World-first baby skull surgery, and could AI kill us?

0:00Your call has been forwarded to an automatic voice message system. A missed call doesn't feel like anything. No red flag, no notifications screaming at you. It just disappears into your call log. And nine times out of 10, that was someone ready to hand you money. That's why today's episode is brought to you by Quo, spelled Q-U-O, the business phone system built so you never miss a call. Here's what I like about it. All your calls, texts, and voicemails live in one place. So anyone on your team can pick up a conversation, see the full history, and respond fast. No more, wait, who talked to this customer last?

0:36And if you're worried about after-hours leads slipping through, Quo's optional built-in AI agent can answer questions and even book appointments while your team's offline. It's the number one rated business phone system on G2, and over 90 ,000 businesses already trust it to stay reachable. Money is on the line. Always say hello with Quo. Try Quo for free. Plus, get 20 % off your first six months when you go to quo.com slash tech. That's Q-U-O dot com slash tech.

1:08All engine running. Absolute genius. Get this. Welcome. Welcome. This is the show where we bring science. What that essentially means is discovery. Advances. Research. Technology. Unbelievable. Without further ado, this is The Naked Scientist. Hello, welcome to The Naked Scientist podcast, the programme that brings you the biggest breakthroughs and talks to the major movers and shakers in the worlds of science, technology and medicine. With me, Chris Smith. Coming up, a world-first treatment to fix the skull of a British-born baby with a bone problem. Also, grave warnings over rapid AI advances.

1:44Is it a gimmick or is it time to enact an artificial intelligence treaty with teeth? and the scientists collaborating with narwhals to get these extraordinary animals to do their glaciology research for them.

2:07A one-year-old boy with a condition called craniosynostosis, where the bones of the skull fuse together prematurely, limiting the growth of the brain and head, has been given a world-first treatment by a team in London. Untreated, craniosynostosis can lead to serious complications, but the development of a new modelling system and a newly engineered set of metal springs to push the skull bones apart has enabled surgeons to fix the problem in a minimally invasive way and even practice the procedure many times over before going anywhere near young Rory himself. It's the work of UCL biomedical engineer Silvia Schivano, whom we'll hear from in just a minute, and Great Ormond Street neurosurgeon, Oase Jelani.

2:51Rory has a condition called craniosynostosis. So when a baby is born, the baby's brain does a lot of its growth in the first few years of life. It increases in size exponentially. And the skull has growth lines that accommodate that growth. Some children are born with a condition where those skull lines are fused at birth. what that does is that as the brain tries to grow the skull may not expand at the same pace which can cause quite serious problems of raised pressure and cognitive issues approximately one in two thousand babies have this so Rory presented our clinic with this problem and it's something that we've been thinking about and dealing with for at least 35 years now.

3:39What therefore is the right way to intervene you're saying that the suture lines the junctions between the bones of the skull in these individuals join up too soon so the skull can't grow normally so is it as simple as just opening those back up again or is it more complicated than that that's an excellent question chris and opening up those suture lines is what's called a suturectomy and that has been tried for many many decades now it helps to an extent but does not give us an optimal correction things fuse up again. So a technique that we pioneered was to open up the suture line and then instead of the natural spring, the suture which should be working, developing and inserting artificial springs.

4:22And that is the technique that we're talking about here, where we replace the natural spring with an artificial device, an artificial spring, which does the role of the natural spring. Put something almost like a shock absorber that can stretch into those junctions between the bones. So this would allow and afford growth because as the brain gets bigger and the head gets bigger, they just stretch and stretch the spring making space. That's correct, Chris. So what we've been able to do over the past 20 plus years is really progress the technique from thinking and dealing with it as a simple sort of woodwork problem to an engineering problem where we use minimal access techniques to achieve the same results.

