Three reasons AI could ‘save us all’: This is why

0:02Sky News, the full story first.

0:11Last time we talked about AI, we explored the ways it might kill us all. Now, I'm told it could save the world. I have questions and this is why.

0:29Online sadistic groups are targeting vulnerable children. The sheer amount of potential victims is astounding. It's a very real threat, and it is here in the UK. It will always shock me. A young boy could commit such awful offences. Some of the content has been the worst I've ever seen.

0:52Tom Clarke:Parents need to be aware of what can happen on these sites. Find The Calm on Sky News YouTube.

1:02Hi everyone, Neil here, and I'm trying something new. I'm going to be positive about AI. Yeah, I know, let's see how long it lasts. Because the warnings about AI are coming thick and fast. It's coming for our jobs, it's going to fool us, we'll lose control of it. And if you listen to some of the people who understand this stuff best, the conversation gets rather more alarming than that. In fact, you might recall I asked our technology correspondent Roland Manthorpe how AI might bring an abrupt end to the human race. We are under attack. But I pray you've forgotten his disturbingly organised answer.

1:39I think we can sort of split this into three, right? three ways that AI could kill us all. One is a human using AI to do something that wasn't possible before, like create a super virus. Two is some kind of Terminator Skynet scenario. And the third is some variation on what people call the paperclip problem. Three ways. Yeah, I'd have been quite content worrying about one. Now, you all know me as a beacon of positivity. So having spent that episode contemplating the extinction of humanity, I thought it was time to ask a slightly different question. What if AI actually does some good? Tom Clark is Sky Science and Technology Editor.

2:22Tom, as you well know, I'm very much a glass half full type of guy in general. But of course, when our colleague Roland was on last time, he did issue something of a warning about AI. I mean, first and foremost, where do you genuinely stand on this phenomenon?

2:37Tom Clarke:It's a bit of a relief for me, actually, because I deal with climate change and pandemics and a lot of doom. And I've just been reporting on how we regulate dangerous AI and those risks are definitely there. But I have to admit, with my science background and generally being positive about what technology has done for us and will continue to do long into the future, hopefully, there's just so much potential in AI. There's so much good it can do. and it's I think definitely the case that certainly the recent re-ignition of concerns about AI safety have sort of dampened all that down. We are sort of taking our eye off a bit what potential this has and some of the benefits it can bring.

3:16Tom Clarke:So I'm more than happy to talk about it. Let's go for it. A few areas I think we can probably run through on that podcast today. But of course something that every single individual considers and may well be whilst listening to this in the throes of something related to this topic. I mean, our health, our health is paramount. And AI, even I know, is making some ground in that area. I mean, in fact, some of the most significant early advances in AI have been in that whole space. Like, AIs are very good at spotting patterns in data. So for medical imaging, we had quite early advances showing that AI tools, very limited ones, not your big chat GPT, you know, super intelligent brains, but things designed to screen medical images like a breast cancer scan, for example, could be just as good, if not better, than humans.

4:02Tom Clarke:And provided you come up with a system that's got a human in the loop, they can sift through a lot more information, a lot more quickly, save a lot of time, get referrals quicker. That's just the start, though. Another early advance was using AI to solve what's effectively a mathematical conundrum. How do proteins fold up? And it won the Nobel Prize last year for Google DeepMind, which was a UK AI lab acquired by Google back in 2015 that pioneered all this research. Wait, wait, wait, wait, wait, one second. Yes. Proteins folding. What are we talking about? We're talking about protein folding. Absolutely.

4:36Tom Clarke:It's a bit, it's a bit esoteric and a bit niche, but it's really fundamental to everything. Yeah, proteins, we need to eat them because they're the building blocks of life, but because also they're the building blocks of life, they're the things that make cells work. They're the things that make diseases happen as well, because they're the receptors on the surfaces of viruses. They are the machinery within bacteria that allow them to become resistant to drugs. They are the targets of cancer medicines and maybe cancer medicines themselves. You've probably heard about using immune proteins like antibodies to treat cancer.

5:06Tom Clarke:They're proteins. And if you can understand the shapes of them and how to tweak those shapes, you've got drugs. You've got huge advances in biology. Before, to get the shape of a protein, they're 3D structures. They fit together in reality and life. Am I right to think of like, you know, a kind of virus as a jigsaw piece and the kind of the solution to it as another jigsaw piece fitting together? Yeah, you might remember from COVID, you know, we know that there was a shape of a protein on the surface of the COVID virus that allows it to connect, bind to your cells and therefore infect you. That same process is repeated in biology throughout all of life.

