AI in health care: an update from the US – with Andrea Palm and Andrew Bindman

Health Foundation (2025). Event: AI in the NHS 2025. 

Health Foundation (2024). Priorities for an AI in health care strategy. 

NEJM Catalyst (2024). Ambient AI scribes to alleviate burden of clinical documentation. 

Kaiser Permanente (2025). AI in healthcare: 7 principles of responsible use. 

Jennifer Dixon:

AI is going to have a huge impact on health and care. With innovations coming thick and fast, how do we move forwards as intelligently and quickly as we can? Well, the US is often ahead of the game. That's where investment development and use of AI in health care is most advanced. So what can we learn from stateside? 

With me in the studio to discuss all this, I am delighted to welcome Andrea Palm, who until January this year was President Biden's Deputy Secretary of the Department of Health and Human Services, where among other things, she led the development of the federal strategy on AI in health and care. And Professor Andy Bindman, who is Executive Vice President and Chief Medical Officer at Kaiser Permanente, one of the US's largest integrated health care systems. Welcome both.

Andrea and Andy, really nice to have you on the podcast. Let's just step back for a minute and think about artificial intelligence in health and health care. It's a very fast-moving scene. It's almost like trying to run in a fog, isn't it? The developments are coming thick and fast. Perhaps turning to Andrea to kick us off, I mean clearly the Department of Health and Human Services needed a strategy. Can you say about how that came about and the ultimate shape of it, to try to help the US progress further faster?

Andrea Palm:

The President, the White House, were very focused on the need to try to stay as close to caught up with the developments in AI as we could, that as a government, we didn't want to fall behind in our role of leadership and our role of oversight, making sure that it was used safely and responsibly across the whole suite of the government enterprise.

At the US Department of Health and Human Services, my job was to help lead our effort to develop a strategic plan. One of the real privileges of being at the US Department of Health and Human Services is the breadth and depth and scope of the place. It's an almost trillion-dollar organisation. At the time, global footprint with over 80,000 employees really cradle-to-grave in how you think about the health and well-being of the population that you serve. So all the human services, all the public health, the entire NIH research enterprise, we oversee the safety and effectiveness of all the medical products that get into the marketplace.

Not to mention the actual health care system, right? It was my role to think about the ways in which we could ensure public trust in the system, that we were on the front end enough to have governance in place that was really driving safe and effective and responsible use of AI products, but also that was setting a vision for the future. What do we want the health care system to look like? How do we take lessons we've learned over time from a very fragmented health care system and think about the way public health and population health need to be married better to the health care system, how human services, social determinants of need, disability programmes, behavioural health, the American health care system, we still have just kind of bolted substance abuse and mental health services onto health care. How do we take this opportunity to fully integrate, to smooth those frictions, to reduce those silos? And so that's really how we tried to think about this strategy. How do we leverage all of the assets of the department to drive better health and well-being for the whole population?

Jennifer Dixon:

I mean, if you stand back from this strategy, by the way, those of you listening, I really recommend you have a look at it. I thought it was a very intelligently written and comprehensive strategy. But Andrea, clearly there would be conceptions about what the role of the state is, the government is in this. How did you conceive where the state should leave off and where guidance should just be there for others?

Andrea Palm:

I think there's two things here. One, an assumption that the authorities that we currently had in statute were the right ones for this new generation of technology. In some places that's probably true and in other places it's not. For example, our statutory framework for data privacy and security probably will match pretty well with how you think about what needs to happen for data security and privacy in AI. But on the flip side, we regulate medical products in a way that hasn't really caught up with generative AI. And so how do you think about what is the framework? Because you can't bring products to the market that aren't safe and effective, and AI brings a different meaning of what safe and effective is to the medical product space. I think you build the plane while you're flying it. You use the tools that you have, but you need to quickly decide which ones aren't adequate and then who should be responsible for them. The federal government and the HHS particularly, we oversee such a large portion of the economy and we have such an important role in ensuring that the American people get what they deserve and what they expect from their government, that there's no way not to have some central theory of how they need to operate and then who needs to augment them.

