Loading summary

Transcript34 lines

[00:02]
A
Lemonada.
[00:11]
B
Welcome to that Can't Be True, a show that sorts fact from fiction, especially on issues impacting our health. I'm Chelsea Clinton. There's a question that's quickly becoming part of everyday life for many people. Can AI help us make sense of our health and for us as a society? What happens when people start turning to AI for health advice, health guidance, maybe even diagnoses and suggested treatments? Today we're talking with Dr. Ashwin Vasan about how AI is already shaping the way that we experience care, the way that providers deliver care, the way insurance companies think about what care to cover and what all of that means for trust, access, accountability, and the future of Public Health. Dr. Vasan is a physician, a public health leader, a senior fellow in health policy and Global affairs at Yale University, and the former New York City Health commissioner. I am really grateful that he's here with us today. Dr. Vasan, Ashwin, welcome.
[01:17]
A
Thank you for having me. It's great to be here.
[01:19]
B
So I thought we'd start with AI because you and I have talked about that before in different places and spaces and it seems like such a inevitably kind of ubiquitous topic these days in anything related to medicine or public health and then kind of move into other areas. But I thought to get us started today, here's a clip from a couple of weeks ago on the Today show. We always have a that Can't Be True segment and sometimes it comes later in the show, but I thought we'd start with it because I think it will help us get to just how quickly things have changed, but also maybe how little has changed when we think about AI and health. Start telling CHAT all of your symptoms. It's the new way to get medical advice quickly and for free. Use Claude to fix your sleep problems. Ones AI tools providing advice on everything from symptoms to possible treatments. OpenAI says more than 230 million users globally ask ChatGPT questions every week about health and wellness. But how accurate are these tools when it comes to helping everyday users with their medical problems? So, Ashwin, 230 million just on one platform actually wasn't surprising to me. Is it surprising to you? And what do you think we would have heard if we'd listened to the rest of the segment? How much are these AI large language models actually helping people?
[02:42]
A
I think that's the, what do they call it, the $50,000 question or the $2 million question? I always forget the dollar depends on your game show. Yes, I always forget the game show's name, but that's the central Point I'm not surprised. To answer your first question, you know, a recent Kaiser survey showed that one out of three Americans are regularly using AI chatbots for healthcare questions for their own clinical care. And when I. When you go deeper and you start to look at the research and even just what people are saying in clinics, in hospitals and patients that you cross paths with as a provider, they're solving for a range of kind of, let's call them, pain points that everyday people feel with the American healthcare system. That same Kaiser survey showed that about 20% of those surveyed were using AI for simply the means of cost. They find it care unaffordable. They want to get some answers without the fear of financial hardship. This is in part why we focused so much on medical debt in New York City and why it's such an issue across the country, because people are facing this inordinate financial burden of simply just getting the care that they need. So that's one important set of issues. I think we must be honest about issues around things like speed. People want answers and they want them immediately. You know, and that may be an instinct that's. It is an instinct that's very human, but it's certainly not one that we can always satisfy in American medicine. Frankly, no medical system can satisfy the kind of immediacy that AI is able to bring. So it's that speed. I think in some cases it's issues of access. Depending on where you are in the country, the certain types of doctors, certain types of specialists just might be hard to find. And the fact that you can go and ask some questions to what I would describe as nothing more than kind of like a learned intermediary of some kind, then I think it's not surprising that people are turning to this. I mean, when you look at some of the other use cases, it, it also is an issue of empowerment. You know, if you're dealing with the chronic illness or God forbid, a catastrophic illness, and you're going into these doctor's appointments feeling kind of helpless. We are hearing a lot about patients using AI to prepare for those in person, doctor's visits and so on and so forth. I mean, I think every single one of those use cases is something where the analog system, the human based healthcare system, is not delivering what people are wanting and needing in some cases. And so it's not surprising to me that folks are using it in this way. And I guess the question is then, all right, how do we balance the issues of safety and accuracy? You know, there's some research that shows that, you know, these LLMs misdiagnose things like sending a person to the emergency room over half the time they will continue to get better, no doubt. But that's a pretty risky and cautious that that is an endeavor that's worth some caution. Right? Like I would caution every single patient of mine and every single person I'm talking to to use AI, but to use it as an intermediary, use it as something to build up your learning, to build up your knowledge and to take to your conversations with an actual human being, an actual provider. Because this is not yet technology that's able to replace a provider. And that's a really important, I think, ground truth, at least today that's what I would counsel people where we are in a year from now, I don't know, maybe these models do get strong enough, accurate enough to where they become more valid and reliable and reproducible over time. But today I would advise a hefty dose of caution.
[06:44]
B
You know, Ashwin, I think while thus far we've talked about kind of individual people interacting with the LLMs and the chatbots, we actually know that there's AI permeating multiple parts of the healthcare system. And some of what we could think of as AI that's actually been around for decades in helping with clinical analytics. So I guess, you know, maybe just for someone who's newly aware of AI in the last year or two, what do you want people to know about how AI might already be showing up in insurance portals, you know, provider decision support tools, hematology or radiology reads. Just what should people know as they're trying to navigate kind of in this new old world?
[07:33]
A
It's a great question and I think your point is exactly right. We talk a lot about AI as if it was turned on in November 2022 or so when ChatGPT was released. It's been there. Yes. It maybe hasn't been in these learning large language models, but there's been predictive analytics, machine learning and forms of AI in healthcare for years, largely in places that patients can't see. And so the way that I would think about it for the everyday person is there's kind of three principal uses of AI in everyday care. One is person facing, consumer facing, the everyday person facing. It's the Apple Watch I'm wearing and the software enabled in that. It's the app that you're using that connects to your medical record and that connects to your hospital system or healthcare provider. Most of these tools are underpinned by some layer of artificial Intelligence that is both learning from the data that's coming in, but also predicting and prompting and pushing out recommendations and advice on the basis of what it knows and is learning across all of the information that it's scraping from the web, from the online world. A lot of the growth is then in the second part, provider facing, which is how is AI being used to support the everyday performance of doctors, nurses, social workers, paramedical professionals, EMTs, everyone from soup to nuts in the healthcare system. And that's some pretty powerful stuff because what we are, while AI makes mistakes and it most certainly does and should not replace the judgment of a doctor or a provider. As I, as I mentioned, it's really good in many, many use cases. It's very accurate in many, many use cases. And certainly for a person who's trained and licensed to practice medicine or any other related field, having a co pilot with you, having someone in the passenger seat providing you with information that helps advance your thinking, is a very powerful diagnostic reasoning tool. It has a very powerful opportunity to improve the accuracy and the quality of your clinical care. And so we were seeing a lot of applications around that. We're seeing it with nurse practitioners in primary care. We've leaned heavily on nurses and non physician clinicians to extend primary care, particularly in rural areas and underserved areas, because there just aren't enough doctors being trained one and two going into primary care professions. And so we're seeing a lot of applications of AI to extend that care and improve the care that's being delivered already at the front lines. Two, it's things like radiology, as you mentioned, and diagnostics, lab diagnostics, and particularly imaging diagnostics, where the technology is so advanced in many ways it can even better the human eye in terms of what it's able to find on an mri, on a CAT scan, on an X ray. And that's a very powerful application that's being used already all around the country. And then there, of course, there are applications of AI in the provider context that help with provider efficiency. Right. We doctors and providers have complained for years about the paperwork burden and the time spent on care that has very little to do with human beings and the practice of care, but has everything to do with the administration of care. AI is already being used to reduce that paperwork burden, whether it's through charting or whether it's through, you know, sending through forms, referrals, insurance, all of that. One of the most powerful tools that's already being used in most advanced health systems are ambient scribes. We've all had the experience of going to a doctor, and the doctor hardly makes eye contact with you because they're charting everything you say. They're typing often as you're speaking. That's a pretty disorienting interaction. And it's certainly not the reason I got into medicine. I got into medicine to look someone in the eye, to accompany them through illness, to be a source of counsel and support and advice, and hopefully wisdom, but if nothing else, accompaniment in some of their most difficult times. And we've digitized care to such a point where doctors and providers are spending so much time at their terminals, entering data and reading data, that they've gone away from the human