MATT LUNGREN: Well, welcome everyone,

to another episode of AI and Healthcare

series sponsored by Stanford Online,

and we are just absolutely thrilled

to have our guest today, Dr. Eric Topol.

He really needs no introduction, but you may recognize him.

He's the founder of Scripps Research Translational Institute

and Executive Vice President at Scripps Research.

He's a major voice in AI, genomics and precision medicine,

and a prolific author of deep medicine, The Patient Will

See You Now.

And, of course, his new book, Super

Agers, talking about longevity.

And we're going to talk a little bit about that today.

He's also just generally an awesome colleague and friend

and an advocate for tech-enabled health, I think, has done a lot,

at least for me in my career, I'm sure, Justin, you agree.

Inspiring us to really lean in and try

to spread the word about the things

that we know are happening in technology

and how they may impact health.

So welcome, Eric.

Thanks for being here.

ERIC TOPOL: Oh, it's great to be with you, Matt and Justin.

I know we'll have a fun conversation.

JUSTIN NORDEN: Well, one of the first things I was curious about

was, how do you do it all?

How do you get all these pieces done?

How do you write all that you do, speak all that you do?

And the context being there's this New Yorker article that

came up recently of are you just cheating your way

through college using AI from a recent headline.

And we can talk about some of the educational pieces.

But how do you do it all?

And we can talk about the education piece in a second.

ERIC TOPOL: Well, it's a combination of two things.

One is being an info war.

I'm just trying to ingest lots of information and then

the other being a lunatic.

So, yeah, I mean, I don't have any secret.

You can ask my wife of 46 years.

She'll tell you that I'm a lunatic.

So yeah.

MATT LUNGREN: Well, I mean, but this is like, have you--

I mean, you've been this for a long time.

And again, I feel like I've been following your work

and Just and I, we try to bridge--

we try to stay up on healthcare and AI obviously

in our day jobs, but it's also I would say our passions as well.

When we see this technology coming,

I keep running into evidence, what Justin was just showing.

Hey, this is really, really good.

This technology is really, really good

at tutoring you on a subject, your education, your grade

level, whatever it is.

I don't hear about folks using it for that much.

And I have to admit, I've just started

to try to have a conversation with the model about maybe

a new concept or new paper.

I wasn't doing that as much, and I

feel like I'm missing out on some

of the capabilities of the models

to up level my own skill set and try

to get my arms around all the new things that keep coming.

I don't know how you have approached

this technology in your own personal work life,

but it seems like a lot of folks aren't really

using it for that purpose.

ERIC TOPOL: Yeah, I don't think people generally appreciate,

and even I'm sure a lot of your listeners

how extraordinary capabilities are,

and they just keep getting better,

like on a weekly, whatever basis.

It just you want to do a--

generate a report on a topic, it could take weeks.

And here you have it in a matter of minutes and you want to--

in the case of healthy aging and promoting the Super Ager story,

the data across so many different layers

at an individual level that no human would ever

be able to assimilate.

But here you have models that will do that.

So the capabilities are just so powerful,

and I just keep getting blown away.

I mean, first, we all had the ChatGPT moment

in November of 22.

Some months ago I started using NotebookLM

and saw podcasts being generated from up to 50 PDFs.

I mean, what?

The kinds of things you can do now are mind blowing.

And I don't think most people realize that because they just

haven't been testing it and more-- there's more awareness

of confabulations and errors and propagation of biases,

all sorts of things like that, but not

enough awareness of the power.

I mean, it's just beyond comprehension.

People talk about have we achieved

artificial superintelligence or AGI.

And I recently talked with Tyler Cowen who said, look,

this is it.

I know porn when I've seen it and this is it.

And I mean, I don't know.

It doesn't really matter what it is, but it's a lot.

MATT LUNGREN: Well, we've had this discussion, Justin,

in other episodes too, about, there's--

the capabilities are moving so quickly

that I'm not sure I would know if we crossed

whatever line we're talking about for that moment of AGI

and--

I mean, there's-- it's seems like there's just so much we can

do with what's already here.

And then again, when it comes to interacting

with the models in it, education way.

What does this mean for healthcare in general?

