UK AI & science-optimised pasta
Artificial intelligence is the big talking point of the week, with UK PM Sir Keir Starmer announcing a drive to unleash its full potential.
It’s already being used in healthcare, but recent studies have exposed both strengths and weaknesses.
We’re joined by Dr James Kinross, a surgeon and researcher at Imperial College London, to discuss the positives and the pitfalls.
Also this week, we talk all about what a Trump presidency might mean for science; why powerful winds are driving the deadly fires in LA – and anyone for science-optimised pasta? Marnie is helped by a physicist to make the perfect Cacio e Pepe.
Spoiler alert: Recipe below!
Presenter: Marnie Chesterton
Producers: Sophie Ormiston & Gerry Holt
Editor: Martin Smith
Production Co-ordinator: Jana Bennett-Holesworth
To discover more fascinating science content, head to bbc.co.uk search for BBC Inside Science and follow the links to The Open University.
Science-backed Cacio e Pepe:
For two servings:
- 240 g pasta
- Black pepper
- 160 g pecorino cheese
- 4g corn starch in 40ml water
Dissolve the corn starch in water and heat until it forms a gel. Let this cool before combining it with the cheese and black pepper. Cook the pasta, then drain, keeping some of the water. Let it cool then mix the pasta with the sauce. Enjoy!
Listen and follow along
Transcript
BBC Sounds, Music, Radio, Podcasts.
Welcome to the podcast for BBC Inside Science first broadcast on the 16th of January 2025.
I'm Marnie Chesterton.
Hello, coming up, winds of change in America, both political and the ones driving the LA fires.
We explored the science of both.
Also, I try to cook a tricky pasta dish.
Can adding physics make it foolproof?
But we start with artificial intelligence.
It's coming into our lives apace and on Monday the Prime Minister declared it had vast potential for rejuvenating the UK's public services.
We're going to look at that potential in healthcare.
It's already being used in the NHS to aid diagnosis and in analysis of scans but a couple of recent studies have exposed AI's strengths but also its weak spots.
Dr.
James Kinross is a surgeon who works on developing AI for use in medicine, and I asked how often he uses it day to day.
Well, I'm beginning to use it a lot more than I did.
That is absolutely the case.
I think it's also fair to say, though, that much of my job is still pretty analogue.
I think surgery is quite an interesting area, though, for the development of AI because surgery is still an inherently dangerous undertaking, and AI allows us to, I think, make surgery quite a lot safer.
So we're very excited about the use of AI in surgery, and we are beginning to adopt it into
our regular practice.
And what about AI in the NHS generally?
It's an interesting question to answer because I think there's the official line and what we really use it for day to day.
And then there's how it's really used in practice.
So I think what we see anecdotally is quite a lot of doctors, physicians, nurses using everyday tools that everyone has access to, like ChatGPT.
But I don't think we've got a really good measure of that.
So actually day to day, if you come into the average hospital, not many services really do use AI at the moment.
moment, but I think that's going to change very rapidly.
Now, there was a recent study at the University of Lübeck in Germany where AI was introduced into breast cancer screening.
Can you tell me what sort of AI was used and how?
So this study was using a very specific type of deep learning and they were using a deep convoluted neural network as its official title.
And what that really means is that they had built training sets using AI to identify breast cancer based on around 200,000 different previous images of breast cancer.
What they then did was they took this tool and they gave it to frontline clinicians, about I think it was 119 radiologists.
They said, you can use this tool to help you make decisions about whether or not we can find breast cancer or you don't have to use it.
And what they found was that if radiologists did use the tool, they were much more likely to detect breast cancer, but also that they were much less likely to unnecessarily recall patients back in to have further tests, which is quite important in a screening programme.
So the significant thing about this work is that it was done in a lot of women.
So I think it was well over 400,000 women.
So it was a really well-powered study of AI in the wild, if you like, and the results were pretty impressive.
And do you think that giving the radiologists the choice to use AI was important?
I do, because I think it reflects how we are more likely to use AI in the real world going forward, because AI is there to augment the way we make decisions, not necessarily to replace them.
There is another recent study that suggests that AI isn't going to replace doctors anytime soon.
And
this is attempts to use different AI models to diagnose patients based off those initial conversations that we'll all have with RGPs where they try and work out what's wrong, right?
Yeah.
What you're referring to is a very particular study where researchers trained an AI chatbot, interestingly, by running thousands of different patient cases through a learning model.
And then they applied it to various clinical scenarios by asking the LLM or large language model to ask questions from a different synthetic data set, which was also trained on a large language model.
