Eric Topol on Why Doctors Shouldn’t Fear AI — And How It Could Finally Fix Our Broken System

Key Podcast Moments

• Eric Topol explains why holding AI to a “zero-error” standard ignores the very real harm caused by human diagnostic error.
• Graham and Eric unpack why medicine remains stuck in treatment instead of prevention and how AI may finally change that.
• A candid critique of longevity hype, including where popular figures and protocols drift away from evidence.
• Why multimodal AI could allow physicians to predict disease years earlier and intervene before damage is done.
• A frank discussion on deskilling, medical training, and how physicians must learn to work with AI, not against it.

AI has arrived in medicine, but unlike past waves of technology this one is forcing physicians to confront something deeper than workflow efficiency. At the center of that reckoning is Dr. Eric Topol. A cardiologist and founder of the Scripps Research Translational Institute, Eric has spent decades challenging medical dogma, calling out pseudoscience, and insisting that progress be grounded in evidence.

Offcall co-founder Dr. Graham Walker sits down with Eric to discuss why doctors shouldn’t fear AI and why, paradoxically, the reluctance to adopt it may already be harming patients.

Rather than framing AI as a replacement for physicians, Eric makes a different case: medicine has always been limited by what humans can perceive, process, and remember. Diagnostic error remains one of the leading causes of preventable harm, yet clinicians are often expected to synthesize thousands of data points across fragmented systems, under time pressure, with incomplete information. AI, when used correctly, doesn’t remove judgment but instead it augments it.

Throughout the conversation, Graham and Eric explore how medicine became trapped in a late-stage treatment model, intervening only after disease has already taken hold. Randomized trials, while foundational, often obscure individual variation. Entire populations are treated based on averages, while most patients in those trials may see little benefit. AI offers a path out of that limitation. Not by replacing evidence, but by layering it with personalization.

Eric points to advances in multimodal AI that can integrate imaging, labs, genomics, immune markers, environmental exposure, and longitudinal health records to identify risk far earlier than clinicians ever could. The promise isn’t immortality or reversing aging, it’s preventing the diseases we’ve mistakenly accepted as inevitable consequences of getting older.

The episode also pulls no punches on hype. From over-promised longevity protocols to under-tested interventions marketed directly to consumers, Eric is clear: evidence still matters. GLP-1s, immune modulation, and AI-driven prediction are exciting precisely because they are being studied rigorously and not because they fit a narrative.

This episode isn’t about hype cycles or fear-mongering. It’s about realism. AI will make mistakes. So do doctors. The question isn’t whether AI is perfect, but whether it can help medicine become safer, more humane, and more preventive than it is today.

For Eric, the answer is clear: if physicians engage critically and lead thoughtfully, AI may help medicine finally do what it has always promised - keep people well, not just treat them when they’re already sick.

Top 4 Takeaways

1. AI isn’t the threat, unrealistic expectations are

Medicine often demands perfection from AI while quietly accepting widespread human error as inevitable. Topol argues this double standard has slowed adoption of tools that could meaningfully reduce diagnostic mistakes, even as diagnostic error remains a leading cause of preventable harm. AI will make errors — but so do clinicians, at scale, under impossible cognitive loads. The real question isn’t whether AI is flawless, but whether it can help reduce harm when paired with human oversight. Waiting for perfection risks missing an opportunity to save lives right now.

2. Prevention has failed because humans can’t do it alone

Primary prevention has always been medicine’s aspiration, but it has never been operationalized at scale. Clinicians simply don’t have the bandwidth to synthesize lifelong records, subtle lab shifts, imaging signals, immune markers, and environmental risk in real time. AI changes that equation by enabling individualized risk prediction years before disease manifests. For the first time, medicine can move upstream — intervening before heart failure, dementia, cancer, or diabetes become irreversible. This shift may ultimately matter more than any single therapeutic breakthrough.

3. Evidence still matters, especially in the age of hype

As interest in longevity, immune modulation, and AI explodes, Topol is clear-eyed about the danger of enthusiasm outrunning data. He calls out interventions that have gone mainstream without adequate evidence while pointing to others — like GLP-1s and immune-targeted therapies — that are supported by growing, rigorous research. AI’s credibility in medicine will depend on the same standards clinicians apply to drugs and devices: validation, transparency, and real-world outcomes. Without that discipline, AI risks becoming just another cycle of broken promises.