aspect. And now AI has come in and said, well, actually, we're going to listen to you as you speak to the patient and chart for you. We're going to populate your note, your record, with information that we glean from this conversation. Is there risk associated with that? Absolutely. Could things be missed or confabulated or misconstrued? Are the subtleties of human communication captured by the model? Probably not at this stage, but it is a good scaffolding. And my hope is that technology like this can actually get doctors like me back to the practice of human medicine and the humanity and authenticity of care. And then lastly, I would think about all the stuff you don't see as a patient, which is the AI being applied in the administrative context. Most people know we spend way too much on healthcare and get not as much as we should for that spend. But what a lot of people don't know is that about a third of what we spend, we spend about $5 trillion. Almost a third of that is spent on healthcare administration. Healthcare administration, which means insurance companies sending you bills, providers sending bills to insurance companies to get paid, pharmaceutical benefit managers negotiating drug prices and negotiating with you directly on what you pay for your costs. It is the, you know, not to use a fancy word, but it's the arbitrage of healthcare where everyone's seeking to pull out profit and pull out rent and kind of pointing the finger at everybody else. And so there's a kind of colloquial phrase being used now called the bot wars, which is to say that each of these different actors who are incentivized, perfectly incentivized, to do exactly what they're doing, are using AI to just do more of it and do it better. Which a good example is we've all experienced prior authorizations, right, where either you as the patient or the provider, I've experienced it Both sides, where you have to actually call someone at an insurance company to get the care that you need to get the care that either I'm recommending for a patient or that I need as a patient and that is recommended by my provider. And you have to negotiate with some faceless, nameless doctor working for an insurance company to justify something that was between you and your doctor. It's an awful experience for anyone who's been through it, which is most of America. Now these prior authorizations are being automated and you might guess that we're seeing rates of denial go up as a result. Right. Because what is the incentive of the insurance company but to provide less care, to pay less on the provider side, we're seeing this software used to increase billing. Most Americans may not know that every time you go to a doctor, we as providers are obliged to attach a code to what we're doing. Whether it's a diagnosis or a procedure or a test that we're running. Everything has a code with it. And then those codes get billed. These are standardized national codes that then get billed to insurance companies with AI. Those the volume of codes being sent for billing has gone up. So insurance companies are really feeling the pinch in their side of the world. And so everyone's trying to engage in this circular firing squad in these early days of AI deployment. So I guess the take home point is that for everyday listeners is that the system is kind of working in its perfectly inefficient way and frankly, unsatisfactory way for most people because there are very little in the way of guardrails. There are very little in the way of either rules or frankly leadership to say this is why we're doing this and this is the strategy behind AI in America. This is, we've complained about health care costs for so long and our health outcomes for so long, we're going to deploy it strategically to actually help or not. But that's going to require a real public conversation. It's going to require that communities that are often left behind are a part of that conversation so we don't just reinforce the disparities and inequities that we see. And it's going to require real leadership from elected and appointed and frankly community leaders to say we need to do this to make Americans healthier and safer and do so at a more sustainable, in a more sustainable and effective and equitable way. We're a long ways away from that, it would appear.
[17:00]
B
You just mentioned though, the communities that too often have been kind of left out, even failed by medicine and public health alike. Talked about kind of rural patients. I also think about historically black patients in black communities, Native American patients in tribal communities, certainly also women who were largely excluded from clinical trials. And so much of the grist of data that has been fed into these LLMs. And so I have been concerned, admittedly for years, as I know you and others have, that the models, while clearly remarkable in so many ways, don't have a real reflection of Americans. And I know this concern is reflected you in many other countries around the world. What do we do about that? I know many of the large language model companies themselves have been quite open about how they've tried to supplement with synthetic data and otherwise. And what should people just know about what some of the gaps in the models might be and what should we be advocating for?
[18:05]
A
Yeah, this is a very important question that I think doesn't get enough attention in this sort of race to the top mindset that we have. Let's go fast and far and move fast and break things. However, you want to talk about what I would describe as a somewhat callous and careless way of putting forth a pretty transformative industrial technology. I think there's two ways in which we need to think about representation or representativeness of the process of AI's role in healthcare and beyond. One is representativeness of the data, which you alluded to. I don't think that there's enough that can be done quickly enough through synthetic data means to get our substrate that these models are trained on to be reflective of America and its diversity and its complexity and its inequity and its. Even its federalism. You know, I mean, no one's even overlaid the policy environment in which care is delivered and how it can affect access and quality and outcomes. So I think we have a lot of work to do then to make an intentional effort to deploy these technologies in partnership with communities now. And what I mean by that is really just setting aside an entire area of work that is about technological education access, including cost, adoption and improvement of these technologies using the lived experiences, the interactions with and the, of course, the data, with permission from people who have historically never been at the table when these, either these technologies are designed or deployed. And I don't see any signs that that's going to happen again without some industrial leadership. We need to start treating these technologies as infrastructure and as an industrial in need of an industrial architecture that can only really be designed through some combination of federal and state leadership and some market incentives that create the capital conditions for this to happen. Right. So in other words, I totally agree with that. Crying out for a roadmap. We're cr out for a strategy for AI's design and deployment. And the second piece is really deployment and diffusion of technologies. When I look at a place like China, which has really taken AI as infrastructure and in a very top down way said these are its applications across the different key sectors of our economy and our society, including healthcare, what is clear is that they know what they're trying to solve for. And what AI is lacking, just as healthcare is lacking and health is lacking in this country is an objective function, which is a mathematical statement. Right? That term comes from math, but it is essentially the thing for which you are solving that can be measured and tracked over time and against which you can decide whether the work you've done to get there is correct or not correct.
[21:18]
B
Ashwin, do you have a view about kind of what you really think AI should or shouldn't do? Or is that just an impossible question to answer given the pace of change in the different realities confronting patients in rural upstate New York versus here in New York City where you were our health commissioner.
[21:40]
A
I'll take it back to where I started our conversation, which is I believe AI should be used to solve for any parts of the health care system that are not delivering for people right now. So AI predominantly should be, you know, it's those same things. It should be used to reduce cost, particularly cost in the pockets of people, everyday people. It should be used to improve access to care, extend care, provide care where care is either hard to find or there are inordinate delays or barriers to care. I think it should be used to enhance speed of care. You know, I just went to my annual visit, I'm waiting on my labs. You don't think that there's like a tiny voice inside of me that's like, I wonder what those labs are going to show. And you know, I'm a little nervous about that. Imagine doing that with a more severe, catastrophic, acute set of issues. We need to be able to deliver information in a timely way. I also think that empowerment and information and reducing this information asymmetry between medical providers and everyday people I think is a good thing. So those are the ways in which I think we should be focusing AI. Now, those aren't the things that have the most economic value or are necessarily
[23:00]
B
the sexiest, candidly, or they're not the sexiest. Doesn't sound the most exciting.
[23:03]
A
It's really not. But it is important, I think. And who gets to Step in and say, hey, this is the direction in which we want the market to go. This is the direction in which we want you to build for the good of the American people, for the good of society, for the good of the globe. We need you. We don't just want you, we need you to build in these ways. Well, I can't think of another model of doing that in our economy and in our society other than the government stepping in and saying, we care about this, we want to engage people in this, we want to incentivize and resource this kind of building. And here's why, here's what's at stake. If here's the cost of inaction, here's the cost, here's the future scenario, if we just let this technology run loose and here's the opportunity on the table, if we come together and actually try to do this the right way. I'm not confident that's going to happen in the next two years, but I think whoever is picking up the mantle in 2028 should do that. Like, this should be job number one.
[24:04]
B
I totally agree.
[24:05]
A
Get people together. Let's come up with the national commission, let's design the future of AI in healthcare and health and public health, and let's do it with the national mission in mind to make us healthier, safer, more competitive, more secure and more equitable in the process.