So it's kind of two AI systems talking to each other.
And they simulated tens of thousands of different patient scenarios and asked whether or not the AI bot could make a diagnosis.
And what they found was that when you have a conversation that's more representative of a real human interaction where there is nuance or open-ended questions, that actually the AI chatbot is much poorer at making accurate diagnoses.
Yeah, because it feels like in this study, what the AIs were good at was if you give them multiple different boxes, like a multiple choice, they can tell you more accurately what the right answer probably is.
But once you start introducing the messy nuance of human-to-human interaction, you're getting accuracies of at most sort of 25%.
Exactly.
You know, when I was at medical school, we were taught always to look for something called the hidden agenda.
So some patients will come to your surgery and they will pretend that they were there for one problem, but actually the real thing that they want to talk about is hidden away because they're they're embarrassed about it.
And the question is, is can models like this identify that and tease that out of a conversation?
And the data that we've seen here suggests that that may not be so easy.
But I do think it's interesting that other AI studies in this field have shown that actually some patients do want to engage with these AI models.
So Google, for example, had a different system in a different study.
And actually they found that users were more likely to engage because it was able to simulate empathy.
That's fascinating.
Are you saying that there are patients who are more comfortable with an AI chatbot than a human clinician?
Well, I am saying that.
And I think one of the problems that we have in clinical medicine is that if you go to any frontline GP service, they've only got, you know, a few minutes to have a conversation and to make a diagnosis.
Actually, with these systems, patients can have a lot more time and these systems can be programmed to demonstrate good listening and empathy.
I think there is kind of one other really important point, which is that our conversation and these data assume that large language models are based on really robust data.
But actually if you seed these data sets with a very, very small amount of falsified information, maybe just 0.001%,
you can seed falsified information into those answers.
So the key word in healthcare is trust, because patients have to be able to trust these data and they have to be incredibly robust, but so do clinicians.
And I think that is is still an open question as to whether or not we really can trust them to make diagnostic decisions using conversational AI.
Are there ethical considerations?
Yeah, I think this is an ethical minefield, to be frank with you.
And my concern slightly is that the ethical car is very much separated from the horse here, which has bolted away.
So this technology is being deployed and we're trying to catch up with it to create the ethical groundwork.
And then, of course, what the legal framework is for when they inevitably fail.
If I'm operating using an AI-based system to make interoperative decisions and it goes wrong, what does that mean for the patient?
Who is responsible?
And how do we create a learning system that ensures that those things never happen again?
And this week, the PM announced plans to unleash AI across the UK.
So his idea was it's going to drive growth in many different areas.
What kind of impact might this have on healthcare?
Well, I think the NHS, I mean, it's faced significant existential problems for some years now, and these are well reported.
And I think the hope with AI is that it allows the healthcare system to become more efficient.
So, for example, how and when you book an appointment.
And actually, if we can get our doctors and nurses to work more efficiently, then that will certainly improve the quality of care that we're able to deliver.
So, all the students today who are going to be the doctors and nurses of the future need education in AI?
Oh, 100%.
In the same way that I have to understand how a drug works and what its side effects and risks are, I need to understand what the side effects and risks and benefits are for an AI algorithm in the future.
So our medical students need to have a fundamental grounding in these technologies if they're to use them safely.
And that's definitely the way that we're going.
Well, as well as your day job as a surgeon, you're also working on a project to build trustworthy AI.
What's your aim with that?
So we were very lucky to be funded to start a project called Indicate.
And really what we're trying to do is to find a new way to leverage AI to help support the process of what is known as peer review.
So peer review is when expert colleagues or peers in the same subject analyze and interpret your data to make sure it's robust.
And that's a big problem because when it fails, we see that the wrong information gets into the world and then that becomes effectively a scientific truth.
It's very, very difficult to challenge.
So Indicate is an attempt to try and solve the infodemic by using AI to create these ways of autonomously performing systematic review on all of the world's medical information continuously.
Great.
So if AI is kind of seeding the misinformation, you can also use AI to correct it.
Exactly right.
What we want to be able to do is to use tools that have come out of Indicate to feed into the models that we've discussed today around diagnosis, for example, so that actually you know you've got trusted information that's very, very high quality, that's been continuously appraised as being fit for purpose in healthcare.
Thanks Dr.
James Kinross, reader in surgery at Imperial College London.
Powerful winds have been a crucial ingredient in the deadly Los Angeles fires and after a brief respite on Wednesday they're forecast to return yet again next week.