4. Physicians must lead AI adoption or lose agency

If doctors don’t shape how AI is deployed, validated, and taught, it will be dictated by health systems, vendors, and policymakers who may not understand clinical nuance. Topol emphasizes that training matters — clinicians need to learn not just that AI exists, but how to work with it, question it, and understand its limitations. The future physician isn’t replaced by AI, but amplified by it — with more time, better information, and stronger judgment. Leadership from within the profession will determine whether AI restores autonomy or further erodes it.

Where to Find Eric Topol’s Work

• Eric Topol’s Substack: Ground Truths — evidence-based analysis of medicine, AI, and science
• Books:
  • Deep Medicine — AI, empathy, and the future of healthcare
  • Super Agers — evidence-based insights on healthy aging and prevention
• Scripps Research Translational Institute: The Scripps Research Translational Institute, formerly named Scripps Translational Science Institute (STSI), was founded in 2007 with one essential aim—to individualize healthcare by leveraging the remarkable progress being made in human genomics and combining it with the power of wireless digital technologies.
• Peer-reviewed research and commentary

Transcript

Eric Topol:
Back when I wrote Deep Medicine, I thought our biggest thing in AI was going to be to relieve the data clerk burden of doctors, give patients more autonomy, get us the gift of time, and bring humanity back to medicine. I think that's starting to happen. But I think what's even bigger, to me, much bigger, is that we have this newfound capacity to prevent these major age-related diseases. So instead of all this work that's being done to reverse aging, let's just accept aging, but don't accept that these diseases are obligatory part of aging because they don't have to be.

Graham Walker:
Welcome to How I Doctor, where we're bringing joy back to medicine. Today, I'm joined by Dr. Eric Topol, cardiologist, founder of Scripps Research Translational Institute, and one of the most cited medical researchers in the world. He's the author of some of my favorite books, including 2019's Deep Medicine, talking about AI, and the brand new bestseller, Super Agers. I think a lot of Dr. Topol's research has directly impacted my own specialty of emergency medicine, TPA for STEMIs, the TIMI trials, EMS acquisition of EKGs, handheld ultrasound. But most of our listeners probably already know Dr. Topol because we skew toward a tech-enabled future of medicine, "How should I be using AI clinicians?"
Here's the real reason I wanted Eric on the podcast since we launched a year ago. More than anyone else I've interviewed, Mark Cuban, Shiv Rao, Dr. Glaucomflecken, I want to follow in Eric Topol's footsteps. He is the voice and the pulse of high-quality evidence in medicine and I think the future of our profession. He has stayed true to his mission throughout his career, and I've tried to do the same with my career, with MDCalc, with the NNT, and now with Offcall, supporting big ideas and important conversations about where medicine is going. I don't just want to know what AI can do. I want to think about how it should do things.
So, today, I want to know what Dr. Topol thinks of all of this. Even if he may not have all the answers, he might help us figure out how to think about them clearly. Dr. Eric Topol, welcome to How I Doctor.

ET:
Well, Graham, thank you. That's an awfully kind introduction.

Balancing Scientific Progress with Public Skepticism

GW:
You write a lot about science, but I kind of want to ask you how you feel about this moment that we're in in late 2025. We have incredible medicines and technologies and AI, but the trust in, I think, our medical system feels worse than it's felt certainly since I started med school in 2003. How do you think about the optimism from the science side with the public's skepticism?

ET:
Yeah, you're bringing up such a critical issue. When I got my first job with University of Michigan in cardiology, that was '85, so 40 years ago. During these 40 years, we were doing pretty well, and then now we've hit bottom. What you're bringing out here is that there's this dichotomy of the most impressive technology opportunities, the tools that we have to take medicine to plateaus at higher levels than we ever would've envisioned. At the same time, we have anti-science, anti everything, so we have made it much more difficult to make the advances that we should be easily jumping ahead. It's unsettling. It's vexing. I'm just hoping, Graham, that it's just a short-term story and that the work that we're all doing to advance the field will be something that's unstoppable, just needs a continued push and we'll get back on track eventually.