Jennifer Dixon:

And so I'm hearing there's a few big areas where the state obviously has a role. Data security, for example, safety I think are two big ones, aren't they? But then other things are left to health systems and professional bodies or whatever to fill in the gaps.

Andrea Palm:

Often in the US, [we] regulate by floor, these are the minimum standards and to comply with law, you meet those minimum standards. If you choose to go above that, then that's fine, but we set the ground rules and the guard rails and then people can fill in and personalise in ways that make sense for their particular system. But you've got to have that floor if you really want people to trust what is happening in this space.

Jennifer Dixon:

Andy, you obviously heard that you are operating within these guard rails, but then there's a lot of room for manoeuvre locally, I guess particularly in Kaiser and other health systems like that. How have you approached this world west of an AI system and all these new innovations bouncing into the system?

Andy Bindman:

We have about 12 and a half million members, and we are trying to think about the tremendous opportunity of tools like AI to improve the care experience, to improve our ability to identify risk in our population, to try to improve the quality of the experience, not only for our members, but also for our providers in terms of making their life easier. So we see great promise, but our number one approach to this first of all is how do we implement this in a responsible way? Just like any intervention we do in health care, do no harm, right? Making sure that we are both moving efficiently and effectively, but doing it in a safe and responsible way. So we do try to have a set of principles that guide the work that we're doing. We use a little acronym called PROTECT.

The P is for privacy protections related to our members and making sure that we safeguard in accordance with laws, regulations and ethical practices.

The R of PROTECT is for reliable that we develop and use AI tools that are consistently stable and secure because one of the things that we've learned about these tools is that they don't always necessarily perform in the same way each time. They're very different than other kinds of tools in some way.

The O stands for outcomes because we're very outcome driven in our work, and we're trying to really leverage AI specifically to improve our outcomes.

The T is for transparent, so we try to ensure that our patients are informed and give their consent when it's appropriate.

The E stands for equitable because we're really mindful of how as we try to harness AI to promote health, there's the risk of, as we've seen with other technologies of widening disparities as opposed to addressing them.

The C stands for customer focus. So we address the needs and outcomes that matter most to our members.

And then the final T of PROTECT is trustworthy. We're trying to build trust in how we use these tools with our members and with our providers and so forth. And reflecting on some of the comments that Andrea made as well, we're also as both a developer and a customer of these tools, are trying to think about how do we use them in ways that are responsible but also that meet the goals of what we're trying to do for our organisation. You have said it's a little bit of a murky situation to try to navigate all the things that are going on there. In my role as the chief medical officer, I easily receive 50 unsolicited messages a week of different vendors saying, ‘Try this tool. It'll change the life of your members,’ and so forth. And I have to really think about what is it that I'm trying to do on behalf of our members to improve their health and to make it work with our system.

And I think we need to talk about the ecosystem and how well it is aligning with the workflow and how we do our care in a way that these tools are an adjunct and not taking our attention away from the most important things you can get lost if you aren't really thoughtful about the North Star of what you're trying to do in this work. And that's why it was so incredibly helpful to have frameworks that really are trying to help us think about on a population level, what are we trying to do here? What can these tools really do to enhance our care delivery model and not distract us from, isn't this a really interesting card trick that these tools can do?

Jennifer Dixon:

How do you decide your North Star? What process do you have at Kaiser?

Andy Bindman:

The sphere of AI of course is not uniquely in the clinical space. It also involves things like supply chain or involves administrative tasks related to the organisation, but particularly things that interface related to clinical care. We have a governance council that involves both clinicians, as well as members from the health plan that we jointly review and evaluate tools. And we do this in part because we have about 24-25,000 physicians in our organisation. And we are trying to be mindful of how do we move collectively together toward the tools that are going to be most helpful for our membership. Clinicians help us evaluate what they prioritise as important areas, but also bring the perspective of the plan and the population issues as well to mind. We evaluate together and make choices together. And then even when we've made a selection, we have a process of responsibly implementing where we're learning along the way together, getting input.

