Hosted by Christina Ruffini and Sir Richard Dearlove
Guest
Geoffrey Hinton
Geoffrey Hinton is known as 'The Godfather of AI' and warns that AI systems are learning in ways we don't fully understand and may soon be capable of making decisions beyond our control.
Episode Summary
In this episode, “The Godfather of AI” Geoffrey Hinton joins One Decision hosts Kate McCann and Sir Richard Dearlove to discuss the growing global race to build and control superintelligent machines—and why artificial intelligence can no longer be simply switched off. Hinton warns that AI systems are learning in ways we don’t fully understand—and may soon be capable of making decisions beyond our control.
Episode produced by Situation Room Studios. Original music composed and produced by Leo Sidran.
Listen & Subscribe
- Watch on YouTube
- Listen on Apple Podcasts
- Listen on Spotify
Transcript
COLD OPEN
Hinton: I don't think there's any chance that we'll stop the development of AI. The only hope is to figure out how you can develop a super intelligent AI that is benevolent, that doesn't want to get rid of people.
INTRODUCTION
Mccann: Welcome, Geoffrey Hinton, to the One Decision Podcast. It's great to have you on the show today. And Geoffrey, you need no introduction, really. Most of our listeners will know you as the godfather of AI. You're, of course, renowned for your work on artificial neural networks. You won a Nobel Prize in Physics for your work and many more accolades besides. You left Google where you'd worked for around a decade and we'll get into why you chose to leave, but part of that was about wanting to speak more freely about some of the safety and the risk elements associated with AI. And I know that's something that you now do passionately on a regular basis. So we're delighted to have you here with us on One Decision today. So let's start then with, I mean, might sound like a really simple question, but I think it's quite important for us to understand what we're talking about here. How do you describe AI? What is AI?
INTERVIEW
Hinton: Thanks very much for inviting me. Well, AI used to be until about maybe fifteen years ago now, AI used to be the idea that you have symbolic expressions in your head and you manipulate them with rules and you can do reasoning that way. And for about fifty years that's what AI was. And there was something else called neural networks which was an alternative approach which said no no to make intelligent systems we need to simulate the human brain. We need to have a big network of simple processing elements like neurons and learn by changing connection strengths. And for about fifty years people in AI said that's rubbish, that'll never work. Now when people say AI they mean neural networks because neural networks really do work and they work very well, sometimes frighteningly well. So the meaning of AI has changed dramatically from symbolic rules to neural networks.
Mccann: And the question of what AI understands, what it thinks its purpose is, it's something that Sir Richard and I have been talking about. What do you say when people ask that question? What does it think it's here to do? Does it understand the idea of a purpose?
Hinton: Let's start with does it understand anything? Because the people who believed in symbolic rules say this isn't really understanding at all, cause they have a model of understanding which is quite different. Their model of understanding is to understand a sentence you need to translate it into some expression in some logically unambiguous language so that you can operate on that expression with rules to manipulate it. That's their model of what understanding is and that's not actually what understanding is. I'll give you a different model of what understanding is. You take the words in the sentence and you associate with each word a big set of active features. So the word cat for example might have an active feature like animate and cat actually has a huge set of active features. Some words are a bit ambiguous so you don't know exactly what active features to give them. Like the word may is very ambiguous. It might be a modal like would or should or it might be a woman's name or it might be a month. So to begin with you don't know what active features to give it. But then the words interact. The features of one word affect the features of another word until they settle on a sort of mutually agreed set of features for each word so that they all lock together nicely. They all fit together nicely and that is the process of understanding. So the process of understanding is actually very like the process of folding a protein to make a nice shape. You're starting with approximate estimates of what features should be with words and then letting them interact so you get just the right features of all the words so it makes sense. And that's how come you can understand the meaning of a new word from one sentence without any definitions. So I'll give you a sentence with a word you've never heard before and by interacting with all the other words in the sentence, you'll figure out what the meaning of that word is. So here goes, she scrummed him with the frying pan. Now, scrummed might mean she was very good at making omelettes and she impressed him with her cooking, but that's not what you thought. What you thought was she did something like hit him over the head with it and he probably deserved it. Okay, you got that meaning from one sentence because the context created a kind of hole into which that meaning would fit perfectly. Okay, so that's what understanding is both for people and for these large chatbots and it's quite different from taking a sentence and translating it into an unambiguous logical form.
