 00:50:35.680] These are only going to be a few companies.
[00:50:35.680 --> 00:50:38.480] There are going to be many that will try in a couple different ways.
[00:50:38.480 --> 00:50:42.880] But I think what will happen in this space is that a few will really stand out.
[00:50:42.880 --> 00:50:45.440] And these are the ones that will be transformative.
[00:50:45.440 --> 00:50:51.440] We review like thousands of companies before we invest and in a year.
[00:50:51.680 --> 00:51:02.440] And so there's many brilliant, hardworking entrepreneurs in this area, but making this type of change is something that only a few people could do and only a few companies will do.
[00:50:59.680 --> 00:51:03.960] And those are the ones that we're looking for.
[00:51:04.600 --> 00:51:12.840] And what do you think, both of you, around your vision for healthcare and what are the big disruptive innovations that are really game changers for us coming up?
[00:51:13.160 --> 00:51:25.640] I'd love to hear your perspective because you, like I said, you have these sort of crystal ball looking at the future and seeing what's bubbling up and also understanding the complexity of healthcare and understanding the challenges and looking for ways to really shift.
[00:51:25.640 --> 00:51:28.120] So I'd love to kind of hear your vision for the future.
[00:51:28.120 --> 00:51:31.000] Maybe I'll take one area and Daisy can take another.
[00:51:31.000 --> 00:51:39.720] So and we can list more, but like if I were to pick one that is the one that's been on my mind is AI.
[00:51:39.720 --> 00:51:44.200] And when you think about healthcare, what are the big issues in healthcare right now?
[00:51:44.200 --> 00:51:50.120] I think if I were to name the top three, I would call them cost, quality, and access.
[00:51:50.440 --> 00:51:52.840] And AI has a hope to address each one of those.
[00:51:52.840 --> 00:51:54.440] What about outcomes?
[00:51:54.760 --> 00:51:56.680] That's the one I care about as a doctor.
[00:51:56.840 --> 00:51:59.480] I put that in terms of quality, like the quality of outcomes.
[00:51:59.480 --> 00:51:59.880] Yeah.
[00:52:00.200 --> 00:52:11.160] You know, in terms of cost, I think one thing that we're already seeing is that AI is a pilot for doc co-pilot for doctors today and may take on more and more tasks.
[00:52:11.160 --> 00:52:20.760] That's something that can actually, what's exciting about it is that when it can be trained from the very best doctors, it can give access effectively of the very best doctors to everyone.
[00:52:20.760 --> 00:52:22.520] And that's something that we just don't have today.
[00:52:22.520 --> 00:52:26.440] And that democratization of medicine, I think, would be very exciting.
[00:52:26.440 --> 00:52:28.680] So that would be cost and access.
[00:52:28.680 --> 00:52:39.880] And then in terms of quality, you know, when we saw a similar arc in other areas, like in, let's say, on Wall Street 20 years ago, people were talking about using computers to do trading.
[00:52:39.880 --> 00:52:42.200] And the reaction was like, that's ridiculous.
[00:52:42.200 --> 00:52:45.000] Being an expert trader takes like decades and decades, right?
[00:52:45.520 --> 00:52:48.880] And there's no way a computer is going to beat a human being.
[00:52:48.880 --> 00:52:50.640] You know, like, there's no way.
[00:52:50.640 --> 00:52:53.280] And then 20 years later, it's like, well, that's ridiculous.
[00:52:53.280 --> 00:52:56.240] There's no way a human being is going to beat a computer.
[00:52:56.240 --> 00:52:58.080] You know, and we saw this in chess.
[00:52:58.080 --> 00:53:00.480] We saw this in so many different areas.
[00:53:00.480 --> 00:53:10.160] And I think it's the flip that we're in the middle of now is that it feels like hard for some to imagine that, you know, a computer and AI couldn't do what a human being can do.
[00:53:10.160 --> 00:53:13.120] But sometimes you think about what we're asking doctors to do.
[00:53:13.120 --> 00:53:22.640] We're asking them to be machines to grind through all of this information, all this medical data about me and about the world, and instantaneously come up with the answer.
[00:53:22.880 --> 00:53:25.120] That's a lot to put on somebody's shoulders.
[00:53:25.360 --> 00:53:30.560] But I think the hope was that AI working with doctors will be the best of both worlds.
[00:53:30.560 --> 00:53:37.440] And the future of, in terms of cost quality and access, would be dramatically improved.
[00:53:37.440 --> 00:53:41.440] Yeah, I think that's a beautiful vision because I think those are three elements.
