Can AI make healthcare human again? with Eric Topol (Transcript)

Listen along

The TED Interview
Can AI make healthcare human again? with Eric Topol
September 29, 2022

[00:00:00] Steven Johnson:
Welcome to the TED Interview. I'm your host, Steven Johnson. The emergence of the novel coronavirus SARS-COV-2 roughly two and a half years ago posed a formidable challenge to many aspects of our lives: our immune systems, our healthcare institutions, the basic routines of everyday life. But in a real sense, it was also a significant challenge to our understanding.

Over the course of the pandemic, we've been forced to make potentially life-or-death decisions that revolve around the latest data on vaccine efficacy, or the evolutionary dynamics of virus variants, or the epidemiology of transmission rates. And needless to say, most of us don't have any training in wrestling with these kinds of questions.

They aren't generally included in the standard high school curriculum (though perhaps they should be), and that lack of expertise has meant that a critical new group of explainers and sense makers have arisen during the COVID crisis. Scientists and doctors and public health experts who've started writing for wide audiences on Twitter and Medium and on op-ed pages, or appearing on podcasts or the Sunday morning talk shows.

Now, don't get me wrong, there have been plenty of charlatans and science deniers out there too, but they've been matched by an inspiring group of true public intellectuals who've stepped up to help us make sense of this bewildering and sometimes terrifying new world. And to my mind, one of the most important figures in this class of sense-makers is our guest today on the TED interview: Eric Topol.

Now I've been following Eric's Twitter feed since the early days of the pandemic, along with a half million other people. He's been an invaluable source in translating the latest scientific papers and clinical trial results into a language that a lay audience can understand. He recently started a Substack newsletter called Ground Truths that’s also required reading for anyone trying to get a handle on where the pandemic is headed.

Eric's principal scientific focus has been on the genomic and digital tools to individualize medicine. He's the founder and director of the Scripps Research Translational Institute, Professor of Molecular Medicine, and Executive Vice President of Scripps Research as a researcher.

He's published over 1200 peer-reviewed articles with more than 290,000 citations, which makes him one of the 10 most cited researchers in medicine. His latest book, Deep Medicine: How Artificial Intelligence Can Make Healthcare Human Again, came out in 2019, and it actually connects with a number of themes we've been exploring this season in our series of episodes on the future of intelligence.

As you'll hear, Eric has his concerns about where we are in our fight against COVID, particularly with the pace of vaccine innovation right now, but thinks that there's a better future for us in healthcare, thanks to innovations like immunotherapy and AI. A deep dive into both the present-day crisis and a more encouraging near future with one of the leading minds in the world of health. That's next on the TED interview.

[BREAK]

[00:03:25] Steven Johnson:
Eric Topol. Welcome to the TED Interview.

[00:03:28] Eric Topol:
Great to be with you, Steven.

[00:03:30] Steven Johnson:
I think it's important that we jump into the kind of current situation with COVID right outta the gate here. And it seems to me that one of the things that really separates the, the kind of the first year of the pandemic from where we are now is the challenge of the variants. You know, it seems like in that first year it was really all about a, a single virus and how long it was gonna take us to develop vaccines, really, to deal with it. And there, there didn't seem to be much public conversation, at least, about a whole range of variants evolving that would develop such skills that an immune escape and so on.

Um, do you think that the scientific community was surprised by the rise of the variants in kind of year two and three, or did we just have so much to focus on, just dealing with the outbreak? Uh, uh, you know, in it’s in—initial form that, that just took up all of our attention.

[00:04:22] Eric Topol:
Well, such a great question Steven, and uh, thanks for your support of the trying to try this pandemic, which is a challenge. Um, I think we were somewhat lulled, uh, in the first year ‘cause as you said, there was no substantive change of the virus. I mean, it was… There were lots of variants. There's obviously lots of mutations, but they didn't lead to anything that was swaying us from the idea that if we got vaccines, which we did by November, 10 months in since the sequence, that we would be home free, that we would have this so-called herd immunity and put this behind us, uh, quickly.

But what happened was not only did we get, um, the successive waves of variants, Uh, Alpha, Delta worldwide. But then we got hit with Omicron, which was really the surprise factor because it had 57 mutations. It was a whammy, you know, it just, and that couldn't have been precisely predicted, that the accelerated evolution of the virus and then unleashed us through infections and super spreading and then worldwide.

And now we're suffering from the fact that that has led to new Omicron variants, which probably each deserve their own Greek letter because they're serious and, um, we don't really have a way to predict where we're headed from here. That is, will we have BA.x or will we have another Omicron, you know, out of the blue?