[24:21]
B
I totally agree.
[24:23]
A
When that becomes clearer, all these point solutions, these building, these questions we're asking about this chatbot or this chatbot become clearer, I think. And right now we're in the early innings. And so I hope that we'll be able to get some leadership in this space.
[24:38]
B
I agree. Ashwin. Before we move to our Fact or Fiction segment, where I throw out various claims and you tell me whether they're fact, fiction, or there's nuance, I wonder if there are other either general experiences or specific experiences you had as our health Commissioner here in New York City that you think are salient to this conversation or just really help shape how you think about public health going forward.
[25:10]
A
Yeah, I won't go on too long, but a lot of what I learned about, or let me say it this way, why did I get into public health in the first place? Because I was inspired by people working on problems that I felt were both solvable and really required extraordinary ambition and collaboration. I started my career in HIV and trying to get HIV meds around the globe at a moment when that was a very hard task and we managed to do it. You and the Clinton foundation played A huge role in that, and then moved on to things like mental health, another sort of crisis, and in many ways like a second pandemic that has emerged both before and after. Covid obviously worked on Covid, but it's these singular challenges that I think I get very inspired by and I want to spend my time on. And so when I became health commissioner, that was my first question, was like, what's our challenge now? I came in, Covid was still raging, but, you know, we could maybe see the end in sight, what was going to be the next set of challenges? And we identified two main things. One was our falling health spans and lifespans and the fact that not only had Covid been this major insult to how long and how well we live, but that we had experienced basically flatlining in the decade prior due to a whole host of things which we don't need to get into today. But I felt like that was something to organize around, and particularly because I was so inspired by public health leaders like Tom and others and Mike Bloomberg and Josh Sharfstein and many others around this country who had made life expectancy this central organizing function, as I described, an objective function of our. Of whatever system they were in control. So I made lifespan and healthspan the objective function of New York City. We passed it into law with healthy nyc. We established really clear numerical goals, not just for lifespan and health span, but for preventable mortality across the leading conditions. And I'm pleased to say that we not only hit that goal, but we experienced really steep drops in things like overdose deaths, obviously COVID deaths, but even across heart disease and suicides and screenable cancers, which had been quite stubborn for some time in this city. So I learned a lot about what that could be in a local environment where you don't control all the levers. I didn't control hospitals. I didn't control all of the health care budget. I was running a public health agency. But what I had was data. What I had was community engagement. What I had was policy on my side. And what I had was a pulpit. And I was able to kind of. We were able to kind of organize people around it, but with more levers at my disposal or anyone's disposal, you can do more. That I learned a ton from the other was mental health and how essential that is to bring into the center and the mainstream of our public health discussion, like health. I think mental health tends to be a little bit hard to crystallize for people in terms of what we're solving for and why. But I Think with how much our teens are struggling and young people are struggling. And what we've seen over the course of the last basically 15 years since the advent of smartphones and social media, I think we can say pretty clearly that not only is it a crisis, but it is something that public health needs to increasingly organize around. So we made that central to our work in terms of suing the social media companies, issuing executive orders around social media use, and of course, good research, locally driven research about social media and digital health and well being and then affordability, which is on everyone's mind. We spearheaded the relief of $2 billion of medical debt in our city, which is still ongoing. It seems like an easy thing to do, particularly when you're buying debt off of the secondary market. But it's what people kept telling us was a huge pain point, particularly coming out of COVID was that my bills are racking up. I delayed my care throughout the pandemic. Now I'm going back, I'm sicker than I thought. I can't pay these bills. I lost my job, I lost my insurance. Medicaid's not paying as much as it
[29:17]
B
used to, which is only going to get worse.
[29:19]
A
It's going to get worse. So, you know, medical debt relief was a huge thing in affordability and then hospital price transparency. You know, I think Americans are getting increasingly turned on to the role that hospitals are paying. I think we know what insurance companies are doing. We've talked a lot about what pharmaceutical companies do, good and bad, but we haven't talked a lot about what hospitals do to drive up the price of care. And one of the ways we thought about addressing this was to bring some sunshine to that and to transparently publish, mandate the transparent publication of hospital pricing in New York City just so that people can make sense of the fact that they're paying $5,000 for chemotherapy at one place and $25,000 at another place for the very same drug. Why? So those are the kinds of things that I learned. And so when I think about AI, I start with this first order question. What's the problem we're trying to solve for? And does technology actually have a role to play in solving it or not? I mean, I think there are fundamental problems that are human problems. I don't think AI is going to solve our coverage problem. I don't think it's going to solve for the fact that 30 million people don't have health insurance and that more will lose insurance over the next years to come. AI is not going to solve that. Right. But we need to have a kind of conversation where we're thinking about it much more strategically. I agree.
[30:39]
B
Well, Aswin, thank you for all your time. Before we leave, though, I am going to throw out a few claims. You're going to tell us.
[30:45]
A
No, I said it wasn't going to be long winded.
[30:46]
B
No, no, no. Whether they're fact or fiction, some about AI and some not, some we've already touched on, but I want to put a finer point on them. AI is already reducing healthcare costs. Fact or fiction?
[31:01]
A
Fiction. And in fact, the Peterson Health Technology Institute has already looked at this. And year on year, healthcare costs are up 9% in the AI era. So absolutely not.
[31:12]
B
AI is, is 100% objective and neutral.
[31:17]
A
Oh, definitely not false. Yeah. Again, you know, whatever you put in is what you get out. And unless you can claim that the data it's training on is neutral, then the output's not going to be neutral. Which is why the representativeness point is
[31:33]
B
so important, I think worth asking, because you just made reference to it as you reflected on your time in New York City as our health commissioner. The biggest drivers of life expectancy are mostly outside the doctor's office. Fact or fiction?
[31:47]
A
Oh, 100. True. I mean, where we all learn this from is it's such an intuitive point like that. You just get that. Right. And paradoxically, it's something that I think the MAHA movement in its own strange way has touched upon the very basic human understanding that your health is shaped outside of healthcare and outside of the clinic and as a provider. I used to think about my patients in terms of the 23 hours and 30 minutes that they weren't with me and what was their life like. So that's a very. It's both a very lived and sort of intuitive point, and it's also backed up by data. Right. Sir Michael Marmont is sort of the godfather of social determinants of health, coining that term, because he realized, and he showed through large scale longitudinal population research, that frankly, most of health outcomes is shaped by the environment you're in. The social factors driving that environment and shaping that environment. Yes. Some by genetics and of course, by the behaviors that flow from your environment, your social factors, and some from your genetics. And only about 10, 15% is shaped by medical care. It's an important percentage, but it's not determinative. And if you were an alien landing in America in 2026, you would think that healthcare is what it's all about, because that's where the conversation has always been. Because it's something we can feel and touch and monetize. And there's people in white coats and there's buildings that we know that are hospitals. But if we really wanted to do health in this country, we would start at home, in communities, in schools and everywhere else.
[33:30]
B
But the hospitals, well, in Ashwin, that is so much of what we did, you know, 100 or 150 years ago. Right. Was we thought about clean water systems and we thought about what housing requirements should be and what neighborhoods should look like.
[33:45]
A
We're a victim of our own success in a way. You know, public health has really done wonders.
[33:50]
B
Yeah. Hopefully we'll learn to walk and chew gum at the same time. Ashwin, thank you so much for your time and it's just always such a pleasure and I always learn so much and I'm incredibly grateful.
[33:59]
A
I'm grateful, too. This was a lot of fun. Thank you for having me.
[34:01]
B
Thank you. You can follow Dr. Ashwin Fasan on LinkedIn and at his new website, ashwinvasan.com. thanks for listening. Talk to you next week. That Can't Be True is a production of Limonada Media and the Clinton Foundation. The show is produced by Katherine Barnes, mix and sound design by Johnny Vince Evans. Kristin Lepore is senior director of new content and Jackie Danziger is VP of Narrative and production. Maggie Kral Shore is our managing director of Partnerships. Executive producer is are Jessica Cordova Kramer, Stephanie Whittles, Wax and me, Chelsea Clinton. Special thanks to Erica Goodmanson, Sarah Horowitz, Francesca Ernst Kahn, Caroline Lewis, Sage Falter, Barry Leary Westerberg, Emily Young and the entire team at the Clinton Foundation. You can help others find our show by leaving us a rating and writing a review. And if you can think of someone who might benefit from today's episode, please go ahead and share it with them. There's more of that can't be true with Lemonada. Premium subscribers get exclusive access to bonus content when you subscribe on Apple Podcasts. You can also listen ad free on Amazon Music with your prime membership.

