← Solutions with Henry Blodget
Solutions with Henry Blodget
Could AI Doctors Be Better Than Humans? This Physician Thinks It’s Possible.
00:53:57
Loading summary
Transcript54 lines
- [00:01]
Dr. Robert Wachter
Support for this show comes from Vanta.
- [00:03]
Henry
Vanta uses AI and automation to get
- [00:06]
Dr. Robert Wachter
you compliant fast, simplify your audit process and unblock deals so you can prove to customers that you take security seriously.
- [00:14]
Henry
You can think of Vanta as your
- [00:16]
Dr. Robert Wachter
always on AI powered security expert who scales with you. That's why top startups like Cursor, Linear and Replit use Vanta to get and stay secure. Get started@vanta.com Vox that's V A N T A.com Vox vanta.com Vox
- [00:38]
Henry
support for this show comes from Indeed. If you're looking to hire top tier talent with expertise in your field, Indeed says they can help. Indeed Sponsored Jobs gives your job the best chance at standing out and grants you access to quality candidates who can
- [00:52]
Dr. Robert Wachter
drive the results you need.
- [00:54]
Henry
Spend more time interviewing candidates who check all your boxes. Less stress, less time, more results. Now with Indeed Sponsored Jobs and listeners of this show will get a $75 sponsored job credit to help get your job the premium status it deserves@ Indeed.com VoxBusiness just go to Indeed.com VoxBusiness right now and support our show by saying you heard about Indeed on this podcast. Indeed.com VoxBusiness Terms and Conditions apply. Hiring do it the Right Way with Indeed. Will your next medical appointment be with an AI doctor? Should it be with an AI doctor? Are AI doctors already better than human doctors? These are questions that many of us patients and doctors are asking and the answers are surprising. Today's guest, Dr. Robert Wachter, thinks AI can improve our healthcare system for professionals and patients alike. Dr. Wachter has written a best selling book called A Giant How AI Is Transforming Healthcare and what It Means for Our Future. In addition to being a physician himself, he's a professor and Chair of the Department of Medicine at UCSF. He has written six books and practiced medicine for more than 40 years without spoilers. AI is already widely used in healthcare and it will be increasingly so in the years ahead. And that's good news, not bad, says Dr. Wachter. Here's our conversation. Dr. Wachter, it's great to have you. Thank you so much for joining us. I heard you recently and I was very excited when I heard you sound like what I would describe as an AI in healthcare Optimist. And you've written this book where you interviewed a hundred people on what's going on today. So is it fair to characterize you as an AI in healthcare? Optimistic?
- [02:53]
Dr. Robert Wachter
Yeah, I think that's fair. I guess I want to make sure people understand my credentials because I wrote a book 10 years ago that is incredibly grumpy about digital healthcare. And so I am fully capable of being on all parts of the spectrum here. I think I am an AI optimist in health care and I think should confine it to that. I'm a little bit more worried about it outside of health care, but I think the reasons are fits health care like a glove. The issues that we have been struggling with. You know, the first time I used ChatG on November 30, 2022, the light bulb went off and said this will solve a lot of real live day to day problems that we're struggling with. And the second is sort of a corollary of that, which is healthcare is massively broken. And despite the fact that we have wonderful doctors and nurses and people who work very, very hard, and I work at one of the great healthcare organizations in the country at the University of California, San Francisco, terrific places and yet incredibly hard to get an appointment to see a primary care do. And there are medical errors all the time and lots of friction marbled through the system. So it was the combination of these tools being incredibly good and different than anything we'd ever seen before and getting better very fast and a healthcare system in desperate need of transformation. And so that's really the source of my optimism.
- [04:15]
Henry
We'll get into the specifics in a minute, but I think that the vast majority of Americans will cheer when they hear you say, as a doctor, yes, the healthcare system is broken and it's just incredibly frustrating it seems, for all involved in including doctors and nurses and companies and everything else. But so before we get into what's working and what the concerns are, let's take us back because one of the things you do in your book is say, hey, this isn't the first time we've heard about AI in healthcare. And also we've just had this big electronic record revolution and actually it's created a lot of problems too. So this is the setup for the adoption of AI now. So talk about that.
- [04:53]
Dr. Robert Wachter
Yeah, there's a few pieces of history that I think are important to understand as foundations to the moment. The first is, as you say, Henry, AI is not a new concept in healthcare. In the late 70s, early 80s, some computer science, very smart folks, many of whom were also doctors, built AI systems and they were not ready for prime time. They were of the if then variety, which works fine. If you have a sore throat. If you have a sore throat and you have swollen lymph nodes and a fever, you probably have strep throat or mono. But it falls apart in the face of the complexity of real medicine, which is more complicated than if then statements. The other thing that. But they blew it on. And they were very smart folks. They weren't naive to the challenge here, but they decided to tackle. The first problem they decided to tackle was diagnosis. And I've interviewed some of the founding fathers, they were pretty much all guys at the time and said why? And they said we weren't naive about how hard that is as a nut to crack. We just thought it was the most interesting thing to work on. And I can get that it was. They're building artificial intelligence. What do doctors learn in four years of medical school and three or four years of residency? To be intellig to take in a bunch of facts and make a diagnosis. But it's also the hardest problem and the problem where if you get it wrong, you can kill somebody. And I think the lessons learned there. A, it wasn't ready for primetime. B, our data wasn't computerized. And so there was really no. The logistics of getting it done were using AI were kind of impossible. But C, don't start on the hardest problem. Start on the most tractable problems. You have to win over the hearts and minds of patients and leaders and policymakers and the incumbents. And I'm talking mostly about doctors and nurses. So those were very important lessons. AI went basically to sleep in healthcare for 40 years. It had a brief comeback when IBM Watson beat the Jeopardy champions and IBM announced the first industry they were gonna tackle was healthcare. And they completely did marketing before quality product. That's never happened before. No. First time ever in history. But somebody had to try. Was all polished, no shoe. It was not ready for PrimeTime. And in six months it was clear it had no future. So those were the prior histories with AI which left people in healthcare quite skeptical that maybe this is too hard a problem for digital transformation. And then the other piece of history that I think is important is until about 15 years ago, I was recording, if you saw me as a patient, I would record your information on a piece of paper and stick it in a paper chart and would go to the medical records department. And it was funny, I was just watching the pit last night where their computer system went down and they had to resurrect the paper. We've all had that experience. And the young people had no idea. Like, what do you mean? We're gonna write things on pieces of paper? And somebody who's using a felt pen on a triplicate form that of course didn't work. And you know, the geezers knew exactly what this was about. And the young people said, how do we do this on paper? That's what we all did until 15 years ago. So healthcare was the last industry to digitize. And I think the flaw in my thinking at least was that that was it. Once we digitized, once we all converted our information into electronic form with these big software systems called electronic health records, we were done. That was going to be nirvana. And it turns out it was just the foundation we had to do it to get to where we are now. But it didn't solve the problem. And as the point you made, and I wrote a book 10 years ago about this, it actually created a of new problems and a host of new bureaucratic and paperwork burdens. Because now as we were recording our data on computers, a lot of people who cared about what the doctor does could now be electronically looking over our shoulders and say, Doctor, click these 32 boxes because that will produce a better bill. And patients noticed, and doctors noticed. We became very expensive, very grumpy data entry clerks.