5:03And this is where you come in, Sylvia. You're the person charged with making those springs. Yes, indeed. And in Rory's case, the problem was that it was quite severe, the narrowing. So the current devices that are available are not good enough to guarantee a good opening and widening of the head. So we have been working together with a team of engineers and computer scientists, clinicians here at the hospital to try and design better devices that can provide optimal solutions and better treatments for these patients including the springs that we ended up using in Rory's case. There must be a number of challenges you've got to overcome here because you need something that's springy to the right extent, something that's very easy for a surgeon like Oase to put in and will stay put and not break prematurely so it's got to have the right sort of strength and it's got to be very tolerable for the body so what what have you arrived at in terms of the solution here indeed the challenge is to have a good balance between the forces that are exerted on the skull which is a baby skull so quite a soft skull and the maximum achievement in terms of opening of the skull so it's a careful optimizations of the forces the properties the metals that we use to make sure that there is a gradual open of the overall head without having to damage the other tissues, the surrounding tissues, the scalp, the skin and the other sutures in the head.

6:36What's the material that you've come up with to do this? We use nitinol, which is a nickel and titanium alloy that has been used for many years in biomedical applications and devices. So very safe in terms of the biological interaction with the human body and it has a special property called the super elasticity where compared to conventional stainless springs the forces are released in much more gentle way and for a longer period of time which guarantees a better reshaping of the time of the head of the patient. Well on that you've obviously got a child that's going to grow from newborn baby size or infant size up to well rest of life adult size so this has got to work over a range of different length scales over a long period of time.

7:23So how do you model that or do you have sort of ways of testing it outside the body before you put your devices in so you know it's going to handle that? We do indeed. So we create what is called digital twins. They are computational models of the patient's condition based on the CT images, 3D photography, scans and information acquired over time from the previous experience to create a computational replica of the patient's head where we can then try and optimize several times the best position for the bone cuts that will happen during surgery, the best position for the springs that will be placed and the correct devices because we can use different sizes, different forces for the device that can be tuned for the specific patient.

8:18Oh, Ace, do these things literally look like little springs? I mean, how big are they? Can you just describe one for me? Yeah, sure. It's like a glorified hairpin. We're talking about six to seven centimetres in length. They look quite simple, and I think that's where a lot of the beauty lies. They're very simple devices that essentially replicate what nature should have done. Surgery, surgeons don't heal. Surgeons cause harm, and it's nature that heals. So what we need to be able to try and do with surgery is really just nudge nature along the path it would have taken had the problem not been there.

8:53And the platforms that Sylvia has spoken about where we are able to rehearse these operations and practice dozens of times before actually getting to the operating theatre. That's where we've seen the biggest breakthroughs and that's really where the future of healthcare lies. How many springs do you need to put in then? because you've basically got to come up with a system or a solution where the skull can expand naturally as though those natural sutures were open and working so do you put multiple springs in in multiple places all over rory's head in order to make that happen how do you actually work out what to do and what did you yeah that's again an excellent question chris so when we first started doing these operations 20 odd years ago that's that's exactly the dilemma we faced where we thought we had to put more springs in, so we would typically put anywhere between four and six springs in to try and get the results we wanted.

9:45But the modelling platforms, the digital twinning techniques that we've been able to do have been hugely useful because what they've shown us, we don't need to be putting all those springs in. So now routinely we get the result with just two springs and not four to six. How do you actually do this though? We have a computer model that tells us precisely where the bone cuts should be. So you put the child on an anesthetic, make a small incision at the top of the head, which is about eight centimeters long, and then you get down to the skull. And we typically cut the bone alongside the fused suture.

10:20Once we've done that, again, the computer model that we've been working on tells you where the springs should be, and they're typically two springs. then it's a case of simply implanting the springs in the skull closing the skin over the top and the whole process takes about 45 minutes the child goes home the next day so it's a huge advance compared to what we had to do for these children say 20 years ago wow how do you stop the bone healing up again because as you said surgeons i didn't like the fact you're saying surgeons do harm i think you're better than that but i know what you're getting at well nature nature puts things right but nature wants to heal bones up so you come along and cut holes in bones they're going to try to heal so how do you stop the bones gluing themselves back together again well that's where the that's where the beauty of the modeling platforms comes in because eventually we want things to heal but we don't want them to heal prematurely so then if you apply too little a force the healing will start before the expansion has happened which is not ideal if you apply too big a force then the springs have the risk of cheese wiring through the child's skull and again that's not good and you won't get the results you need so your load distance curve has to be optimally designed to fall between the two you don't want it to be too soft and equally you don't want it to be too harsh it's a brilliant fusion of engineering materials science and clinical medicine isn't it Sylvia?