The Santa Ana winds, as they're known, have reached speeds of up to 80 miles per hour.
That's hurricane force, fanning the flames to devastating effects.
But what's causing these winds and why are they they happening in Southern California?
Climate scientist Ella Gilbert, who studies mountain winds, joins me now.
Hello, Ella.
Hello.
So tell me about the Santa Ana winds.
What causes them?
Fundamentally, winds happen when you've got a pressure difference.
So air flows from one part of the atmosphere to another.
And the bigger that difference, the faster they go.
The Santa Ana winds are particularly Santa Anner-y.
They're particularly dry, they're particularly warm because that air gets forced over these steep mountains.
And as that happens, the moisture that is contained in the air condenses out into clouds.
Sometimes it rains, sometimes it snows.
And then on the other side, it means they're much, much drier.
And as the air gets forced down slope, it gets squeezed by the atmosphere because the atmosphere is denser at the bottom than it is at the top.
And it's like pumping up a bike tire, if you like.
So you can feel it.
If you pump it up, it gets hotter.
And that's exactly what's happening in these mountain winds.
I've seen the videos of these things like a industrial bellows that's just permanently on blow and I guess that's what's making it very difficult to put out the fire.
Exactly and these winds they're super fast and they're really hard you can't defend against them and they're literally just fanning the flames.
And is this phenomenon unique to this terrain?
They happen all over the world anywhere that there is mountains.
I study them in Antarctica where I would call them Fern which is the German name which is also what they're called in the Alps.
Is another German for hair dryer?
That is exactly it and you have to any German speakers will have to excuse my pronunciation but it's fern yeah.
But in the Rockies they're called the Chinnook, in the Andes they're called the Zonda and essentially it's just different names for the same phenomenon and the specifics here are that you've got this high pressure over the desert over to the northeast of LA and that's particularly dry and then as it gets forced over the mountains especially there are little gaps in the mountains and just just like if you have rapids in a river, they get faster because the water gets shallower.
If you force air through a narrow gap, it also gets faster, which is why you get this really speedy 80 mile per hour in some cases wind.
So what are they doing in Antarctica?
Is it the same sort of thing that is going on in Southern California?
In some senses, yes.
It's still that it...
produces warmer and drier conditions but instead of causing wildfires in Antarctica we're more concerned with it driving the melting of ice.
So wherever they hit, it seems that they're causing problems.
They're a natural part of our climate and our weather picture, but in some cases they can be the straw that breaks the camel's back and it can drive some pretty dramatic consequences.
And Ella, have they changed over time?
So in California, there's some evidence to say that they are happening in slightly different seasons now.
So I've heard fire chiefs talking about this previously being a bit of downtime for firefighters in California, but now they're seeing a sort of shifting fire season and particularly the Santa Ana season is starting to change.
So whereas previously it was in kind of autumn, it's now shifting towards being more in December or January instead.
And is that climate change?
The changing picture of the winds, it's unclear whether that's related to climate change, but the impact of climate change is undeniably in the picture when it comes to wildfires.
It sort of sets the scene, if you like.
Whilst climate change, of course, isn't holding a box of matches, it's not going to trigger the fires themselves.
It's helping to create the conditions for them to be more powerful and to cover larger areas when they do occur.
And climate change sort of preconditions the land, if you like, so that those fires are more intense, more prolonged, and that they might happen more frequently.
Well, thank you, Ella Gilbert, climate scientist from the British Antarctic Survey, and we'll watch the news with our fingers crossed, I guess.
On Monday, Donald Trump enters the White House as the 47th President of the United States, his second go at the biggest job in politics.
In his intray, a whole array of challenges, and at the top of the pile, the climate crisis, space exploration, and AI.
It's clear science is going to matter.
So, with issues like these so central to the political agenda, what might a Trump presidency mean for science?
I spoke to Jeff Tulofsson, who's long reported on US politics and science for nature, and we started with a topic that's looming larger than most, the climate.
You know, you have to think about this on two fronts.
There's the domestic front, where climate policies that have been enacted under Biden this time, those will probably get shelved or rolled back or weakened.
So long-term efforts to curb emissions are going to probably take a back seat, at least for four years.
On the international front, Trump is likely to pull out of the Paris Agreement again.
This doesn't matter for U.S.
emissions so much, but it does matter in terms of global cooperation on climate.
And a lot of the policy experts I've spoken to say that if the U.S.
pulls out again, that will give political cover for other countries to do less.