Because we have so many people out there that are immunocompromised, any one of them could lead to a, a major, uh, new family of variants. And, you know, a lot of people think that, um, it, the things have failed, the vaccines have failed, or whatever's failed. It's really not. It's the virus that's just succeeded so well in evolving. And I think that lost perspective, uh, you know, expecting too much of vaccines, when in fact the virus has just become, you know, in so much more of a beast than it used to be.

[00:06:21] Steven Johnson:
And the estimates are that the vaccines have saved something on the order of 20 million lives. I think there were some studies on that. So there is—

[00:06:29] Eric Topol:
Yeah. Yeah.

[00:06:30] Steven Johnson:
—you know, 20 million people alive who would've been dead is… That’s success, at least in my book.

[00:06:34] Eric Topol:
And I think we, we keep having this illusion that the virus can't get worse, it can't get more immune invasive, it can't get more transmissible. And then it does. So we ought to just start to say that there's still lots of room for this virus to evolve further and start to really plan aggressively to be ahead of it.

That's one thing, Steven. We've just never gone ahead of it. We're just reactive. We're coming up with these variant-specific chasing strategies for vaccines, but we could do so much better than that.

[00:07:08] Steven Johnson:
We really had one of the great triumphant moments in, in, in the history of medical science with the mRNA vaccines. Um, you know, at the end, I mean, just we've talked about it many times. Some, some on this podcast and other places that basically the kind of the architecture of the vaccine was set in place, you know, a couple of days after they, they actually sequenced the virus and, you know, we got shots into arms at this record pace.

Um, so it was, it was in the midst of all this terror and tragedy, there was this extraordinary story of success. It does seem like, and you've written about this a little bit, that the last year or so has been more characterized by stagnation in terms of vaccine development, um, both in terms of, uh, vaccines that would, um, be more appropriate for Omicron, pan-coronavirus vaccines, nasal, um, vaccines, uh, all these potential tools that seem to be also on the table haven't showed up yet. Um, what is going on?

[00:08:10] Eric Topol:
Yeah. Such a, a vital point here, and that is, you know, I do view the, uh, 10 months it took from the sequence of the virus to having huge trials, 75,000+ trials, two vaccines, 95% efficacy, you just never see this. Never. This is unprecedented. You know, normally it takes somewhere close to 10 years to get a vaccine once you have the pathogen.

And we have a lot that we never had a, a vaccine. So how did that happen so fast? Well, of course there were decades of mRNA and nanoparticle work that were the foundation, but it really took this Operation Warp Speed, uh, this government, uh, industry collaborative effort, 10 billion dollars, which is a trivial investment when you think about trillions that are being spent for the pandemic.

And that's what did it, because it de-risked the companies. It said, “Make… make a gazillion doses. Even if it doesn't work, at least we'll be ready, and we'll have a distribution plan, and we'll do all this stuff so that we can actually start vaccinating people and get us out of this pandemic.” So it's an amazing…

Now what is shocking to me is that, that relatively trivial investment, and it could be much smaller, we haven't done the same for next-generation vaccines that are pan-coronavirus or the nasal vaccines, which we need desperately now because our vaccines are basically leaking like a sieve for transmission and infection. So we know the remedies here, but we're unwilling to do another one of these Operation Warp Speeds.

We've called, uh, for Operation Lighting Speed for nasal vaccines, and hopefully, someone in Congress will hear about it, but we just don't, we can't get any traction in terms of an investment at this point.

[00:10:01] Steven Johnson:
And the argument for nasal vaccines is that that really gives you a, a much better defense against infection itself. So the, the vaccines we should point out, continue, particularly with the booster, continue to be miraculously effective at preventing severe disease, even, even with the latest strains. That's, that's correct. Right?

[00:10:18] Eric Topol:
Uh, there's a little slippage in the severe protection as we're seeing with hospitalizations, um, with BA.5. Uh, hopefully, we won't see more slippage, but you're right. I mean, the biggest problem we developed as the variants marched along is that we've lost our infection protection, transmission protection, and we've got some nasal sprays that are in late trials. We've got others that are not far behind, but they've all been on their own.

That is… The day that one of these nasal vaccines hits, which I'm optimistic, we don't have millions of doses ready. These are companies that are relatively small. They're not like a Pfizer, you know, or Johnson and Johnson. So we, we don't have the kind of, uh, the, the big bet made, and it is not big enough to be ready. That’s the problem. And we're just not taking the, the, the, the status of the pandemic or its potential future as serious as we should.

[00:11:11] Steven Johnson:
Can, can we just spend a second explaining the connection between Omicron's, um, you know, those 57 mutations and an immunocompromised person that we believe it might have evolved in? What, how does, how does that process work?

