AIs impact on medicine Nobel Week Dialogue 2025 Health For All

Scientists won the Nobel Prize
12 followers
2 Video Views·Mar 28, 2026
AI Generated·May contain errors
Last year, the Nobel Prize in physics and the Nobel Prize in chemistry was awarded to
AI achievements and the chemistry prize was an incredible achievement
looking at protein folding and solving the challenge of protein folding and given to
Demis Hassabis and John Jumper at Google DeepMind for their contribution to this work.
Now, they're just a stone's throw away from you, Paul, at your institute, but tell us
why was that such a watershed moment?
Well, I'm sort of familiar with it because Demis and John are personal friends, actually,
the DeepMind labs are in the Crick, the wet labs, and it's been an interesting story because they
built it on chess and then playing games on Go and used many of the things they learnt
in those endeavours and applied it to protein folding. I'd watched it over the last five or
eight years and it was just a remarkable performance. It was a problem that was absolutely
suited to the present strengths in AI. It has data available, huge amounts of data, publicly
generated data, I should say, primarily by discovery research but then open and available
to the entire community. That's critical because that is what works. It is what I call thin and
widely spread data, which is perfect for this type of analysis. This problem has been around
since the beginning of molecular biology, since the 1970s, and the modelers have been working
on it for 20 years, trying to do it bottom up. AI, of course, provided a way of looking at it
without understanding the principles behind it but producing a way of actually predicting
how the protein molecules, which is a chain of amino acids that then folds up in complicated ways,
of predicting that having known what other molecules already did. You don't even fully
understand quite how it works, in fact, but it works. It has taken over this type of molecular
biological research in an extraordinary way. All of this, including me, who's not an AI expert,
including me, that we just do it routinely. We discover a gene, it encodes a protein,
we have a sequence and we can see what it looks like, which would have taken a year or two of
experimental studies to get there. It's a great achievement. It isn't yet always completely
precise, so if it really matters you have to follow it up with actual experimental work,
but it is an enormous achievement and perfect for the present status of AI.
And has it opened up your mind to the potential of AI application in other areas of biomedical
research and medicine generally? Well, it made me think about it.
What they produced is a way that allows you not only to look at protein structure but other small
molecules and think of drug development and all of that. What I'm interested in, and I talk to
Demis constantly about it, is can we produce improvements on the methodology, a more general
AI approach that would allow us to understand the complexity of living organisms and how they work.
That is a much more demanding problem, which I think is going to be critical for
the next hundred years in medicine. But if we just take a cell, what we have going on in there
is thousands of reactions, all coordinated and connected in some way, to produce overall purpose.
And that's then magnified again in the living organism. This is an immensely complex problem.
We absolutely will need some tools equivalent to AI, but I think beyond what we now have available.
For me, a combination of that type of analysis with the right sorts of experimental analysis
is really going over the next 30, 50 years, it's a long time, to help us understand what life is.
And now then maybe over to you, Alison. You're doing a lot of work with the Fleming Institute,
the inaugural director there on antimicrobial resistance and these big issues. What's exciting
you around the opportunities with AI in that work?
So two really exciting opportunities that I'd like to describe. One is around the application
of AI. But before I get to that, I just want to say the multidisciplinary working is also
absolutely critical. So to embed fellowships, however they're funded, whether they're Google
DeepMind fellows, or whether they're research fellows from other programs, to actually embed
them with clinicians, epidemiologists, bioengineers, chemists, clinical trialists, etc. It's just a joy
to see. So that people are working from the very beginning, informing each other and understanding
the potential of this from the most basic research to the kind of policy level type of work.
And then just the second area that I wanted to highlight, is that I think it's important that
we think about AI beyond discovery and beyond fundamental research, which is absolutely
critical. It's really exciting, because I don't think we're using it enough in health services
research. I don't think we're using it enough in shaping clinical trials. I don't think we're
using it enough to think about how we protect patient pathways, how we use it for healthcare
resilience, and also how we address an issue of representation in data. So I think there's a huge
opportunity for the application of AI within health services research, safety, resilience,
but also how we can use the data from that to inform better care, better safety, but also
generate the research questions that need answering as well. And so there's some barriers there that
you don't think this is happening enough, that maybe there's bigger investment in the
fundamental research and less in this other space maybe? Well I think there's an issue about
having representative data. I also think within healthcare there's an issue about being able to
follow individuals across the whole patient journey, rather than different areas of that
journey, and to have joined up and linked data that can support AI. But also, I mean, which is only as
good as the data that goes into it, we need to remember. So I think there's many opportunities
there that we're not actually using adequately, and also the issues about regulation as well.