[00:53:41.440 --> 00:53:48.400] On the quality bucket, I would put the paradigm shift that's happening too in medicine because, you know, we can do the same things better, right?
[00:53:48.400 --> 00:53:49.360] Which needs to happen.
[00:53:49.360 --> 00:54:09.360] And often when I hear about quality-based care, value-based care, it really to me is often about improving things around the margin, like improving medical efficiencies, reducing errors, care coordination, better EMRs, better tracking of data, maybe better preventive screening, but it's still diagnosing the same diseases, prescribing the same drugs.
[00:54:09.680 --> 00:54:15.040] How do you think AI can play a role in really disrupting the medical paradigm itself, the scientific paradigm?
[00:54:15.040 --> 00:54:22.720] Not just the practice of medicine and getting people access and democratizing it, decentralizing, and bringing down costs and improving all of that.
[00:54:22.720 --> 00:54:25.840] But how did it really change the scientific paradigm?
[00:54:25.840 --> 00:54:28.320] Yeah, I think we talked about the data analysis part.
[00:54:28.320 --> 00:54:29.520] I think that's part of it.
[00:54:29.520 --> 00:54:38.680] But then I think part, and you would know better than I, but like, I think the part of making medicine successful is giving the right care at the right time at the right place.
[00:54:39.160 --> 00:54:44.840] And AI helping doctors and helping medical systems make sure that happens.
[00:54:44.840 --> 00:54:47.240] And this is a win for providers.
[00:54:47.240 --> 00:54:54.840] You know, doctors want to make healthcare better, but it's also a win for payers in that if we can do that, we can keep people healthier.
[00:54:54.840 --> 00:54:59.880] And healthier patients are obviously less expensive, which is the win-win.
[00:54:59.880 --> 00:55:05.160] We think about what healthcare will look like in 20, 30, 40 years, and then we work backwards from that.
[00:55:05.160 --> 00:55:13.720] And we have invested in a lot of companies who are taking on pieces of that puzzle to build us, you know, toward a better tomorrow.
[00:55:13.720 --> 00:55:21.400] But I think, you know, 30 years from now, we probably 90% of healthcare is delivered via your phone.
[00:55:21.400 --> 00:55:35.560] So we're going to have amazing wearable devices, both, you know, in terms of watches, rings, et cetera, but also subcutaneous that are monitoring all sorts of molecules and things happening in our bloodstream in real time.
[00:55:35.800 --> 00:55:37.480] We're going to all be doing functions.
[00:55:38.040 --> 00:55:41.320] We're going to have at-home, you know, blood collection by then.
[00:55:41.320 --> 00:55:43.080] We probably won't meet up with Otomist.
[00:55:43.080 --> 00:55:44.520] We'll have a device to do it.
[00:55:44.520 --> 00:55:48.200] And so we'll have a real monitoring of our health.
[00:55:48.200 --> 00:55:56.440] And you were describing this earlier, but we're going to have all of our health data in this one place and you're going to be able to chat with your phone and say, I have a stomachache.
[00:55:56.440 --> 00:55:57.000] What's going on?
[00:55:57.000 --> 00:55:59.400] Does anything seem weird in my body right now?
[00:55:59.960 --> 00:56:01.560] We'll ask you questions, right?
[00:56:02.200 --> 00:56:02.840] Yes.
[00:56:02.840 --> 00:56:08.600] And we're all going to have access to like the world's best AI and human doctors through our smartphone.
[00:56:08.600 --> 00:56:14.120] And then probably 10% of healthcare will be you know going to the hospital for procedures.
[00:56:14.120 --> 00:57:45.480] But more and more every year is going to be something that's you know you can do at home um with you know and then we'll have you know drug delivery into the home so i i think it's going to look very different um you know 10 20 30 years from now and i hope it happens faster rather than it seems like the costs will then come way down i mean it seems like the the the costs in healthcare are just kind of crazy and i don't i wonder if you're seeing any technology companies that are creating transparency because you know i can send a patient of mine i did this not too long ago who before function uh who wanted to get some lab work done i wanted to sort of check a bunch of things and i i did kind of an abbreviated panel of what's in function and she she her insurance didn't cover it and she sent me said mark like i don't know what to do like the bill is like ten thousand dollars and i'm like oh i'm sorry um let me call the company and so i i call the lab like hey you know like this is not our pricing like you give us a different pricing and so there's such variability in elasticity in the marketplace you can go to one hospital and get a scan for my knee for four hundred dollars another scan is another hospital it's twenty five hundred dollars for the same scan the same machine and and the consumer doesn't know any of this and they're completely confused i went to go get a a knee uh knee exam and i need a knee brace for something like messed up my knee and i get a call from the hospital today they said oh uh just let you know your insurance didn't cover that knee brace and it's a thousand dollars i'm like a thousand dollars for a knee brace i gotta got a new knee you know and and so the the elasticity and pricing is is and the lack of transparency in pricing you know leaves the healthcare so padded with costs.
