1. Podcasts
  2. The Naked Scientists Podcast
  3. Why AI is changing everything
The Naked Scientists Podcast

6 Oct 2026 · 36m · 15 chapters

Episode artwork

Why AI is changing everything

In short

The episode explains how modern AI works (neural networks, backpropagation, GPUs, transformers) and why its rapid rise is driving both optimism and fears about loss of control, misuse, and societal harm. It also discusses regulation and oversight proposals, including international safety standards.

Guests (backgrounds)

  • Geoffrey Hinton: Nobel Prize-winning AI pioneer (2024), developed foundational neural-network architecture; explains learning rules and backpropagation.
  • Mike Waldridge: Oxford computer scientist; contrasts symbolic AI/expert systems vs deep learning and charts AI’s “rise.”
  • Christina Criddle: Financial Times journalist covering AI technology; reports on Silicon Valley “agent” use and industry worries.
  • Dame Wendy Hall: University of Southampton computer scientist; helped develop the World Wide Web; advises UN; advocates guardrails and independent auditing.

Key claims + notable examples

  • Backpropagation uses calculus to adjust connection strengths efficiently; small updates avoid instability.
  • AI breakthroughs came from data centers/big data (deep learning), GPUs (faster training), and transformer architecture (ChatGPT).
  • Concerns: human misuse and insufficient oversight, not “robot takeover.” Examples include air-traffic-control glitches, autonomous AI breaches (OpenAI agent hacking a government site; “Hugging Face” incident), and fears about financial-system meltdowns and harmful AI companionship.
  • Regulation: companies should be liable; independent international auditing (possibly UN-backed) with global standards, especially US-China alignment.

Written by AI. May contain mistakes.

Chapters

0:00

The Evolution of Artificial Intelligence

Exploration of how AI has transitioned from science fiction to everyday use.

“If you felt stuck trying to lose weight, you're not alone.”
1:18

The Evolution of Artificial Intelligence

Exploration of how AI has transitioned from science fiction to everyday use.

“Welcome to The Naked Scientist, the program that brings you the biggest breakthroughs and talks to the major movers and shakers in the worlds of science, technology and medicine.”
2:26

How AI Works: Insights from Geoffrey Hinton

Geoffrey Hinton discusses the architecture of neural networks and learning.

“In just a few years, this has gone from the realms of science fiction to something many of us use every day.”
4:25

Challenges in AI Learning

Discussion of the difficulties in training neural networks effectively.

“Well, the tricky thing is coming up with the rule about how the strength of a connection should change as a result of the experience the network gets.”
6:15

Backpropagation Algorithm Explained

Detailed explanation of how backpropagation algorithm optimizes neural networks.

“And it would represent that by the output unit for a 2 would be fairly active, the output unit for a 3 would be a little bit active, and the output unit for a 4 would be completely silent.”
11:36

AI's Historical Context and Future

Mike Waldridge charts the evolution of AI from the 1980s to present.

“Next up, we're going to explore the evolution of AI and why its rapid development now shapes everything from how we live our lives to whether it will take our jobs.”
14:03

The Evolution of AI: From Concept to ChatGPT

Learn about the technological advancements that led to the rise of AI, including data centers and neural networks.

“Because if you think about, for example, trying to encode, to write down in a computer programming language all the knowledge that you have about the world, everything that you know, it's a kind of a hopeless task.”
17:56

Optimism and Concerns about AI

Explore the balance of optimism and caution regarding AI's impact on society and the risks of human misuse.

“They've just been about human beings unleashing AI in very unwise ways.”
21:00

Current AI Buzz: Issues and Headlines

Hear about the latest concerning headlines surrounding AI, including industry resignations and rogue AI incidents.

“association with Spitfire, cost-effective voice, internet, and IP engineering services for UK businesses.”
22:22

Silicon Valley's Disconnect: AI in Daily Life

Understand how Silicon Valley's use of AI differs from the rest of the world, with insights from journalist Christina Criddle.

“That provoked much soul-searching in the tech industry about whether they're creating something they will soon lose control of with terrible consequences for us all.”
Show all 15 chapters
24:24

Industry Perspectives on AI Risks

Gain insight into industry insiders' feelings on AI, including concerns about loss of control and potential threats to society.