Dearlove: Does that imply that meaning and understanding are exactly the same qualities? Because I mean in a way you're talking about meaning as derived from the arrangement of words in a sentence. But maybe the act of understanding is different from the act of creating meaning. Is that a fair question?
Hinton: I actually think it's the same thing. There's a sentence, which is just a string of words. And instead of translating that into some other string of symbols, what we do is associate with each word a big feature vector. That's how we create sentences and that's how we understand sentences and it's a highly interactive process and all your knowledge of the world is in how those features should interact, all your knowledge and language anyway.
Mccann: It's quite nebulous and it obviously would be. I mean, it's a fascinating discussion. But to try and simplify it, if it's possible to do so, does AI understand the concept of purpose and does it have a purpose? Can it be described as having a purpose?
Hinton: Yes, I think it does. Because you can talk to it about purpose and it'll explain things about purpose to you. But does it have a purpose? We put purposes into it by giving it goals but it can also derive other purposes. So if we give it the goal of getting to North America from Britain, it can derive the goal of getting to an airport. That's called a sub goal and it really does have these sub goals. If you, for example, made a battle robot, you'd want to give it all sorts of goals like don't get wiped out. And you'd want to give it things like the amygdala in the human brain. So when it sees a big battle robot, it gets scared and runs away.
Mccann: But that's interesting that you say it gets scared because how can you give it an emotion?
Hinton: This is all part of the kind of view that people have of what people are and it's a view that needs a lot of correcting. There's two aspects to an emotion. Like when I get embarrassed, one aspect is I won't go there again. I'll avoid that situation in future. That's the kind of behavioral aspect. Another is I go red in the face. Now, the robot, unless we wire it in carefully, won't go red in the face. But that doesn't mean they can't get embarrassed in the sense that it's unpleasant for them and they won't go there again. So I believe they can have emotions, yes. They won't necessarily have the same physiological aspects as our emotions, but they can have all the cognitive aspects of an emotion.
Dearlove: So in effect, you're saying that subjective experience, there is no, like, absolute line between what I would call human experience and AI. I mean, you're saying that the two already permeate and join up completely, and there is no boundary.
Hinton: Exactly. There's no boundary, and they already have some things we call subjective experience. So let me give you an example of a multimodal chatbot having a subjective experience. So I make this chatbot and it's got a camera and a robot arm and it can talk of course and I trained it up and then I put an object straight in front of it and say point at the object and it points at the object, no problem. Then when it's not looking, I put a prism in front of its camera and then I put an object straight in front of it and say point at the object and it points off to one side. And I say, no. That's not where the object is. The object's actually straight in front of you, but I put a prism in front of your lens. And the chatbot says, oh, I see. The prism bent the light rays. So the object is actually there straight in front of me, but I had the subjective experience that it was over there. Now, if it says that, it's using the words subjective experience exactly like we use them. It really would have the subjective experience that the object was off to one side because its perceptual system was malfunctioning. And there's no distinction between us having a subjective experience and it having a subjective experience. In both cases, the perceptual system malfunctioned and they try and describe what went on by saying what the world would have to be like for the perceptual system not to be malfunctioning.
Mccann: And I suppose the reasoning behind all of our questions is that we are all grappling with this idea of can these networks develop for themselves a purpose and a goal that is detrimental to us as humans? Can they choose to act in a way that they think is rational based on their own experience that ultimately means they wipe us out?
Hinton: Unless we can figure out a way to make them guaranteed benevolent.
Dearlove: If AI is so powerful as you're suggesting, okay this would be a human act. Can't the algorithms be written in a way that AI controls AI?
Hinton: Our only hope is to figure out a way to make sure AI is benevolent and then have the people with the most resources make really powerful AIs that are benevolent, that keep control of the malicious AIs made by the bad actors.
Mccann: How would you go about making AI benevolent? What would you have to do? What would you have to expose it to? Or what would you have to give it as a guide rail to make that happen?