Are Americans concerned about that?
Polling has long suggested that Americans are concerned about climate, by and large.
And if you ask them simple questions, you know, would you support climate policies that do X in general, the answer is often yes.
But for whatever reason, those types of concerns have not translated necessarily to action at the political level.
I mean, I have heard that when it came to the US coal industry, which Trump has publicly backed, he talked the talk, but he didn't actually do anything to save it last time round.
So is there a possibility that he's just going to support whatever is good for the economy, which might well be renewable technology?
Trump spoke a lot about trying to revive the coal industry the last time around, and that did not happen.
Coal continued its decline, it's been declining for at least a couple of decades, rather sharply over the last decade.
And probably that is going to continue.
Basically, the bigger issues are that natural gas is cheaper, and clean energy, like wind and solar, the prices have plummeted for these technologies.
So they are gradually taking over the market.
And those basic trends are not going to change in the next four years.
Coal is probably going to continue to decline and be replaced with natural gas and renewables.
So U.S.
emissions might even continue to decline.
But that doesn't mean that the Trump administration doesn't have control over policies that affect investments that lay the groundwork for decades to come.
you know, investments in clean energy technologies and RD.
And also, they can roll back regulations that are designed to
hasten
these market shifts toward clean energy.
So they can slow down the pace of progress in some ways.
Moving on to AI, what's Trump pledged to do on AI?
What he has done is he's brought in a lot of people who are interested in these issues.
In particular, David Sachs, another billionaire tech investor.
He's going to be the new administration's new AI and crypto czar.
So people within the technology industry are very well placed within this administration.
Also, don't forget
Elon Musk and Tesla and other companies.
So the tech industry has asserted itself under this incoming administration, and it looks like they're going to have a lot of influence.
The administration has talked about trying to pave a way for that industry to move without being burdened by burdensome regulation.
It's the simple way to put it.
So I would expect that whatever is done, it will be friendly to this budding industry.
What are AI experts and scientists actually saying?
There are those who worry about this and who want to see more rules of the road, who want to see checks to make sure that we understand this technology before we unleash it.
Keep in mind that Elon Musk has been skeptical and has said he's worried about unregulated AI or, you know, AI without rules.
So we'll have to see how this debate plays out.
We've talked before on this programme about the Donald Trump-Musk dynamic and the fact that they're both quite enthusiastic about space exploration.
Do you think space exploration is going to be an area of push during the next four years?
You've got Elon Musk, who owns SpaceX, and Trump's appointment to head NASA, Jared Isaacman, is another tech billionaire and a space enthusiast who is not only friendly with Elon Musk, but has also flown in space twice
on his spacecraft.
So there's every reason to expect that human spaceflight and the space program at NASA is going to do well.
There are questions out there about, you know, NASA is developing its
one spacecraft for human spaceflight and SpaceX has another.
It may be that there are going to be pressures for NASA to drop its version and go with SpaceX.
But human spaceflight writ large is likely to do well.
And Jeff, we've been talking about the big science topics.
All in all, when it comes to science, how different will this Trump presidency be to his first?
Has his position on any of the major issues changed?
By and large,
his kind of big picture views have not changed.
One thing that clearly has changed is
the organization on this front.
Keep in mind that, you know, during his first term, it took him a long time to identify and appoint a science advisor and really get up
the infrastructure that scientists are used to when it comes to advising administrations.
This time, these appointments have come very early on in the process, and it looks like this administration wants to get this process up and running from the get-go.
So, this is already a big change in terms of the way that scientists may interact with this administration.
And I think most scientists would be pleased with that.
Who are his key appointments to science?
The main one is Michael Kratzios, who will serve as his science advisor and will help lead his administration's policies across the board on science.
Kratzios was involved in his administration the last time.
He is not a scientist per se, but he has experience in the tech world.
And the scientists I've spoken to are generally appreciative of his work in this area.
What this administration, what Trump and Musk and a lot of their colleagues have proposed is massive, massive cuts in terms of funding and staffing.
It's kind of clear that if you get the kind of cuts that they're talking about making, it's going to be very hard for science to kind of move forward unscathed, as it will be hard for...
any other field to move forward unscathed.
Donald Trump is just doing exactly exactly what he was elected to carry out doing, right?
Yeah, he's been very upfront about this agenda.
Jeff Tollefson, thank you so much for coming on to Inside Science and sharing your expertise.
Thank you, Marnie.
Every six months or so, I have another go at cooking a famous Italian pasta dish called Caccio Pepe.
