BEST OF: Will AI Fix Health Care? Robert Wachter Weighs In

This episode originally aired on February 3, 2026.

Health Affairs' Rob Lott interviews Dr. Robert Wachter, Professor and Chair of the Department of Medicine at UCSF, about his new book A Giant Leap: How AI Is Transforming Healthcare and What That Means for Our Future. Wachter reflects on his own daily use of AI as a clinician, the reasons he has grown optimistic about its potential, and the challenges of regulating fast‑evolving technologies.

Currently, more than 70 percent of our content is freely available - and we'd like to keep it that way. With your support, we can continue to keep our digital publication Forefront and podcast for everyone.

Rob Lott: Hello, and welcome to a health podocy. I'm your host,

Rob Lott. Friends, it's time for another very special episode of

A Health Podicy, an episode where instead of interviewing

the author of a recent health affairs paper, we talk to a

guest from the broader universe of health and health policy,

someone helping shape the day to day discourse about our health

care system and its future. Well, I couldn't be more excited

about our guest for today, Doctor. Robert M.

Wachter. In addition to his role as professor and chair of the

Department of Medicine at the University of California San

Francisco, doctor Wachter has done and accomplished a lot.

Back in the nineties, he made his mark by helping coin the

term hospitalist to describe what was for a time the fastest

growing specialty in American medicine. He has been a trusted

and influential voice on topics of national importance ranging

from patient safety to digital medicine to our national COVID

response. And today, he has a new book all about artificial

intelligence.

It just hit the shelves at bookstores everywhere, and it's

called a giant leap, how AI is transforming health care and

what that means for our future. I cannot wait to hear all about

it on this latest episode of our Humble Podcast. Doctor. Robert

Wachter, welcome to a health podocy.

Robert Wachter: Thank you, Rob. It's just a joy to be here. I

admire what you do and the journal is fantastic.

Rob Lott: Great. Well, let's just, dig right in. And I think

maybe before we talk about the book, I thought we'd start a

little bit with your own personal experience with AI. Do

you use AI today as a practicing physician, as a teacher, a

researcher? What is its role in your life today?

Robert Wachter: Constantly. I was just on the wards at UCSF

last week. And in the old days, like a year or two ago, I'm a

generalist. And a couple of times in the morning, I would

get a curbside consult, which means I would run into my

favorite ID doctor and say, you know, can I take you aside? I

have this case, it's tricky.

Didn't need a full on consult. I didn't need to see the patient.

I just needed advice. Now rather than getting two or three of

them, probably get about ten or twelve in the morning, but I

don't get them from a human. I get them from usually from open

evidence, sometimes from Gemini or GPT.

And the answers are pretty darn good. They're not perfect. I'm

glad I'm there to look over its shoulder, but it's better than

anything I've had before. It's better than what I used to get

from electronic textbooks. It's better than what I certainly got

from Google.

So I use that all the time. I'm using an AI Scribe, so when I

see a patient now I can actually look the patient in the eye and

listen to them rather than looking down at my keyboard. I'm

using a chart summarization tool. Here's an interesting

factoid. One out of five patients' charts are longer than

Moby Dick.

Oh my. More than 600 pages. So the idea that I'm gonna be able

to read that in three minutes before I see a patient is of

course ludicrous. So those are the kinds of tools I use in my

life as a doctor, in my life as an administrator and leader. I

use them sometimes to draft memos.

I use Claude to help edit documents I'm writing. I think

it's a terrific writer and editor. So it's become nearly

ubiquitous in my life.

Rob Lott: Okay. Well, let's talk about the subject of your book.

You call it a giant leap. That's a pretty optimistic title. You

didn't use the title grave reservations, for example.

Was that hopefulness that you, are espousing about, artificial

intelligence something that you had in mind when you set out on

this book project, or was it something that surfaced during

the course of your research and writing?