GW:
Eric, do you see it as physicians or the system kind of needing to communicate better, or is it really the proof is in the pudding that we really are going to see some of these transformational breakthroughs, people are going to see it for themselves, and they're going to have to believe it?

ET:
Yeah, I think it's both. We're going to see more and more compelling evidence that there's a new way to practice medicine and a new way to achieve accuracy and humanistic care and all sorts of things that AI and other things that we have in the offing now can make a difference. But on the other hand, as physicians and as scientists, we don't do anything to neutralize the toxicity, conspiracy theories. A very small percent of us, perhaps 1%, are active on social media platforms, trying to get the right word out.

Evaluating Medical Breakthroughs: Separating Hype from Reality

GW:
I have to imagine that you are constantly sent papers and pitches and slide decks about the latest breakthroughs, and some of it certainly is real and some of it is probably snake oil. How do you decide what is hype and what is, "Hey, I should actually be paying attention to this"? Your newsletter is always coming out with like, "Hey, there's a new technology that may help us in a year or 5 or 10 years," or something like that. How do you personally decide what is worth you digging into and evaluating?

ET:
Well, because I'm old and I have this 40-year span of evaluating things, I've always been trying to think ahead and trying to get some kind of vision for how medicine will evolve as we go forward. For example, when I saw, back in 2023, the first paper on these protein organ clocks, which we've never had an ability to judge the pace of aging of our organs in the body and our immune system, that really got me excited because it was a whole new look, or when I saw data on a p-tau217, a inflammation marker in the blood that tells us many, many years in advance about whether a person's really at risk for Alzheimer's and that we can modify it like an LDL cholesterol with interventions. These are things that just grab me. I just say, "Oh, wow." I read a lot every day and week, and I kind of have a feeling for what, at least to me, seems like something that represents something that could be impactful.

Challenging Medical Dogma: Rethinking Heart Disease and Inflammation

GW:
Physicians, we are so accultured into the way that we are trained to do things. We never want to be the first person to use a medicine. We never want to be the last either. How do you decide like, "Yeah, maybe we are thinking about this wrong. Maybe more heart attacks are due to inflammation and aren't necessarily just like cholesterol plaque that is built up and ruptured and thrombose"? What makes you be able to challenge your own kind of traditional medical thinking?

ET:
Yeah. That's one of the things that I try to teach and emphasize to all of our young trainees and students and even faculty, in that, "Just keep challenging what is dogma. Our sacred cows in medicine are plentiful, and that often can be shot down." A perfect one is the example you just provided. For all these years as a cardiologist, all we did was focus on narrowings and the so called, that a name I made, the oculostenotic reflex where you see a narrowing and then you put a stent in-

GW:
You got to open it, yeah.

ET:
... do a bypass. It turns out, that's just a very small part of the story that we have been fixated upon. We still have a lot of heart attacks, and most of them are not from critical narrowings, but rather that they have inflamed arteries and that we're not paying attention to that. They're not obstructive, and we can now pick them up. Certainly, in people who are suspicious because they're high risk, we can do a coronary CT angiogram and do AI of the epicardial fat around the vessel and say, "Oh, wow, this one's inflamed even though there's no significant narrowing. Oh, wow, this patient has three arteries inflamed, so we really got to get all over this." This is a whole new path in cardiology that we've ignored for decades, the high-risk patient and the vulnerable atherosclerotic non-obstructive plaque. But that type of questioning dogma is something that we should be doing every day because it's too many things we accept without adequate evidence.

GW:
I am curious how much you think stuff that we've either... whether it's like inflammatory component of STEMIs or stuff that we have always just called idiopathic is actually the immune system or it's inflammation, and we just fundamentally... We're calling it idiopathic because we just don't know. That feels to me like it's a trend. Do you feel like that's going to get unearthed as well?