Summary6 min read

Podcast Summary:

That Can't Be True with Chelsea Clinton
Episode: How AI Is Quietly Reshaping Healthcare with Dr. Ashwin Vasan
Date: July 30, 2026
Host: Chelsea Clinton
Guest: Dr. Ashwin Vasan, physician, public health leader, Senior Fellow at Yale, and former NYC Health Commissioner

Overview: Main Theme

This episode explores how artificial intelligence (AI) is already changing healthcare in visible and invisible ways—from individual patients turning to AI for medical advice to insurers, hospitals, and providers quietly adopting automated systems. Chelsea Clinton and Dr. Ashwin Vasan break down where AI has impact, why so many people are now using it, what’s at stake for trust and access, and why leadership and strategic direction are urgently needed. Special attention is given to concerns about bias, equity, and what authentic, community-centered solutions might look like.

Key Discussion Points & Insights

1. The Rise of AI in Everyday Healthcare (00:11–06:43)

AI is rapidly becoming ubiquitous in health: Over 230 million people globally use ChatGPT for health and wellness questions every week ([00:31]).
Drivers for AI adoption:
- Cost: 20% of patients use AI to avoid financial hardship ([02:57]).
- Speed: People want immediate answers which human healthcare systems often can’t provide ([03:20]).
- Access: AI fills gaps where specialists or services are hard to find ([03:44]).
- Empowerment: Patients use AI to feel prepared and more confident in interactions with providers ([04:10]).
Cautionary Advice:

“I would caution every single patient… to use AI as an intermediary… but this is not yet technology that's able to replace a provider.”
— Dr. Vasan ([05:44])
Pitfalls: LLMs (large language models) often err—sometimes recommending ER visits too quickly—posing real risks ([06:28]).

2. AI’s Hidden Hand: Institutional & Systemic Uses (06:43–17:00)

Three uses of AI in healthcare:
- Patient-facing: Apps, devices (like Apple Watch), and portals that use AI for personalized prompts, reminders, and recommendations ([07:46]).
- Provider-facing: Clinical decision-support, diagnostic tools (especially radiology and imaging), and ‘ambient scribe’ technologies to reduce paperwork ([08:55]).
  - Example: “Charting eats up doctors’ attention, but an ambient scribe listens and transcribes, giving physicians more time for real human care” ([10:46]).
  - Risks: AI may miss nuances or make documentation errors.
- Administrative/Back-office: AI used by insurance, billers, and pharmaceutical intermediaries to optimize/automate tasks—often increasing denials or billing ([13:00]).
  - “There’s a colloquial phrase being used now called the 'bot wars' … each of these actors … are using AI to just do more of it and do it better.” — Dr. Vasan ([13:51])
Current shortcomings:
- The system is “perfectly inefficient and frankly, unsatisfactory for most people because there are very little in the way of guardrails … or frankly, leadership … a real public conversation is needed.” ([15:45])

3. Bias, Representation & Equity in AI Healthcare (17:00–21:18)

Underrepresentation in Data:
Dr. Clinton highlights that communities historically marginalized—Black, Native American, many women—are not well represented in the data that LLMs train on ([17:00]).
Synthetic Data is Not a Panacea:
Synthetic data might help but cannot quickly resolve the deep representational gaps or account for lived realities and policy differences ([18:20]).
Need for Structural Solutions:

“We need to start treating these technologies as infrastructure … only really designed through some combination of federal and state leadership and some market incentives.”
— Dr. Vasan ([19:48])
Strategic Purpose Lacking:

“What AI is lacking … is an objective function, which is … the thing for which you are solving that can be measured and tracked over time.”
— Dr. Vasan ([20:27])

4. Defining AI’s Purpose & Public Good (21:18–24:37)

How Should AI Be Used?
Dr. Vasan says AI must address where healthcare is currently failing:
- Reduce costs to patients
- Improve access, especially in underserved areas
- Speed up diagnosis and information sharing
- Empower patients, narrow information gaps
Who Should Lead?