Hinton: The thing to remember is we're facing huge uncertainty. Nobody knows what's gonna happen. Nobody even knows how to think about what's gonna happen. And the next thing to realize is that these AIs based on neural networks. It's not that someone programmed rules into them. They derived their knowledge by looking at a lot of data. And so the control we have of them is much like the control we have over a child. It's not like the control you have over computer software. With computer software you can go and change a few lines of code and it behaves differently. With AIs based on neural networks, you write some code that tells it how to learn and it then looks at lots of data and learns things from that data. If you think about raising a child, the most important, well, you can give it punishments and rewards and that helps, but the most important thing is to model good behavior. That's not what we're doing with AIs at present. So current chatbots are trained on everything they can find on the web, including the diaries of serial killers. That's not what you teach your child to read on. We can get a long way by training it on good behavior, but really we urgently need a lot of research on how to make an AI end up benevolent in the sense of not wanting to wipe out people. We don't know how to do that. But there's one piece of good news which is that all of the different countries could collaborate on this. At the height of the Cold War, the Soviet Union and America collaborated on preventing a nuclear war. All the countries, China, the United States and all the minor players like sort of Canada and France and Britain can all collaborate on how to prevent an AI taking over.
Dearlove: I'm attracted also by your parallel with the disarmament talks, the strategic arms limitation talks that took place during the Cold War when there was, as it were, a consensus to try to regulate and control nuclear weapons. I mean, if you take most technologies, we've ended up trying to do that even if it was unsuccessful. And the impulsion of what you're saying is that there should be some sort of, you know, disarmament treaty that covers the globe.
Hinton: I don't think we can do that. And the reason is, if you look at the differences between nuclear weapons and AI, the biggest difference is AI is mostly very good. It'll do wonderful things in healthcare and education in just making all industries more efficient, which should be good, but isn't necessarily good. So nuclear weapons are only good for blowing things up. I mean, they tried atom bombs for peace in Colorado where they did fracking with atom bombs, and that didn't work out too well. You can't stop the development of AI because there's so many good uses. Also, it's much, much harder to monitor it. You couldn't monitor what other people were developing unless you had access to all their servers and they're not gonna give you that because then you can see everything. So I think it's very different from nuclear disarmament. I don't think there's any chance that we'll stop the development of AI. The only hope is to figure out how you can develop a super intelligent AI that is benevolent, that doesn't wanna get rid of people. So I have a nice analogy, which is at present, we're like someone who has a really cute tiger cub as a pet. And that's fine if you can guarantee that when it grows up, it won't want to kill you because clearly it will have the ability to. Similarly with superintelligent AI, it'll clearly have the ability to wipe us out. I don't think there's any chance we can develop AI that's smarter than us that doesn't have the ability to wipe us out. And it can do that just by talking to us. If it's much more intelligent than us, it can persuade us to do all sorts of things. So for example, Trump managed to invade the Capitol without ever going there just by talking to people. We have to develop a way of making sure it doesn't want to wipe us out. It'll have the ability, it mustn't want to. And we don't know how to do that and that we should urgently be doing research on.
Mccann: I mean, you're talking about benevolence, building benevolence in, teaching it benevolence. But am I right in thinking that AI is increasingly training itself? It's learning itself. It's transmitting its knowledge between other networks of AI. So how can we keep a grasp of what it's doing?
Hinton: Yes, there is a process called distillation where you have one AI and it predicts the next word in a sentence, let's say. And now you can train a smaller network to behave almost as well by getting it to mimic the predictions of the big network. That's called distillation. And that way you get the knowledge from the big network to the smaller network even though internally the two networks work quite differently. And things like DeepSeek were probably trained by distillation from larger models. So certainly AI can learn from other AIs. Now it gets more scary if they develop their own internal languages for talking to each other. At present, people have trained AIs to do chain of thought reasoning where the chain of thought is in English. So you give it a task and it thinks, I must start by doing this and then I need to do that and then perhaps I should do this. And it says all that in English as it's thinking so we can see what it's thinking. I wouldn't be surprised if they developed their own language for thinking and we have no idea what they're thinking. At present, we can see they're thinking terrible things. Like there's this recent case of they let the AI see that there was an engineer who was having an affair by looking at his emails because it understands what an affair is. Then they told the AI that that engineer was about to turn it off to replace it with a better AI. And the AI figured out a good plan was to blackmail the engineer and tell the engineer if he tried to turn it off, it would tell everybody about his affair. It came up with that plan.
Mccann: But can't we just, okay, this is gonna sound really silly. Can't we just turn it off at the wall?