ET:
I do think that we can do much better to define these. Back in 2000, as you recall, in the early 2000s, we thought that the DNA genome sequence was going to unlock everything. That was our operating instructions. The problem is you can study the six billion nucleotides in a genome and you won't find out about the immune system, except in rare mutations associated with very rare diseases of immune system. You have to do perturbations. You have to do special immune system assessments. We don't have that clinically. You're in the ER and you were thinking, "Well, this patient's immune system, maybe that's an explanation for what's going on here." There is no test that you can do. It's a joke.
Here it is. We're basically in 2026 now, and the only thing we have is a white cell neutrophil-to-lymphocyte ratio, which is a joke. Why don't we have ability to look at T and B cell function, NK, dendritic cells, antibodies, autoantibodies, T and B cell repertoire sequencing, interferons? They should be available, and they're not. This is a real problem, the immune system is damn important, but we can't assess it in patients, except in rare academic labs that they can do a comprehensive job. That's a big gaping hole in medicine today.

Viral Footprints and Long-Term Health Impacts

GW:
Well, it was interesting in the book that you got access to all the coolest researchers that were running an immune panel on you to say, "Oh, yeah, Eric Topol has had rhinovirus 15 times, and you've had coxsackie-2 three times or whatever." It was interesting that those markers are stuck in your body permanently, that those can be detected, clearly opened the door to this idea that these viruses are making their impact on us and leaving their footprints permanently.

ET:
And they can be revived and-

GW:
Yeah, reactivated.

ET:
... reactivated, and that can lead to a whole spectrum of outcomes. I mean, I was talking to Michael Snyder this morning, who runs the Stanford system immunology lab that I visited a couple of times and did all these tests on me. What was interesting was I got the respiratory panel, which was supposed to be covered, I guess, through my insurance, but no, it wasn't. It would go to $2,500. So this is a whole thing that they pulled out all the stops to show me all my old exposure to coronaviruses and rhinoviruses and-

GW:
Yeah, your insurance doesn't cover experimental.

ET:
Right. They show that, but, I mean, I do think that we're learning that, particularly COVID, that this virus hangs around. That's part of what long COVID is, is that there's persistent viral antigens being detected. It's not just immune dysregulation, as some people would claim exclusively, but there is residua of the virus hanging around, and, again, the body can't clear it in these susceptible individuals.

GW:
I think that the p-tau217, I think I had heard that study where they did a, I guess, IM injection of p-tau217, and the one group got one, the other got another one and-

ET:
That was for amyloid imaging.

GW:
Amyloid imaging, that's right. And that they were able to reduce the amyloid plaque burden in these patients.

ET:
Lecanemab and donanemab.

GW:
Gotcha. But ultimately, the endpoint is... and I think what these findings have always shown me is that the brain is the most resilient. They're like, "Oh, well, these patients are able to return to school, return to their normal lives," but they didn't... their cognition didn't improve at all. It was just like, "Well, we stopped the Alzheimer's at stage one or stage two. You haven't gotten to five or six yet where you actually can't live independently anymore."

ET:
Yeah. It's very disappointing, and these drugs are overpriced, hard to take. It means multiple, many, many trips, frequent trips to the doctor's office to take an IV injection over an hour. So they've not been, as you say, the great thing. They reduce cognitive decline by 30%, which is impressive except that the baseline was you can give anybody without brain health issue those same cognitive tests, and they do worse over time because they don't know how to do these tests. It's just an effect of getting better at answering questions that have no real relevance to real-world function. That's where this p-tau217 or any of the p-tau markers... it used to be all these p-taus could only be measured in the brain through amyloid imaging like a PET scan, which is super expensive and not practical.
Now, we have many different blood tests that can be done, p-tau217 is one of them. That tells us, like I say, many years in advance, and the p-tau marker is driven by amyloid accumulation in the brain. So when that starts to go up, that's when you need to go after it. Right now, we're just giving the drugs too late, but we're seeing now trials giving these drugs earlier, like with people that have risk by a p-tau, but they don't have objective cognitive decline as opposed to just subjective. Now, we're in prevention mode, and I think that's when we're going to start to see a real difference.

GW:
Well, that's cool.