5:44Tom Clarke:It's very much a three-dimensional enterprise. That's how drugs work. That's how viruses interact with our bodies. That's how our immune system fights diseases. So the shape of proteins, this fundamental issue in biology, it used to take a very complex and extremely skilled process to get the shape of a protein, something called x-ray crystallography. And it takes sometimes years to get the actual 3D structure of protein nailed down. A lot of scientific effort and help. and here along comes an AI that given the sequence of amino acids those are the building blocks of proteins it can basically guess and then learn and through reinforcement learning figure out what the shape of protein will be it has the potential to revolutionize biology we are discovering protein structures that we couldn't have guessed that before it also offers the potential if you understand how a protein can unfold you can work backwards and say I want a protein that's this shape, AI designed me the amino acid sequence I'd need to make it.

6:44Tom Clarke:So if you're trying to come up with an innovative new drug, something to fit a new target, the model can do that for you. Now, there is a bit of a caveat here. That modelling ability has been around for a few years. We are yet to see those first AI drugs go through clinical trials to actually get approved. We haven't yet, as I understand it, gotten a purely AI designed drug that's gone through a safety regulator. But we do have them in trials. And it would be very naive to say that the potential for significant new advances and discoveries are made using these new techniques. And then you've got, yes, because you can do a lot of the, what used to be trial and error in developing a new medicine, screening chemicals, whether they'll be suitable and whether they'll cause harm in a human, what AI allows you to do is radically reduce that pipeline, speed things up.

7:32Tom Clarke:The idea being you can winnow out the kind of the toxic ones, the ones that aren't likely to bind particularly well. So in drug development, the kind of pipeline they talk about in the pharmaceutical industry, AI is definitely already being used to accelerate that process. I think you hear lots of scientists, and these are sensible, rational people who aren't AI drum bangers, saying, you know, there is the potential here for huge advances. And I think we do have to just give a little bit of time for them to work through. We could be talking potentially millions of lives being saved by this piece of technology.

8:04Tom Clarke:Absolutely. And it could be much more personalised. You hear about personalised medicine, but it does conjure the idea that that could be much more honed to you. And we could see much better, potentially much cheaper, more accessible medicines using this approach. Well, let's talk about another area of science, because again, it just feels to me that this is where AI could really reap the benefits. Before artificial intelligence came along, the existential threat that you and I most enjoyed talking about on the podcast was, of course, climate change. I mean, where are we with that? I mean, could AI save us?

8:36Tom Clarke:Now, of course, there's a little bit of an irony here in that we could have to burn quite a lot more fuel or get a lot more energy from somewhere just to power the AI to solve some of these problems. But, you know, there is hope that that process will get more efficient. I think that market forces are driving it to get more efficient anyway. So I personally don't think AI is going to be the thing that drives us towards a much, much worse climate, provided, and this is the caveat, if we divert some of that AI brainpower, some of that knowledge towards solving the most important problems in climate.

9:08Tom Clarke:Some people are already doing that. We've seen AI models of our global atmosphere that are much better, by some measures, at predicting the weather. Now, predicting weather is really, really hard. It takes a lot of computer capacity. You've now got these huge data centers running these very clever models that can do that for us. And that could be important because we know, just look at what's happening with El Niño or our summer this summer, the earth is getting hotter. One of the key questions is, well, how's that going to affect our lived experience, the weather we experience because of a warming climate?

9:38Tom Clarke:It's very hard for scientists right now to predict. AI promises to be able to much more accurately model those things. And it's already being deployed in satellite imagery, being trained on that kind of stuff to give us much better answers. What's the utility of that beyond pointing out where things are going catastrophically wrong because of man-made climate change? Given it's already happening and given we're pretty much certain to get two degrees of warming now, you know, if you're a farmer, where are you going to be able to grow certain crops? Where is the rain going to fall? How much can I expect?

10:06Tom Clarke:If you're trying to predict how we adapt, how should we shift what we do, the ability to forecast that, predict it, couldn't be more important right now. But there's another area in which, you know, the potential to help, and this is going back to that kind of basic science, we're talking about protein folding and sort of using AIs to model stuff. AIs are also proving themselves to be very good at understanding how chemicals fit together, right? So material scientists, they don't get a huge amount of press, I have to admit. It's a fairly - It's not the sexist. I mean, Ed Conway is doing his bit.

10:35Tom Clarke:Ed Conway, you get him, he could sell you a material scientist, possibly much better than I. But if you think about lithium batteries, LEDs, the kind of coatings and linings they put inside engines and nuclear reactors. Those are all specialist materials. You are going to need a lot more of those if we're going to radically change the way we get energy. If we're not going to burn fossil fuels, how are we going to get cheaper, more available energy? And one key thing in that, batteries are a really good example, is you need new materials that have new properties. You want them to be lighted. You want them to restore more energy, tolerate more heat.