Like, how we train physicians and all the--

I don't think we've really gotten our arms around this.

And I think as folks are starting

to use this in their high school and undergrad

and what happens in med school, are we still really going back

to sitting in classes and we all know the data on retention

from live lectures and all this stuff?

Are we still going to do that or are we

going to have to reinvent this?

And I know you know you're involved.

But both Just and I too, in educating the next generation.

I'm not sure how to incorporate this into the curriculum.

ERIC TOPOL: Yeah, it's a great question.

I gave the commencement address at Mount Sinai

last week at their med school.

So it made me think more with--

at the moment things are so dynamic.

What's it going to be for a medical school,

and it's going to have to undergo some radical changes.

And the whole idea of memorizing stuff and what

would constitute the curriculum today

is already outdated, terribly.

There's no AI in the curriculum, which is amazing.

I mean, wow.

I mean, not that we want every physician to be a computer

scientist, but they should at least

feel very grounded, facile, comfortable, understand

the nuances and limitations.

And there isn't a medical school yet today,

which is embarrassing at 150 plus in this country that

has really gotten this into their curriculum.

So a lot has to change there.

Also, selecting who's going to be a doctor.

MATT LUNGREN: Yeah.

ERIC TOPOL: It doesn't have to be the GPA

and MCAT score anymore.

We really want to get people who are exuding the most

human and humanistic qualities.

So we don't do that very well right now.

We're not breeding people for those qualities

as much as we need to in the future.

JUSTIN NORDEN: Where do you think

we are in medical education or even health

more broadly, because you have these conversations with leaders

all across the country are invited to go speak with them.

What do you think is the current understanding

of the capabilities of the technology

at the leaders of healthcare across the country today?

Do they-- obviously we're--

Matt and I are on and you're on one end of the spectrum of what

we think this technology is doing and reading and keeping

up.

But where is the average healthcare leader

of a medical school as well health system

today in your mind?

ERIC TOPOL: I think it's the earliest phase.

So it's being considered for back-office operations

for coding and things like that and operations,

but also maybe for trying to get the ambient conversations

and reduction of keyboard and data clerk work for clinicians,

but it isn't far reaching at all.

There isn't the interest of getting

patients powered and charged to do more with their data.

There isn't really wide-scale implementation.

It's mainly in images related.

And there isn't any multimodal AI

that's really made its way through regulatory,

that's gotten into a health system at this point.

So that's where everything is right now

in terms of the excitement.

It's going to take a while.

The problem, I think, that is driving

the slowness which is characteristic

of the medical community.

I've always thought it's been sclerotic or ossified.

But what really is the lack of compelling data.

So, I mean, if we just go back to imaging,

which has several year history now,

you've got studies of over 100,000 women in Sweden were

randomized, where the mammogram interpreted with a radiologist.

And AI was clearly superior, picking up

25% more breast cancer and significant breast cancer

and reducing the time.

But what about the United States?

How many-- what proportion of our mammograms

are using AI with radiology readers?

As an example, how long is it going

to take with the most compelling data

that we have to actualize it in daily practice.

And we have RadNet that's charging $35 to the women.

I mean, this is crazy.

This should be routine.

So it isn't a good precursor to seeing what I'm envisioning,

which is Super Agers and the future of healthspan.

I do think we're going to get there,

but it takes much too long when the opportunities are

so remarkably great.

MATT LUNGREN: Well, this-- and there's something to that,

right.

So like while we are in for probably

a whole laundry list of reasons, which we could go into,

but just why we're laggards and why a lot of this stuff

has been challenging.

While that's going on, consumers or patients

have access to the literally the greatest AI technology

on the planet, on their phones or on their laptops and Justin,

if you have that quote from that--

there's now-- there's anecdotal evidence everywhere where folks

are putting in their medical data into the models

on their consumer--

on their consumer app and they're

getting, in some cases, some pretty powerful results.

And this is leading folks to say, well, now,

if the patient is able to use this powerful technology

and achieve a health outcome or a diagnosis,

and they do that independent of the physician who decided not

to use that or cannot use that.

This was a comment from of a malpractice attorney

and an article, very provocative thought.