The recipe only needs three ingredients, pasta, pecorino cheese, and black pepper.
How hard can that be?
Well, most of my efforts fail as my sauce balls into hard lumps that I then have to chisel off the whisk.
So it was with joy that I spotted that physicists at the Max Planck Institute in Germany claim to have created a foolproof science-optimized recipe.
So over in my kitchen with spaghetti at the ready, I call up one of the researchers, Ivan de Talitzi, to talk me through it.
So I'm going to put the spaghetti on.
Wait, wait, wait, wait, wait.
Before the spaghetti, you should do the jelly-fied starch.
Ah, you have to do that bit first.
So, um...
Three grams of cornstarch.
You need, let's say, 40 grams of water and put it in a small pot on the fire and you wait until this mixture turns from cloudy and white to
viscous and translucent.
Okay.
The idea is that among the ingredients of pasta la cache pepe, you have have pasta water.
The pasta water is the stabilizant of the sauce and is nothing else than jellified starch in water.
So what we are doing here is scientifically control the amount of starch by doing the jellification of starch directly.
Is that why
caccio pepe is so hard?
Yeah.
Because I mean for the uninitiated it involves melting the cheese with some of the pasta water to make this smooth sauce.
So what it's not telling you is that there's this secret ingredient that you need to think about, the starch.
So starch really stabilizes the mixture of water and cheese so that it really prevents the proteins to clamp all together and form lumpy sauce that is not good.
So this is why your paper spoke to me because I try and make this dish every so often and most of the time I'd say I end up with all of my expensive cheese that I've spent ages grating has just clumped itself onto the whisk and then I get angry.
You said something that to which I feel really attached to because imagine you are an Italian, you invite some German guest to your home to try this wonderful Italian dish.
and then you mess it up and maybe they are too polite to say no this is not good.
They say, no, it's good, it's good.
But you know.
It's fine, it's fine.
You know they're judging you.
Exactly.
So
do you see the gel forming already?
It's still cloudy.
Okay, so you wait a little bit more.
Okay, well, while we're waiting, tell me a bit more about this paper.
We started talking about this idea because in our institute we study also phase separation, which is this physical theory that explains how mixture are stable with respect to temperature, we said, look, we know the cooking process, we know the physics, we should put it together.
And your paper contains something called a phase diagram?
Yeah, you can find on a lot of physics books the phase diagram of water, which very trivially tells you if the temperature is above 100 degrees, then you have vapor.
So the phase of water is vapor.
If you go below 100, then you have the liquid state.
And below zero, you have the solid phase.
Of course this can be generalized to other substances and we wanted to do a phase diagram for the sauce of cache pepper as a function of starch and temperature.
So we have these phase diagrams where you have a region for low starch concentration and high temperatures where you have mozzarella.
This is what we call the mozzarella phase.
and another region where the sauce is homogeneous instead.
Okay, so to stop your cheese from clumping you need a certain amount
of starch.
Exactly.
Talking of which, my cornstarch and water has turned into this jelly-like consistency.
So should I put the spaghetti on?
Yeah, so while you cook the pasta, we are going to blend the cheese with the jelly-fied starch and we are going to do it at low temperatures.
Because if you go above 60 degrees, the proteins will denaturate and they will stick together.
That's what's happening.
I think when I put my cheese in the pan, it's just hotter than 60 and
it's all clumping.
And that's a chin reaction, like you cannot stop it.
It's looking a lot like cheese sauce.
Nice.
So listen, are you going to tackle pizza next or are you back to statistical physics?
We are thinking about gnocchi.
Mm.
I look forward to the paper on that.
Thank you so much, Ivan, for coming onto Inside Science.
You're very welcome.
I've got to say goodbye because I need to drain my pasta now.
Okay, and let me know how it was, honestly.
I will do.
I will report back.
So, thanks to Ivan de Talitzi, and my dinner awaits.
Yeah, that's good.
Yum.
So, if you want to try this foolproof caccio pepe with added physics, the recipe will be on the Inside Science programme page.
And if you have any food science questions, or indeed any questions, please do email us.
insidescience at bbc.co.uk.
I'll be back next week.
Till then, goodbye.
You've been listening to BBC Inside Science with me, Marnie Chesterton.
The producers were Gerry Holt and Sophie Ormiston.
Technical production was by Rhys Morris.
The show was made in Cardiff by BBC Wales and West.
To discover more fascinating science content, head to bbc.co.uk, search for BBC Inside Science, and follow the links to the Open University.