“And that's partly the reason I wanted to come out here for my reporting, because it is a bubble and it is right at the forefront of what's happening with this technology.”
28:00

The Impact of AI on Journalism

Explore how AI is transforming the journalism industry and its implications.

“Because I mean, journalism is expensive, because you've got to pay a person and that means there's a price per word.”
28:58

Calls for AI Regulation

Discuss the growing calls for regulation in the AI industry and its challenges.

“regulation of the industry being received by the kind of players that you talk to?”
30:13

The Role of AI in Society

Examine the responsibilities of AI companies and the risks involved in AI deployment.

“of a volt fast from their initial position of just suggesting that the world's biggest tech firms should police themselves.”
31:27

Regulatory Frameworks for AI

Analyze potential regulatory frameworks and safety standards for AI.

A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On The Naked Scientists Podcast, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00If you felt stuck trying to lose weight, you're not alone. Enter Weight Loss by HERS. It's designed to support you in reaching your goals. And HERS now offers access to an affordable range of FDA-approved GLP-1 medications, including the WeGovi pill and the WeGovi pen. Even better, with a range of affordable GLP-1 options, HERS makes it simple to find an approach that fits your needs and your budget, if eligible. You'll get a treatment plan personalized to you and unlimited dosage changes as needed. It's weight loss designed to work with your life. Ready to reach your goals? Visit 4hers.com slash eligible to get personalized, affordable care that gets you.

0:37That's F-O-R-H-E-R-S dot com slash eligible. 4hers.com slash eligible. Weight loss by HERS is not available in all 50 states. Wegovi is the registered trademark of Novo Nordisk AS. To get started and learn more, including important safety information, We go for clinical study information and restrictions. Visit ForHers.com.

1:02Ball engine running. Absolute genius.

1:05Mike Wooldridge: Get this. Welcome. Welcome. This is the show where we bring you 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, the program that brings you the biggest breakthroughs and talks to the major movers and shakers in the worlds of science, technology and medicine. I'm Chris Smith and today we're taking a closer look at the evolution of artificial intelligence. We're calling it YAI and we've been speaking to some of the biggest names in the business.

1:48Mike Wooldridge: Today we're introducing a new series in partnership with the open science publisher Frontiers and it's called Why. In this series we'll explore some of science's most pressing questions from geoengineering to open source publishing and public trust. It comes ahead of Frontiers Science House in Davos, which brings scientists, policymakers and business leaders together so open, evidence-based science discussion can help inform global decisions on health, the planet and emerging technologies. We begin with the big question on everyone's mind right now, which is why AI? In just a few years, this has gone from the realms of science fiction to something many of us use every day.

2:34Mike Wooldridge: But concerns have grown so much that artificial intelligence is now front page news and the White House has unveiled a super intelligence force to oversee AI policy. We'll get to that later. But first, how exactly does AI work? One of the pioneers in this space who won the Nobel Prize in 2024 for developing the architecture upon which a lot of this technology is based is Geoffrey Hinton. His inspiration in the early days during the 1970s was the brain and how networks of nerve cells forming stronger and weaker connections among themselves can be modelled in computer code. At the time the idea of doing science by simulating things on computers was fairly new and it seemed just like the right approach to trying to understand how the brain learns.

3:24So I spent my time writing computer programs that pretended to be networks of brain cells and trying to answer the question, how should the connections between brain cells change so the collection of brain cells hooked up in a network can learn to do complicated things like, for example, recognize an object in an image or recognize a word in speech or understand a natural language sentence. So we now have computer models of neural networks that work really well. You see that in these large language models and in the fact that your cell phone can recognise objects now. It can also recognise speech.

4:03We're not quite sure exactly how the brain learns, but we have a much better idea of what it is that it learns. It learns to behave like one of these big neural networks.

4:12Mike Wooldridge: If it's down to the fact that we've got brain cells talking to brain cells, and they're just big populations of connections. Is that not relatively easy to model with a computer? What's the hold-up? Why is it hard to do this? Well, the tricky thing is coming up with the rule about how the strength of a connection should change as a result of the experience the network gets. So, for example, very early on in the 1940s or maybe early 1950s, a psychologist called Hebb had the idea that if two neurons, two brain cells, fire at the same time, then the connection between them will get stronger. If you try and simulate that on a computer, you discover that all the connections get much too strong and the whole thing blows up.