But the position that he's promoting here is that listen,

it's negligence if you--

there's a tool that is better for patient care

if you could use it and you choose not to use it.

I think that's a very interesting--

this has never happened before, to my knowledge.

Like, there was never someone did

such an amazing Google search that it could make some really,

really powerful diagnoses that the physician may

have overlooked or that seemed much further away.

But now we have literally a model that's going to reason

over your healthcare data, find connections and potentially give

you a answer that maybe your clinician didn't--

couldn't do.

Anyway, it's a fascinating--

ERIC TOPOL: Every day, I know.

Today, in The Wall Street Journal,

there was a young fellow who wrote about how he was disabled

and he used Claude Anthropic, prompting with all his data.

He had all this sensor data and it fixed him.

I mean, it's like, wow.

After he'd seen several specialists and whatnot.

I mean, it's a common story we're seeing now.

And we're going to see a lot more of that.

But unfortunately, the medical system

doesn't support individuals to have more control.

They want to control everything.

They don't even want to have all their data.

I mean, it's their data, but they only

can get a portal access and even that's limited.

So this is a real imbalance that shouldn't be tolerated.

But we don't see enough of health systems

that are promoting AI in their patient base.

I mean, another striking example is the first multimodal AI

for patients.

Besides we go back to the diagnosis of arrhythmia

through a smartwatch, right.

But more recently, the diagnosis of sudden death

through a smartwatch.

FDA cleared-- it could save lives

that if people have that watch with an app and they keel over

and it brings the paramedics to them immediately, I mean,

that's amazing.

So yeah, there's also-- and that's obviously

a multimodal story of detecting pulselessness

without and motionless and having a baseline.

So there's-- the opportunities here are just extraordinary.

They're limitless.

And we need to give patients more charge.

They're eager to have it.

It has to be practical, inexpensive, well proven.

It's easier to move the patient side

than it is the clinical side.

JUSTIN NORDEN: So it's interesting you bring up

that piece on consumer empowerment and consumers going

to find this information.

And it's interesting this week I'm

actually out here in Washington DC,

right now, where we've had a bunch of announcements.

But relevant to this on an RFI from CMS talking

about wanting for more exchange of data,

wanting for more consumer access to that,

and then a real call for technology, candidly, in a way

I haven't seen before from administrations coming in here.

And so it seems like there's pressure on that side.

The other thing actually, I'll pull up,

which was some interesting examples of private companies

starting to take a more aggressive view

towards empowering consumers.

And I was hoping to find more talk about this in your book.

But what's happening now of-- a controversial within medicine

of hey, we're just going to scan healthy people.

And now using AI, we can do it faster.

We can look for more, as well as run tons of blood tests,

combine AI on top of that.

And it's clear investors are super excited about this

as a concept.

And so with these companies combining to do all of this

together.

And so how do you put-- how are you

putting those pieces together?

ERIC TOPOL: Well, it is in my Super Agers book.

And it isn't good about doing a total body MRI,

I can assure you.

This is a bad thing to have for healthy people

because it's a recipe for false positives.

And I've already seen patients who, for example, had a liver

nodule, then they had a biopsy and they almost bled to death

and/or had a pneumothorax from a chest nodule.

All these benign things that are being found by total body MRI.

And I just did a podcast with Mark Hyman, who

is the co-founder of Function, and they just

acquired Ezra, the second largest purveyor of total body

MRIs.

They reduced the cost and reduce the time.

So what?

It shouldn't be done.

And I made it clear that none of these tests

should be done, unless there's a good reason.

That is a person's risk.

And finding a cancer in a mass on a scan

is not as appealing as finding it at a microscopic level

before it ever shows up on a scan.

And we can do that in patients.

So one of the three major age-related diseases

that I go through in Super Agers is cancer.

We can get ahead of cancer.

It takes 20 years for most cancers

to fully take hold and get to a point of metastasis.

Well, if we use a person's electronic health

record, structured, unstructured text,

if we get setpoints of their labs,

that is labs in the normal range,

but trending in the wrong direction

that we can't see because we're just

looking at normal or abnormal.