- [08:52]
Henry
I loved how you went through that because this is, as a, as somebody who goes to the doctor, this is something that was sort of happening in the background over the last couple of decades for me. And I have noticed in the last many years as I go to the doctor, that, yes, the doctor spends most of the appointment sitting at computer in the corner typing. And it is disconcerting. You talk about how human beings actually like eye contact and you want the
- [09:15]
Dr. Robert Wachter
doctor to talk to you.
- [09:16]
Henry
Yes, exactly. And that is not happening. And then the other thing that you say that anybody who works can relate to is the fact that the poor doctors are now besieged in email all day and they can't get to it during the day, they do it at night.
- [09:31]
Dr. Robert Wachter
Yes, we call it pajama time. That it's after you put the kids to sleep. You have two hours of electronic work that you did not have before. And briefly, the story there was the electronic health record originally was a doctor tool and a nurse tool. And then about seven or eight years ago, all the companies opened up a patient tool, which is terrific. It's called a patient portal and most common is called MyChart. And patients could be on this portal. And then by federal law five years ago, patients got access to everything that I have access to, their lab tests, their X ray results, and including my note, sounds great, but we gave them absolutely no help in interpreting any of it and made it even impossible to make an appointment. I mean, it's much easier to make a restaurant reservation than a doctor's appointment. And so patients saw. What do you mean? My magnesium is low and my EKG is abnormal. But the electronic health record companies did one very helpful thing, at least we thought. They put a little button at the bottom of the screen. It said, click here to send a message to your doctor. And lo and behold, patients, being normal human beings, clicked there to send a message to their doctor. And the result was Doctors were getting 100, sometimes 200 messages a day. And we hadn't given a moment's thought to how this is going to work. What's the workflow model? What's the payment model? And so just another example of completely unanticipated but pretty negative consequence of digitization.
- [10:51]
Henry
Yeah, and I get it on both sides again. I mean, I had some blood tests recently. You go and you get these incredibly detailed breakdowns of all these different things that are being looked at. And everything looks. It's kind of okay. And then this one's off the charts, flashing red. You immediately want to call the doctor,
- [11:07]
Dr. Robert Wachter
but they're just, oh, my God, I'm going to die. Oh, my God. And, oh, there's this button. Great. Yeah, of course, of course. And I'm not arguing that we shouldn't have that button. I am arguing that we got to figure out how to make this work. And we did not figure it out at all.
- [11:20]
Henry
All right, well, so let's go to one of the more. I was going to say futuristic, except it's 2024, so it's actually way deep in the past aspects of that, which is you spent some time talking to a doctor who had a digital twin who could handle all the questions when the doctor was busy. Where are we on that? Is that something that's actually in real world now?
- [11:41]
Dr. Robert Wachter
I start the book off with sitting with the CEO of the Mayo Clinic, who shows me a video of a. A very thoughtful, compassionate doctor talking to a patient. And about 15 seconds into the video, in walks his twin behind him, standing up behind him, and waves awkwardly to the camera, and then walks off stage left. And it turns out that the doctor talking to the patient was fake, was a digital twin of the real doctor. And in many ways, we've seen these things before. I think people have now gotten used to it. And obviously, there's a super dark side of that, because for every digital twin that the male clinic clinic produces that you know is going to be reliable and trustworthy and use Mayo Clinic data, not data from Reddit or from the Onion to produce his or her pronouncements. There's another digital twin that could have me right now saying to you, you should never get vaccinated. They will kill you. And it will look equally believable. So there's a dark side of it, but I think digital twins are a little premature. It's a little futuristic still. But really all a digital twin is is putting a human face and body on what is happening all the time, which is we're using artificial intelligence now to take in vast amounts of data, to be able to take in the kind of query that in the old days I would have had to ask a human doctor. And I think the epiphany for me, when I was Speaking to the CEO of Mayo, he said, 30 years ago, when, you know, Mayo has probably the best brand in healthcare and it had a campus 30 years ago in Rochester, Minnesota, that's where you'd have to go. When they wanted to expand 30 years ago, what did they do? They built a campus in Scottsdale, they built a campus in Jacksonville, Florida, and they partnered with organizations in London and the Middle east, et cetera. He said, today we probably wouldn't do that. As we think about expanding the reach of the Mayo Clinic to more people in more places, we would think, how do we do that using AI, using technology like the digital twin? And I think that is the metaphor here. You used to go to your local bookstore. Now you buy your books probably from a company and it doesn't matter where they are. You get your product where you need it. I think we're going to see more healthcare deliver to more people. And I think this is net a good thing because lots of people have no access to doctors or to the appropriate specialists. More people will get their healthcare through AI enabled sources. And we've got a lot of stuff to figure out there. But I think that creates the opportunity for access and access to high quality medicine. That's greater than anything we could possibly do. If the only way you could see a doctor is that we put a human being in an office somewhere.