So one of the big things that we implemented together, for example, was an AI scribe programme. And in doing that, we worked closely between the health plan and the clinicians about the possibility of this. We did a small pilot of it. We then expanded that pilot as we learned more together, got feedback from the clinicians about it, and ultimately scaled up over time and learned together and also made a contribution to the literature related to our experience. So we could hopefully help others along the way, but it really is a partnership between the plan and our clinicians about what tools can be helpful. And I don't think it would work with just one side trying to drive it and the other side not involved because when these tools are developed in isolation, as I was mentioning, they aren't mindful of the workflow and what the clinicians need to do and how they're going to use the information. It's so important to have that connection so that people understand, oh, we're going to discover more diseases way.

We have to be prepared on the health system to respond to it, or if we're going to create new workflows, how that's going to work. And I think that's the biggest challenge with a lot of vendors that are approaching us, is that they have isolated tools that are not fitting into how the system actually needs to them and think about making use of them on a population level.

Jennifer Dixon:

And Kaiser is unusual in that it's a big operation, it's well respected. It’s been going since 1945. There are lots of other hospitals in the US and medical groups that simply don't have the resources. So I'm just looking at Andrea, how does then the federal level learn from the big centres to help the smaller ones in the US? Was that part of your strategy?

Andrea Palm:

It's got to be wired into all the ways in the US that you think about the health and well-being of people. Because we always have small providers, rural frontier communities, Native American populations suffer huge health disparities in the US. And if you aren't conscientious in these efforts, fill in the blank, which one it is, this one is AI. But if you're not really conscious about how you make sure they don't get left behind, they get left behind. And so it's part of the reason why it was so critical that our strategy incorporate human services as well as public health because they have historically gotten left behind in the technology revolutions. I think when you have big systems like a Kaiser and others, our academic institutions, people are working really hard to be contributors to the responsible development and deployment of AI and how you partner with them.

Jennifer Dixon:

What we've had in this country is quite a lot of stuff poking through the national press where there's been an individual vendor who's making claims about a system, about the impact. And often the NHS is part of that claim because there hasn't been sufficient rigorous evaluation of it, or there's some initial results with a small group of people and then suddenly it's all over the press. It's very difficult for then to adjudicate. So part of this issue, presumably is to just get a proper system going of really a bearing truth as to what's actually happening, and yet to do this in a fast way that are normal evaluative techniques don't do. If you're waiting for an RCT, it's like three years is now two years, $2 million later. So I don't know whether your focus has been on really fast assessment that's good enough to keep you going until there's more of a formative assessment for later. Can you say a bit more about that?

Andy Bindman:

First of all, one of the risks here is if these tools are coming in through all the doors and windows of your organisation, it's very hard to then keep track of it, to learn in a constructive way and to make sure that you're achieving your goals related to equity and ways that you're systematically getting the benefits. So we have put a strong system of governance in place to evaluate which things we're doing. We create sandboxes, if you will, where people can experiment and try to think about things that they might develop from the front lines, but in terms of deploying them at scale, really try to have a system in place of governance councils that ensure that we're not having these tools sort of just show up in different parts of our system and then we don't know what is going on. So we are trying to work collectively, and I think that's really an important part of what needs to be considered here.

And then, yes, making a commitment to learning rapidly, and it's not always in the form of a randomised trial. Sometimes there are other ways that you can implement in a real world way to try to learn both about the safety aspects, but also what some of the benefits are. And again, one of the ways that we think it's important to contribute to the general knowledge around this is to be doing this evaluation and publishing on it so that not every organisation has to reinvent the wheel and go through this evaluation. But I do think there are rapid ways to be able to kind of learn. And we stood that up. For example, with this AI scribe, we've done similar things. In some cases they are randomised trials. We've done a randomised trial, for example, of an alert monitor system that we developed that was to help us identify and risk stratify patients on our medical and surgical wards who might be at risk of decompensating and needing to use higher-level care like an intensive care unit.