And if you add to that the cancer susceptibility

genes, the pathogenic mutations, the gene-- whole genome

sequence, the polygenic risk score and then things

like a methylation clock, organ clocks,

you can get to the point I know this person's high risk

for this type of cancer.

And then if that person irrespective of age,

you might get a tumor DNA sample of the plasma so-called liquid

biopsy.

And only then if the AI of that doesn't tell us

where the organ is, then you would get a total body MRI.

But right now, everything is all screwed up.

You got promoting Functional--

Functions and many of the companies

are selling, including particularly Prinova,

selling total body MRI to wealthy people with no data.

And in the book I go through [INAUDIBLE] case,

which he's a New Yorker writer and a great physician

at Cornell, and he goes along with getting

a total-- free, total body MRI.

Guess what happens to him in his 40s?

Healthy guy.

They find a prostate abnormality.

Now, he has to get prostate biopsies every six months.

That's a don't sign me up for that.

I'm sorry.

And his life is completely been affected by that.

And there's a great quote by a colleague in the book about it.

Look, this is terrible.

This is predatory.

Predatory.

And we shouldn't let this happen,

but it's unregulated jungloid space.

It's great to have MRIs cheaper and quicker.

Who wants to go hang out as a patient in an MRI machine,

but use it for the right people.

And it's just like mass screening for cancer.

We have everybody treated like cattle.

They're all the same.

Just if you're a woman, you go for your mammogram.

88% of women will never have cancer.

Why do we put all of them through this

and have so many false positives and false negatives?

So everything we do is so dumb and wasteful

and it just is what it's doing.

It's really good at is getting people anxious for no reason.

And it's just we got to do better than this.

It's really-- I wrote a long substack on these 12 longevity,

quote, longevity companies.

And there's not much there.

I do think they're on the right track,

that they want to get better data on people,

but when they promote things with no proof, that's

where I draw the line, because we

do have proof for a lot of things and we're not using it.

JUSTIN NORDEN: And tell us-- tell us more about that side.

And I know-- and I know you got to see a bunch of really

positive things.

I loved your synopsis of the book of lifestyle

plus of areas of--

it's great because I think we all

have these conversations with family members, colleagues,

friends.

Hey, what would you recommend for these things.

It's a beautiful synopsis.

I think of what you said at the highest quality evidence things.

But talk to us just about a few of those

and how you see AI playing a role in a positive way.

ERIC TOPOL: Yeah.

So I mean, we don't get people to-- again,

here we are prescribing the same lifestyle for all people,

and not just diet, sleep, exercise.

You got many other layers of that.

That's why I call it lifestyle plus.

Anyway, it doesn't work that well.

You need to have that at an individual level.

And who's likely to adopt these things.

It's people who know they're at risk.

And they're at risk, way in advance.

And we have this opportunity we've never had

before for primary prevention.

I mean, this has been a fantasy for millennia in medicine

that we would prevent diseases from ever occurring.

We don't do that today, but we can.

Now, lifestyle is a big way to do it.

And that I think--

I think once we get into this, you

accept that there's these three age-related diseases,

not that we're trying to reverse aging.

We're accepting aging.

We just want to prevent the diseases that

are pegged to aging process, very different lean--

different position.

Anyway, so once you have these three,

then the lifestyle factors are common for all

if exercise is this essential, if paramount.

If not-- if there's one thing and it's not just aerobic,

it's resistance and strength and balance.

I mean, all these things are critical as we get older.

The diet is a very important not to have a pro-inflammatory diet,

ultra processed food, lots of red meat, things that we know

are not good for you for that.

And then, of course, sleep--

getting deep sleep.

When we clear these waste products

from our brain that are just bad,

we got to get them out of there.

We need more deep sleep as we get older,

and we can track it now easily with a smart watch, with a ring

and pretty accurately.

So those are some fundamentals.

There's many other layers.

But once you know you're at risk for let's say,

Alzheimer's 20 years in advance and you know things can help.

And you get a P-tau217 biomarker,

which would be abnormal 20 years in advance.

That's what's amazing.

And it responds to exercise and lifestyle.

And you can bring it down 40, 50, 70%.

You're motivated now.

You can see your lifestyle kicking in

just like you would an LDL cholesterol with a statin.