[00:11:28] Eric Topol:
Right. So, you know, we would, when we went from, you know, the original virus to Alpha, Delta, we saw a limited number, a few key mutations. But what happened here is just unbridled evolution on a really fast path. And it just… There was no check on it. It just evolved quickly. And we have documented that in many immunocompromised people where, you know, every few weeks there's a whole bunch more mutations.

And that's the problem is that, um, it doesn't take much for something like this to happen. It's just a combination of, of those mutations in the, in the wrong place for people. Or for our animal reservoirs. I mean, that's another potential source where in the future we could deal, we've already seen spillovers from deer and, uh, hamsters and mink and cat, and that, that's another way you could see accelerated evolution in an animal reservoir that could come to people.

[00:12:29] Steven Johnson:
It's probably, and another important point to make about protecting and taking our own precautions to protect immunocompromised people in the population, right? So that, that they are vulnerable in, in ways that many of the rest of us are not. And so they deserve protection and we need to think of them in, in our actions and the, and the measures we take.

But there's also a collective good here, which is there's likely to be more mutations, uh, if you have more immunocompromised people, uh, infected. So, uh, it, it should be a, a reminder that, that, that, that we keep at the top of our heads.

[00:13:04] Eric Topol:
Absolutely.

[00:13:05] Steven Johnson:
Now, in your latest Substack post, uh, you have a line which really struck me, which is that you said, “Why do I remain optimistic? Because SARS-COV-2 is a much easier virus to prevail over than influenza.”

So we've been talking about some of the negative news of late. Explain what that means, the contrast with influenza and, and tell us some more reasons for optimism, ‘cause I think we’d all like to hear those.

[00:13:32] Eric Topol:
Yeah. So you know, I would say this virus, SARS-COV-2, is a lightweight compared to influenza. Uh, influenza, we talk about mutations. That is the most hyper-mutating virus, you know, that we've seen. And that's why we can never keep up with it. Our vaccines are pretty, uh, you know, uh, low efficacy. I mean, we're usually, with all the multiple, this quadrivalent vaccines that we use for flu each year, try and anticipate the new strains, which we never do right.

I mean, it's just a moving target, you know. It's just extraordinary. Uh, and so our efficacy… Maybe 30%, 40% for flu vaccines, right? The main pill we, the best we ever had was Tamiflu. It's a weak hitter, right? And the, so the properties of influenza, which you'll remember back in 1918, it can be a real nasty virus. Right?

Uh, it's, it's a much bigger bear, worse challenge. Now, SARS-COV-2. Yeah, it mutates, but we've, we've only had, you know, a few that had global dominance. Uh, and the mutation speed is nothing like influenza. We have a Paxlovid pill which makes Tamiflu look like nothing. It's really effective, and you know, it has, it may have rebound problems, but it has, you know, about 90% reduction of hospitalizations and deaths and people at, at significant risk.

So we have proof. You know, there has never been a flu vaccine that had 95% efficacy. We were there for all the way through Delta, right? So, you know, we can conquer, we can get ahead of this virus. I'm confident of that. We just don't see, I don't know why people don't see that, you know, what we learned from influenza is it's much more challenging. The head of the influenza virus, no less the, uh, stalk are just so bigger challenges than the virus we're confronting today.

[00:15:30] Steven Johnson:
I want to turn a little bit to your career and, and the broader kinda landscape of, of the public understanding of science and medicine. Um, you were voted by physicians and number one most influential physician leader, um, uh, which is a great honor.

You are an influencer despite the fact that your Instagram account may not be so good. Um, but you've also written about, and you wrote an op-ed in The Guardian about COVID denialism and, and the challenges of our, kind of our collective inability to really integrate what we've learned about the virus and the pandemic.

Um, I guess looking back over the last two or three years in, in terms of the pandemic itself, what has surprised you—maybe disappointed you, I suspect—about the, the, the public understanding of what we've been going through and, and the reaction to it?

[00:16:24] Eric Topol:
I have to say the biggest shock was, um, the intensity of anti-science and anti-vax. And we knew there was some of that out there, obviously, um, particularly in the United States, but I had no idea that it would build so, you know, exponentially and vehemently, uh, and, you know, get supported and organized and basically, you know, make us much more vulnerable. I think the, the, the, uh, rest of the, uh, public, um, has been supportive of science, you know, but I, I think that was a surprise because, uh, I still believe that, you know, the momentous, just, validation of the vaccines is singular in life science history.

Um, I mean, I'm also obviously into genome editing as a, a major breakthrough. No less the AI, but to discount, how important, as you said, Steven, 20 million lives at least were saved and more to come because of vaccines. Um, it hasn't settled in how science has been so extraordinary during the pandemic, has advanced us to a, a great degree. And, unfortunately, we, you know, we have this counterforce that I never thought would be as large, as intense as, um, um, as it is right now.