“Who gets to step in and say … we want you to build for the good of the American people? … I can’t think of another model … other than the government stepping in.”
— Dr. Vasan ([23:03])
Call for a National Commission:
Both guests agree on the need for a national mission and commission to roadmap AI’s future in health for “healthier, safer, more equitable” care ([24:04], [24:21]).

5. Learning from Public Health Leadership (24:37–30:38)

Lessons from NYC:
As Health Commissioner, Dr. Vasan made lifespan and healthspan the city’s explicit goals, with measurable targets—leading to real drops in preventable deaths ([25:35]).
Mental Health as Core Public Health:
Teen mental health struggles and social media harm have become fundamental public health issues. NYC’s approach integrated lawsuit, policy orders, and research ([27:40]).
Affordability & Transparency:
NYC bought and relieved $2B in medical debt, and mandated transparent hospital pricing, helping patients understand wild cost variations ([28:50]).
Philosophy for AI:

“When I think about AI, I start with this first order question: What's the problem we’re trying to solve for? And does technology actually have a role to play in solving it?”
— Dr. Vasan ([29:59])

6. Fact or Fiction: Rapid-fire Myths (30:38–33:30)

AI and Healthcare Costs

Claim: AI is already reducing healthcare costs
Dr. Vasan: “Fiction. In fact, the Peterson Health Technology Institute has already looked at this … costs are up 9% in the AI era.” ([31:01])

AI’s Objectivity

Claim: AI is 100% objective and neutral
Dr. Vasan: “Definitely not false. … Whatever you put in is what you get out.” ([31:16])

Determinants of Health

Claim: The biggest drivers of life expectancy are mostly outside the doctor’s office
Dr. Vasan: “Oh, 100. True. … Most of health outcomes is shaped by the environment you're in. The social factors driving that environment…” ([31:47])
Chelsea Clinton: “That is so much of what we did 100 or 150 years ago … clean water systems, housing requirements, what neighborhoods should look like.” ([33:30])

Notable Quotes & Memorable Moments

On AI as a “Learned Intermediary”:
“Nothing more than kind of like a learned intermediary of some kind … it's not surprising people are turning to this.” — Vasan ([04:02])
On Systemic AI Use:
“We've digitized care to such a point where doctors and providers are spending so much time at their terminals, entering data and reading data, that they've gone away from the human aspect.” — Vasan ([10:37])
On Strategic Leadership:
“We’re a long ways away from that, it would appear.” – Vasan, on AI being used purposefully and equitably in healthcare ([15:45])
On Public Health’s Real Impact:
“If you were an alien landing in America in 2026, you would think that healthcare is what it's all about … But if we really wanted to do health in this country, we would start at home, in communities, in schools and everywhere else.” — Vasan ([32:37])

Timestamps for Key Segments

AI’s adoption and the $2 million question: 02:42–06:43
Three layers of AI in healthcare: 07:32–13:51
Bot wars, administrative inefficiencies: 13:51–16:45
Representation, bias, and the road ahead: 17:00–21:18
AI’s true purpose & call for a national mission: 21:39–24:37
NYC public health lessons: 25:10–30:38
Fact or Fiction segment (rapid-fire claims): 30:38–33:30

Overall Tone and Takeaway

The tone is both realistic and urgent—celebrating the potential of AI without hype, warning of its real dangers, and calling for intentional, community-centered, government-led direction. Both Chelsea Clinton and Dr. Vasan stress that while innovation is exciting, the path forward must be about solving concrete public health problems, advancing equity, and restoring the “human” in healthcare.

Summary prepared for listeners who want the depth and context of the discussion, without missing the original spirit, insights, or cautions voiced by the speakers.