- [14:15]
Henry
And as you say that, I can feel the fear of people who say, come on. And I have to say, just having classified you as an optimist, you do spend dozens and dozens of pages going through in great detail the hallucinations and the biases and the other risks here. But you also talk about something I found fascinating, which is people fear change. We are resistant to new technology. You have a wonderful example, which is the driverless cars. In San Francisco and elsewhere, where every time there is an accident, it is held up as this just proof that this is a terrible idea. And yet any study you look at is very clear. These cars are much better drivers than humans. And if we replaced all cars with driverless cars, we'd be much safer and thousands and thousands of people wouldn't die and so forth. And yet there's always this resistance. And then 15 years later, it's totally taken for granted and we all love it and so forth, or it's at least used. But here you have, I think, both actual healthcare professionals and patients just using what's there and doing just what you described, which is, okay, well, maybe before I go to the doctor, I'll actually talk to ChatGPT about it and see what ChatGPT has to say.
- [15:32]
Dr. Robert Wachter
Yeah, yeah. And the sun also rises. 100 years ago, Hemingway wrote, what character went bankrupt? And another said, how does a man go bankrupt? And the answer was, oh, two ways. Gradually and then suddenly. And so this is, I think, our suddenly moment. It really is a little shocking that healthcare, the fastest industry, adopting AI because we tend to be, we're laggards, but laggards in a funny way, we actually love technology and are pretty rapid adopters of technology. But we are, when we have a very specialized technology that we can plunk down in the radiology department or in the operating room and use it in very specialized way that we're pretty good at. And we tend to adopt very quickly what we've never done for a lot of reasons. But in some ways the complexity, the regulatory environment makes this difficult and the stakes are high. What we've never done successfully is adopt a technology that transforms the nature of our work. And the electronic health record experience in some ways augmented that fear. But I think this time it really was everybody saw the needs. I mean, for a physician, I can tell you that when GPT came out, it was just obvious to me that, that even if it's not perfect, the ability that I now have to, if I'm a generalist taking care of patients mostly in the hospital, I'm walking around and I have a question that I really could use an answer from an infectious disease specialist or a nephrologist or a cardiologist. And until three years ago, I would just hope that I would run into my buddy in the cafeteria, which is a pretty inefficient way of organizing a high stakes practice. Today I will pull out my phone and use often a tool called open evidence, which is sort of GPT for doctors. And it's not perfect, but it's damn good. It's better than I, it's smarter than I am. And so, and then as we were talking, you know, the doctor who isn't making eye contact with the patient, we rolled out a tool called an AI scribe, which basically it basically audio tapes the conversation that I'm having with the patient, but then does something that is somewhat magical, which is turn that into a properly formatted doctor's note. Within six months of that thing being available, everybody wanted one because it solved a very important problem that doctors felt and patients felt and did it in a relatively low risk way. In other words, we weren't starting with it, making the diagnosis, certainly not autonomously. We were starting with low hanging fruit, solving real problems that everybody could see and winning over hearts and minds. And I think it's very important that we didn't bypass that stage exactly because of the Waymo problem that you talk about. You know, there used to be two driverless cars companies in San Francisco. One was Waymo and one was Cruise, which was a GM company. And Cruise about two years ago ran over a woman, actually a car, a regular car, ran over a woman. The Cruise didn't see her, ran over her. She didn't die, but she was severely injured. Six months later, big lawsuit, lots of headlines, and a year later they were out of business. So in that case, a single bad accident was enough to take the company down. It's going to be, there will be that in medicine. There will be times where the AI gets it wrong and we kill somebody. That's just inevitable. But I like Biden's quote in the book. I say don't compare me to the Almighty, compare me to the alternative. And in many cases the alternative is you can't find a doctor or the doctor's not very good. And we're human and we make mistakes and we didn't keep up with the literature. And so I think we need to build this reservoir of trust because something will go wrong. But I think in the end we'll probably be like Waymo, that this will be better and safer and probably cheaper and more convenient for patients with AI than we are without. And that's the, that's the North Star we should be looking for.
- [19:12]
Henry
And given the fact that healthcare is out of reach for so many people in this country and a lot of people are uninsured, or it's so expensive, or you just don't want to go to the doctor even if you can afford it, is it Reasonable, from your perspective, for the patients to actually be consulting ChatGPT first. Is that, is that a reason?