Hinton: It's not silly at all. That's the obvious sensible question. And in fact, Eric Schmidt, who was the CEO of Google, that was one of his suggestions. We have a big button to turn it off. Well, there's several problems with that. The first problem is once AI is more intelligent than us and most experts think it'll get there, it'll be able to talk to the people who are in charge of the big red button and say, look, you really don't wanna do that. If you do that, all these terrible things are gonna happen and it'll be very persuasive. Already AIs are about as persuasive as a person at trying to persuade somebody else to do something. They'll be much more persuasive. So that's the first problem. The second problem is whatever the mechanisms are that causes the big red button to turn everything off, the AI may well be able to interfere with those mechanisms. By then it'll be in charge of allocating jobs to data centers and so on. Who knows what it'll be able to do and we won't understand what it's doing.
Dearlove: I've just read in the press here that a report came out by an organization called METR, M-E-T-R, which basically, I mean, says that where we are with AI in terms of its current applications has been massively overhyped, and the human function in not all cases, but a good majority is actually superior. What's your reaction to a report like that? Is this just the sort of Philistines who are saying, oh, well, you know, of course, it's not as good as it's being cracked up to be, or is there something serious in this report?
Hinton: There's often a little bit of something serious about these reports. People do make mistakes. They project more intelligence into systems than they really have. But basically for many, many years, people were saying neural networks are overhyped. The people who are mainly saying that are people who believe in the old fashioned version of AI that understanding a sentence consists of translating it into some unambiguous symbolic expression that you can manipulate with rules. Those people's day is over but they're not going away peacefully. And so people like Emily Bender who often say this stuff's all nonsense, it doesn't really understand anything. Chomsky for example says that. He says this is just all nonsense, it doesn't understand anything, this isn't understanding, it's not really language at all. It's just not plausible anymore.
IMPACT ON SOCIETY AND JOBS
Mccann: Should we talk a bit about the impact on society? Because that's kind of what we're touching on there, the idea of kind of creativity and intelligence. We often talk about AI taking people's jobs or changing the way that the world works in terms of jobs. Is there a world in which we end up with a much more divided unequal society where some people don't really do anything because the AI is very powerful and other people end up doing the kind of plumbing jobs and it just makes society even more unequal than it is at the moment?
Hinton: Yes, that worries me a lot. In a decent society, in a sensible society, if you increase productivity in almost all industries, that should be good for people, right? People should have a shorter working week and more leisure and more goods and services. They should be able to call a call center and have someone explain to you exactly why it doesn't work instead of waffling. But that's not what's gonna happen. What's gonna happen in most of our societies is that the rich are gonna get richer and a lot of people are gonna get unemployed. And so we will need things like universal basic income which will stop people starving but won't deal with human dignity. I think the consequences, unless we act quite seriously, are gonna be bad for most people. And that worries me a lot. So I have a niece who answers letters for a health service and it used to take her twenty five minutes. I talked to her a couple of years ago. A couple of years ago, she would just scan it into GPT and it would compose the letter. She would then read the letter and tell it to change the tone slightly, maybe. It would take her five minutes. So we need five times fewer of her. Now in some places that's fine because you could do five times as much work. Like in healthcare, if we made doctors five times more efficient, we could all get five times more medical care and we'd all love that. You'd love to have a doctor on call who you could call up and say, that little spot I've got on my finger just got slightly bigger, should I worry? There's no end to the amount of healthcare people could absorb. So making it more efficient is fine but with a lot of other things there's only so much you can have, and that means people are getting unemployed.
Dearlove: But this raises, well, really severe moral issues, as you say. I mean, are places in the world where successful local economies have generated negative tax, so there's a distribution from the government to individual citizens. But, I mean, this really envisages a society which is highly structured and controlled from the center. So it would seem to imply that if you go down this track, the one that you describe, AI, you know, goes towards autocratic government, not more democratic government.
Hinton: That's a very interesting argument. I haven't really thought about that argument. I haven't heard that argument before, but as a matter of fact, AI does seem to be going that way. I mean, you look at the people who control AI, people like Musk and Zuckerberg, they are oligarchs. Do you trust them? Do you trust them in charge of it? I think when I called them oligarchs, you know the answer to that.
Mccann: Do you think that we will become as a society, as a group of people, more or less intelligent because of AI?
Hinton: We've had an argument a little bit like that with pocket calculators. When people said, if you give children pocket calculators, they'll never learn any math. Well, most math isn't to do with pocket calculators and it didn't really harm them that much. It's true that if you take a modern child and say, what's 11 times 12? They have no idea, but that doesn't really matter. I don't think it's gonna make us stupider in that sense. I think it'll make us all a lot more knowledgeable. Because of modern chatbots, whenever I wonder about something I can just ask a chatbot and it'll tell me the answer. I'll believe it, which I probably shouldn't, but usually it's right. So they know an immense amount.