Robert Wachter: Yeah. The latter. I think it evolved

organically. I I am fully capable of writing a grumpy book

because I wrote one ten years ago. It's called The Digital

Doctor and it really was about electronic health records.

It was about healthcare going from paper to digital. And I

wrote it out of frustration. I wrote it because all my

colleagues were moaning about the electronic health record and

how it ruined their life and this term pajama time. I came

home at night and had two hours of work to do and I'm not

looking my patients in the eye anymore because I'm so busy

checking boxes on the screen. All that stuff, none of which

anyone anticipated.

Then patients got their patient portal and now they could click

a little button saying, send a message to your doctor. And so

they did. There's a chapter at the end of the Digital Doctor, I

think it's 27. People just came up to me and said, who was your

ghostwriter for that one? Because after 26 chapters of

grumpiness, there's this hopeful, I think this is gonna

work out chapter.

And it's like, no, I can actually see how this gets us to

a much better place, but not yet because we didn't really think

about how to implement these tools in ways that would make

things better. And I think in retrospect, we didn't have the

tools that we needed, that the EHR succeeded in digitizing our

information, solved a lot of problems, including doctors'

handwriting, but didn't really do anything to provide us help

in dealing with all the bureaucratic stuff we have to

do, the prior auth writing, any help in decision support. And I

guess the thing on top of that was my optimism is partly borne

by these tools are remarkable and can do things we never could

do before, like read a note, like deal with unstructured

data. I can go to my tool, my AI and say, I've got an 84 year old

patient who's got CLL who comes in with a fever, shortness of

breath, chest pain, has a hematocrit of two, and a

creatinine of 3.7. What do you think is going on?

You know, impossible. Any prior tool would have sputtered on

that. So the tools are amazing. But I think the other part of

the optimism, and this is sort of perverse optimism, is three

years ago before GPT came out, it wasn't like I was sitting

there saying, Google's terrible. I can easily imagine something

better than Google.

Google was great. Couldn't imagine anything better until I

used ChatGPT the first time. And I said, oh yeah, this is better.

Do you know anybody who says the healthcare system is fantastic?

I don't.

I don't know any doctor, nurse, administrator, or patient who

says the system's great. So it really is a combination of these

tools being remarkably good and the system desperately needs

them, needs to have the kind of transformation that I think AI

promises that I think is impossible without it. So my

optimism came from both those things and really emerged

organically from, I did 110 interviews. The more I learned,

the more I thought about it. It's not like I'm naive, not

like I don't think there will be unanticipated consequences or in

some ways negative consequences, but I think the net chance for a

positive outcome is much, much higher than the things I worry

about.

Rob Lott: Okay. You talk about that grumpiness and then the

shift to the optimism at the end of The Digital Doctor. I'm

wondering if you aspire to sort of reconcile those two attitudes

with this. You sort of look back on the ways you were, as you

say, potentially naive about, the shift to electronic health

records. Here, you're sort of walking the tightrope, if you

will.

How do you recommend other folks in this space kind of reconcile

those two attitudes?

Robert Wachter: Well, I'd say the first thing is to admit that

we can't possibly deliver what our society and our patients

need without digital help. And that this is the form of digital

help that we needed. We needed tools that can do the kinds of

things that AI can do. I think also it's not like we in the

world of healthcare have been static for ten years. So now

that we're trying to implement this new technology, so ten

years ago, fifteen years ago, we implemented this other

technology called electronic health records.

None of us really had the infrastructure to do that

thoughtfully. Didn't really understand how to think about

return on investment for digitization, how to weigh

privacy concerns against the value of having data and data

fluidity. I think we could easily get snookered by a

company pitching us something. If it looked pretty on the

PowerPoint slides, we'd say, great. Think now as we implement

these tools, first of all, the company is building them, I

think are more thoughtful.