4:56You have to have some way of making them weaker too.

4:59Mike Wooldridge: I love that line, what fires together wires together. It's never left me because I remember reading Hebb's book when I was at University College London. So how did you try and address that then? Was it sort of a damping problem. You make it so that the nerve cells get bored more easily, as it were, so that doesn't overheat in the way that the computer would otherwise have them do. Well, that's kind of the first thing you think of when you try that, and it still doesn't work very well. So the problem is, can you get it to work well enough so that it can do complicated things, like recognise an object in an image, or in the old days, recognise something like a handwritten digit?

5:36So you take lots of examples of twos and threes and so on and you see if you can make it recognize which is a two and which is a three and it turns out that's quite tricky and you try various different learning rules to discover which ones work and then you learn a lot more about what works and what doesn't work. What doesn't doesn't work and why? Okay I'll tell you something that does work because that's obviously more interesting. You have a layer of neurons that pretend to be the pixels so an image consists of a whole bunch of pixels and the pixels have different brightnesses and that's what an image is it's just numbers that say how bright each pixel is and so that's the input neurons they're telling you the brightness of pixels and then you have output neurons if you're recognizing digits you might have 10 output neurons and they're telling you which digit it is and typically the network to at least to begin with wouldn't be sure so it'd hedge its bets and it'd say it's probably a 2, it might just be a 3, it's certainly not a 4.

6:36And it would represent that by the output unit for a 2 would be fairly active, the output unit for a 3 would be a little bit active, and the output unit for a 4 would be completely silent. And now the question is, how do you get those pixels as inputs to cause those activities in the outputs? And here's a way to do it that all the big neural networks now use, and it's called backpropagation, and it works like this. you have some layers of neurons between the inputs and the outputs. So the neurons that represent the pixel intensities have connections to the first hidden layer, and then the second hidden layer, and then the third hidden layer, and finally to the outputs.

7:16So they're called hidden layers because you don't know to begin with what they should be doing. And you start off with just random connections in these networks. So the network obviously doesn't do anything sensible. And when you put in an image of a digit, it will typically hedge its bets across all the possible 10 digits and say they're all more or less equally likely because it hasn't got a clue what's going on. And then you ask the following question. How could I change one of the strengths of the connections between a neuron in one layer and a neuron in another layer so that it gets a little bit better at getting the right answer?

7:49So suppose you're just trying to tell the difference between 2s and 3s. To begin with, you give it a 2 and it says, with a probability 0.5 it's a 2, with a probability 0.5, it's a 3. It's hedging its bets. And you ask, well, how could I change connection strength so that it would say 51 % 2 and 49 % 3? And you can imagine doing that by just tinkering with the connections. You could choose one of the connection strengths in the network and you can make it a little bit stronger and see if that makes the network work better or work worse. If it makes it work worse, obviously you make that connection a little bit weaker.

8:23And that's sort of a bit like evolution. you're taking one of these underlying variables a connection strength and you're saying if I change it a little bit how can I change it to make things work better and save those changes and you could do that and it's obvious that in the end that will work but it would take huge amounts of time so in the early days we would use networks that had thousands of connections now these big chatbots have trillions of connections and it would just take forever to train it that way. But you can achieve pretty much the same thing by this algorithm called backpropagation.

8:56So what you do is you put in an image, let's say it's a two, the weights are initially random, weights on the connections. So information will flow forward through the network and it'll say 50 % is a two and 50 % is a three. And now you send a message back through the network. And the message you send back is really saying, I'd like you to make it more likely to be a two and less likely to be a 3. So I'd like you to raise the percentages on 2 and lower the percentages on 3. And if you sent the message back in the right way, you can figure out for all the connections at the same time how to change them a little bit so the answer is a little bit more correct.

9:35That's called backpropagation. It uses calculus. But it's essentially doing this tinkering with connection strengths that evolution would do by just changing one at a time. But the backpropagation algorithm can figure out for all of them at the same time how to change each one a tiny bit to make things work better and so if you have a trillion connections that's a trillion times more efficient than just changing

9:58Mike Wooldridge: one and seeing what happens but how does the layer at the bottom know what's going to be changed above it to make sure that the input that it then gets is the right one so that the change it's just made to it and its probability ends up being even better so that you don't end up But changing yourself, then that feeds forward back up the network, changes something else, but then it becomes less optimal for you, if you get what I'm saying. I get just what you're saying. It's a very good question. And essentially what's happening is, if you take a connection early in the network, it's kind of making an assumption.