[00:57:45.480 --> 00:57:54.200] You know we we spend twice as much as any other developed nation and get much worse healthcare outcomes uh you know, we're like the bottom of the pile of developed nations.
[00:57:54.200 --> 00:58:04.280] So, how do you see kind of this evolving and us actually using technology and ai to help create transparency and kind of more democratize healthcare?
[00:58:04.280 --> 00:58:06.120] Because it's so messed up right now.
[00:58:06.280 --> 00:58:07.880] Yeah, it's funny, Mark.
[00:58:08.120 --> 00:58:13.800] We all work in healthcare, and I think none of us understand how the pricing works or what we're going to get.
[00:58:14.440 --> 00:58:15.960] You know, what kind of bill we'll get in the mail.
[00:58:15.960 --> 00:58:19.080] I was actually trying to figure out if I'd hit a deductible today.
[00:58:19.080 --> 00:58:22.680] And it is purposely very confusing.
[00:58:22.680 --> 00:58:26.200] But I think there's a lot of promising changes on the horizon.
[00:58:26.200 --> 00:58:29.560] We're getting some regulatory changes around price transparency.
[00:58:29.560 --> 00:58:37.640] We're investors at a company called Turquoise that's helping consumers and other entities in healthcare understand what everyone's pricing is.
[00:58:37.640 --> 00:58:52.280] And so I do think we're starting to see, and you have a lot of people moving on to high-deductible health plans, which is probably not a great trend in healthcare where you have to, you know, you have to pay out of pocket for the first $5,000, $10,000, $20,000 before your health insurance kicks in.
[00:58:52.520 --> 00:59:00.680] But the silver lining of that is I do think it enables some more free market dynamics where people are going to start shopping for their care and comparing prices.
[00:59:00.680 --> 00:59:04.360] And we are, we're definitely seeing some of that in consumer behavior today.
[00:59:04.360 --> 00:59:06.760] And we actually saw it in relation to function.
[00:59:06.760 --> 00:59:09.080] I think we saw, you know, $500 a year.
[00:59:09.080 --> 00:59:12.280] Is that something that most Americans are going to want to pay?
[00:59:12.280 --> 00:59:21.320] And what really struck us when we were going through all of the customer surveys is how many people were like, this is amazing value.
[00:59:22.360 --> 00:59:23.320] Something's wrong with my health.
[00:59:23.320 --> 00:59:26.840] I'm bouncing around the healthcare system trying to figure out what's going on.
[00:59:26.840 --> 00:59:30.680] And I know these tests would cost me $10,000 elsewhere.
[00:59:31.320 --> 00:59:35.640] And so you guys are obviously doing amazing things for cost in healthcare.
[00:59:36.040 --> 00:59:49.840] But I think to the question about AI, we also obviously, it's funny, BJ and I have talked about this a lot, but AI has way worse margins and it's way more expensive than traditional software, but it is way cheaper than human services.
[00:59:50.160 --> 00:59:52.880] And healthcare is a $4 trillion industry.
[00:59:52.880 --> 00:59:58.240] That's like 90% human services and a lot of expensive human services in doctors.
[00:59:58.240 --> 01:00:01.280] And so I think we're going to see a lot of cost reduction from that.
[01:00:01.280 --> 01:00:20.960] Yeah, I mean, it is striking to me how the value we're getting is so low in terms of the diseases going up, people getting sicker and sicker, you know, rising costs, rising hospital burdens, rising disease burdens, and we're spending more and more than any other nation and getting less and less.
[01:00:20.960 --> 01:00:22.960] And that can't stick.
[01:00:22.960 --> 01:00:30.080] And, you know, I meet with senators and congressmen, and I work in Washington on food policy and healthcare policy.
[01:00:30.080 --> 01:00:32.880] And I don't think any of them even have a clear v.
[01:00:33.200 --> 01:00:43.680] I said to one of the other night, I said, you know that $1.8 trillion of the entire federal budget is spent, which is about a third of the entire federal budget is spent just on healthcare.
[01:00:43.680 --> 01:00:49.920] And not just through Medicare, but Medicaid, the Department of Defense, the Indian Health Services, VA, I mean, you name it, put it all together.
