“ChatGPT, MD”: Author Says AI-Empowered Patients, Doctors Take Control

Mark Masselli (00:04)
Our guest is renowned healthcare leader and author Dr. Robert Pearl, who says, artificial intelligence will soon
be as common in healthcare as the stethoscope.

Dr. Robert Pearl (00:14)
What we're gonna find is the technology is 10% better than the average clinician in an advice center or the
average nurse in a chronic disease management program. But for that matter, potentially even the average
physician doing inpatient care when there's five or six different doctors taking care of the same patient and
they're not effectively communicating.

Margaret Flinter (00:36)
Dr. Pearl is a professor at both the Stanford University School of Medicine and the Stanford Graduate School of
Business, which gives him a great vantage point to assess the challenges and the opportunities for healthcare in
2025.

Dr. Pearl (00:50)
If you are a patient with chronic disease, it can effectively manage your chronic disease by having you have
wearable monitors that connect with the device that can do calculations and tell you, are you getting better or
worse?

Margaret (01:04)
And this is conversations on healthcare.

Mark (01:17)
Well, Dr. Pearl, welcome back to Conversations on Healthcare.

Dr. Pearl (01:21)
My pleasure to rejoin you. I look forward to our conversation today.

Mark (01:25)
Well, we do too. And Margaret, it was back in 2014, we had Dr. Pearl on. Lots has changed since then. At the
time you were the executive director and CEO of the, Permanente, medical Group. It was really a quite a
fascinating, interview because you discussed the battle between traditionalists and technologists and medical
care. Medical care. I'm wondering if it's fair to say 11 years later the technologists have won.

Dr. Pearl (01:55)
Well, I think there's a new technology and that's generative AI. And it's important that we differentiate it from
everything that has come before, because this is the first technology in the history of medicine that I can think of
that actually empowers the patient. Not designed for the clinician, but for the patient, or wasn't designed for that,
but applicable to the patient. And in that way, you can say that the technology has evolved in a way that has led
it, I think, to be able to be an equal partner with the human.

Margaret (02:33)
Well when look at that, interview, and, and actually I remember this and I'm reminded again. You had great
enthusiasm for the future. I think enthusiasm is always good, in, in healthcare and life generally. But when we
spoke to you, none of us, could have anticipated, I don't think the pandemic, and with it, the growing skepticism
about science that has come into the society. And I wonder, what are your thoughts about that? What's your
advice for navigating this post Covid era, where there is such skepticism about science out there in general?

Dr. Pearl (03:08)
You're absolutely correct that the, well, the pandemic could not have been predicted for the year that it would
come. But I think a future pandemic could obviously have been predicted and we were not adequately prepared.
But I would say the, I say the shift away from science today is very regrettable. It's going to harm people, cost
lives. All you need to do is look at Texas. We're currently, there's over 500 cases, 600 children hospitalized and
two deaths, right? That's one in every eight afflicted children with the disease requiring an inpatient admission.
And that number is four children per thousand dying. But as you know, the historical numbers have been one to
two or one to three. And that's what we're gonna see if this unscientific approach to a very contain very
controllable chronic disease, sorry, controllable infectious disease persists.

Mark (4:11)

You know. I wanna get back to your book because it has a great, chapter titled, chat, GPT md, how AI
empowered Patients and Doctors, can take Back Control of the American Medicine. And you say you wrote the
book with the assistance of Open AI's popular chatbot. Tell us a little more about how that, transpired.