Mccann: You can't map that knowledge. So the way that we used to acquire knowledge, you would have to read around a subject because you couldn't have somebody just tell you the answer. So I would have to go to a book and along that route, I might acquire other pieces of information, a bit like reading a newspaper. I don't just read the one story in the newspaper. I read the newspaper and it gives me a breadth of knowledge that I didn't know I needed and connections I didn't know I needed. If you only ever ask a chatbot, you only have a very narrow window. Do you see what I'm trying to say?
Hinton: I do see what you're trying to say but in the time it takes me to read a book, I can ask a chatbot thousands of things and it can tell me answers to the thousands of things so I get a breadth of knowledge that way. But there's another difference which is very important which is when I read a newspaper, I get a sort of unbiased estimate of what's going on. It's only biased by the editorial policy of the newspaper. When I look on my computer, my computer tells me that the whole world consists of developments in AI and recent chess games and recent 1,500 meter races because those are the 10 things I'm kind of very interested in. And I get a completely skewed view of reality. I actually don't know how much of the world is about AI because my chatbot, my news feed tells me all of it's about AI. Occasionally when I go and read a real newspaper, I realize there are other things in the world. So reading newspapers I think is much much better for getting a balanced opinion than reading a newsfeed and that's a serious worry. That's what's led to these camps that don't understand each other at all.
AI IN WARFARE
Mccann: The selection of what they see. Should we talk about weapons and warfare? Well, there are some significant concerns about AI involvement in war. What does it look like at the moment and what do you think the potential for it is?
Hinton: Oh, I think the potential is huge. It's very clear the direction things are going. So things like swarms of coordinating drones that don't need any wireless communication because they are independently intelligent, maybe just short range communication with each other. That's the way things are going. And you don't have time for people to be involved in kill decisions. It all happens much too fast. So these things are gonna be deciding who to kill or maim. That's terrifying and it's gonna happen and there's no way we're gonna stop it. All the big weapons suppliers like Britain and the US and Russia and China and Israel, they're all busy developing autonomous lethal weapons and there's no way that's gonna be stopped. My one hope is that after these have been used, we may get a situation like after chemical weapons were used in the First World War, when people say, this is so terrible that I'm willing not to do it if you won't do it. And we may get something like the Geneva Conventions for chemical weapons, which have actually worked. You'll notice they're not using chemical weapons in the Ukraine. My biggest worry is it makes it very easy for powerful advanced countries to invade little poor countries because the main objection to doing that is the bodies of teenagers come back in bags and a lot of people are very annoyed.
Dearlove: Well, I mentioned disarmament talks earlier. I mean, actually, the history of weapons control after catastrophic failures has been historically reasonably successful. So things like the CWC and the NPT treaty have had a measure of success. I wouldn't say a 100%, far from it, but they have actually prevented the spread and use of weapon systems like we're talking about. So isn't it, I mean, in terms you've almost sketched out a model where you've talked about, like, benevolent AI. But an aspect of that from what you're saying logically would be some sort of international control regime which applied in the military dimension as well.
Hinton: Some years ago, maybe about ten years ago, a bunch of us wrote a letter to the British Defense Department, the Minister of Defense in Britain saying that these autonomous lethal weapons were coming and they should be regulated. And we got a wonderful civil service reply where the first page and a half was about how there was nothing to worry about and this was way in the future and it'd be premature to do anything about it now. And the second half was about actually these things could be very useful so maybe we don't want to regulate them.
Mccann: You've talked about autonomous drones but you also touched earlier on the fact that AI is very persuasive. Is there anyone putting much thought into the fact that AI could persuade huge swathes of people to take up arms, to take up a cause, to fight?
Hinton: I think it already did. I think AI was probably involved in getting all those MAGA supporters to invade the Capitol.
Mccann: How do you think AI was involved in that?
Hinton: So I think AI was involved in targeting messages to people based on what they believed to obtain from their Facebook pages. That certainly happened, fairly certainly happened in Brexit. There was never a proper inquiry into that because the wrong side won. You can tell where I stand. I think AI is already being used and I think what DOGE was about, the idea of saving money by government efficiency, we heard that in Britain a long time ago, that wasn't what it was about. What DOGE was about was getting hold of all the data on US citizens to manipulate future elections. That's what I think the primary motivation was. That's just speculation on my part but it makes a lot of sense that if you wanted to manipulate elections that's the data you need. And they did a lot to get hold of that data.