So this is what makes these three diseases.

And each has their own strategy.

Like, I mentioned P-tau217 and another neuro inflammation

markers.

This is a unique time in medicine.

We've never had this.

We had to have the AI to integrate all the data,

and we had to have the organ clocks and these biomarkers

and liquid biopsies and all these things.

We didn't have those until recent times, some very recent

time.

That's why we're so set up right now.

It's a convergence of the science

of aging with the advances in large language models

and large reasoning models.

MATT LUNGREN: Yeah.

This is so spot on, I think, in how

AI is coming at the right time in a lot of different ways.

But for this in particular, it's--

again, ties back to what we were saying

about consumer empowerment.

But it's not just about hey, let's throw you

in as many scans as possible, and it's

collecting the right data.

It's addressing some of the things

that we know are evidence based that are leading

to chronic disease and tied to healthy aging, those behaviors.

But there's also an aspect of this

too, that you also mentioned, which

I'm just seeing a lot more, I feel maybe

it's just confirmation bias.

I feel like there's a lot more interest when

you hear about AI drug discovery and identifying ways

to now address some of these age-related processes.

And again, I'm not saying there's

the fountain of youth in a pill, although I'm

sure there's startups that will be pitching that.

But there is some aspect of this that now that we have a better

understanding of the surface area, of the things that we know

are like definite contributors to disease,

are we now able to look at the therapeutic side

and does AI also help us there.

And again, this is my crystal ball

is a little foggy in this space.

But it does seem like with this rapid ability

to identify targets and identify functions.

And now that we're pointing it at things

that are related to longevity, talk about that a little bit,

because I know that you touched on this in your book,

and it's just an area that I don't know if I understand well

enough and I haven't seen, I guess, the breakthrough

that maybe I would expect already.

ERIC TOPOL: Yeah, I think it's coming.

I mean, I think the ability to identify targets and project

their efficacy and safety profile.

I mean that's all going to accelerate.

There aren't that many great examples yet,

but I think we'll see them.

And even not just predicting protein structure 3D and high

resolution, but designing the proteins and small molecules

and antibodies and all the biomolecules.

So that's in the works.

But what I think is something to get some perspective is

the GLP-1 drugs, all right.

So this a whole chapter on that.

And if we would go back 20 years ago when the Novo Nordisk only

wanted to develop this for diabetes.

And if they had asked AI should we just should

we look at it for obesity.

I quipped in the book, yeah, I would have told them,

but it took a human to keep pressing them there in Denmark,

Lotte Knudsen, for years.

We got to try this for obesity.

We got to try it.

And it's amazing because the reason they didn't want to do it

is because the people with diabetes

didn't lose any weight, like 3 or 4 pounds.

People with obesity lost 40, 50, 80 pounds.

We still don't know why.

And I even asked AI why and it didn't give me an answer.

Nobody knows why, this theory is, right.

There's lots of theories, why do women live longer than men.

Nobody knows that either, really.

Anyway, we are in a time when all these gut hormones.

We're just getting started.

If you think that these drugs are, miracle drugs, Ozempic

and [INAUDIBLE] and Mounjaro, you ain't seen nothing yet.

We got triple receptors, we got pills,

and we got a lot more gut hormones coming.

Anyway, the gut hormones talk to the brain.

They talk to our immune system.

They're potent.

They're well tolerated.

We don't know any serious side effects, really,

when you think about it.

This is going to explode.

And that's just talking about gut hormones.

So now we're just--

if you-- you guys are young, but we've been trying to crack this

obesity case for a long time--

MATT LUNGREN: Long time--

ERIC TOPOL: Many decades.

And now look at this.

I mean, it's just we got people taking these drugs just

to lose 5 pounds to whatever.

It's ridiculous.

But it's amazing what's happening here.

And I don't want it to continue in terms of a drug--

a forever drug.

I'd like to see people take advantage of this

and get off of them and be very inexpensive

and keep getting better.

But that should tell you, we're in a new age of drug discovery

itself, just one class of drugs.

And it's broad.

I'm not just talking about glucagon

like peptide or glucagon.

I'm talking about all the gut hormones

because there's a lot of them.