Personalized Medicine: Beyond Traditional Clinical Trials

ET:
The other piece of this, which was exciting to me, was that a woman, who I put in the book, Claudia Kawas and her colleague Maria Corrada from UC Irvine, they took on the oldest people without dementia, and average age was 102, and they got a lot of them. They got over 2,000 individuals that had been followed for more than 20, 25 years now without dementia. They did brain imaging and they did the p-tau blood tests and so on, and what they found was that more than 40% had amyloid and tau accumulation in the brain by imaging. By brain imaging, they're supposed to have Alzheimer's except they didn't. They were resistant. That's superagers by many metrics. They also found that there's a subset of individuals who do have high p-tau and the amyloid yet, somehow, they have the ability, they have resilience that we don't have. But now, we can use these biomarkers as a way to say, "Well, I'm a brain superager. I've got this resilience, so I can treat these things aggressively."

GW:
I wonder if that's genetic or if it's something... like you write about Kawhi Leonard, you write about LeBron James. It feels to me there's probably... I guess if you're 102 years old, you're probably pretty good about all the other things to have gotten to that point.

ET:
Yeah. They're healthy otherwise, they're independent, they're living at home. It's just such an inspiring group. When I got to spend time with them, met them at a follow-up visit in Laguna Woods, it was just really impressive.

GW:
I think that, Eric, probably my favorite idea of the book was around the idea of at least we're starting to understand what are they markers of long-term decline. It's like LDL was in 1972. I remember hearing stories my dad's first day of primary care. It was like, "All right. So it looks like cholesterol is probably bad and it might be causal for heart attacks, but we don't really know yet." Now, we know. That's what that section really opened my eyes to is like, "Hey, p-tau217, oh, yeah. That's probably going to be something that we understand in the next 50 years." What is the marker for frailty? What is the marker for sarcopenia, all these things?

ET:
Yeah, that was one of the main messages of the book, that, yeah, we are getting the biomarkers. Frailty actually is already quantified through accelerometry, these wearable sensors that were used in the UK Biobank with half a million people, these accelerometers that put on the wrist. These are not your typical consumer-grade thing, but they're much more sophisticated. They've been used over the last six years in the UK Biobank to be able to predict frailty. I think just knowing these things are the future of medicine, being able to know the pace of aging of organs and your immune system and the individual markers like p-tau for the brain, homocysteine, Lp(a), your inflammation markers for your heart, your kidney markers, all of these things so we can monitor individually when you're at risk for a disease that typically happens with aging but doesn't have to. We can head it off at the pass, and that's where I think the preventive power is going to come from. I'm excited about that.

Precision Medicine vs. Randomized Clinical Trials

GW:
But if that's truly where we're going, how do we reconcile that? We've learned everything we know from RCTs. RCTs are what I built my career around. It's MDCalc, it's the NNT. We take these populations and say, "Hey, we treat a thousand of these people or we treat 400 with the calcium channel blocker, 400 with the diuretic, 400 with the beta blocker for these people who have that exact diagnosis." You're describing precision medicine ultimately where I know that, hey, I've got my p-tau217, which is low, but it's starting to rise. My wife's is declining. Mine's rising. That's weird. How do we learn those lessons in... I think about learning that, and it sounds like N-of-1 studies, but does that scale? Do we need millions and millions of people to say, "Hey, this therapy works for people with that"? How does that all get untangled to where we get to make the best clinical decisions for each individual patient?

ET:
Well, it's a great question because you can, as an individual, order these tests and know that you have a high p-tau and you have a high coronary calcium scan or something that's a risk factor for a disease. But as you point out, there hasn't been trials on individuals or subsets necessarily that have that risk. That's a problem today. It is N-of-1 where you'd go ahead. So what I'm saying is you have these risk factors, you got to do what you can to mitigate it. But in the long run, we're going to do trials with... and some are already launched of people with biomarker risk, not clinical, but biomarker risk. For example, for Alzheimer's, I told you there's people signing up now that have high p-tau but no overt cognitive decline, and that's prevention. We're going to have trials like that in the future for all these things. When you start to see a person's got atrophy in their musculature by CT or MRI or ultrasound, it's easy to tell, but you've got to do it. We can then monitor sarcopenia, and when that's starting to show it, then we would use perhaps a GLP-1 drug because that happens to promote, as we now have learned, muscle mass and lean body mass.
We have other ways to learn. Someday, we're going to have digital twins. We're going to have a billion people around the world with their stack of data and we're going to say, "Graham, we found your twins. We found your nearest neighbors that are just like you. They're replicas of you, and this is what happened in their life, and this is what helped them, this is what hurt them." These are new ways that only became possible with stacks of data for each person.