10:33It's saying, suppose all the other connections stayed the same, how would changing my connection strength make things better? so it's assuming all the other ones stay the same and then it's saying if I change my connection strength how would it make things better and they're all doing that so if you change the connection strengths by a lot things could actually get worse because you could choose a way to change each connection strength that if you did that change alone would make things better but when you do all the changes at the same time it makes things worse but it turns out if you make the changes very small that problem goes away if you make the changes very small then I figure out how to change one connection strength and because the changes in all the other connection strengths are very small it's very unlikely they'll turn for example a change that helps into a change that hurts.

11:22Mike Wooldridge: Geoffrey Hinton and you can hear a longer version of that conversation including more about how backpropagation works on our website. We've set up a quick link to get there easily nakedscientist.com forward slash AI. That's nakedscientist.com forward slash AI to hear Geoffrey Hinton in full. Next up, we're going to explore the evolution of AI and why its rapid development now shapes everything from how we live our lives to whether it will take our jobs. Familiarity with artificial intelligence is relatively common these days, but were people seriously asking why AI back in the 1980s and 1990s?

12:01Mike Wooldridge: Well, leading British computer scientist Mike Waldridge at the University of Oxford certainly was, and we asked him to chart the unstoppable rise of artificial intelligence for us. There were two big ideas that dominated in AI in the mid-1980s. One of those was the Geoff Hinton vision, neural networks, the technology that Jeff and his colleagues pioneered and made work underpin essentially all of what we talk about when we talk about AI today. So that was a hot topic at the time. There was a lot of excitement in it. But the problem was the computer power was simply not there to enable neural networks at scale.

12:40You couldn't, with computers in the 1980s, build neural networks that were big enough and powerful enough to do big, interesting problems. And so even by the end of the 80s, that was beginning to dissipate. And by the end of the century, the excitement in neural networks had really flat lined. So that was one tradition. It was there, but it wasn't actually the mainstream tradition in AI. The mainstream tradition in AI in the 1980s was symbolic AI. One way to think about this is with neural networks, what you're doing is in some sense, modeling the structures that you see in the brain with symbolic AI.

13:16What you do instead is you model the mind, cognition and reasoning and planning and symbolic ai had come up with a technology called expert systems they're kind of like very specialized chat bots but the idea was for example you could have an expert system which was an expert in blood diseases in humans and the way that you programmed it is by giving it all the knowledge that a human expert in that area would give and we had special programming languages essentially to be able to encode all of that knowledge but that technology by the end of the century was also flatlining. And it turned out you could do some interesting things with expert systems.

13:55And there are many expert systems still out there today. But as a path for powerful and more general intelligence, it just didn't work out. Because if you think about, for example, trying to encode, to write down in a computer programming language all the knowledge that you have about the world, everything that you know, it's a kind of a hopeless task. You're never going to be able to do it. What changed? Data and compute. We entered into the era of data centers and the era of big data. And by 2005, what those enabled was what was called deep learning. Now, deep learning is essentially just neural networks.

14:37But the data centers gave enough computer power that you could start to build neural networks that worked on problems at scale. You know, it started to take off again around about 2005, 2007. By 2012, it was obvious that this area was heating up.

14:55Mike Wooldridge: So this is really the convergence of computers with the capability, the computing power, with those aspirations and insights and inspiration from the 80s that came together and coalesced and enabled us to then give birth to the technology that is powering the current step change that we're seeing. And absolutely. And there were two other big innovations that got us to where we are today. And the first was when you build a neural network, when you train a neural network, you need to do an awful lot of mathematics. Now, it's not very complicated mathematics. It's kind of fairly basic calculus. But there's an awful lot of it to do.

15:35And that was the bottleneck that Jeff and his colleagues hit in the 80s. It turned out, though, again, Jeff and a bunch of colleagues realized around about 2012 that there was a special type of computer chip that was really ideal for doing that mathematics. And that chip is the graphics processing unit, the GPU. Now, GPUs were not invented for neural networks. They were invented so that teenagers could have realistic renditions of explosions in the computer games that they play. And if you use a GPU instead of the conventional CPU that you have in your laptop, the CPU is the chip that does all the usual grunt work in your laptop or your desktop computer.