[00:17:41] Steven Johnson:
Yeah. And one of the things that I've always felt I've tried in some of my own work to counteract this a little bit is, is that we were set up for that backlash that you're talking about in part because we do a bad job of reminding people of the progress we've made in the past. Right? So the analogy I always use is, you know, every kid knows about the moon landing, but how many kids are taught about smallpox eradication? And yet smallpox eradication absolutely had a more, you know, transformative effect on, particularly on childhood and, and people's experiences of being a parent um, given how many young children died of smallpox over, over millennia. Um, and, and so, you know, and that was the triumph of vaccines, right? It was the triumph of global cooperation and, and vaccine science, um, that really took arguably the biggest killer in, in our history and, and just wiped it off the face of the earth.

I mean, that's an extraordinary achievement, but it's not part of the canon of great breakthroughs that everyone learns in school. And so when you have a new threat that involves a virus and vaccines, people don’t have that history, you know, accessible to them in this way. Don't you think that's kind of a problem?

[00:19:00] Eric Topol:
I, I couldn’t agree with you more. I think that's very insightful. There's nothing more important as far as a medical intervention than preventing a disease. Right? So we don't do a good job of, of making that, you know, inculcated in everyone's mind as they, as they go through school because it's, it is, it's so true what you just reviewed.

[BREAK]

[00:19:31] Steven Johnson:
I wanna turn now to something we've talked about a lot on this season, uh, of the TED interview, in which you've written about and thought about a lot over the years, right before the pandemic changed everything. You published a terrific book, uh, called Deep Medicine, um, which talks a lot about the, the growing relationship between health, medicine, and artificial intelligence. Um, now this season on the podcast, we, we've talked to Chess Grandmaster Gary Kasparov.

[00:19:58] Eric Topol:
Yeah, oh yeah. Love Gary.

[00:20:01] Steven Johnson:
Um, AI visionary Demis Hassabis, who—

[00:20:03] Eric Topol:
Oh, you talked to Demis too. Oh. Oh, he's a hero of mine.

[00:20:07] Steven Johnson:
Yeah. We talked to Demis. Yeah. So you're, you're a good company. He's a fascinating guy, right?

[00:20:11] Eric Topol:
Yeah.

[00:20:12] Steven Johnson:
And, and you know what's funny about both of them is that they, they both got in, in a way, into artificial intelligence through games, um, and, you know, chess and video games and, and things like that. Um, I, I'm curious, where did, where did your interest in, in artificial intelligence originally come from? When, when there was, was there a turning point when you really realized the significance that it was gonna have for medicine?

[00:20:36] Eric Topol:
What really led me to this was the fact that, um, you know, I've been so much into digital medicine and genomics that we were getting so flooded, overwhelmed by data that we needed something to help with it that, you know, no, no way, we were gonna be able to deal with it. And I think that, um, that's why I started to turn to what could AI do to rescue us. And then I realized, whoa, there's something in the deep learning phase, which really, Demis was one of the main leaders of forces of that, along with Jeff Hinton and a few other people.

But they really set us to have machine eyes. And I could see—with a connect between, you know, looking at images in medicine, which is of course a central part of whether it's x-rays and films or looking at electrocardiograms or retina or whatever—that if you had machine eyes that could see things that humans can't see, this could really change the course of medicine. So I think it was really for me, the realization that, um, we had gotten to a point where all these layers of data were being generated with sensors and our electronic health records and, uh, uh, of course, our genome, uh, microbiome, whatever, that we needed something to be able to process that data. We didn't have it.

And so, you know, some five, six years ago started to turn to AI, started to really talk to people like Gary, uh, Demis, um, and so many others. Fei-Fei Li, and all these other new friendships that I've made over the years to try to get us some grounding of where AI could possibly take us in the years ahead.

[00:22:19] Steven Johnson:
So for years, when, when people have talked about the potential threat of kind of job losses from, from future forms of artificial intelligence and, and particularly kind of high-income job losses, we’re not talking about factory workers. Um, the, the shorthand was always that radiologists are gonna get put out of business.

Like that's, that became kind of a cliche in this, in this world that, you know, we're still gonna have plumbers, but, you know, don't, don't, don't, you know, if you're, if you're a radiologist, you may wanna start, you know, thinking about going back to grad school and something else. Um, and you know, so far that has not been the case.

Um, radiologists seem to be continuing to do their jobs and have not been replaced by algorithms. You've got a fascinating take on that in, in Deep Medicine. Uh, what, what is the future of, kind of, expertise, uh, as these technologies develop?