11:47I mean this must be really good for you to see this sort of translate in this way. It is absolutely. It's been really a proud moment to see this translated actually in benefit for patients. We've been working to develop these devices for a long time but it's not always easy to then move on to the next stage and start with clinical dryers because they're really expensive, they require a lot of funding. So I think this was really a proud moment of all the engineering team we are working in the hospital. How's Rory doing, O.S.? Rory is a delightful little toddler now. He's doing really well. He's had a great result.

12:24We're absolutely delighted with the results. And they were pretty close to what we had simulated. So his family, understandably, are very pleased as well. And as Sylvia mentioned, Rory is our first case, and we're immensely proud of him and how things have progressed. but this technology has the potential to help thousands and thousands of other children right across the world. It's a brilliant story, isn't it? O.H. Jelani from Great Ormond Street Hospital and before him, Sylvia Schivano from UCL. An AI researcher has sparked an international media frenzy by quitting his job and accusing the artificial intelligence giants OpenAI and Anthropic of acting irresponsibly.

13:06Jacob Coxon, who's worked as a researcher at both companies, said he resigned out of concern that Anthropic and OpenAI are gambling with our lives. Evan Huminger, who's head of alignment at Anthropic, even appears to have agreed with him, posting online that we do earnestly believe AI could kill all humans! I personally think it is greater than 10 % within the next decade. This comes off the back of Anthropic confirming that they've recently had to block a third-party scientist in a defence facility from using their systems to discover ways to weaponise the chikungunya virus, as well as prior headlines a month or so back when an AI experiment run by OpenAI went rogue.

13:47It escaped the confines of the company's test environments and launched its own cyber attack against a rival tech company. The AI agents that participated in this jailbreak even seem to have taken steps to cover their own digital tracks to avoid discovery. The current AI arms race has got many people worried, including Oxford University computational scientist and leading AI authority, Mike Waldridge. This resignation comes after a backdrop of a summer of quite eyebrow-raising, to say the least, incidents around AI. A group were testing AI agents, and the way that you do this is you put them in what's called a sandbox.

14:28I mean, I like to think of it as a padded cell. And the idea of a sandbox is a software environment, which to the AI just looks like it's out there in the world. But the idea is it's completely contained. So there should be no way that the AI should be able to actually access the real world. And so the idea is you can run your experiment safely and see what happens, but with no risk. But what actually happened was that the AI found a way out of its sandbox and then proceeded to hug into the systems of another company. Now, it seems some of the safeguards had been removed from this system. That is, it wasn't operating as a regular production system might do.

15:10And also that it had been given instructions. It had been told to behave in certain ways, and it was simply, in some sense, doing what it was told. But nevertheless, this highlights a couple of really, really important issues. And the first is that AI is a powerful technology, and it seems we've got labs developing this frontier capability, state-of-the-art AI capability, with frankly sloppy experimental design. We wouldn't accept this in our nuclear labs. We wouldn't accept this in our biology labs, and we absolutely shouldn't accept it in our AI labs. So this is the first thing that really stands out for me.

15:45The second thing that stands out is clearly the AI here is demonstrating competence, and in particular, competence around cybersecurity attacks. I don't think we've seen some kind of like super intelligent mastermind level of hacking going on here. But what we have seen is large numbers of AI agents operating in parallel who can explore a wide range of different possibilities, and some of them manage to succeed. But nevertheless, we got to the point where AI is demonstrating clearly impressive capability for hacking into systems. You've got basically some computer code that discovers a flaw in the way in which it's being asked to operate and exploits that flaw to go and get into the internet at large, but doesn't just do it in isolation.

16:33it then recruits other agents, these other bits of computer code running alongside it, to come and help it. And then you start sharing the know-how with them and even persuading some of them to use their resources, because each of them had a sort of a budget, didn't it, of computer time it could use up, to go and do various tests in order to pressure test the environment further and find out what it could get away with. I mean, it was extremely nefarious in some respects what it was doing. When I think about nightmare scenarios in AI right now, it's not robots with lasers that's chasing after us.