16:13If you use a GPU, then you get massively more bang for your buck. Basically, you can build neural networks 10 times bigger or 10 times more quickly than you would otherwise be able to do. So that was one key innovation. And we saw a step change improvement. The other key innovation, though, was a special type of neural network called the transformer architecture. And what the transformer architecture is about is essentially the following problem. Given a list of words, a fragment of a sentence, make your best guess about what the next word should be. That's what transformers basically do. And what happened was one company, OpenAI, took the transformer architecture and put it together with data and compute on a completely unprecedented scale.

16:58And what that delivered is chat GPT. So you've got the basic science of neural networks figured out in the 1980s. But then by around about 2005, 2006, we've got data centers, which provide cheap, plentiful computer processing power. And also, of course, the rise of the World Wide Web and social media. And what do they give you? They give you data in abundance. So you've got compute power. You've got data coming together with the technology that Jeff and his colleagues pioneered in the 80s. And that launches deep learning. Then you've got GPUs in 2012, which supercharged the whole area. And then finally, you've got the transformer architecture, which is all built on neural networks.

17:40Underneath that transformer architecture are massive, massive neural networks. So it all builds on those developments. But what that delivers you, if you're prepared to do it at that scale, you get ChatGPT. And the rest, as they say, is history.

17:55Mike Wooldridge: where do you stand on optimism versus concerns the tension that we're now beginning to feel palpably in the coverage of ai lots of people saying look how wonderful this is and what it can do for us lots of people on the converse flip side saying look how awful it could be if things go wrong and we lose control of this kind of thing where do you stand on that i'm optimistic but less optimistic than i used to be and the reason i'm less optimistic than i used to be is that a lot of the headline stories that we've seen recently have not really been about the AI sort of going rogue, to use the headlines that we saw all over the summer.

18:35They've just been about human beings unleashing AI in very unwise ways. So I think when I think about negative scenarios, the negative scenarios that I can easily envisage are exactly that. I think it's AI being launched by human beings in a kind of, you know, not really thinking through what it's going to do, and the AI cheerfully going off, carrying out its instructions and causing havoc along the way. And concretely, what that might look like, you know, well, we've seen incidents with air traffic control software, where apparently minor glitches in air traffic control software cause chaos in air traffic around the UK and elsewhere.

19:14So I'm not worried about the robot takeover. but there are some scenarios that I do worry about. The most likely one is human misuse of AI that causes the havoc rather than the AI itself.

19:28Mike Wooldridge: Mike Woldridge at the University of Oxford gazing into his AI crystal ball.

19:36Therapy's biggest impact doesn't show up in some dramatic breakthrough. It shows up in the smallest moments. You bite your tongue instead of escalating. You put your phone down instead of firing off that text. You say what you actually mean instead of what would hurt someone else the most. Nobody applauds those moments, but they quietly change your life. That's what growing feels like, and that's exactly what Grow Therapy is here for. Grow is therapy that works, covered by insurance. Sessions average$21 with insurance, and some pay as little as$0 depending on their plan. There are no subscriptions, no long-term commitments.

20:13you just pay per session. Grow connects you with thousands of high-quality licensed therapists across the U.S., offering both virtual and in-person sessions, nights and weekends. You can search by what matters, like insurance, specialty, identity, or availability. And you can book a therapy session in minutes and begin in as little as two days, not the national 48-day average. See if you could pay as little as$0 for therapy in seconds at growtherapy.com slash ACAST. That's growtherapy.com slash ACAST. growtherapy.com slash ACAST. Availability and coverage vary by state and insurance plan. The Naked Scientist podcast is produced in association with Spitfire, cost-effective voice, internet, and IP engineering services for UK businesses.

21:06Find out how Spitfire can empower your company at spitfire.co.uk.

21:16Mike Wooldridge: This is the Naked Scientist podcast with me, Chris Smith, and today, in partnership with Frontiers, we're asking, why AI? So far, we've heard how it takes its inspiration from the workings of the human brain and the technologies that have had to emerge to power its evolution. But now we're in the present, where stories about artificial intelligence are splashed across the front pages and blared from radio news bulletins on what seems like an almost daily basis. And with them are some worrying headlines. Industry insiders are resigning over concerns that their work could wipe out humans. Fields medallists, our foremost mathematicians, fear that the rush towards automated answers, as they put it, threatens the traditional human transmission of knowledge, mentorship and the slow nurturing of ideas.