[00:23:14] Eric Topol:
Yeah, I think this idea that, you know, radiologists or any specialty of medicine is gonna be, uh, made, uh, obsolete or unnecessary is silly. Uh, what's gonna happen is adapting. Uh, so radiologists, you know, largely live in the basement, in the dark. They look at images all day long. But if you talk to them as I have when I've been a patient, they yearn to see patients. They wanna have human interaction. They didn't go into radiologists so they're living in the basement in the dark, you know?

And the other thing is they are acutely aware of how many of these images that are ordered that are totally unnecessary, silly, waste that lead down a rabbit hole, incidentalomas, that sort of thing. So radiologists, uh, would very much like to morph their profession, um, and have patient contact, become a more active participant in how patients are cared for.

And one of the other things that's noteworthy, Steven, is that, you know, when you get a diagnosis, you, you, it's hard to find an honest broker. You know, you have, you talk to the surgeon, you know, “when in doubt cut it out”, right? Yeah. But the radiologist doesn't have any interest in supporting the surgeon.

The radiologist is gonna tell you, “You know, this kidney stone that I see, it's very likely you're gonna pass it. Don't let this, don't let this urologist go near you.” You know, kind of thing. So, um, that's another thing that radiologists, they have wisdom. I mean, they, they have a lot of clinical correlation and, and so, they’re not getting that patient interaction and that independent assessment that patients can benefit from.

[00:24:51] Steven Johnson:
So what would be… Let, let's just imagine the scenario, let's say five years from now, and the progress that we have seen in, in, you know, machine learning, deep learning, um, algorithms is determining whether a particular image on a scan is cancerous or not. Um, let's say that progress continues. So what does this scenario look like? You go in for a scan. Um, that partnership between the radiologist and the machine and the patient. Tell us what that might look like.

[00:25:23] Eric Topol:
Well, for one, uh, they would do a lot of scanning yourself. So, um, you know, one of the biggest things that we had never foreseen was what the retina tells us about the whole body.

When you take a retinal photo, you're getting, uh, a handle on your blood pressure control, your glucose control, your, uh, kidney disease, potential for Alzheimer's, hepatobiliary disease, heart disease risk, a long list of stuff. Right?
[00:25:52] Steven Johnson:
Wow.

[00:25:52] Eric Topol:
And it's, it's actually kind of amazing that, you know, first we, we found out that, hey, you know what? Machine eyes can determine the gender by the retina. You know the retinal experts, they're, they, they get a 50% right? Male or female? The, the, the machine eyes: 97%, right?

[00:26:11] Steven Johnson:
Yeah.

[00:26:11] Eric Topol:
Then, then we learned about all the other parts of the body that it's a window to. So you're gonna be, five years from now, you'll be taking your phone. Take a picture of your retina and getting algorithmic output about what is it showing about, you know, so the idea that you have to check your blood pressure three times a day, every day to find out if your blood pressure’s under control, you'll, you'll see a whole different look on that. Um, but in addition to that, what other thing that really excites me, I'm really into smartphone ultrasound.

So the idea is that, you know, this, you pop a, a transducer, uh, probe of the ultrasound to the base of your smartphone. You can image any part of the body except the brain. When I first did this, I guess I got so excited. I did a whole total body medical selfie. I mean, everything. I did, I did everything for my, my carotid artery sinuses all the way on my left foot. I mean, it was just incredible.

[00:27:06] Steven Johnson:
Yeah, but it doesn't really get you a lot of followers on Instagram though. That's the problem with those selfies. It just, it's not. It's not a winner on social media.

[00:27:13] Eric Topol:
No. No, but you know what? What I, what I see now, which is so extraordinary. So in the hinterlands of Africa and Indian places, you know, you could auto capture the heart, the echocardiogram, beautiful images you don't even know you had them, and then give you an interpretation.

So what AI can do in imaging, we’re just starting to see the earliest, uh, capabilities. And it'll take a while, but one of the things that particularly excites me is the potential to reduce inequities and, uh, you know, this whole idea of doctorless screening because, you know, we can help decompress the, the burden of the medical community with a lot more charge for those people who are willing or need to, to use, um, this screening capability.

[00:28:02] Steven Johnson:
Wow. That's amazing. Well, the other thing that I, I think is so striking about your perspective on AI and, and this actually, you know, in some ways echoes stuff that Gary Kasparov, has written about as well is, well, it's really in the subtitle of, of Deep Medicine, your, your book, which is “How Artificial Intelligence can Make Healthcare Human Again”, right?