- [19:30]
Dr. Robert Wachter
I think it, I think it's reasonable with a little bit of buyer beware. One of the things I think I had wrong and got smarter about this I was writing and researching the book is I love using ChatGPT or Gemini to do the kinds of things that patients do. If I woke up this morning and had some weird symptom I didn't understand, I would have no hesitation putting it into a large language model. But one of the things I underappreciated is there's something magical that happens when you go from novice to expert. I mean, and you would do the same in a business conversation. There's stuff that I might have no idea what to do and you would just do, like breathing. It's second nature. When I pick up the phone or go into my computer and say, I might have 100 facts at hand about a patient in front of me. I know their medication list, which has 12 medications in their past history and they've had four surgeries in the past and they had a pulmonary embolism and a heart attack and now they come in with a headache or shortness of breath or whatever. I pick up my phone and I say, this is an 81 year old man with a history of CLI who comes in with a fever, a white count, an infiltrate on chest X ray, and an abnormal alkaline phosphatase that is all completely Greek to you. And what I've done there is take seven facts out of a hundred and distilled it down into a narrative that is an intensely cognitive act that takes a lot of expertise. And then when the tool comes out and says, well, this sounds like it could be A, B, C and D. I look at it and say, oh, I didn't even thought of A, that's great. B and C, I'd already thought of D. That's stupid. I'm going to ignore that again. I can do that as an expert. You as a patient can't. So when the patients use these tools, and if they're using the same tool, we're asking them to do those two things, which they can't generally do, understand what the key facts are to put into the prompt and then interpret the results, some of which may be great and some of which may be completely wrong. So I think the tools for patients are not ready for primetime. I think that better tools will act much more like your doctor. They will know your past history. That's relevant, they'll have culled it down into the relevant nuggets that are relevant to this particular problem. And then you will say, I woke up this morning, I have a headache. And it will say to you, tell me more about it. What part of your head? How long has it lasted? Does the bright light hurt your eyes? Does it hurt when you bend your neck? The exact questions that I would ask if I saw you, those will be the tools. But the question you ask, is it ready for primetime today? The answer is no. But I still think it's probably better than nothing. In some circumstances. What I would do would be give it as much information as you can. What I would do is ask it to ask you, are there any other things you need to know in order to diagnose me correctly? You would ask it, if this is a terrible thing, what would it be? Or are there certain things I should really worry about with this? And then I might ask for a second AI opinion, meaning if I did it on GPT, I'd probably go ahead and do it on Gemini. I can't prove that if they're both in sync, that it's more likely to be right, but that's my instinct. And when I played with it, that seems to be the case. So. So I think the tools will get much, much better and over time. I don't think there's any question that there are lots of things that you go to see the doctor for today that you don't need doctor level expertise or doctor level expense or doctor level inconvenience to do. I think five years from now, having your cholesterol medicines managed, or even prescribing a blood pressure medicine or maybe even diabetes medicine will be done by AI probably with some sort of triage protocol that says no. There are things about you that you really do need to see a doctor. Let me get the doctor on the phone or on the computer. But I think there's a lot of things that we right now say, oh, that's the thing. I have to see the doctor for that. I think that's the AI gets better, you won't. And I'm not sure the AMA would agree with me, but I think that's good. I think if we can, you know, we make it better and cheaper and more convenient, that's great if it works.
- [23:21]
Henry
And, and that sounds great. And I. I'm with you. And yet I think one of the things that people are so frustrated about, particularly the United States healthcare system, is that it really is unequal. You've got much better healthcare for people with a lot of money who work for big companies that have great insurance plans. And it's really rough. And so I think one of the fears that people have when you talk about no, it's going to expand access, it's going to be great, is sure, if you're rich, you get to see Dr. Bob Wachter. If you aren't, hey, the AI doctor will see you anytime. And is that a concern to you?
- [23:55]
Dr. Robert Wachter
Of course. But I actually am more concerned in the other direction. I think that today if you don't have resources trying to get any healthcare, trying to get in to see a primary care doctor or mental health professional is so unwieldy and impossible and inconvenient that I think if, you know, when I was back in my high school debating team, if I had to take either side of the argument, this will improve access to people who don't have means or it will harm access. I would take the improve access side. I think over time, as the tools get better, sure, the person who can afford it or has complex needs sees a real live accountant to get their taxes done. But a lot of people use TurboTax and fine, it solves important problems and does it less expensively. And the same thing is true for how you plan your travel. And I think if we get this right, and that's a big if, I mean, these are challenging questions. If we get this right, I think the ability to provide better access to people at a lower cost and therefore make it more accessible even to people that don't have resources. I would bet on this over the alternative, which is trying to figure out ways of getting doctors into, you know, into poor places in the country or into rural America or around the world. I think this has a greater possibility. And then over time it may be that the AI version of this is the preferred version. I could tell you now when I pick up my phone in San Francisco and say look and say, oh, there's an Uber two minutes away and a Waymo eight minutes away. I wait for the Waymo. I prefer it. I don't particularly want to talk to a human. I'll sometimes sing Springsteen in the backseat with my Nobody's looking and it's clean and it's nice. So I can actually see a world where this is actually quite good healthcare. But we've gotta be very careful. Obviously, if you have a really complex disease, I don't want the bot to tell me I have cancer. I'd like no one to tell me I have cancer. But if someone has to, I'd like it to be a human. I think we've got to figure out the relationship between the AI and the humans and get it right. But I think in terms of access and convenience and expanding good medicine to more people, I would bet on this. Adobe Acrobat Studio your new foundation Use media spaces to generate a presentation. Grab your docs, your permits, your moves. AI levels up, your pitch, gets it in a groove. Choose a template with your timeless cool. Come on now, let's flex those tools. Draft, design, deliver, make it sing. AI builds the deck so you can build that thing. Do that, do that, do that with Acrobat learn more@adobe.com do that with Acrobat
- [26:49]
Sponsor/Advertiser
Support for the show comes from Public, the investing platform for those who take it seriously. On Public, you can build a multi asset portfolio of stocks, bonds and options options and now generated assets which allow you to turn any idea into an investable index. With AI, go to public.com podcast and earn an uncapped 1% bonus when you transfer your portfolio. That's public.com podcast paid for by Public Investing Brokerage Services by Open to the Public Investing Inc. Member FINRA and SIPC Advisory Service by Public Advisors, llc. SEC Registered Advisor Generated Assets as an interactive analysis tool Output is for informational purposes only and is not an investment recommendation or advice. Complete Disclosures available at public.comdisclosures it's true that some things change as we get older. But if you're a woman over 40 and you're dealing with insomnia, brain fog, moodiness and weight gain, you don't have to accept it as just another part of aging. And with MITI Health, you can get help and stop pushing through it alone. The experts at MITI understand that all these symptoms can be connected to the hormonal changes that happen around menopause. And miti can help you feel more like yourself again. Many healthcare providers aren't trained to treat or even recognize menopause symptoms. MIDI clinicians are menopause experts. They're dedicated to providing safe, effective, FDA approved solutions for dozens of hormonal symptoms, not just hot flashes. Most importantly, they're covered by insurance. 91% of midi patients get relief from symptoms within just two months. You deserve to feel great. Book your virtual visit today@join midi.com that's join m I-I.com all right, so you
- [28:31]
Henry
talked about AI scribes, which is doctors don't have to sit there typing or taking notes anymore because the AI will do that. What else is really working and helpful already in medicine.