Loading summary

Transcript34 lines

[00:02]
A
Lemonada.
[00:11]
B
Welcome to that Can't Be True, a show that sorts fact from fiction, especially on issues impacting our health. I'm Chelsea Clinton. There's a question that's quickly becoming part of everyday life for many people. Can AI help us make sense of our health and for us as a society? What happens when people start turning to AI for health advice, health guidance, maybe even diagnoses and suggested treatments? Today we're talking with Dr. Ashwin Vasan about how AI is already shaping the way that we experience care, the way that providers deliver care, the way insurance companies think about what care to cover and what all of that means for trust, access, accountability, and the future of Public Health. Dr. Vasan is a physician, a public health leader, a senior fellow in health policy and Global affairs at Yale University, and the former New York City Health commissioner. I am really grateful that he's here with us today. Dr. Vasan, Ashwin, welcome.
[01:17]
A
Thank you for having me. It's great to be here.
[01:19]
B
So I thought we'd start with AI because you and I have talked about that before in different places and spaces and it seems like such a inevitably kind of ubiquitous topic these days in anything related to medicine or public health and then kind of move into other areas. But I thought to get us started today, here's a clip from a couple of weeks ago on the Today show. We always have a that Can't Be True segment and sometimes it comes later in the show, but I thought we'd start with it because I think it will help us get to just how quickly things have changed, but also maybe how little has changed when we think about AI and health. Start telling CHAT all of your symptoms. It's the new way to get medical advice quickly and for free. Use Claude to fix your sleep problems. Ones AI tools providing advice on everything from symptoms to possible treatments. OpenAI says more than 230 million users globally ask ChatGPT questions every week about health and wellness. But how accurate are these tools when it comes to helping everyday users with their medical problems? So, Ashwin, 230 million just on one platform actually wasn't surprising to me. Is it surprising to you? And what do you think we would have heard if we'd listened to the rest of the segment? How much are these AI large language models actually helping people?
[02:42]
A
I think that's the, what do they call it, the $50,000 question or the $2 million question? I always forget the dollar depends on your game show. Yes, I always forget the game show's name, but that's the central Point I'm not surprised. To answer your first question, you know, a recent Kaiser survey showed that one out of three Americans are regularly using AI chatbots for healthcare questions for their own clinical care. And when I. When you go deeper and you start to look at the research and even just what people are saying in clinics, in hospitals and patients that you cross paths with as a provider, they're solving for a range of kind of, let's call them, pain points that everyday people feel with the American healthcare system. That same Kaiser survey showed that about 20% of those surveyed were using AI for simply the means of cost. They find it care unaffordable. They want to get some answers without the fear of financial hardship. This is in part why we focused so much on medical debt in New York City and why it's such an issue across the country, because people are facing this inordinate financial burden of simply just getting the care that they need. So that's one important set of issues. I think we must be honest about issues around things like speed. People want answers and they want them immediately. You know, and that may be an instinct that's. It is an instinct that's very human, but it's certainly not one that we can always satisfy in American medicine. Frankly, no medical system can satisfy the kind of immediacy that AI is able to bring. So it's that speed. I think in some cases it's issues of access. Depending on where you are in the country, the certain types of doctors, certain types of specialists just might be hard to find. And the fact that you can go and ask some questions to what I would describe as nothing more than kind of like a learned intermediary of some kind, then I think it's not surprising that people are turning to this. I mean, when you look at some of the other use cases, it, it also is an issue of empowerment. You know, if you're dealing with the chronic illness or God forbid, a catastrophic illness, and you're going into these doctor's appointments feeling kind of helpless. We are hearing a lot about patients using AI to prepare for those in person, doctor's visits and so on and so forth. I mean, I think every single one of those use cases is something where the analog system, the human based healthcare system, is not delivering what people are wanting and needing in some cases. And so it's not surprising to me that folks are using it in this way. And I guess the question is then, all right, how do we balance the issues of safety and accuracy? You know, there's some research that shows that, you know, these LLMs misdiagnose things like sending a person to the emergency room over half the time they will continue to get better, no doubt. But that's a pretty risky and cautious that that is an endeavor that's worth some caution. Right? Like I would caution every single patient of mine and every single person I'm talking to to use AI, but to use it as an intermediary, use it as something to build up your learning, to build up your knowledge and to take to your conversations with an actual human being, an actual provider. Because this is not yet technology that's able to replace a provider. And that's a really important, I think, ground truth, at least today that's what I would counsel people where we are in a year from now, I don't know, maybe these models do get strong enough, accurate enough to where they become more valid and reliable and reproducible over time. But today I would advise a hefty dose of caution.
[06:44]
B
You know, Ashwin, I think while thus far we've talked about kind of individual people interacting with the LLMs and the chatbots, we actually know that there's AI permeating multiple parts of the healthcare system. And some of what we could think of as AI that's actually been around for decades in helping with clinical analytics. So I guess, you know, maybe just for someone who's newly aware of AI in the last year or two, what do you want people to know about how AI might already be showing up in insurance portals, you know, provider decision support tools, hematology or radiology reads. Just what should people know as they're trying to navigate kind of in this new old world?
[07:33]
A
It's a great question and I think your point is exactly right. We talk a lot about AI as if it was turned on in November 2022 or so when ChatGPT was released. It's been there. Yes. It maybe hasn't been in these learning large language models, but there's been predictive analytics, machine learning and forms of AI in healthcare for years, largely in places that patients can't see. And so the way that I would think about it for the everyday person is there's kind of three principal uses of AI in everyday care. One is person facing, consumer facing, the everyday person facing. It's the Apple Watch I'm wearing and the software enabled in that. It's the app that you're using that connects to your medical record and that connects to your hospital system or healthcare provider. Most of these tools are underpinned by some layer of artificial Intelligence that is both learning from the data that's coming in, but also predicting and prompting and pushing out recommendations and advice on the basis of what it knows and is learning across all of the information that it's scraping from the web, from the online world. A lot of the growth is then in the second part, provider facing, which is how is AI being used to support the everyday performance of doctors, nurses, social workers, paramedical professionals, EMTs, everyone from soup to nuts in the healthcare system. And that's some pretty powerful stuff because what we are, while AI makes mistakes and it most certainly does and should not replace the judgment of a doctor or a provider. As I, as I mentioned, it's really good in many, many use cases. It's very accurate in many, many use cases. And certainly for a person who's trained and licensed to practice medicine or any other related field, having a co pilot with you, having someone in the passenger seat providing you with information that helps advance your thinking, is a very powerful diagnostic reasoning tool. It has a very powerful opportunity to improve the accuracy and the quality of your clinical care. And so we were seeing a lot of applications around that. We're seeing it with nurse practitioners in primary care. We've leaned heavily on nurses and non physician clinicians to extend primary care, particularly in rural areas and underserved areas, because there just aren't enough doctors being trained one and two going into primary care professions. And so we're seeing a lot of applications of AI to extend that care and improve the care that's being delivered already at the front lines. Two, it's things like radiology, as you mentioned, and diagnostics, lab diagnostics, and particularly imaging diagnostics, where the technology is so advanced in many ways it can even better the human eye in terms of what it's able to find on an mri, on a CAT scan, on an X ray. And that's a very powerful application that's being used already all around the country. And then there, of course, there are applications of AI in the provider context that help with provider efficiency. Right. We doctors and providers have complained for years about the paperwork burden and the time spent on care that has very little to do with human beings and the practice of care, but has everything to do with the administration of care. AI is already being used to reduce that paperwork burden, whether it's through charting or whether it's through, you know, sending through forms, referrals, insurance, all of that. One of the most powerful tools that's already being used in most advanced health systems are ambient scribes. We've all had the experience of going to a doctor, and the doctor hardly makes eye contact with you because they're charting everything you say. They're typing often as you're speaking. That's a pretty disorienting interaction. And it's certainly not the reason I got into medicine. I got into medicine to look someone in the eye, to accompany them through illness, to be a source of counsel and support and advice, and hopefully wisdom, but if nothing else, accompaniment in some of their most difficult times. And we've digitized care to such a point where doctors and providers are spending so much time at their terminals, entering data and reading data, that they've gone away from the human