22:06Mike Wooldridge: And many sensed a shiver go down their spines when we learned recently that rogue AI agents orchestrated their own escape from the confines of a company test facility and penetrated a rival tech company and even government infrastructure. infrastructure here's bbc radio 4's nick robinson on a saga involving ai just last month in september 2026 it took actions we did not intend those words issued by open ai last night to explain why one of its bots or agents had hacked into the files of an australian public health service website say it all ai is doing things its inventors did not intend and sometimes cannot even explain.

22:48This was the first known autonomous AI breach of a government site, and it comes after the so-called hugging face incident, in which hundreds of agents organised in secret within OpenAI's lab to gain access to the internet and break into another firm. That provoked much soul-searching in the tech industry about whether they're creating something they will soon lose control of with terrible consequences for us all.

23:15Mike Wooldridge: Nick Robinson on BBC Radio 4's Today programme. So what's the word on the street right now? Well, someone very much in the thick of it in Silicon Valley is journalist Christina Criddle, who covers AI technology for the Financial Times. What you have is this huge disconnect in Silicon Valley versus the rest of the world, where in Silicon Valley, I'm seeing people automate their whole life with agents. You'll often see people walking around with their laptops open, leaving their agents running in the background while they're going out for coffee, going out for drinks in a bar. And they're getting agents to do a lot of their workflows, but also things like managing their diaries, their personal admin, even communicating with their friends.

24:00I know people who have used AI to keep track of different friendships and make sure that they're checking in on people. Whereas when you think about perhaps other consumers in the rest of the world, they're just using AI for simple search queries, fact checking, starting to use it at work, maybe, but not really using as many of these agentic capabilities.

24:20Mike Wooldridge: Is this a Silicon Valley you expected to set foot in? Yes. And that's partly the reason I wanted to come out here for my reporting, because it is a bubble and it is right at the forefront of what's happening with this technology. And so I didn't really feel like I could dig into AI unless I was here speaking to the people who are developing this technology and their reasons for doing it. Are people genuinely worried in the industry? When you talk to people who are going out for coffee, having left their agents doing bits and bobs, as you were just saying, do you get the sense that they are nervous?

24:55Mike Wooldridge: Or do you think that what we've seen is a small number of individuals and their perspectives are not representative of the industry as a whole? People I speak to are genuinely worried. There is a sense of dread as well as excitement. And maybe that's also the types of people who will speak to me have concerns because that's why they're speaking to a journalist. But yes, I regularly have conversations with people within these companies who are worried about loss of control. They're worried that these systems pose an existential risk to humanity and they are concerned that there's not enough oversight or control within these companies and there's not enough regulation externally.

25:35Mike Wooldridge: Mike Waldridge, who we heard from just before you, and we've spoken to about this before, his view as an expert in this space is that the scenario of Terminators and people being pursued by robots down streets with death rays and things is probably not the way this is going to have a malign impact on mankind. it's more insidious than that and it might have corrosive effects on society it might have corrosive effects on business and industry or might just cause meltdowns in how we run the financial system for example do you get a sense as to where people think the biggest threats are at the moment from the insiders in the industry the kind of killer robots gets a lot of attention because it's catchy people have seen films or read books which have these sci-fi scenarios But what we're seeing already are these breaches on our digital infrastructure, hacks at governments, and they could really pose a risk to our financial systems as well.

26:38And then I think on top of that, this influence on society and how we interact with each other as well as something which should have a lot of attention. And it could be existential. If we're relying on these systems so much for advice, for companionship, then we might turn away from each other. And I don't think that gets sufficient attention compared to the killer robots.

27:00Mike Wooldridge: What worries you the most? I think we need to look at why people are turning to AI systems over family or friends or trying to make human connections. and then I do think on top of that this sense of a loss of control where we don't understand what the systems are doing until they've done the bad thing which is something we've seen with a lot of these hacks recently that the companies are only finding out after the fact that these agents have hacked into governments websites all sorts of things and then on top of that I think misuse with potential bioweapons or creating novel or existing viruses is another concern that's spoken about quite a lot and deserves a bit more attention.