And we did that as a randomised study to partly, again, bring along our own community of clinicians and others to say, is this going to be valuable? Because everyone was like, ‘Well, of course I'm monitoring my patients,’ but in fact, we demonstrated that there was statistical benefit in terms of being able to reduce morbidity and mortality associated with using tools like this. So I do think it's really important, but part of it is to not get distracted on every possible one of these tools, but to really think about what value are they offering on a population health basis? What is the North Star for the outcomes that you're trying to pursue? And to make sure that in fact, these tools are doing that at the same time that they're building trust with your patients, they're building trust with your clinicians and so forth. So in some ways, we're trying to think a little bit about what are the things that have the most broad-based utility for us and really going at scale for those. So when we did the AI scribe, we rolled it out to all 24-25,000 physicians over a short period of time. We saw the value quickly with that as opposed to spreading ourselves over many, many tools.

Jennifer Dixon:

And I saw the New England Journal AI article, I think it was initially 5,000 clinicians that had rapidly assessed. Can I ask, how rapid was that?

Andy Bindman:

Oh yeah. It was just a handful of months that we were able to do this, and we had a steering committee. We would meet weekly, a team of us in which we would rapidly gather the input related to what we were hearing back. We actually learnt some interesting oddities about the tool. We're able to tweak that along the way.

Jennifer Dixon:

So troubleshoot, adapt, iterate, and understand if there's sort of glaring safety issues.

Andy Bindman:

Correct, like a small thing that might come up. For example, if the patient was in the room with a clinician and then there was a family member and so forth, and so there were three people talking, the tool might initially sometimes be confused about that. And so we were able to work with the vendor and help to identify that. And it got much better over time in terms of being able to do it and also able to speak to our clinicians to give them certain hints about how they could help the tool to be more effective at being able to sort that out for the medical note writing.

Jennifer Dixon:

So Kaiser is funding this by itself because it's part of its business. And if it's a private vendor that's giving you the AI thing that you are then adapting, does Kaiser get any monetized benefit from the fact that you are enhancing that product for 12 million people?

Andy Bindman:

It's not developed that way. I will say that by of course being a large purchaser, you can think about anything that you can purchase that you're able to be more effective as a purchaser when you're buying it for 20,000 physicians than you are if you're going out and trying to buy it as a small practice by yourself. But that's been less of our focus. We really have been trying to think about which ones of these tools will lead to the benefits we're getting, which is more efficient workforce, a happier workforce, happier patients. One of the things that has been most striking to us around this scribe tool is that as we've continued to try to recruit primary care physicians to our organisation, what we're hearing is that these primary care physicians are saying they won't even consider anymore going to organisations that don't offer these kinds of tools because it's become something that now is seen as really fundamental to making it possible to do the work of primary care.

Andrea Palm:

I think that the scribe technology is the perfect example for governance. It's critical at the health system level that exists. It's critical at the federal level that exists. But I think it also speaks to the way in which we're going to have to change our thinking about oversight because this is a classic example of something that would never fall under federal oversight. It's not a medical device. Nobody is at the federal level at the HHS is going to be thinking about its appropriate deployment to minimise whatever risk there might be, right? It is a low-risk application, but it is a perfect example and there will be thousands of them of things that fall outside of traditional health care governance from a regulatory and oversight perspective. And what is it going to provoke us to think about to make sure that there aren't inadvertent consequences? And again, not everything should be regulated.

I am not suggesting that at all. But this is a generation of technology in which we are going to need to rethink its interaction with the health care system considering how high-consequence it is for people's health. And again, the low-risk stuff is not what I'm worried about, but this is something that is going to get deployed rapidly in the system and that nobody is going to be thinking about unless you have governance at the health system level that is well-developed and it is good at it, which again brings back to these small providers who aren't going to have the expertise and the capacity. And so how do we make sure that low-risk unregulated tools are dispersed in a way that they are vetted and who are those partners? And it's a whole different piece of oversight that is going to be really critical and the government should have a role in helping to figure that out. Governance is foundational to all of this.