But there's such an opportunity for bringing in, for instance, diagnostics and decentralized
diagnostics and the data that we can get from that, that can cover the whole kind of patient
journey and think about how we can use it to ensure that we have good healthcare that actually
protects patients rather than amplifies infection or is a reservoir for infection as well.
And I mean, I'm talking about this from the perspective of infectious diseases, but actually
integrating data from diagnostics across the lifespan and patient journey is also critically
important for all sorts of non-communical diseases as well. And also it will inform, as I say,
opportunities in clinical trials, targeted interventions and policy.
And this is already happening. I think during COVID, there was some examples of AI being used.
Yes, that was really interesting. And again, this was an AI PhD working with us who did work
linking diagnostic data with geospatial and patient movements so that you could identify
where the maximum nodes of transmission of COVID were. And bearing in mind that one in five
patients that went through healthcare acquired COVID from within healthcare, then came in with
it, that type of data is really critical. And we need to be able to use that in real time.
Maya, maybe I'll come to you then. So as Adam said, you're State Secretary for Health here,
but you are also an expert on AI policy and governance. And a lot of your research has
been on that. And you've just recently moved back to Stockholm or to Sweden from the EU,
where you were in the AI office. And you published a book on AI and the power over decisions.
So we want to talk a little bit about the governance issues around AI and particularly
relevant to the medical professions and some of these issues. What do you see as some of the big
challenges on governance? I think it's nice that we're starting this AI and creativity. And I also
think it's interesting in the 1990s, if you were in a panel, I wanted to say you're a feminist,
you always had to start by saying, but I love men. And nowadays, if you're in a panel on AI,
you always have to start by saying, I love technology. And I do love technology. But
I think it's interesting when we use AI as public actors. And I think that's something
different, actually, than using AI as a tool for creativity. I think very many of us
use AI just to pose a question or to play with something, sometimes that results in
groundbreaking research, sometimes that just results in a less boring dinner speech.
But when public actors use AI, they use the power that we delegated to them as a part of our
democracy. And as citizens, I think we should, for example, demand transparency from that. And
there are a number of interesting and actually quite appalling examples from European politics
when public actors have been using AI or algorithms. There's one example from the
Netherlands where the Dutch government used AI to find people who were trying to commit fraud with
child allowance. And they specifically targeted people with double citizenship. So this became
a racist algorithm that tracked down very vulnerable families, who in the end showed
out that 95% of them did not do anything wrong. So there's a lot of problems, I think,
or there should be a demand of transparency when we hand over power to algorithms as a part
of our democratic system. And with the governance and the EU on AI, I mean, there's
AI legislation coming in, there's pushback against that legislation. But how does it,
you know, are the risks that it will slow down the innovation? Or are these essential safeguards
you have in other places, you know, the US and China, very little regulation in these areas?
Personally, I'm often annoyed by this narrative that innovation and regulation stands against
each other. I would say that's a false dichotomy. I mean, we have a number of examples where
regulation is actually a stimulated innovation. And we have other examples. I'm thinking of,
you know, the car, for example, we have speed limits. And does speed limits slow down the
innovation of fast motors? Probably. Do we think that's a problem, you know, big enough to take
away speed limits? No, we don't. So I think that's a concerning narrative, actually. And I
do think that that narrative is also being pushed for political reasons. And I think that there's
also a tendency, I mean, the less regulation we have on the EU level, the more responsibility
is being delegated to the end user or to companies, I mean, to the one who is actually
using the AI. And that's concerning. I think that that puts us all in a vulnerable position.
So I'm concerned about that. Now, Paul, how are you using AI
on a day-to-day basis? And how's all that going? Well, I listened to my PhD student
to be more precise about it. But we shouldn't forget, sometimes people use the term everyday
AI. And we do use everyday AI. That's things that are not blockbusters. They're just
sensible things. Like we analyse images, microscopic images. And what would take us a day
or two to do, we can do in minutes. It saves us, you know, an awful lot of time. We can do
literature surveys rapidly. Instead of sending somebody to the library for four or five days,
then they can get it back in half an hour. Now, what they get back is a bit dull. It's a bit
conventional. It's not that creative. It sort of reads like it's been written by a robot. But of
course, it has been written by a robot, basically. So it's bound to read like that. But we shouldn't
sneer at it for that reason. I mean, it gives you a foundation upon which it can work. I emphasise
these things because actually, that does actually help us. We see, of course, an immense amount of
hype in the markets about it. And that's almost certainly going to collapse because there's
almost no actual real applications coming from much of that investment.