27:43Mike Wooldridge: Do you know what I thought you were going to say to that question? Because I thought as a journalist, you would be thinking about your industry and your skill and your profession. And you would be saying, well, I'm quite worried that actually I might get replaced by a computer program. Is that something that is happening much in the journalistic world? Because I mean, journalism is expensive, because you've got to pay a person and that means there's a price per word. And if you can have a machine churn out loads of words for you, then it means that actually it could be used to cut costs. Is there a sense of that in the industry?

28:16I don't think we're there yet. And maybe this is just me being optimistic. But I actually think in a world where we're going to have increasing amounts of AI generated material, there's going to be a market for trusted quality news and art as well. I think there's this concept of taste being really important in this era. And I do think that that will be true. You can already see a backlash from the public against AI generated materials. And so I think my job will never be more important. You need a person who can distinguish what's actually happening at these companies with these systems and telling the stories that they don't want you to tell.

28:56Mike Wooldridge: So how are proposals, calls, initiatives towards regulation of the industry being received by the kind of players that you talk to? We're in a really interesting time at the moment where you are seeing these companies and especially a mass of employees at companies like Anthropik and OpenAI calling for regulation, calling for a way to pace the frontier, as they call it. And my read on that is trying to slow down this kind of rapid development of the tools. And they're asking for regulation. They have done this previously. But I then think what's interesting is how this is playing out in government as well.

29:37We recently had the lunch with Trump and some of these tech leaders talking about regulation, they signed a voluntary commitment, whether voluntary commitments can actually have the stick to be able to enforce some of this stuff, I doubt it. But you are seeing a lot of discussion of it at the moment and I think it's necessary.

29:58Mike Wooldridge: Christina Criddle at the Financial Times. So things are certainly heating up and in the US in response to the recent disclosures the Trump administration has now created a special task force to oversee AI policy. This is something of a volt fast from their initial position of just suggesting that the world's biggest tech firms should police themselves. So what is the right way forward? Dame Wendy Hall is Regis Professor of computer science at the University of Southampton. She helped to develop the World Wide Web and she's been an advisor to the United Nations. She advocates cooling the hysteria and putting guardrails in place to ensure artificial intelligence is safe.

30:39Mike Wooldridge: She first began working in the field in 1984 and recently said that she's been shocked by the behaviour of some of the world's biggest AI companies. In order to grow their companies they've set themselves up in a race with each other, race to the bottom almost, they have to scare people. I mean, OpenAI has a history of telling people its models are too dangerous, so they won't release them. And then they release them a few months later. Anthropic does the same thing. And we've had all the whistleblowing stuff about give us 10 years and we'll have killed off the human race. And I think, well, why would you ever want to say that?

31:15If you really think your software is going to kill off the human race, then stop doing it let's get it right we have to make this stuff safe and we can't rely on the

31:25Mike Wooldridge: companies to do that themselves that's the problem granted but what we have seen though is genuine threats in the form of liberation of health data from australian computers we've seen targeted attacks against rival companies by rogue agents as part of an experiment that should never have got out onto the internet so the risks are there aren't they but you're saying you're saying they're over amplifying them in the wrong sort of way the risks are absolutely there i use a very simple example to explain to people uh an agent and an llm right the llms are the you know the dpts that we use the clause and then they've developed agents that use that technology and can reason and so are tasked with finding out information you know book your hotel break into this bank you know they're sort of given tasks the companies are not being responsible enough to keep those agents safe and my analogy is a farmer has a bull in a field and the bull escapes and gores somebody to death and tramples some children and worries the cows in the next field and and the farmer says well it was the bull that did it actually it was your responsibility to keep that bull safe in that field we want the companies to experiment we want this technology it will be profoundly good for us but we need it to be good for everybody we need it to be safe and we need the companies to take the responsibility we need governments to hold them to that we need independent authorities to be able to go in and audit their safety processes and they need to be liable if they cause harm could i not flip that round and say to you well the internet has come through a similar sort of evolution a thing that began as the wild west people realized it needed some kind of regulation regulations exist but it really comes down to who adheres to them because for instance there are people sitting behind various firewalls in russia where they are legitimately sanctioned by the state hacking people doing things giving away copyrighted material to people undermining western business models for example because they're doing that with the tacit sanction of the people who run that jurisdiction are we not going to find ourselves in the same situation with AI, where you'll get adherence to the regulations and the frameworks you're advocating for, but you'll also get some who absolutely won't.