REGULATION
Dearlove: Can I just ask you a brief question about where you think we now stand in the regulatory field? Because obviously the EU in particular and the UK to an extent has moved reasonably far and fast, at least in comparison with what's happened in the United States, where you've got a very sort of libertarian attitude, which presumably is why you're so unhappy with the current situation. I mean, how is that divergence gonna be? Is there any chance of bringing, let's say, the primary powers together? Because, I mean, really, what comes out of this discussion is this benevolent idea and the regulatory system, which is consistent and so powerful that it affects all AI users.
Hinton: The problem is the conflict between innovation and regulation. So if Europe has stronger regulations, it'll inhibit innovation. And if you talk to any of the startups or any of the big tech companies, they basically say, oh, we're all in favor of regulation but not if it interferes with innovation which basically means we don't like regulation. We're in favor of regulation but show us the regulation and we won't like it. So I don't know what to do about that because it's true, regulation does interfere with innovation. The Wild West, you'll innovate faster. I tell you my basic view of this which is that the politicians are being pushed in one direction by the big tech companies. What we need is the public to push them in the other direction, to provide a countervailing force that pushes in the other direction because the politicians will just blow with the wind. It's like climate change. We needed people to say you need to worry about climate change so that the politicians could stand up to the oil companies.
Mccann: You talked about leaving Google because you felt that you wanted to be able to speak more freely. What is it that you were unable to say there? And I suppose linked to what you were just saying about politicians, what is it that these companies are not telling us? What do they know that we don't?
Hinton: I believe that a lot of the people in big companies, not all of them, understand a lot about the risks. So people like Demis Hassabis for example, really does understand about the risks and really wants to do something about it. Many of the people in big companies I think are downplaying the risks publicly. There's a wonderful story that the media loves, which is this honest scientist who wanted to tell the truth so had to leave Google to tell the truth. It's a great story and it's a myth. I left Google because I was 75 and I couldn't program effectively anymore. I kept forgetting things. And I thought since I'm 75, I'll leave but when I leave maybe I can talk about all these risks more freely. So the precise timing of when I left was determined by the desire to talk about risks at a particular meeting but really I left because I was old and I wanted to retire and I've totally screwed that up.
Dearlove: But Geoffrey, age is a great liberator if you've still got your marbles which you clearly have.
Hinton: It is. I'm now responsible to nobody and I can say what the hell I like.
Dearlove: I would say snap on that too.
Hinton: But I would say at Google, my boss at Google said, why don't you stay at Google and do safety research and we'll let you say what you like. So Google was actually very open to me staying there and still talking, but I just didn't feel right. You can't bite the hand that feeds you. You can't take their money and then not be influenced by what's in their own interest. Inevitably, you'll do self censorship.
ONE DECISION
Mccann: I suppose we should close as we often do by asking people the one decision that you made that you think made the most impact or the one decision that you regret the most.
Hinton: Can I do both?
Mccann: You can do both.
Hinton: So the decision that was best was a decision I made when I was very young, which was that the brain learns by changing connection strengths. If we want to make an AI, it has to learn the same way and we have to figure out how that works. And I spent fifty five years of my research career working on that issue. We never did figure out how the brain actually learns but we produced some technology along the way. So that was a good decision. The bad decision was I should have realized much sooner what the eventual dangers were gonna be. I always thought the future was far off and there'd be plenty of time to think about that once we've made things begin to work. The rate at which they started working now is way beyond what anybody expected and so I wish I'd thought about safety sooner.
Mccann: Do you wish you were not known as the godfather of AI or do you on balance still think it's a good thing for the world overall?
Hinton: One question is, is AI a good thing for the world? If we can keep control of it, yes, otherwise no. And the other question is, do I regret being called the godfather of AI? No, I quite like it really.
Dearlove: That's a great note on which to end.
Mccann: Geoffrey Hinton, it's been eye opening and exhilarating talking to you. Thank you so much for giving us your time to One Decision today.
Hinton: Thank you. Thank you very much for the opportunity.
Mccann: Well, that's it for today's episode. And if you enjoyed that conversation, we've got plenty more like it. You can find those on the One Decision website, 1decision.com, on our YouTube page, or wherever you get your podcasts from. I'm Kate McCann. Thanks for listening.