Jennifer Dixon:

And just going to public health, what excites you about the possibility of AI in population health activities or indeed prevention?

Andrea Palm:

Public health is under-resourced around the globe, and it is a place where the US government, particularly CDC, have been a leader in capacity building in everything from Ebola outbreaks, to trying to eradicate polio. And so this is a place where I think the US government should absolutely continue to lead. People are very interested in the shiny object of predictive analytics and chasing the next outbreak, and that is very cool and really important. But I think foundationally, we as a population still don't have our arms around the biggest drivers of health care costs and illness in our country. Everything from diabetes to cancers, to all of the things that a population health model could really help move the needle on. Not only is this an opportunity for population health, it's an opportunity for public health and for its reintegration into the health care system in a way that really could just so dramatically improve the health of the public.

Andy Bindman:

We've put some of this in place at Kaiser Permanente. One of our developed tools is a predictive model with AI that we use to identify members who have a high level of social needs. So we use different kinds of indicators in our data. I'll just give you at a high level, maybe an individual who's missed multiple appointments to their GP and seeing that as, oh, I wonder, is that an indication of someone who's got a challenge with transportation or is having a challenge of balancing the financial needs that they have related to housing and food with what they need for medication and so forth. And then we connect that with our clinical teams to also then address social health issues so that we can in fact try to improve health outcomes for our population as a whole. So I do think exactly what you're saying, Andrea, in terms of connecting public health strategies with our clinical care and ultimately for the benefit of our patients to improve their outcomes.

And I do think these tools when thought of in an intentional way of what they can do on a population health basis. Unfortunately, I think the marketplace has been largely focused on things that might be purely about generating revenue or driving a certain kind of utilisation pattern. And so while many of the tools are adaptable and helpful, we are also finding places that we need to develop tools that meet the needs of our system and trying to be value-based in our orientation.

Jennifer Dixon:

How far is Kaiser going upstream into primary prevention given all the tools that are likely to come about on personalised risk prediction? How far are you going to get into risk mitigation, particularly for people perhaps, I don't know, age 60 plus.

Andy Bindman:

I mean, this is so fundamental to our integrated model and how we think about population health, which is to, as you say, move upstream and try to prevent health problems when it's more possible to do that at an earlier stage. It's often less expensive to do so, but you can't just be screening and then finding all sorts of things that you go like, oh, that's a shame, because you have to have interventions that can actually make a difference. So we are really mindful of contributing to the evidence around that as we have through our work over the years of looking at the value of screening for cancers of different types and so forth. So we're really intrigued with some of the power of the tools of precision medicine and are doing quite a bit of work to expand the implementation of precision medicine tools to help us move upstream in the way that you're talking about.

But we're evaluating all those things along the way as well. And what our capacity is a system, because if you just discover all these things, suddenly you inundate your system and you have challenges being able to do that. So it's a fine balance between finding things earlier and having the capacity to be able to attend to them. But I do think ultimately, yes, we are looking at a future where we have a greater ability for our patients in partnership with our clinicians to learn more about the health risks and to hopefully attend to them at an earlier, less costly stage.

Jennifer Dixon:

And Andrea, you mentioned earlier something called human services, which we don't use that term here, but what I think you mean, or at least what I read in the strategy were interesting potential use cases of AI by communities that are trying to help build assets locally in health. And we've got this discussion at the moment because clearly public funds are quite limited. There are lots of left behind neighbourhoods, there's a lot of assets in those neighbourhoods that are currently unused. And I wondered if you might just expand on that for us.

Andrea Palm:

When we talk about human services, we should be talking about it from an asset model. Oftentimes in the US, we talk about it sadly from a deficit model. And so we talk about poverty, we talk about things like substance abuse as things that we need to tackle in communities. And so I think because human services is also very underfunded, again, there's an opportunity here to think from an asset perspective how we utilise this next generation of technology to really bring to bear opportunities for folks who are in these circumstances to level up in the child welfare system. For example, caseworkers with huge caseloads, kids who need more attention than can be provided by a really overworked system. How do we think about these tools making the jobs of caseworkers easier, of getting ahead of risks, of helping families cope better with circumstances in ways that we don't lose years of a kid's life, right?