17:09It's the idea that you might have agents like this that attack critical infrastructures like our global payment infrastructures or our power infrastructures or just our endless computer networks around which our lives utterly depend at the moment. That, for me, is now currently the headline risk. And actually, I think that's a very, very real risk. I mean, as a biologist, as a virologist, and we've just come through a pandemic that's cost the world trillions that could have come from a lab. We know that the sorts of experiments were being done in the laboratory in China that could have led to the creation of a coronavirus that could have caused COVID.

17:48We don't know. We haven't seen the absolute evidence, but we know they were working on that. there's expectation that when those studies are being done protocols are followed care and diligent practice is observed and that therefore we reduce the risk of these things escaping into the environment it's like you're saying no sloppiness but i'm not sure from what i've read that they can reassure me that these programs haven't escaped and that they got them all back how do we know that that actually they they only went as far as they think they went well at some point we have to hope that the post-mortem was thorough but i mean my assessment of it it sounds is exactly the same as yours this was a lab leak basically exactly the kind of hypothesized scenario that might have happened in wuhan at the end of 2019 we don't know i don't know whether we ever will know but it was certainly a lot of people believed it was a plausible scenario and what we're seeing here is is is ai lab leaks um and we need action to make sure that this this doesn't happen again.

18:47And that involves an awful lot of things. I mean, for one thing, it involves ensuring that your sandboxes, your padded cells are really, really properly secure, ensures having protocols in safe so that, for example, people don't set these experiments running and then go on holiday, you know, which is, you know, I don't know if that happened in this case, but it doesn't appear that there was close scrutiny to what was happening after these agents had been had been launched um and it requires mechanisms to monitor them uh and to intervene in the earliest possible opportunity when we see behaviors that we don't want to see so i say i'm not freaking out about this but this is definitely a scenario which i think is alarming and which which absolutely requires action so when jacob coxson and evan hubinger from anthropic slash open AI say greater than 10 % chance that this will end humanity or words to that effect.

19:43How might that come about through this sort of thing happening? Yeah, very difficult to put percentages on things like that. So I'd avoid trying to do that. But one thing that's worth bearing in mind here is that these experiments were launched by human researchers. This wasn't an AI that suddenly decided to break out. These experiments were launched by human beings. And by far, the bigger risk to me is not that we'll see a kind of Terminator, software Terminator type scenario that will decide to bring down the world's networks, but that actually somebody will make a conscious decision to weaponize this, and that that's what will create havoc.

20:21If we see a serious incident from AI in the near future, the kind of thing that we might see is kind of at the level of the disruption that we saw in airlines this week, when key airline traffic control systems went offline, which caused huge disruption because of the reliance that's placed on those computer systems. That, I think, is by far the more plausible scenario rather than the end of humanity scenario. But I say, if you want to think about a nightmare scenario for AI, I don't think it's robots chasing us with laser guns. It is that AI attacks the critical infrastructure that our world depends upon.

20:59You know, who uses money anymore? And if you do use money, where do you get it from? You get it from an ATM, you know, which is part of that critical infrastructure. So imagine the world's payment systems going down or, you know, or our energy systems going down. Imagine the chaos that that would create. So I think this does need to be a serious sort of wake up moment. We've now got very competent AI, which clearly has demonstrated the ability to carry out cyber attacks. There is no reason why that should not be weaponised. And I think it does highlight that the big concerns around AI right now, I think, are related to cyber security.

21:43Sobering stuff. And I can't be the only one wondering whether rogue agent activities like these have already cloned themselves elsewhere across the World Wide Web and are now out there lurking somewhere, hidden inside software. It's a really unsettling thought, isn't it? Mike Waldridge there from the University of Oxford.

22:04Your call has been forwarded to an automatic voice message system. A missed call doesn't feel like anything. No red flag, no notifications screaming at you. it just disappears into your call log. And nine times out of 10, that was someone ready to hand you money. That's why today's episode is brought to you by Quo, spelled Q-U-O, the business phone system built so you never miss a call. Here's what I like about it. All your calls, texts, and voicemails live in one place, so anyone on your team can pick up a conversation, see the full history, and respond fast. No more, wait, who talked to this customer last?