aspect. And now AI has come in and said, well, actually, we're going to listen to you as you speak to the patient and chart for you. We're going to populate your note, your record, with information that we glean from this conversation. Is there risk associated with that? Absolutely. Could things be missed or confabulated or misconstrued? Are the subtleties of human communication captured by the model? Probably not at this stage, but it is a good scaffolding. And my hope is that technology like this can actually get doctors like me back to the practice of human medicine and the humanity and authenticity of care. And then lastly, I would think about all the stuff you don't see as a patient, which is the AI being applied in the administrative context. Most people know we spend way too much on healthcare and get not as much as we should for that spend. But what a lot of people don't know is that about a third of what we spend, we spend about $5 trillion. Almost a third of that is spent on healthcare administration. Healthcare administration, which means insurance companies sending you bills, providers sending bills to insurance companies to get paid, pharmaceutical benefit managers negotiating drug prices and negotiating with you directly on what you pay for your costs. It is the, you know, not to use a fancy word, but it's the arbitrage of healthcare where everyone's seeking to pull out profit and pull out rent and kind of pointing the finger at everybody else. And so there's a kind of colloquial phrase being used now called the bot wars, which is to say that each of these different actors who are incentivized, perfectly incentivized, to do exactly what they're doing, are using AI to just do more of it and do it better. Which a good example is we've all experienced prior authorizations, right, where either you as the patient or the provider, I've experienced it Both sides, where you have to actually call someone at an insurance company to get the care that you need to get the care that either I'm recommending for a patient or that I need as a patient and that is recommended by my provider. And you have to negotiate with some faceless, nameless doctor working for an insurance company to justify something that was between you and your doctor. It's an awful experience for anyone who's been through it, which is most of America. Now these prior authorizations are being automated and you might guess that we're seeing rates of denial go up as a result. Right. Because what is the incentive of the insurance company but to provide less care, to pay less on the provider side, we're seeing this software used to increase billing. Most Americans may not know that every time you go to a doctor, we as providers are obliged to attach a code to what we're doing. Whether it's a diagnosis or a procedure or a test that we're running. Everything has a code with it. And then those codes get billed. These are standardized national codes that then get billed to insurance companies with AI. Those the volume of codes being sent for billing has gone up. So insurance companies are really feeling the pinch in their side of the world. And so everyone's trying to engage in this circular firing squad in these early days of AI deployment. So I guess the take home point is that for everyday listeners is that the system is kind of working in its perfectly inefficient way and frankly, unsatisfactory way for most people because there are very little in the way of guardrails. There are very little in the way of either rules or frankly leadership to say this is why we're doing this and this is the strategy behind AI in America. This is, we've complained about health care costs for so long and our health outcomes for so long, we're going to deploy it strategically to actually help or not. But that's going to require a real public conversation. It's going to require that communities that are often left behind are a part of that conversation so we don't just reinforce the disparities and inequities that we see. And it's going to require real leadership from elected and appointed and frankly community leaders to say we need to do this to make Americans healthier and safer and do so at a more sustainable, in a more sustainable and effective and equitable way. We're a long ways away from that, it would appear.
[17:00]
B
You just mentioned though, the communities that too often have been kind of left out, even failed by medicine and public health alike. Talked about kind of rural patients. I also think about historically black patients in black communities, Native American patients in tribal communities, certainly also women who were largely excluded from clinical trials. And so much of the grist of data that has been fed into these LLMs. And so I have been concerned, admittedly for years, as I know you and others have, that the models, while clearly remarkable in so many ways, don't have a real reflection of Americans. And I know this concern is reflected you in many other countries around the world. What do we do about that? I know many of the large language model companies themselves have been quite open about how they've tried to supplement with synthetic data and otherwise. And what should people just know about what some of the gaps in the models might be and what should we be advocating for?
[18:05]
A
Yeah, this is a very important question that I think doesn't get enough attention in this sort of race to the top mindset that we have. Let's go fast and far and move fast and break things. However, you want to talk about what I would describe as a somewhat callous and careless way of putting forth a pretty transformative industrial technology. I think there's two ways in which we need to think about representation or representativeness of the process of AI's role in healthcare and beyond. One is representativeness of the data, which you alluded to. I don't think that there's enough that can be done quickly enough through synthetic data means to get our substrate that these models are trained on to be reflective of America and its diversity and its complexity and its inequity and its. Even its federalism. You know, I mean, no one's even overlaid the policy environment in which care is delivered and how it can affect access and quality and outcomes. So I think we have a lot of work to do then to make an intentional effort to deploy these technologies in partnership with communities now. And what I mean by that is really just setting aside an entire area of work that is about technological education access, including cost, adoption and improvement of these technologies using the lived experiences, the interactions with and the, of course, the data, with permission from people who have historically never been at the table when these, either these technologies are designed or deployed. And I don't see any signs that that's going to happen again without some industrial leadership. We need to start treating these technologies as infrastructure and as an industrial in need of an industrial architecture that can only really be designed through some combination of federal and state leadership and some market incentives that create the capital conditions for this to happen. Right. So in other words, I totally agree with that. Crying out for a roadmap. We're cr out for a strategy for AI's design and deployment. And the second piece is really deployment and diffusion of technologies. When I look at a place like China, which has really taken AI as infrastructure and in a very top down way said these are its applications across the different key sectors of our economy and our society, including healthcare, what is clear is that they know what they're trying to solve for. And what AI is lacking, just as healthcare is lacking and health is lacking in this country is an objective function, which is a mathematical statement. Right? That term comes from math, but it is essentially the thing for which you are solving that can be measured and tracked over time and against which you can decide whether the work you've done to get there is correct or not correct.
[21:18]
B
Ashwin, do you have a view about kind of what you really think AI should or shouldn't do? Or is that just an impossible question to answer given the pace of change in the different realities confronting patients in rural upstate New York versus here in New York City where you were our health commissioner.
[21:40]
A
I'll take it back to where I started our conversation, which is I believe AI should be used to solve for any parts of the health care system that are not delivering for people right now. So AI predominantly should be, you know, it's those same things. It should be used to reduce cost, particularly cost in the pockets of people, everyday people. It should be used to improve access to care, extend care, provide care where care is either hard to find or there are inordinate delays or barriers to care. I think it should be used to enhance speed of care. You know, I just went to my annual visit, I'm waiting on my labs. You don't think that there's like a tiny voice inside of me that's like, I wonder what those labs are going to show. And you know, I'm a little nervous about that. Imagine doing that with a more severe, catastrophic, acute set of issues. We need to be able to deliver information in a timely way. I also think that empowerment and information and reducing this information asymmetry between medical providers and everyday people I think is a good thing. So those are the ways in which I think we should be focusing AI. Now, those aren't the things that have the most economic value or are necessarily
[23:00]
B
the sexiest, candidly, or they're not the sexiest. Doesn't sound the most exciting.
[23:03]
A
It's really not. But it is important, I think. And who gets to Step in and say, hey, this is the direction in which we want the market to go. This is the direction in which we want you to build for the good of the American people, for the good of society, for the good of the globe. We need you. We don't just want you, we need you to build in these ways. Well, I can't think of another model of doing that in our economy and in our society other than the government stepping in and saying, we care about this, we want to engage people in this, we want to incentivize and resource this kind of building. And here's why, here's what's at stake. If here's the cost of inaction, here's the cost, here's the future scenario, if we just let this technology run loose and here's the opportunity on the table, if we come together and actually try to do this the right way. I'm not confident that's going to happen in the next two years, but I think whoever is picking up the mantle in 2028 should do that. Like, this should be job number one.
[24:04]
B
I totally agree.
[24:05]
A
Get people together. Let's come up with the national commission, let's design the future of AI in healthcare and health and public health, and let's do it with the national mission in mind to make us healthier, safer, more competitive, more secure and more equitable in the process.