Dr. Pearl (4:32)
Happy to do so. First of all, I wanna tell viewers and listeners that all prophets go to Doctors Without Borders.
A great charity has had the prophets from all my previous books. I wrote it because as you knew, November
30th, 2022, the Generative AI Chat, GPT was released. Now I named it Chat t md, my book title 'cause it flows
so beautifully. But everything I'm about to say about chat PT is just as true about Gemini from Google, just as
true as Claude from Anthropic. And then most recently released Chinese deep seek model. But what I realized
that this is an exponential technology. And that way it's very different than something like the electronic health
record. An exponential technology actually like a coronavirus doubles every year. And in this situation, the
technology, Chat GPT would be 32 times more powerful in five years. And in fact, if I went through the typical
cycles I had with my previous books of going through a traditional publisher and writing it on my own, it
would've taken about two years. By that time, the technology's four times as powerful, and everything I write
would be outdated. So I said, how do I increase the speed, accelerate the writing and distribution process? And I
decided that the way to do it, and by the way, an interesting way to do it would be to work with chat PT so that I
can firsthand experience this generative AI tool, not just in clinical practice, but also in the creation of the
artistic work. And I approach it the same way I do with a resident or a medical student with whom I'm write a
paper. This, in this situation, I gave chat GPT, the 1.2 million words that I had written and published. I would
normally give a bunch of articles to a medical student or resident. And then we exchanged drafts back and forth,
probably 30 different drafts per chapter. There's six pages in there. I should point out that were written only by
chat, GPT, or mainly by chat, GPT. And I highlight those so readers can see how the technology itself would
work. But all the others were, I'll call it a collaborative effort and represents a combination of the voices that
were there. And as you know, the book was then distributed available on Amazon still today, bestseller both
originally and continue to be so today, on amazon.com.

Mark (07:04)
Well, you know, I wanna point out that the training of an AI, instrument is very important, right? Where it gets
its information. So I'm wondering if you sort of put the version that you were on and if you would go back, go
forward in a year and maybe think about doing another rewrite given how fast, how fast the, improvement is
going in AI.

Dr. Pearl (07:28)
You are absolutely right. You know, when I wrote the, book, the only way to enter data was by typing into text.
And now you have, visual image, you have voice, so many different ways to do it. The time that I wrote it, the
version wasn't even up to date. As I note the book was 18 months behind. Now, by combining the training that's
been done along with the availability of information on the internet, it's up to date right now. But I think maybe
the most important part was just this release of the product called Deep Seek. This is an open-source tool.
Unlike Chat GPT, or Gemini or Claude, it means that everyone can access and understand how it works. And
the way this tool is created was not as you're indicating through downloading, through pre-training on the
entirety of the internet and all material, all textbooks, all journal articles, everything available. It was done
through a process called distillation. That means they took a currently existing model and extracted from it
information, not by copying it 'cause you can't do that, but by asking it millions and millions of questions. And
in that way, what you're able to do is to narrow the focus and go far deeper into whatever you're trying to
accomplish. And then having done distillation, the next generation of generative AI tools is not only gonna have
the large language model capabilities, but it's also gonna have additional training. Think about it this way.
Millions of calls are made to advice centers. Millions of calls are done annually for chronic disease
management. The monitors in every hospital record a terabyte of data annually, and that is 10 with 14 zeros and
97% is thrown away, never looked at again. The opportunities to learn to narrow it, to focus it on diabetes or
narrow and focus on heart failure or inpatient treatment now will create tools that are not only easily available,
easy to use, but they will also be very specific. And I believe that the ultimate acceptance of these tools will be
done because the people will take the model and they will use the model and simultaneously in a research
project, have a human provide the same expertise to the same level of treatment, and then compare the two. And
what we're gonna find that the technology is 10% better than the average clinician in an advice center or the
average nurse in a chronic disease management program. But for that matter, potentially even the average
physician doing inpatient care, when there's five or 60 different doctors taking care of the same patient and
they're not effectively communicating.

Margaret (10:30)

Well, you're making it clear. We're all gonna have to stay around and stay engaged because it's way too exciting
to miss out on, on what is about to come. But I wanna maybe, see if there's a leap we can make from what
you've, just said to, what you referenced in your book, which was that healthcare providers and patients, can join
forces to reclaim from prevailing corporate interests, more, perhaps ownership authority agency. Explain,
explain to our listeners what you meant by that.