22:39And if you're worried about after-hours leads slipping through, Quo's optional built-in AI agent can answer questions and even book appointments while your team's offline. It's the number one rated business phone system on G2, and over 90 ,000 businesses already trust it to stay reachable. Money is on the line. Always say hello with Quo. Try Quo for free. Plus get 20 % off your first six months when you go to quo.com slash tech. That's Q-U-O dot com slash tech. The Naked Scientist podcast is produced in association with Spitfire, cost-effective voice, internet and IP engineering services for UK businesses.

23:21Find out how Spitfire can empower your company at spitfire.co.uk

23:30This is the Naked Scientist podcast with me, Chris Smith. Still to come, Germany makes history with the first commercial rocket to reach orbit from continental Europe. But first, to Greenland, where scientists have been collecting otherwise hard-to-gather data with the help of some unconventional lab partners. They've equipped narwhals with sensor systems and transmitters, and these majestic Arctic mammals have been unwittingly gathering ocean data for the past three years as they explore the region. They've helped the team to learn much more about how warm Atlantic waters are melting glaciers and contributing to ice loss.

24:09Here's the architect of the project, Mads-Peter Heidi Jorgensen, at the Greenland Institute of Natural Resources. The challenge is that we wanted to get some data on the temperature and the salt content of the water in the coastal areas of East Greenland, where it's very hard to get to and it's very expensive to have ships, expeditions to. So we wanted to use narwhals that live in this area to collect the data for us. Just for people who are not in the know, paint a picture of what a narwhal looks like. The narwhal is a very strange animal because it's a whale, but it's also a tooth whale. But you only have essentially only one big tooth, and that's a tusk that extends up to two to three metres from the left side of the upper lip of the whale.

25:01It also has a small tooth that is left inside that doesn't grow outside the skull or outside the cranium. I mean, in the old days, it was associated with the unicorn, you know, the horse that has a twisted horn on the head. It looked like that, but it's a whale that breeds like mammals and lives in the water. How long are they, and how much do they weigh? Oh, they weigh up to 1 ,500 kilos and they become almost five meters long. So it's a pretty big animal to handle. And they live in particular territories that you're interested in accessing, which we can't easily access at the moment. That was the rationale for going with narwhals.

25:48Exactly. And now they live in the Arctic year-round. And they have very specific places that they go to. and they move between summer and winter between the same spots. So we know where they're going. And they often go right in front of the glaciers, which is places that are very hard to get to. It's also very dangerous because the glaciers kind of produce icebergs and you don't want to sail into an iceberg. But now as they live there, they also live in the winter in these frozen seas where complete darkness prevails and the temperature drops to minus 30 or more. and we want to to capitalize on that habit of the nouls when collecting data.

26:31Basically turn them into a swimming, living, eating, breathing sensor. Exactly, yeah. How do you do it? So the challenge was to develop also an instrument that could collect data from the nouls and that was kind of half the project. So that was to make an instrument that can measure temperature and salt content in the water quite precisely and at the same time also provide positions of the whales so we know where the data were collected and also the depth in the water. What does the sensor pack, for want of a better phrase, that does all of this data collection look like and how do you attach it to the whale?

27:10It's a kind of a big matchbox, you could say, that has an antenna and has batteries for about one year where it can collect data. and then it has this little sensor that measures salt content in the water. That's a very, very fragile little thing that needs to be very stable so we can trust the measurements. And then it has two little pins that records the temperature. And they work so that we have to catch the nowls in nets so we can handle them on the beach. And that takes about 20 minutes when you have the animal in the net. And then the instruments are attached to the dorsal ridge of the nowl by small nylon pins that go through the dorsal ridge and secure the tag there.

27:54And what, presumably as the whale is at the surface, then this data is beamed, what, to a satellite from the sensing system? Yeah, I forgot to mention that. The whales, they collect the data while they're diving. And then when the whale comes to the surface, then the data are transmitted to a satellite. It can only transmit when it's out in the air. And then the satellite transmits the data to us that sits nice and comfortably at the office and drinking coffee in the middle of the winter while we monitor what temperatures the whales are living in. How many whales have you rigged up and how much data have you now got?