When I was in medical school, there

weren't any gut hormones, instantly.

Now, there's 50 of them.

Yeah.

JUSTIN NORDEN: Well, there's almost too much

to cover on what's coming next and down the pipe.

Like today, today, if--

a lot of, I think, our audience is

trying to figure this out for themselves,

how do they use the tools?

Where do they do this?

So today, today, what are you using?

What would be a message you would want people to take away

for where they should be experimenting or using

some of these tools now?

ERIC TOPOL: Well, I mean I'll use anything as long as it

works, as long as it helps me.

So I don't know that getting into nitty-gritty helpful there.

But I just would go back that when you--

I think only in a tunnel vision of how can I

use AI to improve health, that's what I think about.

I'm obsessed with it.

I've been obsessed with it for many years,

and I don't know if I'll ever get over it.

So that's what I think about.

And that's why I think the next frontier is prevention.

Demis Hassabis was on 60 Minutes a couple of weeks ago

and he said, you will end of diseases, cure diseases

within the next decade.

I don't see why not, he said.

And a lot of people don't think that's possible.

And I don't think it's going to happen in 10 years.

But we have a way forward now to use these tools

and to keep improving them so that we can really get there.

And that's why we're going to have

a lot more wellderly people.

Now, they're the few than the elderly people, which are

almost all after age 60 some.

We can flip that.

And it's really exciting.

And so that's what I keep thinking about,

and that's what I'm working towards.

And that's what the book Super Agers is all about.

MATT LUNGREN: I feel like your obsession--

we benefit from your-- whether you think it's healthy or not,

your obsession with health and AI because honestly it's

laid out so clearly here in that framework.

And again, in terms of using the different tools,

I think part of it's just the fact that you

encourage that folks do that.

I think it's important because of the accelerating pace

of it's still that jagged edge.

Like, there's still things that hey,

it's not quite there in this one or two areas,

but yet it's so profoundly good in others.

And if you-- but without that intuition,

it becomes hard maybe to even see how that vision plays out.

I know Justin, you have that--

there's a graphic where it's like, what

are folks starting to converge on using these models

for in their day-to-day life.

And it started out with hey, can you answer this quick question?

Can you write this blog?

And now look at this.

I mean, this to me is fascinating.

And this ties into another theme in your book,

superheroes to me because you see

again it editing text, generating

ideas and this little number two category therapy

and companionship from 2024 now is the number one use.

And then we have organizing my life

and then we have finding purpose.

Is this starting to address the idea of social isolation

and loneliness as a contributor to a decline in health

and wellness and is this, again, another place where

unexpectedly, something is coming up from the technology

side to fill that gap or to fill that need?

And anyway, the point is, I feel like folks

are starting to tell us what these models are

more useful for to them in their daily lives.

And it's surprising to me that it's not homework and doing

my TPS reports at work.

These are things that I wouldn't have predicted,

having been in technology for this long, that

would be so popular.

ERIC TOPOL: Well, I mean, you're absolutely right, Matt.

And when I wrote the medicine, I wrote that the machines

will never help with empathy.

I couldn't have been more wrong.

Because now, there's 12 studies to show that patients

perceive empathy from machines more better than their doctors.

I mean, what?

How could this be?

Now, that's of course, doctors are rushed

and machines don't even know what empathy is,

but they can transmit it.

They can be a vector for it just from having learned

been pre-trained with language.

So what we're seeing is something

I never would have predicted.

And it's replicated many, many times

through various different studies.

So I know there's something there.

It also tells us that we humans have to amp it up

if we can be beaten so easily as physicians by machines.

So it's exciting that the unpredictable positive

features that you touched on.

And it isn't like we're plateauing here.

We're going to keep seeing progression capabilities.

And yeah, here's another graph of yours.

JUSTIN NORDEN: Yeah, and you just brought this up.

And it's come up a bunch of times.

But I think what I certainly didn't

predict was that for much longer,

I thought AI plus human performance

would beat out AI on some of these diagnostic tasks.

And this was another--

actually, Matt, do you want to give some more

context on this benchmark from OpenAI

this week on their evaluation set showing again,

actually, these models now can start to perform better

without humans tweaking.