That, in a strange way, the rise of these new algorithms, um, creates a possibility for a more direct human-to-human kind of healthcare. Um, explain that paradox to us a little bit.

[00:28:37] Eric Topol:
Yeah, it really is, uh, counterintuitive. How could technology possibly make things more, um, human-centered? Uh, but it really is the truth because right now, we need the gift of time.

Basically, the patient-doctor relationship has deteriorated terribly, uh, and so over decades. And there's no been any way to get it back except this. And that is if you have all the capabilities of AI that are, um, used and validated, like, for example, taking all the data of a person and getting it all ready for the doctor or nurse so they don't have to go through pages and pages of data.

Uh, if you have notes—made synthetic notes—so there's no, uh, looking at the screen, you're, you're actually together now, and you're actually communicating, and you're doing a full exam, and you have time: the biggest factor, a presence. That you're really connected. The bond, the human bond that has been lost, we can reestablish that, that precious relationship, which is based on trust.

It's, you know, the empathy, the, the humanistic qualities that, you know, are not, uh, techno, uh, capable. So I, I think that's the most overarching, exciting potential for AI in medicine. I, I hope we make that the priority because we already have seen how we can get accurate images complementing, that goes back to your earlier question, Steven, about how you don't wanna rely just on the machine, uh, interpretation. You always wanna have a human in the loop, experts, whatever.

But in addition to that, we've gotta get back the time. That's really what has been lost, you know, with the average visit in the United States. You know, so, and it's worse, of course, in other countries. You know, I remember visiting, um, uh, Korea and it's like less than two minutes in a, in most places in Asia. So we just have lost that, that, that relationship. You need time and, uh, AI can get that back. So that's really what I'm excited about most, uh, for AI.

[00:30:40] Steven Johnson:
There's another fascinating concept in, in Deep Medicine, which, which I believe kind of builds on, on your long interest in, in, in personalized medicine and data, which is this idea of, of using software to find what are called digital twins, uh, of patients, and this is a really, uh, fascinating concept. Explain, explain what that actually means.

[00:31:04] Eric Topol:
Yeah. You know, that's something that I, I also think will inevitably occur. It'll take longer. The idea is that if you had all the layers of data for each person, and you had now a data resource of millions of people with their follow-up, then instead of what we rely upon today, which are randomized clinical trials, the best trial might be 1 in 10 people benefit from the treatment, but we treat all the people because we don't know any better. Now, this is much more precise, granular, because now if someone developed cancer, they would have the multiple digital twins who are the closest facsimilies to them at every layer. But those people had treatment and outcomes that worked, and then we would know much better. And so then you start to apply that for, you know, preventing all sorts of other conditions.

[00:31:59] Steven Johnson:
So the idea is that, you know, in, in, in the global population, let's say you've got a specific form of cancer, um, and you have a specific profile, um, you know, Europe, in my case, 54-year-old man, I'm not asking for a diagnosis here, but, just using this example, 54-year-old Caucasian male who has this presumably kind of genetic profile, like we would know something about, like, my actual genome that we would be able to kinda map in some way. And in trying to figure out the, the proper treatment for this particular cancer, the, the most relevant information is someone very similar to me who had this same specific cancer who went through a treatment that worked, like that's the person you wanna find. You don't wanna find just a randomized control trial of 10,000 people who are all very different from me. You'd wanna find someone who's a twin, in a sense, who had this exact same thing. If you could find that person, that needle in a haystack, that would be incredibly relevant information.

[00:32:58] Eric Topol:
Absolutely. You summarized it well. I mean, we already know. There's so many cases where a drug for, uh, leukemia, the one that worked in a person was a kidney cancer drug that no one would ever have predicted. But instead of these random things, now we actually can match you at, you know, every layer, uh, that makes you, that makes you unique. We find people that are very, very proximal, you know, as close as we can to you with their treatments and outcomes, and I think this is exciting. It isn't being pursued much yet. It's starting to happen in cancer, but eventually it'll go across the board.

[00:33:36] Steven Johnson:
And the idea is that these people, their medical histories are available, but we don't know who they are. There's nothing about their actual name or identity that can be tracked in some way. Right? That’s, that there, there's a privacy protection built in theoretically to this kind of approach.

[00:33:52] Eric Topol:
Yeah. You're bringing up, I think one of the bright lights of AI. There's a few, but one of them is privacy computing. And so, uh, this idea of federated learning, uh, edge computing. You don't have to ever know the identity. This can all be done, um, through AI—new, um, dimensions of privacy computing. So that's another thing is that while today this is a, a sense of a lot of worry, insecurity about the privacy of data, we're gonna do much better on that in the years ahead.