They are more likely to have clinicians engaged in the

process of building them and understand that if it doesn't

work in the real world, it's not gonna work and we're not gonna

buy it and use it. I think internal to healthcare systems,

we've got better processes and sort of more thoughtful and less

fewer naive people who are gonna look and scrutinize. Like,

what's the evidence this thing really works and do the kind of

appropriate tire kicking. The tools are better and they're

just easier to use. AI is not new in healthcare.

We tried it forty years ago, but we started with the hardest

problem. We started with diagnosis. And that violates

change management law one through 57. You wanna start with

the easiest problem, the lowest hanging fruit, the ones that are

gonna gain you buy in. Then once you have buy in, then you say,

all right, let's move to more ambitious use cases.

So rather than starting with diagnosis, what did we start

with with the new AI? We started with AI scribes. We started with

an absolutely clear pain point that everybody complained about,

doctors and patients. And we kind of fixed it with tools that

really do this thing really pretty remarkably well, that

aren't that terribly expensive. And they're not perfect, but

they're pretty damn good.

But even if they kind of screw something up and they miss a

word, it's probably not fatal. And I can tell you that every

doc at UCSF, three or 4,000 docs now has access to an AI Scribe.

Most use it, most love it. And at this point, if we turned it

off, they would all threaten to quit. And there's never been a

technology like that.

Usually it's like, if you turn it on, I'll threaten to quit.

And so I think it has gained buy in and a level of receptivity

for, all right, let's try chart summarization. Let's try having

a draft, my discharge summary. All right, let's carefully try

computerized decision support where the stakes are higher and

if we get it wrong, somebody can be hurt. And let's figure out

which company do we wanna go with.

Do we wanna go with our EHR vendor, which is building all

these tools? Do we wanna go with a startup, which has maybe some

advantages of nimbleness, but maybe won't be in business in

three years? So all those sort of things, I think we're all

more thoughtful, more mature about this than we were. And

part of that was the learning curve of what we learned from

the early experience and I think the painful experience with our

first stage of digitization, which was really the EHR. Great.

Rob Lott: In the book, you write that humans are, quote, awful at

anticipating the consequences of new technology. Obviously,

that's a lesson learned from the move to EHRs. And I'm wondering,

you do a good job in the book of describing how even though this

feels like a very sudden innovation, it's sort of been in

the works for the last five decades. But even if we look

back to just the last couple years of this so called giant

leap, I'm wondering if there are ways that you've been surprised.

Is there anything that's happened in just the last few

years, even since you've started writing this book, that you

didn't foresee?

Robert Wachter: Yeah. I I mean, I I mean, the biggest surprise

was the first time I used ChatGPT. It's like, my god. And

I think most of us had that that holy cow experience where, you

know, this thing can be kind of remarkably human, and that's

both wonderful and part of what makes this so exciting and scary

because it does increase the probability that you will over

trust it. I'd never heard of this term hallucination until

two or three years ago, where it not only can give you an

incorrect answer, but it sort of bathes it in this blanket of BS

so that it just feels credit seems like it's right because it

feels like it's a human talking to you.

I've been surprised by that. Guess I've been But then also

surprised by how quickly they've gotten better. And that's part

of the challenge here. If you used AI two and a half years

ago, you might have said this thing's not ready for prime time

in a high stakes business like taking care of sick people. And

you might have been right at the time, but a year later you

probably were wrong.

And we've not really seen in healthcare a technology that has

evolved that quickly. And that's part of the regulatory

challenge. When the FDA says this pacemaker is good to go or

this medication is safe and effective, they're dealing with

something that's gonna be static. It's gonna be the same

as it is today, three years from now. If they say this AI is not

ready for prime time, it's possible next week it will be

because it will be better than it was.

So the evolution has been, the speed of evolution has been

really kind of remarkable. Big surprises. If you asked me ten

years ago, which comes first? Particularly if the AI gets

better, which comes first? I'm gonna sit in the backseat of a

driverless car, fall asleep and trust that it'll wake me when I

get home, or all of our radiologists are out of

business.