- [28:41]
Dr. Robert Wachter
Yeah, I'd say the things that are working and they're very real in my world. So at ucsf, when I'm on the wards now, as I'm going to be in a couple of months taking care of patients who are sick in the hospital, I will use an AI scribe. I will use a tool that summarizes your past record. There's a study a couple years ago, one out of five patients, their past medical record is longer than Moby Dick. So if you remember from medical school, that was 600 pages. The idea that in three minutes I have, before I see you, I'm going to be able to review that in detail is a joke. And so these tools are very good. They're not perfect, but I am not perfect trying to page through 700 pages in three minutes. So I will use that. I will use a tool that will draft your discharge summary, meaning that some patients have had a hospitalization that lasted three months. There too, they may have 200 pages of medical record, and then I have to write a summary of their discharge. I will push a little button and it drafts that. Is it perfect? No. Do we want me to edit it and make sure it didn't miss anything? Yes. Am I a little worried that if it was right 10 times in a row, I'm gonna be asleep at the switch when I'm supposed to be editing number 11? Yes. I mean, those are all real issues. As we think about systems that essentially what we call them as the human in the loop or the doctored loop. The fundamental dilemma we're going to have in medicine, and maybe everything else for a while, is the AI now is good enough to be very useful, but not perfect enough to be completely trusted. And therefore you're going to have, in my case, a physician looking at your note or the review of your records, or the draft discharge summary, or the diagnoses that are suggested to me by my AI tool and vet them and say, oh, that's right, or no, that's wrong. I need to. To weigh in and edit it. That sounds better than it is because humans will naturally be asleep at the switch over time if they begin to trust the technology. And then there's this other problem we call de Skilling, which is, over time, I'm probably less good at the thing than I was in the beginning. Is that a problem? I have de skilled on map reading. I have no idea how to read a map anymore. And that's fine. Doesn't matter. But there are parts of Medicine, where we're actually quite concerned about that. It's one of the big questions we are struggling with in medical education. As the tools get better, what are the skills that we can actually begin to jettison? But what are the skills that if we do that, in some ways, we created a death spiral where the technology is better than us, in part because we're not as good as we used to be.
- [31:06]
Henry
And you talk about that in the book, specific examples of things like pilots, where you have fly by wire and fly by computer airplanes, where effectively the pilots are so trusting of the aircraft that they don't know how to fly it or what have you. And I worry about that, that in what I used to do, which is research, where you study something and then you come to a conclusion and you write a research report. And now deep research, or for Claude or ChatGPT, can do that in six minutes. And I read it. And is it a staggering work of analytical genius? No. Is it better than I could produce in a month? You know, often. And what. What's missing is producing it in a month where you go out and talk to a hundred people and you come back and you have your view and you write it and you say, that's not quite right. You rewrite it again, and that's how I learn.
- [31:54]
Dr. Robert Wachter
Yeah. No, of course, I think it's probably true. For every novice profession, almost every endeavor is there is this progression from novice to expert. And at least all of us who went through it remember it being hard, painstaking work that partly was iterative and partly was, I tried this. I blew it. I got feedback. I then got better. I did it again. And the idea that we will bypass that and push a button and get the results and somehow not have lost something crucial is really troubling, I think, where the rubber meets the road. You know, medicine is a mixture of a whole lot of different skills and tasks. And you can see ways that AI might make this better in part because of its capabilities. You know, I often, if I'm looking at a medical student and they go in and talk to a patient, there's a decent chance I'm not watching them. What I then watch is they come out and they, quote, present the case to me. And that's a proxy for how they were talking to the patient. They may have present the case to me beautifully, but been awful talking to the patient. The AI now can listen to them talk to the patient and give them feedback in ways that are just more convenient than me sitting there for 45 minutes watching them. So there are things like that that are I think probably going to be quite helpful even in education and kind of the foundational steps. But in skills, for example, a surgeon, the AI can now watch the surgeon's hands and see, oh no, when you twisted your hand that way, that is different than I. You know, I've looked at a hundred expert surgeons and they all do it the other way. And I had a version of that when I took a golf lesson. It compared my swing to Tiger woods and Ernie L's. And obviously if they're different, they're probably right and I'm probably wrong. So there are those things that are very quite helpful. But the big question we're asking is a huge amount of medical education is me learning a ton of facts. You know, the differential diagnosis, meaning the possible diagnoses of various, you know, symptoms and signs and lab tests, abnormalities. I mean, it's years and years of learning a huge corpus of information and then learning a different set of skills. That is taking that information, doing this thing called clinical reasoning, which is how do you then take all of it, take it in, sift through it, and then come up with a diagnosis or a tentative diagnoses and then what are the tests you do to confirm it? These are really hard things to do. And there is a line of reasoning that says we're wasting our time. Why are we teaching the students all this stuff? All of the facts are all in the computer and even the clinical reasoning now can be done by AI and probably as good as them and over time certainly better than them. So we can bypass all those steps. I think that's a mistake now. And I think it just for the reasons I told you before that a first day medical student, as I watched my daughter go through med school, the first day medical student, she is an office, she's a patient essentially, by the end she's an expert. And there's something different about the way experts use AI knowledge tools. That is better. We know what to put in, we know how to interpret the results. And if we bypass that stage because they don't need to know that stuff anymore, then at some point they still novices and maybe the tools at some point get better enough that in fact they don't need to know anything or we don't even need doctors. But I think we're at this funky stage for the next foreseeable future where you really do want your doctor to be smarter than you are as a patient, to know the tools, to know the right questions to ask and the Other issue is time that. I mean, when Google first became a thing, we said, maybe doctors don't need to know anything anymore. They can Google everything. And a. That didn't work. It doesn't work very well. Doesn't work well enough. And baby, you know, I might have 15 questions if I saw you as a patient. I don't have time to Google 15 different issues. So I think that's the thing we're struggling the most. How much do we can we take off their plate? And the answer, I think right now is relatively little.