Dr. Pearl (11:01)
The challenge today, I believe is the economics of healthcare. The fee for service system. This is a pay for
volume. Charlie Menger, Warren Buffett's, partner before he passed, said, show me your incentives and I'll
show you the outcomes you're going to get. And certainly Warren Buffet's gotten great outcomes, but because
the system has aligned around that, that's what I see in healthcare today. That what clinicians are do, are, driven
to doom by incentives, by rewards is just do more and more and more. And as a consequence of that, I believe
they're losing control and patience as a result, maybe getting a lot more things done. They're certainly not
healthier. Life expectancy today is no better than 15 years ago. Maternal mortality in the United States is twice
as bad as any other country in the world. Childhood mortality is the worst of any, of our peer nations. We could
go down an entire list. There are 400,000 people who die every year for misdiagnoses, not 'cause doctors aren't
smart, they just rush. They just have too much to do in a day. So the question becomes, how do you change this?
And this is where I believe that what needs to happen is to move from a pay for volume, fee for service to a pay
for value capitation, a single payment made to doctors and hospitals, not to insurance companies, doctors and
hospitals to provide the care to a population of patients. What's interesting to me, what's very exciting to me is
according to the CDC, if we could effectively control chronic diseases, we're talking hypertension, diabetes,
heart failure, things like that, we would see 30 to 50% fewer heart attack, strokes, cancers, kidney failures. We
would save probably half a million lives a year. The cost of medicine would go down by $1.5 trillion. And why
have we not been able to accomplish that so far? We haven't had the technology. And that's why generative AI
becomes to me so powerful. If you have a patient with chronic disease, it can effectively manage your chronic
disease by having, you have wearable monitors that connect with the device that can do calculations and tell
you, are you getting better or worse? If you're getting worse, you need to have a medication change. If you're
doing better, you don't even need to see the clinician 'cause you're doing well. This is the kind of power the
technology's going to do. It provides expertise to the patient. When they have a new symptom, they can figure
out what's going on and they can decide whether they need to see a clinician based upon the expertise. That's
why I don't talk about AI. I talk about generative AI because what exists for the most part, particularly in the
past, was what's called narrow AI tech. AI developed by taking two data sets, let's say women with
mammograms, 5,000 of which show cancer, 5,000 of which are either normal or benign disease. And then they
compare the two and figure out some discrepancies. And that works well for visual images. But generative AI is
based upon the totality of all the information that we have, every medical textbook, every journal article, and
now it becomes available to the clinician to look up a problem that he or she can't figure out. And it allows the
patient that I could give you dozens of stories of parents able to figure out what was wrong with a child when
the medical system couldn't. Or clinicians who facing a situation could go into the literature and find four cases,
five cases, very obscure cases, and know exactly what to do. And that would've taken them days in the library.
As I say, this is the dawn of a new era, and we shouldn't think of that as just another AI tool. This is as different,
different from what's come before as the iPhone was from the telephone that was in most people's kitchens
attached to the wall.

Mark (15:15)
Well, certainly we want those, important outcomes and results, but we also wanna make sure that there's maybe
a randomized clinical trial, that sort of evaluates the efficacy of those. And I was interested to see that. We
found that when, in a randomized clinical trial, when AI worked independently to diagnose patients, it achieved
92% accuracy while a physician using an AI assistant were only 70, 76% accurate. I'm wondering if you have
some thoughts about, that particular trial.

Dr. Pearl (15:48)
Absolutely. So the trial you're referring to, was published by, researchers at Stanford. It was done in conjunction
with researchers at Harvard. It was a small trial, so I'm not sure the numbers will hold up when we're talking
about thousands and thousands of clinicians. But what it points to is that humans have what's called
confirmation bias. If you come to me and you tell me you have a certain symptom and I immediately leap to a
conclusion as to what the disease is likely to be, I'm likely to ask you questions, to be able to hear information
that reinforce my belief rather than change it. These are the 400,000 deaths a year from misdiagnoses. And the
reason why that happens is that doctors are rushed. You know, there's been some research done that said if
primary care doctors did everything recommended by regulatory agencies for every patient who came to their
office, it would take 27 hours a day. Not possible. So what happens is the doctors have to take shortcuts. It's not

that they don't know what to do or they wanna do something else, they don't have a choice today, and this is why
I say we're on right now, a treadmill where patients, doctors are paid on a fee for service basis that insurers keep
lowering the dollars that are being paid while drug companies raise prices and salaries go up. So doctors have,
they're seeing more and more people a day, they see on the average now a patient every 17 minutes. They
interrupt the patient every 11 seconds, not because they are not dedicated, hardworking, smart, and want to do
the right thing. We just have given them today an impossible job. And this to me, is where the technology
comes in. It can do a lot of the pre-work, it can do a lot of the post-work, it can manage these chronic diseases
that patients have every day on a continuous basis, not episodic. It can provide care 24 hours a day, seven days a
week. I think done well. It could save hundreds of thousands of lives, make medical care affordable for all
Americans, and lower the burnout. That, as I say, is a crisis with half of all clinicians affected by it in the United
States currently.