[24:21]
B
I totally agree.
[24:23]
A
When that becomes clearer, all these point solutions, these building, these questions we're asking about this chatbot or this chatbot become clearer, I think. And right now we're in the early innings. And so I hope that we'll be able to get some leadership in this space.
[24:38]
B
I agree. Ashwin. Before we move to our Fact or Fiction segment, where I throw out various claims and you tell me whether they're fact, fiction, or there's nuance, I wonder if there are other either general experiences or specific experiences you had as our health Commissioner here in New York City that you think are salient to this conversation or just really help shape how you think about public health going forward.
[25:10]
A
Yeah, I won't go on too long, but a lot of what I learned about, or let me say it this way, why did I get into public health in the first place? Because I was inspired by people working on problems that I felt were both solvable and really required extraordinary ambition and collaboration. I started my career in HIV and trying to get HIV meds around the globe at a moment when that was a very hard task and we managed to do it. You and the Clinton foundation played A huge role in that, and then moved on to things like mental health, another sort of crisis, and in many ways like a second pandemic that has emerged both before and after. Covid obviously worked on Covid, but it's these singular challenges that I think I get very inspired by and I want to spend my time on. And so when I became health commissioner, that was my first question, was like, what's our challenge now? I came in, Covid was still raging, but, you know, we could maybe see the end in sight, what was going to be the next set of challenges? And we identified two main things. One was our falling health spans and lifespans and the fact that not only had Covid been this major insult to how long and how well we live, but that we had experienced basically flatlining in the decade prior due to a whole host of things which we don't need to get into today. But I felt like that was something to organize around, and particularly because I was so inspired by public health leaders like Tom and others and Mike Bloomberg and Josh Sharfstein and many others around this country who had made life expectancy this central organizing function, as I described, an objective function of our. Of whatever system they were in control. So I made lifespan and healthspan the objective function of New York City. We passed it into law with healthy nyc. We established really clear numerical goals, not just for lifespan and health span, but for preventable mortality across the leading conditions. And I'm pleased to say that we not only hit that goal, but we experienced really steep drops in things like overdose deaths, obviously COVID deaths, but even across heart disease and suicides and screenable cancers, which had been quite stubborn for some time in this city. So I learned a lot about what that could be in a local environment where you don't control all the levers. I didn't control hospitals. I didn't control all of the health care budget. I was running a public health agency. But what I had was data. What I had was community engagement. What I had was policy on my side. And what I had was a pulpit. And I was able to kind of. We were able to kind of organize people around it, but with more levers at my disposal or anyone's disposal, you can do more. That I learned a ton from the other was mental health and how essential that is to bring into the center and the mainstream of our public health discussion, like health. I think mental health tends to be a little bit hard to crystallize for people in terms of what we're solving for and why. But I Think with how much our teens are struggling and young people are struggling. And what we've seen over the course of the last basically 15 years since the advent of smartphones and social media, I think we can say pretty clearly that not only is it a crisis, but it is something that public health needs to increasingly organize around. So we made that central to our work in terms of suing the social media companies, issuing executive orders around social media use, and of course, good research, locally driven research about social media and digital health and well being and then affordability, which is on everyone's mind. We spearheaded the relief of $2 billion of medical debt in our city, which is still ongoing. It seems like an easy thing to do, particularly when you're buying debt off of the secondary market. But it's what people kept telling us was a huge pain point, particularly coming out of COVID was that my bills are racking up. I delayed my care throughout the pandemic. Now I'm going back, I'm sicker than I thought. I can't pay these bills. I lost my job, I lost my insurance. Medicaid's not paying as much as it
[29:17]
B
used to, which is only going to get worse.
[29:19]
A
It's going to get worse. So, you know, medical debt relief was a huge thing in affordability and then hospital price transparency. You know, I think Americans are getting increasingly turned on to the role that hospitals are paying. I think we know what insurance companies are doing. We've talked a lot about what pharmaceutical companies do, good and bad, but we haven't talked a lot about what hospitals do to drive up the price of care. And one of the ways we thought about addressing this was to bring some sunshine to that and to transparently publish, mandate the transparent publication of hospital pricing in New York City just so that people can make sense of the fact that they're paying $5,000 for chemotherapy at one place and $25,000 at another place for the very same drug. Why? So those are the kinds of things that I learned. And so when I think about AI, I start with this first order question. What's the problem we're trying to solve for? And does technology actually have a role to play in solving it or not? I mean, I think there are fundamental problems that are human problems. I don't think AI is going to solve our coverage problem. I don't think it's going to solve for the fact that 30 million people don't have health insurance and that more will lose insurance over the next years to come. AI is not going to solve that. Right. But we need to have a kind of conversation where we're thinking about it much more strategically. I agree.
[30:39]
B
Well, Aswin, thank you for all your time. Before we leave, though, I am going to throw out a few claims. You're going to tell us.
[30:45]
A
No, I said it wasn't going to be long winded.
[30:46]
B
No, no, no. Whether they're fact or fiction, some about AI and some not, some we've already touched on, but I want to put a finer point on them. AI is already reducing healthcare costs. Fact or fiction?
[31:01]
A
Fiction. And in fact, the Peterson Health Technology Institute has already looked at this. And year on year, healthcare costs are up 9% in the AI era. So absolutely not.
[31:12]
B
AI is, is 100% objective and neutral.
[31:17]
A
Oh, definitely not false. Yeah. Again, you know, whatever you put in is what you get out. And unless you can claim that the data it's training on is neutral, then the output's not going to be neutral. Which is why the representativeness point is
[31:33]
B
so important, I think worth asking, because you just made reference to it as you reflected on your time in New York City as our health commissioner. The biggest drivers of life expectancy are mostly outside the doctor's office. Fact or fiction?
[31:47]
A
Oh, 100. True. I mean, where we all learn this from is it's such an intuitive point like that. You just get that. Right. And paradoxically, it's something that I think the MAHA movement in its own strange way has touched upon the very basic human understanding that your health is shaped outside of healthcare and outside of the clinic and as a provider. I used to think about my patients in terms of the 23 hours and 30 minutes that they weren't with me and what was their life like. So that's a very. It's both a very lived and sort of intuitive point, and it's also backed up by data. Right. Sir Michael Marmont is sort of the godfather of social determinants of health, coining that term, because he realized, and he showed through large scale longitudinal population research, that frankly, most of health outcomes is shaped by the environment you're in. The social factors driving that environment and shaping that environment. Yes. Some by genetics and of course, by the behaviors that flow from your environment, your social factors, and some from your genetics. And only about 10, 15% is shaped by medical care. It's an important percentage, but it's not determinative. And if you were an alien landing in America in 2026, you would think that healthcare is what it's all about, because that's where the conversation has always been. Because it's something we can feel and touch and monetize. And there's people in white coats and there's buildings that we know that are hospitals. But if we really wanted to do health in this country, we would start at home, in communities, in schools and everywhere else.
[33:30]
B
But the hospitals, well, in Ashwin, that is so much of what we did, you know, 100 or 150 years ago. Right. Was we thought about clean water systems and we thought about what housing requirements should be and what neighborhoods should look like.
[33:45]
A
We're a victim of our own success in a way. You know, public health has really done wonders.
[33:50]
B
Yeah. Hopefully we'll learn to walk and chew gum at the same time. Ashwin, thank you so much for your time and it's just always such a pleasure and I always learn so much and I'm incredibly grateful.
[33:59]
A
I'm grateful, too. This was a lot of fun. Thank you for having me.
[34:01]
B
Thank you. You can follow Dr. Ashwin Fasan on LinkedIn and at his new website, ashwinvasan.com. thanks for listening. Talk to you next week. That Can't Be True is a production of Limonada Media and the Clinton Foundation. The show is produced by Katherine Barnes, mix and sound design by Johnny Vince Evans. Kristin Lepore is senior director of new content and Jackie Danziger is VP of Narrative and production. Maggie Kral Shore is our managing director of Partnerships. Executive producer is are Jessica Cordova Kramer, Stephanie Whittles, Wax and me, Chelsea Clinton. Special thanks to Erica Goodmanson, Sarah Horowitz, Francesca Ernst Kahn, Caroline Lewis, Sage Falter, Barry Leary Westerberg, Emily Young and the entire team at the Clinton Foundation. You can help others find our show by leaving us a rating and writing a review. And if you can think of someone who might benefit from today's episode, please go ahead and share it with them. There's more of that can't be true with Lemonada. Premium subscribers get exclusive access to bonus content when you subscribe on Apple Podcasts. You can also listen ad free on Amazon Music with your prime membership.