I would've said the radiologists are toast. I would've said

reading a collection of digital dots and saying, this looks like

lung cancer, or this looks like pneumonia. That struck me as an

easier problem than making a left turn across the Visadero

Street in San Francisco. And it turns out that I take a Waymo

here about once a week. They're spectacular.

And we can't hire enough radiologists in the same city,

in the same completely tech obsessed city. So that surprised

me. And actually it's a good thing. And the reason it's a

good thing is, I'm kind of a student of politics. If doctors

or nurses thought their jobs were at risk, they would fight

against this and it would block it in healthcare.

And we're pretty darn smart. We would not fight against it and

say, I'm worried about this taking my job. We would say, I'm

worried that this will kill you. And a layperson probably would

believe us. And so I think it will take a lot of jobs in

healthcare, but the jobs, sadly, each of the people losing their

jobs will be sad.

But in terms of the political valence of that group of people,

it's not the most powerful incumbents. Who is it? It's the

thousand people we work and have in the billing department. It's

the hundreds of people in the quality department who are

flipping through charts, recording data. It's the people

drafting and writing prior auths and sticking them in fax

machines.

We still use fax machines. Those people, I assume there will be

some labor savings. When you hear from healthcare leaders,

they will engage in the happy talk. That is, we're not laying

anybody off, but we normally would have had to grow our

workforce this much and it's gonna be flatter, maybe. But I

think ultimately, and this turns out to be a big deal because

people ask all the time, all right, you're optimistic.

Does that mean it's gonna save money? And the answer is, I'm

not sure. We are pretty good at figuring out ways of things not

saving money in healthcare. New technologies, as surgery got

cheaper and cheaper, we do more Lasix on your eyes and hip

replacements. So we'll see.

But if it's going to save money, the only two mechanisms I can

think of, one is labor replacement. One is replacing

human FTEs with AI. And the second, at least theoretically,

is if it guides us to more cost effective therapies. I could see

that happening, but I could also see it guiding us to the

therapies that are gonna be the most profitable for the health

system or some other actor. So a lot of this has to do with the

policy sort of assumptions underlying the recommendations

that could go either way.

Great. Well, I

Rob Lott: wanna ask you a little more about those policy

assumptions, but first let's take a quick break. And we're

back. I'm here talking with doctor Bob Wachter about his new

book, A Giant Leap, all about artificial intelligence and the

future of health care. You alluded to sort of the policy

challenges and the the regulatory work that still has

to be done, and I'm wondering if you can say a little bit about

sort of the principles that you'd like to see our

policymakers building on as they shape the rules and regulations

around AI and healthcare going forward?

Robert Wachter: Yeah, I mean, like most regulatory challenges,

we have the usual Goldilocks problem. You want be nimble and

open enough to embrace things that are good and helpful and

improve care and quality and maybe lower costs. At same time,

not approving something that's not ready for prime time and can

actually harm people. I think the structures that we have to

insert this problem of AI into are not, as The UK would say,

fit for purpose. I mean, the FDA doesn't have the foggiest idea

how to regulate a thing that can shape shift as quickly as AI can

shape shift.

They're trying. They're coming up with different ways of

thinking about that. But I think they have a pretty good

structure to say, all right, here's a new tool to help read a

mammogram or a CAT scan. We know how to certify devices and say

whether they're safe and effective. They've come up with

some mechanism by which you can say, Well, I'm going to be

tweaking this thing.

And they'll say, All right, it's certified for now. And if you

come back in a year, we have a pathway to get it recertified.

What they don't really have is the way AI is going to be used

mostly in clinical medicine, is going to be things like chart

summarization, things like an AI scribe. Then probably the real

money here, and I don't just mean money, I mean also stakes

from a quality standpoint, really is in computerized

decision support and AI driven decision support. It's even an

open question of should they be regulating that?