[00:34:25] Steven Johnson:
There’s a, there’s a very powerful chapter in, in Deep Medicine that also has a personal story about your father-in-law, um, that I was really interested in. It was something I'd never really thought about, which is using the kind of the pattern recognition capabilities of machine learning to do something that humans right now aren't actually all that good at in general, which is predicting the, the timing or the overall likelihood of death for terminally ill patients.

Um, and so it's kind of a disturbing idea on some level that you're, you're turning to the algorithm to find out when your loved one might die and making plans based on that. Um, but there is some evidence that the machines can already do that better than, than we can, and that could have a lot of benefits.

[00:35:11] Eric Topol:
Yeah. I think this is, uh, another misperception and that is that, you know, when you talk to a doctor about a very serious illness, uh, and with looking at things objectively with all the layers of data, that's one thing that we can get a better sense of. It, it’s, it's never gonna be perfect, but you can certainly do beyond, uh, experienced physician judgment.

Um, and then the two together, of course, can, like in everything in medicine, can compliment the models, the, the data processing aspects and then, you know, the experience, and the, the wisdom from a given clinician. But, uh, in the case of my father-in-law, you know, it was, uh, when it was like a Lazarus story, uh, you know, he was basically being sent to a hospice to die at our house, and then, you know, just woke up and, and live for a stretch longer.

Uh, and that was completely not predicted by any of the many physicians on his team that were experts in this domain. Uh, so yeah, I, I think we'll see more of that. Uh, there will be questions, uh, about its misuse. Uh, insurance companies. I mean, there's always nefarious aspects of all this stuff. So we have to be leery about how data and models are used and biases and that kind of stuff, because that's some of the downsides of AI that we have to never, um, you know, let up on our guard on.

[00:36:37] Steven Johnson:
It makes such a difference though, even the small, little increments of time, right? You know, knowing that it's likely you actually have another six months when your doctor is saying, “It's time to go to hospice right now.” I mean, that is everything to, to, or, or the reverse of that. And knowing that you, that you're likely to die very soon, and it's time to say your goodbyes. Um, you know that, that, that if there's a way to get better at making those kinds of predictions. Um, which is, you know, that is… Predicting the future is one of the core attributes of, of human intelligence. When we do it well, it, it makes a huge difference. And if we can do it with something as important as, you know, when a loved one is likely to die, that seems, that seems like an important advance.

[00:37:20] Eric Topol:
Well, and you know, as we, you're talking about the digital twins, just think about, you know, decades from now when you would know a lot more about your real risk from your, from your twins. And, uh, you know, so the precision is just gonna keep getting better over time.

[00:37:38] Steven Johnson:
You’re at, at Scripps and UCSD, um, down there in San Diego. There has been a lot of interesting research recently on longevity, radical longevity. This idea of kind of radical life extension. Um, are you involved in that world at all, or is, is, is that something, um, that is gonna become more mainstream? Thinking about actually pushing past that boundary of 105, 110, where humans seem to not be able to live past, how, how interested are you in that, in that world?

[00:38:10] Eric Topol:
Oh, I'm very interested because I'm a lot older than you.

[00:38:13] Steven Johnson:
Well, not that much, but yeah.

[00:38:15] Eric Topol:
Yeah, no, so, uh, you know, it's, it's scientifically, it's fascinating. Um, I think there's been, you know, quite a bit of hyperbole about our chances of changing human lifespan, particularly health span, you know, ‘cause if you keep somebody alive to their 130 and they, the last 20 years, they're demented, that doesn't really help anybody. Um, so I, I think there's lots of things that are happening in this space that are very provocative, um, that makes you think that maybe we will prolong health span.

What we did was we did a big genomic study of we called the welderly, where we did whole genome sequencing of hundreds of people that live past 85 and never were sick to find that if there's something special about their genes. We found a few things, but largely we can't yet account for the biologic, uh, endowed features that, that separate these, these, um, special people that just seem to be able to withstand any illness, even though they're, if you looked at their prediction of heart disease and, and, uh, cancer and all the others, you would've guessed they'd be just like everybody else.

So we, I've studied it. Uh, I think you're, I think right now you're seeing more intense activity about this than ever before. But I wouldn't say we have found any Fountain of Youth or any treatment. You know, there's some that have been advocated, but I'm still quite suspect.

[00:39:42] Steven Johnson:
There seems to be a, I've read a little bit about, there seems to be some connection between social connectivity and, yeah, and lifespan as well. My, my grandmother lived to 104. And she was just one of the most social people I've ever met. She just, you know, uh, to her dying day, she was connecting with people and had a, you know, a, a, a thick set of connections with family and, and close friends around her. And I, I think that was, that was part of her story for sure.