MATT LUNGREN: Yeah, it's called HealthBench.

It was released by OpenAI.

A nice contribution, if I'm being honest,

to the broader community because it is open source data set

5-plus thousand conversations, a couple hundred physicians

on simulated medical questions from 60 countries.

I think it's around 50 languages and across dozens and dozens

of medical specialties.

And then they do these really full evaluations.

And again, I think it's meant to be

a nice new thing for the community to think about

as they evaluate the medical performance of these models

as opposed to--

Step 1 was hard, and I am impressed

that these models do well.

But as we've saturated those benchmarks,

I think we all recognize that maybe that's not

the ultimate goal of leveraging these models to do medical type

tasks.

And so this one I think is slightly more realistic.

I would say some of the great work

that Nigam Shah's group has done at Stanford in MedHELM.

Another way to start to take real-world medical data

and turn those into benchmarks and then be able to rate

and rank the different models.

But it's just another example, though,

where I think part of this is just not--

I still hold out faith despite being accused

of being on the v-node side of the spectrum on some of this,

I still believe that physicians, with these tools,

if we can figure out how that relationship is ultimately

symbiotic in a good way, I think it's going to be awesome.

But right now, it still feels like we're

at that junior high dance where doctors

are on one side, AI on the other a little bit,

and we haven't figured out how to dance together

to make the results that we expect to see, which is

together, that they're better.

ERIC TOPOL: Yeah, it's a good metaphor with the dancing

and junior high.

But no, you probably know Pranav, who you know very well,

[InAudible] and I wrote an op-ed in The New York Times

when we commented on six studies, where AI was compared

to doctors who had AI, and whether it was radiology

or whether it was diagnosis or patient management.

All six studies showed the AI better in performance,

and that was not anticipated, as you said,

that there was no symbiosis there.

Now, is that because doctors are not

grounded with AI, even though they had it to be able to use?

Is it because they have automation bias that they just

don't appreciate it can help?

Or is it because the studies are contrived?

They're not real-world medicine.

We don't know yet.

I hope, like you just asserted or opined

that we're going to get there, that we

will get the best of both worlds,

but we don't have evidence for that yet.

And it looks like I hadn't seen that graph from the benchmark

that OpenAI just released yesterday, I think,

or a couple of days ago.

But yeah, I mean, it's the same kind of thing.

Right now, we don't have evidence

that best of both worlds is working.

We have AI am exceeding the performance, which

hopefully that's not going to be the way it rolls out.

Pranav thinks it will, just so you know.

So he and I have all these discussions.

Of course, he's a computer scientist and I'm a physician.

So we go at it.

And I don't know who's going to be right here,

but he still thinks AI is going to win no matter what.

We'll see.

MATT LUNGREN: Well, he's much smarter than I am.

I can say that I've worked with him for years,

but I'm an irrational optimist.

I think there's got to be some secret sauce of us being

able to leverage these things to really take it

all to the next level.

But I'll remain optimistic until hopefully some evidence backs me

up a little bit on that.

ERIC TOPOL: So I'm with you.

Just so you know, I'm very much with you.

JUSTIN NORDEN: And a lot of people

forget who don't come from medicine,

many of our patient visits start when you already

have the diagnosis set.

And there are guidelines that even get you

to the right treatment.

So often that is the starting point.

And I think a lot of people outside of medicine

go, oh, the job is done if I have the diagnosis.

And often that's the starting point.

And so anyway, it will be fascinating to see

how this evolves.

MATT LUNGREN: Well, thanks for another awesome conversation,

Eric.

Thanks for joining us.

This was tremendous.

I do encourage everyone to please pick up the book

Super Agers.

You won't be disappointed.

There's some pretty big insights in there

that I think that are starting to touch on a theme

that I think we're going to see a lot more of in the future,

which is aging.

Well, I like your I like your term wellderly

because we're all going to get there someday soon.

And in my case, more sooner than some others.

But I do feel like understanding really

where the truth lies amongst all the hype

and the swirling, I think, is critical here

and how AI is going to take us to that next stage.

ERIC TOPOL: No thanks.

Great to be with you both.