I pick up my phone and use a tool that tells me this patient

is likely to have pneumonia or a pulmonary embolism or has a high

chance of being readmitted or needing to go to the ICU or the

best treatment we recommend for this patient's Crohn's disease

is this immunotherapy. But I'm the doctor gonna who's gonna

ultimately sign the order in the chart. Is that regulatable? And

if it is, why is it more regulatable than the textbook I

used to do to do that or the Google search I used to do that?

It's really just taking a knowledge base and delivering it

to me in a more convenient and maybe more updated and more

accessible form.

So even there are very big fundamental questions about

should they be regulating that? Clearly devices and tools that

are high stakes and can kill somebody and particularly where

the thing is gonna act autonomously, or even if it's

not, the clinician really has no good ability to vet what it's

doing like a read of a mammogram. I think it needs to

regulate. There it's a question of who does that and how it does

that. But decision support, think is really an open question

at this point.

And then you get to, should it be the FDA? Well, the FDA

regulates only the manufacturers of the devices. Whether AI works

or not has a lot to do with how UCSF implements it. And so maybe

that's not the regulatory framework. Maybe it should be a

version of the joint commission that says UCSF is doing the

right thing in terms of the way they purchase these tools, has a

structure set up to vet whether it's working on day one,

convincingly has a structure set to vet whether it's working on

day three sixty five.

Some have made the argument, David Blumenthal has made the

argument that we should be regulating AI like we regulate a

doctor. How was it trained? Did it pass the right test? All

that. So I think this is a long winded way of saying, end the

chapter on regulation saying, basically, I have no idea, and

this is a really hard problem and we're gonna need to innovate

as much in the way we think about regulation as the way we

think about AI.

Generally, there's a lot to be learned.

Rob Lott: Great. Fair enough. Well, great fodder for future

health affairs content at least. So, I was a regular follower of

you and your social media posts during COVID, and, I might

describe your vibe in that space as pretty reassuring. You didn't

sugarcoat things, but you also found a way to honestly lay out

the facts and also shine a light toward maybe hope of better days

to come.

And I'm curious if in your work as a practicing physician,

you've had patients confide in you their fear or anxiety about

AI and how that might make navigating an already pretty

scary healthcare system even scarier. What your response is

to someone expressing that kind of fear or anxiety?

Robert Wachter: Yeah, I'm not hearing that much fear, And I'm

hearing mostly positive things from patients who are using GPT

and tools like it to put in their doctor's notes and say,

can you tell me what this means? Or I just went on my patient

portal, it says my magnesium's low and my EKG's abnormal. What

does that mean? I think they're finding it to be a useful tool

to help understand the system, to some extent to navigate

around the system. I think where there's fear is something that's

about to happen but hasn't happened yet, which is if I take

all of my medical record and put it into ChatGPT, which as of

yesterday now can happen as ChatGPT just rolled out

something called GPT Health, do I trust that it's gonna be kept

safe and private?

And so I think patients are now already used to, and I think

accepting of the idea that my medical record is in digital

form, sitting in a cloud somewhere, probably a cloud that

Epic is running or whoever my EHR vendor is. So there's

nothing sort of fundamentally different about the privacy

issue unless you're taking your data and sort of moving it

around to different players. I don't know that most patients

know that OpenAI or other companies like that are not

bound by HIPAA. If they did, they might be more fearful of

doing it. But I think you know, I can tell you where So I think

most people actually like it and feel like it's making their care

better.

It informs them before they go in to see the doctor. I think

most doctors, I think doctors are a mixed bag on that, but

we've already dealt with this. We've already dealt with

patients coming with 20 pages of Google printouts. And I guess

I'd rather them come in with a GPT printout. So I think it's

more likely to be useful to them and accurate.

Where the rubber's gonna meet the road here is as patient

facing tools get better and more robust. I think most patients

are gonna like that and welcome that. To me, some of that's

great and some of that's scary because it will give them some

things that are really smart and other things that are dangerous.