- [35:50]
Henry
And that makes a lot of sense to me. And when I hear the arguments that, okay, that's it, all entry level work is going to be done with and no one will ever get a job again. It just seems crazy to me, both looking at prior tech revolutions where there is a lot of adaptation and people do have to learn new skills, but you actually end up creating more jobs. And just the fact that it's not this. People who are senior are not going to know how to use AI. And to me, you've got an incredibly powerful tool. As a medical student, tell me if I'm wrong, but that tool is helping you and accelerating. And then you have to learn and you have to evaluate the tool and sure, you have to know what it's doing to think about that. So I don't. The doom thing where we're just not going to need humans anymore, it doesn't resonate for me at all.
- [36:36]
Dr. Robert Wachter
And some of it, of course, comes from fear, both of the unknown and fear of losing your job. And that's natural and understandable, I think, in medicine. And one of the reasons I was optimistic in the book is that I think for the foreseeable future, the jobs of doctors and nurses are safe in part because the unmet needs are so vast that even if these tools improve our productivity by 30 to 50%, you're still gonna need doctors and nurses. Now, there will be job losses, but the job losses will be, you know, we've had massive hiring of administrative people to deal with paperwork and bureaucracy. And some of those job losses, they're obviously sad for the individuals that lose their job, but, you know, they don't add much to the clinical care and they add lots of costs. So. And you know, so. But it's, you know, I think for the foreseeable future, you're going to need well trained doctors. And most of them, I think the vast majority of them will keep their jobs.
- [37:30]
Henry
All right, so you've brought up, We've gotten back to Diagnosis, which I want to talk about in a second. And then this concept of empathy. But you mentioned the Pit earlier, so I have to go back to roll back the tape a little bit, which is in this season of the Pit, there is a tiny little subplot where the new modern doctor comes in and saying to all the residents, you must have your AI summary done. And the residents are resisting that and saying, that's awful, it's a waste of time, it's wrong and so forth. But it sounds like from what you're saying on the discharge statements that, hey, you think that doctor might have a point. It's quicker.
- [38:02]
Dr. Robert Wachter
Yeah. I think the thing that they get wrong is the resistance among the young physicians who generally embrace this. They don't get wrong that. There are some of the older docs, in this case, it's Dr. Rabi, who think this is the end of the world, as they know there is that. But it's funny because there was a huge amount of resistance when the electronic health record came in, among particularly some senior doctors here. In some ways, the tools are so convenient and the learning curve is so short that even some of the folks that you might think would be haters have embraced it. They find it useful in their day to day life. And so the level of resistance is low. But yeah, it's fun to watch that as a subplot on the Pit. I think they've. Because they have to create drama. You know, the first time you use an AI scribe, the idea that it missed a crucial medication, which is what happened, that just doesn't happen. I mean, you know, the tools are more reliable than that. They're not perfect. We do have to look at them. But the way they framed it, the first time they use it, the young medical student says, oh, wasn't the patient on drug A? And it says the patient's on drug B. And the pro AI doctor says, well, you know, they're not perfect. They make mistakes, but they're still good. So that's TV for you.
- [39:20]
Henry
Yeah. And as you say, people make mistakes too.
- [39:23]
Dr. Robert Wachter
Yeah, exactly.
- [39:24]
Henry
Yeah. So, all right, so come back to diagnosis, because you started off by talking about how, I guess Watson and the others, in the early days, they focused on diagnosis. It's incredibly difficult, and they blew it. And so that set back AI in healthcare for 20 years or what have you, but now it seems like it actually is really useful. So talk about that.
- [39:43]
Dr. Robert Wachter
Yeah, I think it is. I think that, you know, often what I am doing as a doctor is taking a whole bunch of Vaccines and distilling them into what's going on here. And in the same way, if you use Claude as a writing coach, which I sometimes do, I think it's quite good. I'll say, you know, give me 10 possible metaphors for this thing. I'm trying to say that just helps unlock the brain. I will never count on AI to make a diagnosis, but putting into it sort of the relevant, the salient facts about a patient saying, what do you think might be going on? And then sometimes I'll say to it, what's the worst thing this could be? What would be the thing if I missed? It could really kill the patient. I find it to be a useful prompt for me. And they're just getting better and better. If you used it two or three years ago and you say these things are not ready for medicine because the stakes are too high and if it blows it, somebody could die because it hallucinates, they just hallucinate far less than they did. And if you connect the dots and say, if you go from 2022 to today and how much better they've gotten, what are they going to be like in two or three years? I think they will be extremely helpful. Where things will get much, much better is right now there's not much diagnosis support in the electronic health record. So I am taking the information electronic health record and then porting it into my phone or my computer. Over time that will change. Over time these tools will be embedded in the system that we use. And the same thing will be true of the patient portal, by the way, that you will use these tools will be embedded. So I won't have to say to it that it's a 92 year old patient history of lupus, who comes in with a fever. It'll know that it's actually in. I've just put all that data in this computer and I'll be able to say to it, you know, given this patient's past history and their current complaints, what do you think is going on? It will probably be monitoring the patient with me. Over time it may say, you know, doctor, you have a tendency. You've diagnosed this patient as having pneumonia, started the patient on what seems like appropriate antibiotics, and I notice the patient still has a fever and a high white blood cell count on day four. That seems unusual. And I've looked back at the last 100,000 patients with pneumonia and 98% of them did not have a fever and a high white count on day four. You might want to rethink the diagnosis. Is that gonna be helpful or annoying? I don't know. We've gotta figure out how to get that right. But I think we're gonna be moving to a world of much more robust, what we call clinical decision support. To me, this is the real promise of digitization, that the computer can not only be a huge filing cabinet, which is what it's been up until now, but a source of intelligence and insights and making my work more convenient. And then an analog of that happening for the patient too, where it knows your past history. You say, I woke up this morning with a runny nose or my belly hurts. It asks you the appropriate questions and then some of the time says, you know, you can take this over the counter medicine and I'm not worried about it or and with a very low threshold says you actually do need to see a doctor and I've given your entire history to the doctor and therefore made the whole thing more convenient than it was today. That should happen. I mean there are a lot of steps between here and there, but it seems like the right thing to do. And there's nothing technological that's going to get in the way of that vision.