But for discovery research, we've got a new tool that allows us to do rather dull things much more
effectively. And so people are going to be able to spend more time thinking about it, as long as
they don't lose the capacity to think. And why I'm emphasising that is treat AI as providing a tool
to give you the information. Do not think that is the answer. Use that information to actually
come to more creative outcomes. So I'm in favour of it, but let's not think it's the
solution to the problem. It is a mechanism to get towards a solution.
Okay. And then, indeed. Yes, hopefully it won't stop us thinking. This is the key thing. And
let's go back to AMR. And you're very interested in surveillance and these kinds of issues around
detection and spread of disease. And you see a lot of potential with AI in that area.
Yes, indeed. But I'm just thinking about everyday AI and also AI that can be generated from big
pieces of data, which would be really good for public health and policy. But I want to think
about everyday AI for people who prescribe drugs, for instance, and are wanting algorithms to
support them in the optimal therapy for that individual. So personalised medicine, but also
thinking about individual dosages and polypharmacy and everything. But if your data is based on you
looking at a formulary somewhere, and if that data actually doesn't include you because you
happen to be pregnant and you're not a fit Dutch bloke in his 30s somewhere, but you're actually
a baby or you're pregnant or you're a bit obese or whatever, there is no algorithm or decision
support tool that includes that data because the numbers are not there. So we really, really need
to think about representation in data and how we can use some of this in a kind of global data
resource to inform what we're doing, but of course with the appropriate governance and regulatory
frameworks around that. But that enables us to have it as a kind of everyday tool if you are
responsible for prescribing or thinking about prescribing policy. But we need that representation
so that the decision support and the advice about how to prescribe and personalise and tailor therapy,
we need that there. Okay, and then Samaya, do you see this as a governance issue as well then,
on that biases in the data, how we use it in a way that minimises those biases and how we can
direct the medical research community in that way? And these sorts of issues around data protection,
that you know, is this balance between data protection and data access, how do you see
the governance issue around that? I think it's a multiple issue really and I think you put it very
well Alison, that sort of the data that goes in is definitely part of both the solution and the
problem. And I'm thinking about when to use AI and also not to use AI, which I think is connected
to that. Policy researchers sometimes speak about question zero and that is, should we use AI to
solve this problem? And sometimes the answer is yes and sometimes it's no. But there's also a risk
of not using technology and I think you're pointing to that Alison, and for example in the 1950s
we were bad at detecting heart attack in women, you remember? Because all research on heart
failure had been done on men and I think we risk doing that all over again when we collect data
from very special populations. And if we don't use AI in public healthcare for example, it will for
sure be used in private healthcare and then all the data collected will be from a very particular
part of the population, which will make AI-based medicine bad for the majority of the population.
So there's a lot of build-up in this that can be both dangerous and difficult.
Okay and then, so as a final question, I mean there's huge amounts of investment, huge amounts
of money appearing for AI and there's a risk that, you know, are we spending too much on AI
and then maybe ignoring some other areas of research that we should be prioritizing
or getting, or is the balance about right? Is this the most exciting area and in terms of science
and we should be putting all the money in there. But any reflections on that funding
in just a few words? I said something briefly about it, there is a lot of hype out there that
we just have to admit. And so we have to be careful of that. I actually agree with the
comments particularly you made, that to get progress we have to get strong connections
between the AI experts and the experts are doing the things that it's supposed to help.
If we divorce these two, it just won't be productive and that requires effort. And if we
get that together, instead of just trying to see whether we can make trillions of dollars on
something that is being sold in an investment way, I'm getting tired of that to be quite honest.
I'm getting tired on boards of people telling me that AI is the next revolution and I know enough
about AI to know I don't know enough and they know nothing at all. It just is very irritating.
But getting those who are trying to solve the problems, working closely and interactively
with those who can apply the AI expertise, I think it's the real way forward.
Completely agree.
Then on that note, ladies and gentlemen, a big round of applause for our panel.