Years in a kid's life is a lot of their life and we owe it to them to do it better. It is the most important role of a federal government to lift up, to look ahead, to leverage all of what we have to deliver better. And we cannot leave this particular population behind. They need work supports. They need housing supports, maybe food assistance, things where the cracks really showed in Covid. We should not let that happen again. And this is an opportunity to lift all boats, and it's our responsibility as a government to serve the people that our legislatures have tasked us to serve. And we can do it better and we should do it better. And this is a real opportunity.

Jennifer Dixon:

Andy, we spoke earlier before this podcast about the critical infrastructure of data on which AI platforms are built. There's no point having the flashy stuff until the basics are done. Kaiser is well known. I think you were the first to have the Epic system in the US. I think it was $6 billion. It was a massive investment everyone was making, which for US standards, it was massive. We have something called the data Maturity and quality Index. We do assess quality in the date of the NHS. We have one system and it's patchy, but it's improving over time in today's cutthroat, fast-moving world where any extra money is just looked at, frowned upon because people don't have it. How do you build investment for the long term here? How do you decide how much to put into the capital versus the immediate flashy stuff?

Andy Bindman:

So we have a budget and we have a defined population very similar to the NHS, and that gives us a certain degree of flexibility to try to think a little bit about how to spend our resource. And to your earlier comment, our organisation was one of the very first adopters of electronic health records because it was seen that this could be a powerful way to support population health. And so we weren't on that hamster wheel, if you will, of having to charge fee for service and be stuck in that system. So we were able to make specific decisions around investments in it, and they have paid off tremendously. I mean, I think Kaiser Permanente is known throughout the United States as having among the very best quality because all of our clinicians are working off the same medical record. And we speak a lot about like, oh, when you go to see the eye doctor in Kaiser Permanente, that eye doctor can also say, ‘Oh, gee, Jennifer, you're behind on your colon screening exam. Can I make sure that the FIT test is mailed to your house and we can get you up to speed on that?’

So everyone is part of a team and can see that in a 360 view. And that's a really important part of how we provide care. And it's that platform that became the basis of then having different kinds of clinical decision support tools, which have really allowed our primary care physicians to work to a very high level of service delivery. And I think we see AI in the same kind of way as a further enhancement, but we need to have to your question and your point, our data organised in a way that can really fuel that. So we're continuing to look at where do we have potential silos in our data that we can organise to make sure that the power of what these tools can offer can really be unleashed to the benefit of our members.

So a lot of behind-the-scenes work to get our data organised in that way to make sure that we can fully leverage. And I think it's a critical part. It's probably the most important thing of making sure that we're really ready to take advantage of the revolution that's going to be available through these kinds of tools. So it's an enormous part of it. At the same time that we're building that capacity, we're also trying to build capacity in our workforce to understand how to use these tools and to see them, right? If you sort of provide them and then people are like, oh, is this a direct threat to my job versus something that it's over here and I don't need it, then that's not helpful. So we're trying to maintain a good balance between getting ourselves ready to be able to take advantage of these tools through our data, but also training our workforce that's not just our physicians, but our nurses and other health professionals, how to use these tools to make their jobs easier and more effective and so forth. So I think there's a lot to it. Data is fundamental to driving the value of these tools, but you can't let it get so far ahead of the other parts that people aren't ready to take advantage of those tools either.

Jennifer Dixon:

And related to that, how far did you at Kaiser anticipate, and do you and Andrea at DHSS anticipate quite significant dislocations in work patterns and jobs? It's quite a risk, isn't it? And to bring along people is crucial.