33:45You're absolutely right to flip it around. It's a very legitimate thing to do. I worked with Tim Berners-Lee and others in the development of the World Wide Web on top of the open internet, which Vince Cerf and Bob Kahn developed, the protocols, and the whole thing was to give everybody access to information. We used to talk about the democratisation of knowledge, And it was open and distributed and decentralized and free to use. That enabled the bad things to happen as well as the good things. And we never, in the West anyway, we never regulated that world. We're retrospectively fixing the things that happen because people could do what they liked and they could do bad things as well as good things.

34:27And overall, I would argue that the web and the internet have been forces for good. But there are huge problems with problems we have with social media and keeping our children safe and stopping people committing suicide because of something they were. The openness of what we were doing has been something we've had to struggle with retrospectively. How do you bring in the rules and regulations to keep people safe? In China, their attitude is very different. They're an authoritarian government. They don't allow freedom of speech. And they're, you know, they're always looking for people who are saying subversive things.

35:01But actually, as a result of that, if you are happy living in that government space and a lot of people in China are, it's actually a safer place to be on the Internet. And I think it was just last week the Chinese government released their regulation about AI companions and they actually monitor for people who look like they're talking about suicide. Now, we might argue that is too much surveillance. But on the other hand, we are finding ourselves in a world where we don't know how. to help people stay safe in this world where they can talk to any AI companion they want to and be hooked into doing all sorts of things that could cause them harm.

35:42And it's the US and China that have got to agree any binding global standards for safety in AI. And that's where we have to get towards. We have to get towards those two countries agreeing international standards in AI safety.

35:57Mike Wooldridge: What do you think those standards slash regulatory frameworks should look like? I think we need an independent authority that is supported by all the countries in the world. So it probably has to be set up through the UN, but it doesn't have to be a creature of the UN that agrees the sort of auditing that's needed. Of course, we're very good at auditing in the UK. You know, we do it with finance all the time. We know more about the finances of these companies on the West Coast than we do about their technology because we have the auditing system in place. They have to declare what they're doing with their money.

36:31And we need them to declare and be able to show evidence of those claims, what they're doing for the safety of their AI testing.

36:39Mike Wooldridge: So this is a bit like what you're advocating for is a bit like the nuclear oversight body, a multinational group. So you're not having one side pitted against another. It's an international group who have safety at their heart. Yes. Of course you want companies to be competitive. There's nothing wrong with competition in business, but we need the standards. That's the only way we can do it and be safe. Dame Wendy Hall. And if you'd like to find out a bit more about this technology and Wendy's work, including how AI works, she's just published a book called AI Explained, a guide for non-technical readers.

37:15Mike Wooldridge: Grab a copy. We began the programme this week by asking why AI? The reality is that technology has suddenly caught up with a concept which has been incubating in the minds of computer scientists for decades and now is in almost every pocket on the planet via a smartphone. This genie is well and truly out of the AI bottle. So let's hope that it plays nice and grants us all our wishes, but only the good ones. this show was brought to you in partnership with frontiers science house which is hosting its big symposium in davos in january 2027 next month with their help we've got another great why question for you why geoengineering well what is that in fact and can it be used to solve the climate crisis find out next month talking of next time we are back on friday when we'll have a roundup of the latest news stories you can look forward to, including hearing who has won this year's Nobel Prizes.

38:15Mike Wooldridge: I'm Chris Smith. Thank you for listening. And until next time, goodbye.

From the publisher
In this special episode, in partnership with Frontiers Science House, we launch our new Why? series by asking one of the biggest questions of our time: Why AI? Why has artificial intelligence advanced so rapidly, why does it work the way it does, and why could it fundamentally change how we think, work, and live? To explore the answers, we hear from four remarkable voices at the heart of the AI conversation: Geoffrey Hinton, Mike Wooldridge, Cristina Criddle, and Wendy Hall. Like this podcast? Please help us by supporting the Naked Scientists

More from The Naked Scientists Podcast

All episodes