11:11Tom Clarke:Right now, what a material scientist does, he goes, right, that's the problem I want to solve. I need to just go through a bunch of known materials that we've got from previous things we've used them for and see whether anything fits. Now what we're seeing is the kind of the reverse of that process where you invert the problem and you actually say to the AI, this is the problem I've got. You know how all these different chemical compounds can potentially, the billions of ways they can combine with each other, come up with some for me and they are generating previously unknown materials. And also if you automate your lab and this is what they're doing at Berkeley in the US, get an AI to run your lab for you.

11:46Tom Clarke:It can do all that chemistry and spit out very small samples. You can then go and test and it saves you a bunch of time with a bunch of chemists having to do that. Sorry, you just glossed over a phrase there, previously unknown materials. We are literally inputting a set of circumstances into our AI model saying, look, I need a solution for this. It's having a wee scratch of its chin and it's coming up with your material that we have never seen before. I think that's the idea. That's sort of where we're going. And it's a little bit like the thing with the biology. Essentially, if you want to find out new biological molecules, components, you're kind of restricted to what evolution has come up with.

12:23Tom Clarke:But what the AI allows you is sort of get out from underneath what evolution has given us and start playing around with other things. Now, obviously, organic materials do not evolve in the same way. But, you know, the things that we've come up with using traditional chemistry, we've only been doing it for a few hundred years. AI, it's not constrained by human thinking. It can just do it in a probabilistic mathematical way and churn out new combinations. They still need to be tested. We still need to find out what they are. But the potential for sort of taking the slog out of doing a lot of that science is there.

12:57Tom Clarke:And if, here's a scenario, fusion, nuclear fusion, okay, it's for literally 60 years, they've been saying it's 10 years away and it's going to save us all. This is the way of generating electricity, but without any cost. is basically taking water and turning it into energy. Pretty much. It's the idea of taking what goes on inside a star and somehow containing it on Earth to generate nearly limitless amounts of energy with very little raw materials going in. But it's a real challenge. One of the biggest challenges is how do you contain that star? It's very, very hot, very wobbly. You need to put it in a kind of magnetic bottle.

13:29Tom Clarke:Really difficult science. I've been banging away at it for years. And a lot of those issues of materials, what do you line that container with? and their mathematical modelling of how you contain it. And what people in the fusion industry now are excited about is using AI to build a sort of digital twin of a reactor, you know, do it all in a computer, solve a lot of those hard problems without having to build very expensive prototypes. So when you do build your reactor base on that, you know you're closer to getting something that will work, add in there new materials, and, you know, the timelines of that kind of technological development timeline shrinks a lot.

14:09Again, I don't want to be accused of over-egging the pudding here, but you are telling me that at some point in the future, it is perfectly possible to imagine that AI will come up with a method of containing a star, the reaction of a star, nuclear fusion, and then we would have pretty much limitless free energy.

14:28Tom Clarke:I mean, that's making it sound extremely simple. But we know it works. We've already built, you know, demonstration fusion reactors that can produce energy. They just don't produce more than you have to put in and get them to work. AI, it's a tool like any other. And our ability to, you know, do three-dimensional drawings using computers, massively advanced engineering and design, things like that. If you can apply the same tools, use this as a software to help solve those problems, it will just help advance the process. It's not silly to suggest that. I'm not promising you that within 10 years we'll have a fusion reactor, but it now looks a bit more likely than it might have done 10 years before.

15:07Tom Clarke:And just going back to your climate change point, if you had nuclear fusion, for example, a lot of our worries about where our power is coming from and how much of an impact it has on our efforts will immediately vanish. We might have more power than we ever need. You know, for most people, it's going to be the societal impacts that will be the most tangible to them in the short to medium term. I mean, in everyday life, give me some examples of where this stuff isn't coming after me. I think right now, and again, the headlines about AI safety, you know, the potential of, you know, swarms of agents roaming the web, hacking into things, you know, collapsing civilization have made people think, maybe feel a little bit skeptical about the utility of AI in everyday lives.

15:50Tom Clarke:But, you know, we're already seeing it. I mean, the way my kids are using AI to do research, to understand a new topic, to just, you know, hoover up basic information, was adopted very quickly. People might say that they are sceptical about AI, but I see a lot of people being quite happy to use it in their everyday lives. I mean, I do. I use it for research purposes every single day. Absolutely. But that's you just prompting it, asking it questions. AI agents, things that can roam around and do things. A few more concerns they raise, but think about the potential, the potential utility of something like that.