It may convince some of them they don't need to see a doctor

when they really do. But you can see just in the last week, Utah

just approved that an AI can refill your prescription without

a doctor.

It's not a big deal because it means a doctor prescribed it in

the first place, but you know that that is just the baby step

to ultimately AI being able to prescribe. And like all these

things, it's a mixed bag. I guess the final mixed bag is

deep fakes where, you know, let's say somebody wants to see

me as a doctor and I don't have availability. I think pretty

soon you'll be able to see digital twin of me that looks

like me and sounds like me and says things that I would say

because it's trained on me. That sounds like a good thing.

On the other hand, you could take a video of me talking to

you now and having me say, you shouldn't get vaccinated and you

should have 32 glasses of wine a day and it'll look like me. And

so, you know, almost all of this is just double edged sword, it

can go either way. But I think most patients are pretty

enthusiastic about this from what I've seen. Maybe biased

because I live in San Francisco.

Rob Lott: Fair enough. I mentioned in the introduction

you're considered by many to be one of the fathers of the

hospitalist movement. You helped coin the term with a deeply

influential New England Journal article published thirty years

ago with Lee Goldman. And so maybe to wrap up our

conversation today, I thought I'd hit you with hypothetical.

How would you, or rather how would your take on the emerging

role of hospitalists back in 1996 been any different had

today's AI existed back then?

Robert Wachter: That's a great question. Probably not very

different. I think in some ways what a hospitalist will do is

going to change a little bit, and I would be cognizant of

that. One of the things that we did, I think, quite effectively

and smartly was to frame the hospital's field as it emerged

as being the first field that Remember when we coined the

term, it was really in the middle of the managed care era.

The IOM report on safety and qualities were just about to

come out.

And what we did was we framed the field as the first physician

specialty that was about not only taking care of patients,

but making systems work better. And so I would have pushed the

field to completely embrace AI because there's no question in

my mind that it's the future, to be leaders in your, to really

understand it in a very deep way, to be leaders in your

system and trying to figure out how to make this work for your

system and for patients, I think that would have been

fundamental. What I would not have worried very much about is

that AI is gonna take your jobs. And I was in the wards last

week, and I've done this the last few times I've been on, I

try to use it as much as I can and then ask myself the

question, let's say three to five years from now it gets even

better and better and better. Do I still have a job?

And the answer is yes. The answer is it will take some of

the bureaucratic tasks off my plate. It will suggest diagnoses

and tee up treatments or testing regimens that I can just click,

Yes, I agree with that. But I think people are still gonna

want a doctor being the final arbiter. They're not gonna want

AI to tell them they have cancer, if God forbid they have

cancer.

Not gonna want AI to begin chemotherapy, tell them you need

to go to the OR. And so much of what I do as a hospital is

coordinating really complex teams and who gets involved in

what time, decisions where there's no right answer and

you've got to sort of weigh different, not only different

odds of success, but also the opinions of patients and family

members. So I would have been quite encouraging about you're

going to have a job, but this thing is gonna be a fundamental

part of that job. And you at least, even as a day to day

practitioner, need to be good at it. But I would actually

encourage some of you to become experts in and to be leaders in

it.

And I think that would have been fine.

Rob Lott: Great. And good wisdom for for us today as well. Doctor

Bob Wachter, thank you so much for taking the time to chat with

us today. I really enjoyed it. Thanks so much, Rob.

Really appreciate it. To our listeners, thanks for tuning in.

That was the real Doctor. Bob Wachter, not a deepfake, and, I

hope you enjoyed it. If you did, leave a review, recommend it to

a friend, smash that subscribe button.

If you're watching on YouTube, join us again every week, and,

of course, tune in again soon.

Robert Wachter: Thanks for listening. If you enjoyed

today's episode, I hope you'll tell a friend about a health

policy.