Andy Bindman:

Well, in terms of how we use these tools, there's always a human in the loop, right? We're still learning so much about them in terms of their accuracy, reliability, some of the things that I touched on earlier about how we're trying to think about responsible use, and we perceive that there's so much that could still be done to enhance the health of our populations. It's really less about any kind of trying to see our work. It's trying to really help our workforce work to a higher level and to find more joy in the work of what they're doing in terms of relieving them from administrative tasks and really helping them work to the top of their licence for the benefit of our members. So I really think that has been the focal point of how we're thinking about deploying these tools.

Andrea Palm:

The only thing I would add, we've got just massive workforce shortages in the US and any ways in which this is easing the burden is making for happier health care providers is seen as a recruitment tool, I think all to the good in the places where we are really struggling to attract and retain health professionals.

Jennifer Dixon:

Given what you know about the UK NHS, the fact that it can be coordinated, it's centralised. We have got the single data system, what's the advice you'd be giving us now as we go through into the next couple of years facing AI?

Andrea Palm:

I would say a couple things. One, make sure you think about the application of lessons you've learned, whether it's the deployment of electronic health records, whether it's pandemic lessons, all the things that make us smarter about the delivery of care in our communities and the way in which the system works under pressure and doesn't. In the US, I think we had so much hope and so much in the basket of the deployment of electronic health records and the initial generations of health IT that has not come to fruition because we built EHRs primarily for health care systems and health care providers, and we give a lot of lip service to patient access. It's the patient's data. They're driving their care, their partners in their care, but we haven't really fulfilled that vision. I just would caution all of us not to make the same mistakes twice, not to just think about AI tools as tools for providers and systems for their efficiency, for their work, but how we really think about what we want the health care system to look like and demand that the tools help us get there.

Andy Bindman:

So you need to be very strategic and mindful of what you're trying to do and how to use these tools to pursue the goals and values of what your organisation is. And what I've always admired about the NHS is it's always been very clear about values of equity and population health and lifting up everyone. And I think being really mindful of that as you look around at the opportunities that these new tools offer to not lose sight of that because of the shiny object or the bells or whistles that come with these different things. And I touched on this earlier, but the degree to which these tools fit into a system of care is so important for them to ultimately be successful in achieving those goals. Having a system to build these tools into be able to have them give benefit at scale is a tremendous opportunity. I want to see the NHS be successful here. I want to be able to learn what I can bring back to KP and hopefully more broadly to the US. But I think there's a tremendous responsibility on the shoulders of the NHS and KP to try to do this and to provide that knowledge as a community benefit so that everyone can benefit from it. And we have a responsibility not to get distracted, but to stay focused in that way.

Jennifer Dixon:

Well, I don't know about you, but I thought that was really interesting. And the three big things I learned were: firstly the state's role, certainly in regulation safety and data privacy. The second big area was how important it is to learn from fast rapid evaluation done in really good centres, possibly to a rigorous method that's already been agreed and kite-marked. I think unless we have this syndicated, federated way of learning, we're going to get left behind. So I think what Andy said about Kaiser being one of those systems and that information being sucked up into a national intelligence radar screen, I think is going to be critical moving forwards. Otherwise, we might have a thousand flowers blooming and some of the flowers are actually not flowers, they're weeds. And the third area was the interesting potential uses of artificial intelligence to boost population health, not just on the wider determinants, not just on pandemic preparedness and response, but also ongoing supportive communities that are trying to build assets to improve health in their communities. And I think that's under-discussed and probably underinvested environment that could be very interesting for the future. 

So we must leave it there. Many thanks to my guests, Andrea Palm and Andy Bindman for their insights today. And if you'd like to know more, catch a recording of our AI summit at which Andrea and Andy both featured. Please check the show notes to this podcast for the link or check the Health Foundation's website. Join me next time with Rachel Sylvester, Political Editor at The Observer and our very own Director of Policy, Hugh Alderwick. Meantime, many thanks as ever to Sean and Leo at the Health Foundation, to Paddy and his team at Malt Productions and goodbye from me, until next time, Jennifer Dixon.