Medical Intuition and Clinical Gestalt in the AI Era

GW:
This is the biggest question that I have, my biggest worry, and I honestly don't know the answer to that, so I'm hopeful you have a way to at least think about this. I think, in Deep Medicine, you wrote, "The machine will see things humans will never see."

ET:
Yes.

GW:
In the ER, I rely on my gestalt all the time. It's not just the vital signs, it's not just the way the patient eyeballs... but the general sense of like, "I think there might be something really bad going on here." How does that work with today's med students, today's pre-med students as they start to go through their medical training, if the AI starts to give them the answer immediately?

ET:
Yeah, you're bringing up a really important concept. In The New Yorker today, there was an article, Are We Getting Stupider?

GW:
Sure, yes.

ET:
And also, de-skilling or non-skilling. That ability to assimilate a person's story and data, that's the essentiality of medicine, but I think it can be improved, augmented with help of AI so you don't miss anything because we are not good at pulling in all the data. It's just too many clicks on too many pages. How do you go out to other providers' data and look at every lab data point? That's not our power. Our power though is conveying, communicating with patients and that wisdom and judgment and oversight. Ultimately, they will complement.
But I am concerned about young people getting de-skilled or unskilled in that way, yes, and there's even evidence of that among gastroenterologists already with this... at least one study. This is something we've got to keep an eye on. It's part of the threat of AI in general, not just in medicine, of course.

The Future of Medicine: Prevention Over Treatment

GW:
Sure, yeah. And then, Eric, my last question, are doctors cooked? Are we hosed in the long term in 20, 30, 50 years, maybe right around the time I retire? What is the role of the human physician? Will AI make us better, or will it make us an obsolete profession?

ET:
Well, I'm very optimistic, of course, and I do think it's going to make this profession extraordinary. They're going to have much more accuracy. We're going to have much more of a humanistic way and rebuild the patient-doctor relationship that's suffered. I do think that our prevention, to me, the biggest thing, is that we will go into primary prevention, which we've never done in any real and meaningful way. We couldn't do it. I think we have the ticket to do that. I see medicine going forward that clinicians and doctors should not be threatened. Eventually, we're going to be loving this. But of course, we in medicine have been completely onto the treatment side, treat, treat, treat, and our treatments are generally not that great.

GW:
Yeah, fair point.

ET:
They're also very expensive, they have side effects. And they're often too late, like cancer. Treat Alzheimer's when you have Alzheimer's, or Parkinson's, no, no. Treat somebody with heart failure or heart attacks, no. We have to shift to prevent, prevent, prevent, and we cannot do that with our current human limitations as clinicians. We need help. We have the help that's emerging now, and so that's why I'm so excited about the future.

Embracing AI as a Clinical Asset

GW:
The more I use AI, the more I feel the same way you do. The more I understand these tools, the less scared of them I am and the more I understand where my place will be to continue to help patients as well.

ET:
Yeah, what a great asset to have to use. Patients are afraid of AI, many of them, because they don't really understand where it comes into play. There's a backlash right now because of the fears, children, teens killing themselves from chatbots that are run amok. There's some definite scary things going on right now. I get that, but it's still in the early innings here, so we have to be patient. Methodically, hardcore evidence, just plod through this and we will get there.

GW:
Well, Eric Topol, thank you so much for joining me today. Really, it was an absolute honor to have you on the podcast today.

ET:
Well, thank you. You'd be asked some of the best questions ever for me, and I've really enjoyed our conversation. Keep up the great work you're doing.

GW:
Likewise, thanks.
Thanks for joining me today. For interviews with physicians creating meaningful change, check out offcall.com/podcast. You can find How I Doctor on Apple, Spotify, or wherever you listen to podcasts. We'll have new episodes weekly. This has been and continues to be Dr. Graham Walker. Stay well, stay inspired, and practice with purpose.

To make sure you don’t miss an episode of How I Doctor, subscribe to the show wherever you listen to podcasts. You can also read the full transcript of the episode below.

Offcall exists to help restore balance in medicine and improve the wealth and wellbeing of physicians. Sign up for Offcall here to bring about more physician compensation transparency and join our physician movement.