- [42:56]
Henry
And as you have pointed out and as we see in other fields, AI is really good at vast pattern matching across humongous data sets. And I'm not a doctor, but I imagine that that's a lot of diagnosis is having seen things thousands and thousands of times.
- [43:13]
Dr. Robert Wachter
It is, although it's a little trickier than you would think that when I take your past history and distill it down into a one liner that turns out to be very useful. I was talking to a leader of this company called Open Evidence, which is this GPT for doctors that we all use. And I said, when are you going to embed that in the medical record so it can look at those 600 pages and have that be the foundation of its suggested diagnosis. And he said, said now soon we understand how important it is, but it's not a no brainer that that won't muck things up. In those 600 pages is a vast amount of information that I somehow in the magic of medical training have learned to discount. Things one through nine and thing ten I know is salient and goes in my prompt for the AI now he said there is a risk and we've got to work it through. There's so much, much data there that it will get distracted by it. And so it's just not. He said, we're being very careful not to just turn on the switch and say, ingest. All of this because you had a heart attack 40 years ago is different than you had a heart attack two years ago. And the computer theoretically will know that. And if I know it, there's no reason the AI couldn't know it. But a patient's chart may have 100 things like that and it's a little tricky for the computer to sift through that and figure out what are the salient points.
- [44:35]
Sponsor/Advertiser
This is advertiser content brought to you by Stonyfield Organic. Our cows, them going out to pasture, they love it. They're so excited to go out every day, they wait right at the door. In fact, we milk them and we just open up the laneway and let them just go right out to pasture. I'm Rhonda Miller Goodrich and I'm a dairy farmer in Cabot, Vermont. Our farm is Molly Brook Farm. We're an organic dairy farm and we are a supplier to Stonyfield. Molly Brook Farm has been in my husband's family since 1835. We started our organic transition in 2015. We had 53 acres of corn ground and of course we had to use herbicides and pesticides and the soil was dead really. For all intents purposes, we stopped growing corn and stopped using herbicides and pesticides and we seeded that down to perennial grasses. After that we begin to see biodiversity in that soil again. To be organic certified, our cows need to be in pasture at least 120 days. I think the organic practices really benefit our animals. You know, having good feed, good water, a nice light area. That's what's important to us and that's what's important to Stonyfield. Visit stonyfield.com to find Stonyfield Organic Yogurt near you.
- [46:08]
Dr. Robert Wachter
Avoiding your unfinished home projects because you're
- [46:10]
Henry
not sure where to start.
- [46:12]
Dr. Robert Wachter
Thumbtack knows homes so you don't have to to don't know the difference between matte, paint, finish and satin or what that clunking sound from your dryer is. With Thumbtack, you don't have to be a home pro, you just have to hire one. You can hire top rated pros, see price estimates and read reviews all on the app download today.
- [46:33]
Henry
Support for this show comes from Odoo. Running a business is hard enough, so why make it harder with a dozen different apps that don't talk to each other? Introducing Odoo. It's the only business software you'll ever need. It's an all in one fully integrated platform that makes your work easier. CRM, accounting, inventory, E Commerce, and more. And the best part, Odoo replaces multiple expensive platforms for a fraction of the cost. That's why over thousands of businesses have made the switch. So why not you try Odoo for free@odoo.com. that's o d o o dot com. So empathy again, you go to fear. Ah, I don't want to talk to a machine. I want to talk to a human being. And of course, human beings are even better at explaining to me. They listen to me, they look at to me. One of the most startling and amusing studies that I saw come out recently that of course you talk about in your book is that when, when they test AI versus doctors, AI wins in empathy. So talk about that.
- [47:40]
Dr. Robert Wachter
Yeah, pissed me off, that study.
- [47:43]
Henry
It wasn't you. You are certainly way above average.
- [47:47]
Dr. Robert Wachter
Of course, I live in Lake Wobegon. I mean, the average is the average. And we can all think of, on every skill, diagnosis, empathy, we can think of these iconic people who are, you know, I can't imagine the computer being better than them, although probably at some point it will be. But that's not, you know, we shouldn't be comparing ourselves to some mythical set of perfection because that's not medicine. And you know, the thing that bothered some doctors about those empathy studies is that the computer, the AI, has unlimited time. And often the responses the AI gave were longer in part because they have unlimited time. And the doctor's responses were like, don't worry about it. But that's true. I mean, the AI does have unlimited time and therefore it can give, if people judge the response that's longer and more detailed, to be more compassionate or more empathic, that is what they're judging. So yeah, as I was watching AI kind of reveal itself over the first couple of years of generative AI, it was like, okay, can pass a test, okay, can solve tough cases, but we still have empathy, we have communication. Nobody's going to hear they have diabetes or kidney failure from a bot. And then you saw these studies. I think that I tried to go into this with an open mind, recognizing that I'm human and therefore kind of rooting for the humans. But this is really about how do we deliver the best health and best healthcare at the most conveniently and at the lowest cost to the most people. And I don't think empathy is going to be the great dividing line anymore, that patients are automatically going to prefer talking to a human. Now I do think at the last chapter of the book, I kind of did a little almost test of myself. I was on the wards Taking care of very sick patients in the hospital for 10 days. I tried to use AI everywhere I could and really think about where it's going to go over time. And there was a whole lot of stuff that even as an AI optimist, I think we needed me, you know, talking to patients about their cancer, organizing really complex discharge planning, where I needed to talk to the family, to a social worker, to a case manager, to a discharge planner, all that kind of. Could I imagine AI doing that? Yeah, sort of, but not in the foreseeable future. But, you know, I think one of the tricky things for us, Henry, is going to be, you know, as I say, I think you can get a lot of your basic primary care from AI and get it effectively. And I think ultimately the AI will be given the opportunity to prescribe certain medicines if it can do it safely. There's already in Utah, they just passed a regulation where AI can refill your medicines without the input of a doctor. I think that's just the first step into a world where if the AI is good at diagnosing you, talking to you, knowing what you have the idea, that then. Oh, then now you've gotta talk to a doctor to get your antibiotic. That's silly. That can't happen. Or your hypertension in medicine. So I think that's gonna happen. The question is sort of how we do the triage, because there are gonna be people that have multiple complex problems or really complex social needs, or English is in their first language. Obviously, that's not an insurmountable problem for AI, but adds complexity. You know, the tech companies are very good at building tools for people that have means, whereas when I take care of patients, many patients are homeless or have absolutely no means at all. All these things, I think, add levels of complexity, but I don't think they're insurmountable. I think we've got to figure out systems where the AI does what it does really, really well. And I don't think communication or empathy are off the table. I think it can do those things quite well. And that there is something in a triage protocol that says you see a doctor for this thing because your needs are this complex, or what the doctor needs to talk to you about is really, you know, is. Is. Is hard stuff. You know, the devil's in the details. How we're going to have to try to figure this out. And it's not for the faint of heart. It's going to be hard to get it right.