Margaret (18:09)
Well, I'm gonna, pivot for just a minute from, generative AI and impact and talk about the people side of this
equation and our current moment, in history. And you recently had some positive thoughts about our new HHS
secretary, Robert F. Kennedy Jr., and called him an imperfect leader at a perfect time. Take us through your
thinking on that.

Dr. Pearl (18:33)
Well, I would call my perceptions, and I'll use the word balanced, although I really mean balanced the line
'cause both extremes are equally weighted or both have weight on both ends of this, long seesaw on the, I wanna
start with the negative side. I think that his views, he's expressed in the past around vaccines are unscientific and
right now are contributing to the, problems that we have in Texas and the deaths that we've seen. And I think
there are other areas around some of the cutbacks. These are not driven by, Robert Kennedy Jr. But by the
overall administration that is happening, whether the FDA, the NIH, the CDC that are gonna have long-term,
very negative consequences. Not by the way for the only for the health of Americans, but it's gonna set us back
behind the rest of the world. We've been a leader in scientific advances. So I, I want all listeners to know, I am
not saying that he is the positive answer that other negative side exists, but if there's a silver lining to his
appointment, it's that he has been a very strong advocate for the challenges that our food industry produces. And
what we know today is that obesity is a major problem. Over the past 20 years. It's doubled for about 22% of the
population, of 46% of the population. Its cause is very simple. The food industry, the foods that are being
created are designed to be ultra processed, high caloric. They're designed to be able to give more calories per
bites. And that is how we drive obesity in this country. And why that is so significant is obesity is the leading
cause of diabetes, which is the most common reason why people get heart attacks and kidney failures, 80% of
lower limb amputations. And it is also the cause of 40% of cancers. So we are talking about an amazing,
amazingly terrible, a really terrible type of, impact. And this is where in the article I wrote, we have yet to take
on the food industry every time someone has tried in legislative branch to put up the kind of effective
approaches, what's an effective approach? Taxation. We see that in cigarettes, you raise the tax on cigarettes,
people consume fewer, you raise the tax on sugar soda, they take, they in a lot less, sugar soda than they
otherwise would put a tax on these high caloric, high fat, ultra processed foods and use that money to subsidize
fruits and vegetables and the kinds of things that many Americans simply can't afford. And now you have a
possible recipe to reverse the pandemic. You add to that some things like the GLP one drugs so that people can
start the process of losing their weight along with diet and exercise, lifestyle and food changes. Now you start to
have a more effective way to be able to reverse this obesity crisis. And now you add to it some of the things that
the, health Secretary has said, which is a primary care is very important that we know that it is. And we know
that every incentive in the fee for service world goes in the opposite direction. It tends to drive more specialists,
more inpatient care, more procedures. We are a nation that treats sickness and disease after it happens at a nation
that ignores the ability to prevent these problems before they occur. And I'm hopeful

Mark (22:24)
If you listen to some of his top aides, Calley Means, really sort of, characterizing the health system as utterly
failed. That we spent more money than any country in the world and have worse outcomes than all of them. You
look at some of the things they're proposing on the SNAP program to end the access to buying, soda pop. So
they're very focused on a whole different side. And you just touched on it, which is to prevent disease, increased
longevity versus what you were talking about managing diseases. That's a whole paradigm shift. Forgetting all
the other politics, just sort of thinking about it, what many people have said for a long time, we're just spending
an inordinate amount of money on our healthcare system and not getting the outcomes that, we, we would love
to see, but we're simply maybe taking the wrong approach. Do they have something there?

Dr. Pearl (23:17)

Absolutely. I mean, you've described it perfectly and what I said earlier, you know, we talked about capitation in
1934, that's a hundred years ago. We recognized the problems created by the fee for service system, but at every
point it has been blocked. You know, we've known about the challenges of food, the ultra-processed food, the
way that, obesity has been a rising crisis for 20 years, 25 years. These are not new problems. And yet we can't
get out of that. Yes, I'm hoping that the secretary, along with the people he is appointed to lead various agencies
will take us in that direction. You know, there's an expression I like that, you know, the impossible becomes
possible and then inevitable, this change is inevitable. The only question is how many people need to die
prematurely before we actually make the changes that we know will be effective.