[01:00:49.920 --> 01:00:53.520] It's a ton of dough, and they're not even managing it.
[01:00:53.520 --> 01:00:55.920] They're not even thinking about it as one problem.
[01:00:56.240 --> 01:01:08.960] And so, and the reason I love function is that to me, it's kind of like this little rascal on the outside of healthcare that's trying to give people what they want and bypassing all the red tape, all the confusion, all the lack of transparency.
[01:01:08.960 --> 01:01:15.600] I mean, like I said, I could literally get more than two function memberships for the price of one knee brace.
[01:01:15.600 --> 01:01:18.000] You know, it's like, that's nuts.
[01:01:18.640 --> 01:01:27.120] The other thing that I think anyone who's gotten sick has seen or has loved ones that got sick is that you kind of have to be the one managing that process, right?
[01:01:27.120 --> 01:01:33.320] You kind of like your house is a body and you have to be the general contractor for all the people coming to help fix it.
[01:01:29.760 --> 01:01:35.080] And that's really hard to do.
[01:01:35.400 --> 01:01:42.680] But if you realize that's what's going to happen if you get sick, I think you start having this mindset shift that maybe I can do that while I'm healthy.
[01:01:42.920 --> 01:01:46.280] I don't have to wait till I'm sick to sort of be the general contractor there.
[01:01:46.280 --> 01:01:48.120] I should be thinking about my health.
[01:01:48.120 --> 01:01:49.800] I should be on top of this.
[01:01:49.800 --> 01:01:56.040] And we see more and more people thinking that way with, you know, for all these different reasons, they come to it that healthcare is top of mind.
[01:01:56.040 --> 01:01:59.000] And then they start looking and they start looking for alternatives.
[01:01:59.000 --> 01:02:00.680] And I think that's the opportunity.
[01:02:00.680 --> 01:02:03.160] That's the market opportunity to present those alternatives.
[01:02:03.160 --> 01:02:07.240] If you love this podcast, please share it with someone else you think would also enjoy it.
[01:02:07.240 --> 01:02:09.560] You can find me on all social media channels at Dr.
[01:02:09.560 --> 01:02:10.280] Mark Hyman.
[01:02:10.280 --> 01:02:10.840] Please reach out.
[01:02:10.840 --> 01:02:12.840] I'd love to hear your comments and questions.
[01:02:12.840 --> 01:02:15.160] Don't forget to rate, review, and subscribe to the Dr.
[01:02:15.160 --> 01:02:17.320] Hyman Show wherever you get your podcasts.
[01:02:17.320 --> 01:02:19.560] And don't forget to check out my YouTube channel at Dr.
[01:02:19.560 --> 01:02:22.760] MarkHyman for video versions of this podcast and more.
[01:02:22.760 --> 01:02:24.680] Thank you so much again for tuning in.
[01:02:24.680 --> 01:02:26.040] We'll see you next time on the Dr.
[01:02:26.040 --> 01:02:27.000] Hyman Show.
[01:02:27.000 --> 01:02:34.040] This podcast is separate from my clinical practice at the Ultra Wellness Center, my work at Cleveland Clinic, and Function Health, where I am chief medical officer.
[01:02:34.040 --> 01:02:36.920] This podcast represents my opinions and my guests' opinions.
[01:02:36.920 --> 01:02:40.760] Neither myself nor the podcast endorses the views or statements of my guests.
[01:02:40.760 --> 01:02:47.800] This podcast is for educational purposes only and is not a substitute for professional care by a doctor or other qualified medical professional.
[01:02:47.800 --> 01:02:54.040] This podcast is provided with the understanding that it does not constitute medical or other professional advice or services.
[01:02:54.040 --> 01:02:58.440] If you're looking for help in your journey, please seek out a qualified medical practitioner.
[01:02:58.440 --> 01:03:06.760] And if you're looking for a functional medicine practitioner, visit my clinic, the ultrawellnesscenter at ultrawellnesscenter.com, and request to become a patient.
[01:03:06.760 --> 01:03:14.720] It's important to have someone in your corner who is a trained, licensed healthcare practitioner and can help you make changes, especially when it comes to your health.
[01:03:14.120 --> 01:03:19.280] This podcast is free as part of my mission to bring practical ways of improving health to the public.
[01:03:19.440 --> 01:03:23.840] So I'd like to express gratitude to sponsors that made today's podcast possible.
[01:03:23.840 --> 01:03:25.840] Thanks so much again for listening.