[00:40:06] Eric Topol:
Yeah. You know, I think that is something we have seen that theme of people who are more connected and also more upbeat, you know, more, uh, I think those are qualities that were quite thematic in this… A couple thousand people that were in this special category. So you're obviously right. Doesn't mean that if you're, uh, a loner and that you are, you know, doom and gloom that you know you can't be, but the frequency absolutely.

[00:40:32] Steven Johnson:
One last just general question about machine learning and AI and deep learning. Um, it’s… As I mentioned, it's been about three years since Deep Medicine came out. It has been a very active period in the history of artificial intelligence. Um, you know, the… Famous for having this AI winter for many decades. We, we seem to be in AI summer right now. What's been the, the kind of the biggest surprise for you, the most interesting development since Deep Medicine came out in that field?

[00:41:01] Eric Topol:
Well, what was clear, you know, back a few years ago, um, uh, that you could train, uh, AI to see things that humans couldn’t, has been amped up, you know, to the, to a whole nother level.

Uh, and so I think what we haven't yet gone through from the image category, uh, is just extraordinary. It keeps building, you know, like, almost every day. The, the ability to take that to speech is really starting to take hold now. I've seen synthetic notes from a conversation with patients that make the, our typical notes look like a joke.

Uh, and the idea of keyboard liberation, to finally get rid of this data clerk function of clinicians, which everyone hates uniformly. And it also takes up a lot of off hours, no less from the actual patient, uh, interaction. So that part gets, is getting me excited. And then the, the other frontier, of course, is we are dealing with text, unstructured text, which we're still relatively early in, but you know, we're sweeping across these different forms of data input.

And the other thing that has taken off in the last few years, which I wouldn't have predicted this fast, would be getting, uh, around the need for massive annotated data sets, but using much more self-supervised learning with smaller input data sets. So, you know, I think we're on an accelerated phase.

I agree. This is AI summer, uh, it's still in the medical sphere. It's still very early. Um, and we have a shortage of talent because, you know, the, the, the real great people in this field, uh, although, you know, Demis is one who understands life science well and the medical potential, but most are working in other areas, uh, that are not related to medicine, and it's hard to attract them, particularly in the academic circles, to stay, uh, in our world.

[00:43:01] Steven Johnson:
Maybe all the math majors and, and grad students who were, were mining crypto for the last three or four years will finally start taking all that intellectual superpower and applying it to medical issues. One can hope.

I wanted to ask you one last question here. When you think about the, the outstanding kind of mysteries that are out there that haven't been solved yet, in your, you know, the broad scope of your research, what's the one you are most eager to see a solution for? Like, what's the one you'd really like to fast forward and, and get the answer to?

[00:43:34] Eric Topol:
Yeah, I've been working on that for, well, I dunno, 15 plus years. And I know it's gonna take a lot more than another 15 years, uh, to get to the answer, but it's to find what makes each of us unique. To… It isn't just our genome. Okay?
Uh, you have twins with the exact same genome DNA sequence, and they couldn't be more different in many other respects. Right? So it's understanding uniqueness, individuality at every layer, and then multimodal to pull it all together. That's to me, um, that the, I, I've been on the hunt. It's been a long course.

Uh, I like big challenges, so that'll help keep me going for quite a while. But once we crack that at, at, at a significant level, I do think we can deliver better medicine, better prevention, which up until now has been like a fantasy. It gets us back to the, you know, digital twin concept that we spoke about. So that's what I'm after.

I think there was unfortunately too much, um, of this sense when the genome was cracked back in 2000. The first draft, you know, we got this when in fact, uh, that was just one layer in the beginning. And it's a long arc to get where we need to go.

[00:44:53] Steven Johnson:
Well, it's a worthy question. If it takes you 15 years to get you, we will, we will have you back on the podcast 15 years from now. 2037 if I have the math right. And you can tell us all about it, but we'll also have you back before that. Um, Eric Topol, thank you so much for joining us today on the TED Interview.

[00:45:10] Eric Topol:
Uh, what a joy, Steven. Thanks for having me.

[00:45:16] Steven Johnson:
That's it for the show today. The TED interview is part of the TED Audio Collective. This episode was produced by Kiara Powell and mixed by Erica Huang. Sammy Case is our story editor. Fact-checking by Meerie Jesuthasan. Farrah Desgranges is our project manager, Constanza Gallardo is our managing producer, and Gretta Cohn is our executive producer.

Special thanks to Dan O'Donnell, Michelle Quint, and Anna Phelan. I'm your host: Steven Johnson. For more information on my other projects, including my latest book Extra Life, you can follow me on Twitter @stevenbjohnson or sign up for my Substack newsletter: Adjacent Possible.