Margaret (24:19)
Well, I, agree with so much of that and all the, the universe of spinoff challenges arises immediately as we think
about them, right? So we've all been engaged in, healthcare delivery and also training the next generation of
people who are gonna deliver that care. I haven't, I haven't heard the same kind of clarion calls for revamping,
not just of the education and training of physicians, but the education and training of nurse practitioners, of
social workers and psychologists, of physician assistants, of all the people, the community health workers who
are gonna have to drive this shift. And I wonder what your take is on that. Do you think it's going to be a case of
10 years or 15 years till we really revamp our curriculums to drive towards this kind of focus? Or do you have
some insights into that and what federally we ought to be doing about it?

Dr. Pearl (25:14)
In my second book “Uncaring: How the Culture of Medicine Kills Doctors & Patients”, I look at the educational
process in detail and I point out that it is always at least 20, if not more years behind. Because the people who
are doing the teaching are the ones who received their training 20 years in the past. And they still see that as the
best way. 'cause they see themselves as being the best practitioners. It's just natural. It is the human tendency.
And you know, certainly quite a number of people have written about all of these, I'll call mental mistakes that
are written. People won Nobel prizes for the literature they've created around that. And we just see that very
much. I have little confidence the academic institutions are going to change. You can completely alter the
medical school curriculum. You can make generative AI not just a tool that people are taught around, but they
are forced to use as a medical student or resident. The systems that we have for admission, for evaluation, for
certification, they test memory. Memory was a crucial skill 30, 40 years ago. If you wanted to carry around all
the medical information with you, you need a 50-pound backpack. Now we have tools. Just ask generative AI a
question and you'll have the answer in a matter of seconds. Why should we not be using this tool? But I agree
with you completely. I think the academic institutions are not going to do it. And this is gonna be a classical
type of disruptive change. A change that, has been well described by Clay Christensen, where the people, I'll
call it with the least power, are the ones who start to embrace it. And as they start to use it, and it gets better than
people with more ability and more ability until ultimately penetrates the whole system. And who has the least
power in American medicine today? The patient. And I think it'll be patients who start to use it, then you're
gonna see primary care physicians, then you're gonna see specialists. And it's already happening, by the way.
You know, I can't give a talk and I gave 50 keynotes this year, to conferences about the future of healthcare.
And I have 40 more scheduled coming up. This is the hottest topic out there and I can't go anywhere, right? I
don't hear stories. You know, I had a great story. I was at Stanford this summer teaching a program to 80 CEOs
from across the country. At the end of the program, one of them came up to me and said, Dr. Pearl, I had never
used the tool while you were talking. I registered for Chat GPT, I entered information from my daughter who
had a very baffling disease, that I took her to six different academic centers over 18 months. And I asked for
Chat GPT’s differential diagnosis, and it gave me three diagnoses. The number one was a very rare
inflammatory disease, which was exactly what we learned 18 months after we started seeing academic experts in
at various medical centers across the United States, this is what people are doing. You see it in the surveys and
was noted earlier. What we're seeing right now is this technology is highly reliable and it's excellent. It's not that
it's perfect and people should understand that this doesn't replace a doctor today. When I wrote the book, Chat
GPT md, I fact check everything that Chat GPT told me, 30-page bibliography. And I found only one big
mistake. And that was an expedition to the North Pole that it had hallucinated, but it still was a mistake.

Mark (28:47)
Well, let me, let me get at that. Let me get at that issue of the future of, of medicine, about trust in medicine and
about AI because it seems to me that doctors and other medical workers, fear the falsehood and conspiracy
theories, that are eroding trust in their, in their profession. And I'm wondering if you think AI is friend or foe in
that battle because, there are many other examples where the footnotes in fact half of them were, were
hallucinated. So what's your sense about how we're gonna get at medical trust and does AI help us restore
medical trust? Because if we're gonna have people like yourself speaking, if we have people who aren't listening
because they've lost trust in the healthcare system, that's a real problem.