- [51:42]
Henry
But that sounds like a lot of the Reality of what's happening in other industries and disciplines too, is, is the way the argument is held up as it's human or AI, one or the other, it's binary. And, and in fact, what's happening is it is coming together and we're using AI and it's starting to change the way we do things. But in a lot of cases, it's. And so it's the doctor and AI, the therapist and AI. And on the empathy study, I would say, you know, you go in and you learn that you have cancer and it's very serious. Sure, it's great if you can get five to 10 minutes with the incredibly busy doctor who gets paid the same whether they spend a minute with you or two hours with you and has to go on to other people who need them. Yeah, it's great if they're really empathetic in that, but you're going to want to talk about it for days with everybody. And AI is really good at that, 24 hours a day, just being there. Hey, here's what my doctor said. Did I miss anything or what have you? So, to me, it sounds the way you're framing it, which is. It's good news. This will expand care, it'll expand empathy. All right, so to finish up, take us forward five years based on you've been watching AI closely for the last three or so as it's come out, or the last four. Where do you think we'll be be in five years? What'll be different? What'll be better, what'll be worse?
- [52:56]
Dr. Robert Wachter
Yeah, I think people will still see doctors for, you know, for the bulk, the majority of their problems, but there will be increasing numbers of medical interactions that occur independent of the doctor's visit and the doctor's office. Some of that will be, I think we will begin in five years to have certain problems that are managed primarily by. By AI. You know, you're seeing versions of this now where you get your weight loss medicine by calling some 1-800-number and you really have no relationship with a person. All those things, I think are going to be increasingly automated. They're going to be pathways for patients to get basic care, preventive care, their vaccines through AI, without ever seeing a doctor. I think the patient portal that you will have on your computer will be much, much more robust and much more trustworthy that you'll be able. It will know about you. And so when you have a test result, it will be able to say to you, here's what it means, not in general, but for you given the medicines you're on, the diseases you have, I think the wearable thing is interesting. I mean, as patients are wearing watches or rings that are giving healthcare data out there, but often in ways that are then not integrated with it, knowing about you and the medicines you're on, your past history, I think all of that will come together in a more meaningful way. So our ability to kind of monitor how you're doing with your Parkinson's or your blood pressure, your heart failure, will increasingly not depend on you coming in and see the doctor in the office, but will be through data streams that are being monitored through AI. I think the robotic thing will be interesting. Five years may be a little bit short, but I think in the 5 to 10 year range you're going to be seeing surgery that is markedly assisted by AI. Most procedures where AI is doing a significant chunk of it, there's still going to be a human operator in part because people will insist on that. The legal system, the billing system will probably insist on that. But I don't think it will be revolutionary in five years. But I think you're going to see more and more things done by AI, more and more patients able to get the care that they need by AI. And one of the things I think we have to account for is if I'm a primary care doctor, I look at my schedule and I see easy patient cholesterol, blood pressure and patient number three is the patient from hell. They've got 32 problems and are are 97 medicines. And if my entire schedule is that because all the low hanging fruit has been taken out, we better account for that because that's become a harder. We tried to make it an easier job because we'd taken this stuff off your plate. But if the hamster wheel is spinning just as fast as it was before, but now every patient is massively complex. You could see how that could go off the rails pretty quickly. So there are a lot of these kinds of issues. But yeah, I think it's gonna be net good. The thing I worry about the most is deepfakes. The thing I worry about the most is the idea that anybody can be made to look like they're saying anything. And how can patients figure out what to trust and what not to trust? I think what it means is you're gonna have to find a health system where, you know, I'm seeing this system, ucsf, even though I may not go and see a doctor very often. Cause I'm getting a lot of my care outside of the information I'M getting is from that system that I trust. Cause otherwise was things can be hard to go on YouTube and know what's real and what's not.
- [56:16]
Henry
Yeah. As it already is. And I do think that that ability to have some sense of that is an incredibly important skill to just surviving in the world these days. Dr. Wachter, thank you so much. Your new book is a Giant How AI is Transforming Healthcare and what It Means for Our Future. It's an excellent book recommended to everybody and we are grateful for for your joining us. So thank you so much.
- [56:40]
Dr. Robert Wachter
It was a great pleasure. Thanks for having me.
- [56:49]
Henry
Support for this show comes from Odoo. Running a business is hard enough, so why make it harder? With a dozen different apps that don't talk to each other. Introducing Odoo. It's the only business software you'll ever need. It's an all in one fully integrated platform that makes your work easier. CRM, accounting, inventory, e commerce, and more. And the best part, Odoo replaces multiple expensive platforms for a fraction of the cost. That's why over thousands of businesses have made the switch. So why not you try Odoo for free@odoo.com that's O D O O.com
- [57:30]
Dr. Robert Wachter
Adobe Acrobat Studio, your new foundation. Use PDF spaces to generate a presentation. Grab your docs, your permits, your moves. AI levels up, your pitch gets it in a groove. Choose a template with your timeless cool. Come on now let's flex those tunes Draft, design, deliver, make it sing. AI builds the deck so you can build that thing. Do that, do that, do that with Acrobat. Learn more@adobe.com do that with Acrobat.