16:23Tom Clarke:Something that can do work on your behalf that's specific to you. That's a neat idea. Doing what sort of stuff? Well, so, you know, we've had for a while a fairly basic kind of AI, something like Siri or a chatbot on your phone. You can ask it questions. It can do various things. Send an email to so-and-so. Call my mum. You know, that's an AI working and they're not particularly new. But combine the power of, you know, these frontier AI models to do that kind of task for you. And you've got a lot more power and a lot more utility. Meta recently launched Muse, which is an AI agent run by one of their AI models.

17:00Tom Clarke:And it can do some pretty smart stuff, provided you give it access to some of your personal information. But if you let it have your bank details, it can pay bills for you. It can make phone calls to your insurer to sort out insurance. There's a boring process where you're on hold for ages and you have to press nine and whatever. Get something else to do that for you. That's fantastic. But I also think for, you know, societies and governments that run them, I think there's huge potential for AI as well. I think our government recognised that. I think, was it last, I think 2024, maybe it was sort of Keir Starmer's prime ministership.

17:33Tom Clarke:They made a claim that fully integrating AI across sort of government systems could save 14 and a half billion pounds. Yeah, we need that right now. Just from removing some of the bureaucracy. And you look at some of the areas where an AI could be really helpful, the backlog in the court system, for example. A lot of that is just bureaucratic. It's forms, it's ordering the queue, it's prioritization. It's stuff that you could, provided it was well-designed AI software with someone, a human actually checking it's doing the right thing. But it could save an awful lot of time doing that kind of thing.

18:09Tom Clarke:Look at one of the most active areas of dispute of government. We need to save money. The benefits bill is huge, out of control, and a major drain on financial resources. There's estimates, you know, nine and a half billion pounds worth of fraud in benefits claims at the moment. Because a lot of those are sort of effectively administrative tasks, it's checking names against double applications or an NHS record that claims for disability that might not exist. If you had an AI tool that could sift through a lot of that data and at least flag ones that need investigation, it's something that's being deployed.

18:43Tom Clarke:And DWP, for example, are claiming that they're already saving money through using AI-type tools in that sector. And that's not something to be sniffed at. No, it's not to be sniffed at. But okay, I'm going glass half empty again. Sorry, Tom. We both know government IT projects always overrun, always overspend, always underperform. Why would AI be any different? And I think that's a very rational, sensible approach to this. Because although AI is amazing and do brilliant things and it's very, very clever, it's still ultimately a software project. You still need to find a way of integrating it with a clunky system that already exists, and you need to make sure it's safe.

19:21Do you know what, Tom? You have given me reasons to be cheerful in the way that Roland Manthorpe, when he was on the podcast, gave me reasons to have sleepless nights for a month. But I do find myself just focusing on a couple of things. One, if we are pointing AI in the direction of a problem, it seems to go great guns. If we make it focus on something that we really need to get moving on, it gets the job done. But we're also in a bit of a kind of an AI arms race right now, aren't we? All of these companies appear to be pursuing this thing. You know, super intelligence, you know, general intelligence, basically, you know, machines that are smarter than us across the board.

20:00And I don't know that I want to live in a world like that.

20:03Tom Clarke:Yeah. And I think this gets the heart of quite a few of the issues we're talking about at the moment, right? A lot of the fear and concern around what's going on in AI is we've got these frontier labs, they're called the ones working on the biggest AI models with the most access to computing power, are a handful of companies in the US promising us superintelligence, artificial general intelligence, some sort of AGI, something that's smarter than any human that's ever lived. And this idea that there is a race to get to that ultimate supreme AI that's going to win its all. You heard Donald Trump, who wins AI wins.

20:42Tom Clarke:Is that what we really want? Step back and think for a second, what do we want AI to do? All right, we want it to solve the following problems. But what we're actually sort of seeing at the moment is some sort of mad race to develop something that a lot of people, even in the AI research field, say AI, certainly the way they're currently doing, isn't capable of doing, it might be much better at us at writing computer code and solving math problems as it stands. It's going to take a lot more advances and probably very different kinds of AI to do things in the real world like we do, the world in which we experience that.

21:16Tom Clarke:Let's remember, AI has no experience of because it lives inside a computer and always will. So there's a lot of work to be done on replacing all of the kind of intelligence that humans have. And we're not there yet. If, however, you break it down, what tools do we need? What do we want AI to actually do? And if you look at most of the challenges that we've got and the things that we'd like AI to do, they're better dealt with as a kind of unique problem, design the AI to solve that problem. And it takes away a lot of those other worries. Well, as I breathe a sigh of relief, Tom, many thanks. Thank you.