Dr. Pearl (29:38)
Well, there's trust in the healthcare system and there's trust in the technology. And I'm not sure which are the
two you're most interested in. In terms of the technology. I think it's gonna increase the trust with the clinician.
It's gonna do it in two ways. Number one is gonna give doctors the time that they need to be able to minimize
the errors of today. We don't think about that. You know, we ask about the technology, could it harm someone?
We don't think about that in the context, what we'll do to 400,000 misdiagnoses. What about if it reduces it to
200,000, that's still 200,000 people who’ll die, but it's half of what exists today. But maybe it's 100,000, maybe
it can get it down to 50,000. I don't know how far it can go, but it can create that. And the second thing I think
it's gonna do is it's gonna prove the dialogue. It's gonna actually make the doctor patient relationship better.
Why do I say that? You know, I was on a podcast where the host said to me at the end, we were off air, “Could
you gimme your advice? You’re a doctor, you’re a skier. My husband fell three months ago, his arm was over
his head. He slid down the mountain a hundred feet, his shoulder still hurts and it doesn't work as well as the
other side, the arm, what's going on?” I said, “I think I know exactly what he has, but do me a favor. You said
you'd never use Chat GPT. Why don't you put all of his information in there, see what it says, and then call me
back.” She calls me back five days later. And what is she saying? “Thank you so much. It said he most likely
had a rotator cuff tear, that he needed an MRI to find out which tendon was damaged. And he needed this
orthopedic surgeon in order to have the tendon reattached before the muscle contracted.” And when he went
there, when they went to the doctor, he said exactly that. He said, you probably have a total rotator cuff. He
ordered the MRI. They knew exactly why that was gonna happen. And when he said, you need surgery, then he
said, thank you. And by the way, at the end of the operation, he successfully reattached the tendon. But if he had
said, if he'd come three months later, I probably couldn't have done it. That's the power and the expertise. It's a
better relationship. It's not a more problematic one. Now, in terms of the overall system, the system needs to be
changed. That's why I've written these other books as well, because it is fragmented. Doctors are scattered
across the community. I'm connected with each other. They, and it uses this pay for pay for volume. This type
of, 19th century cottage industry means, piecemeal reimbursement. The technology, by the way, doctors use
today, the most common way they exchange information is the fax machine. I tell my students this never
graduate school of business, they communicate with a fax machine. They say to me, what's a fax? A fax
machine? There's no leadership structure out there capable of making the change. Yeah, we need to change
medicine in order to reestablish that trust, but is the generative AI tool that will empower the patient and the
doctor and just compare that to the difference between doing that and clicking on a Google link where you get a
hundred pieces of information. A lot of it's, by the way, misdiagnosis. It tells you a hundred different reasons
why arms don't work as well as the opposite side. And there's no way to personalize it to figure out what is it
that I have that my loved one has. Generative AI is going to be the tool to accomplish it maybe three years from
now, four years from, or five years from now. But right now everyone should be experimenting with it and
seeing all the things that it can do.

Margaret (33:08)
Well, there's no question about it. It has an important role to play, in the world, in health and certainly towards,
our, guiding light, which is healthcare as a right and not a privilege for everybody. So thank you so much Dr.
Pearl for joining us. I don't think we'll let 10 years go by until our next conversation with you, because
obviously…

Dr. Pearl (33:27)
Yeah, I'll add we're absolutely convinced that a dedicated clinician with an empowered patient and our
generative AI tool will be able to accomplish 10 times more than any of the three alone. Now is the time for us
to begin the process, not the time to totally allow it to end. And in beginning the process, I think that we will
come up with the best models for the future and hopefully change overall healthcare delivery in the United
States and make us once again the best in the world. Thank you for having me today.

Margaret (33:58)
And thank you, Dr. Pearl, for joining us. And thanks of course to our audience, for being here. Just a reminder to
be sure and subscribe to our videos on YouTube. Find us on Facebook and X, but please also share your
thoughts and your comments about this program. Take care, be well.

(34:16)
This copyrighted program is produced by conversations on healthcare and cannot be reproduced or remitted in
whole or in part without the express written consent from Community Health Center, Inc. The views expressed
by guests are their own, and they do not necessarily reflect the opinion of conversations on healthcare or its
affiliated entities.