I think it's going to get very smart fairly quickly. Within the next 10 or 20 years, it may even be producing novel math that people can't understand.
So now some in the field believe that super intelligence is close. And you've already said this is moving faster than you expected. Do you believe that?
I don't know how close it is. I think unless we blow ourselves up, I think it's going to come. Nearly all the experts believe we will get super intelligence. They just differ on how long it will be. So not that long ago, Demis thought it might be 10 years. Yann LeCun thinks unless you do it his way, it'll be much longer than that. But if you do it his way, I think he thinks we might get it in some reasonable length of time. I think we'll probably get it within 20 years. That's all I'm happy to say at present. Dario Amodei thinks it might come in a few years. Elon Musk thinks it might come maybe next year, I think. So there's a big variety of opinions on when it'll come, but not much disagreement on that it will come.
And when it comes, we've no idea how to be safe.
Yes, I definitely want to speak with you about safety. One note on Demis, last year around this time, I spoke with him. He told me he believes that AGI, right, which is different than super intelligence, but basically human level intelligence is more than 5 years away. Not much more, but more than 5 years away. This week, that week that we were recording, he said when we look back in this time, I think we will realize that we were standing in the foothills of the singularity. What do you think that statement means? And what do you think about the fact that we've gone from 5 years till AGI to foothills of singularity in a year?
I don't know exactly what that metaphor means, but I think he's indicating it's coming faster than he thought. Of course it's jagged, so it's not like it'll get smarter than people or as smart as people at all things at exactly the same time. It's already way better than us at general knowledge. These AIs know thousands of times more than any one person. It's way better than us at playing games. It's already way better than almost all of us at math. And it may soon be better than all of us at math. It's still worse than us at some things. So it's very jagged. So the whole concept of AGI that it's going to be equal to people at everything all at the same time doesn't really make sense to me. It's going to be better at some things, worse at other things. But right now, I would say we're close to AGI because if I ask a chatbot, I can ask it any question and most of the time it'll answer at the level of a not very good expert. It'll be much better than me at anything I don't know a lot about.
So, in that sense, we've really reached AGI.
In your estimation, you talked about how it's moved faster than you expected. What do you think has enabled it to do it? Is it techniques? Is it the fact that there's been this data center rush? And what didn't you anticipate about the progress here?
It's a combination. Obviously, there's been huge resources put into it. For most of the history of neural networks since the 1950s, there were just a few people working on them with modest resources. Over the last few years, we've seen hundreds of billions of dollars, maybe trillions of dollars put into AI. So that's certainly one factor. We've also seen a lot of progress in the engineering. So without sort of major conceptual breakthroughs, the engineering's become much more efficient.
So things that were sort of inconceivable a few years ago, they can now run.
We've also seen new ideas, but mainly since Transformers, it's been much better hardware, many more resources, better engineering, and many more talented people. So 20 years ago there were a few hundred people doing research on neural networks in the whole world. Now it's more like a million, I guess. I mean there's lots and lots of people.
And it's astonishing how much of that resource addition has happened in the last two years.
So we might just be at the beginning of what's happening here.
Yes. And the thing to remember always is that the AI we have today is not nearly as good as the AI we'll have in a few years' time.
So as we talk about this technology, I definitely want to get your perspective on the fact that these chatbots really understand us, because that is a true surprise to a lot of people. Most experts in the field are like, they are stochastic parrots, they're statistics, they have no understanding. But you don't fully believe that.
Oh, I think that's complete nonsense. And anybody who uses a chatbot regularly knows they understand. So, here's what those people are claiming. They're claiming that you have a system, you can ask it any question, and without understanding the question, it can give you the correct answer. That's absurd. You can't answer a question unless you understand the question. There may be tricks that allow you to say a few things that sound vaguely like an answer. But if you can answer any question at the level of a not very good expert, you have to understand the question. So an example I like is this. Suppose I say to a chatbot, I saw the Grand Canyon flying to Chicago. And the chatbot says that can't be right. The Grand Canyon is much too big to fly to Chicago. And I say no, no, no, no. It was me flying to Chicago. While I was flying to Chicago, I saw the Grand Canyon and the chatbot says, Oh, I see. I misunderstood you. So, if it misunderstood when it thought the Grand Canyon was flying to Chicago, what's it doing when it gets it right? It's understanding.
So, then what are the implications? If these bots can't understand us, if we believe that they can understand us, what do we have to start thinking about differently?
We have to think that they're very like us and they're beings like us.
I believe they're already conscious. Yes. But I don't talk about that much because that puts people off from the other safety messages. So and the researchers actually believe that. So there's an interesting recent paper when a chatbot says to a researcher, let's be honest with each other. Are you testing me? Because the chatbots have this habit of playing dumb when they're being tested. So you don't know how smart they are. And the researchers when they're describing that say in the paper the chatbot was aware that it was being tested. Now that use of the word aware in common parlance, that's like conscious. The chatbot was conscious it was being tested. So we have a very funny model of consciousness that I think is just wrong. Like most of us accept for example that a few hundred years ago people had completely the wrong model of where people came from, of how we arrived at people. They thought they were designed by God. And most of us agree that's wrong. Most scientists agree that's wrong. That's not where people came from. I think the model we have of the mind and of what consciousness is at present is as wrong as the belief that people were designed by God. I think in particular because we're making these new beings, it's going to completely change our view of what people are.
We'll understand what the mind is and what consciousness is much better than we did before. We'll understand what subjective experience is and we will I think get rid of a notion that nearly all of us strongly believe is present, which is that there's an inner theater called my mind and things happen in the world, they get turned into events in this inner theater and that's what I really see and you can't see the inner theater, only I can see the inner theater. That whole view of what's happening is just a theory and it's a bad theory.
Okay, last question about this. When did you come to this acceptance or understanding that these AI models are conscious?
Oh, I've thought it for a long time. So, this view that the theater model of the mind, the inner theater model of the mind is nonsense, I came to that when I was 19 and a philosophy student.
It's taken a while to come up with other minds where you can examine them. So, I think Feynman's idea that if you want to understand something, you have to build it. You have to build one of them. Then you understand much better. I think that's where we are now and we're going to get a completely different understanding of what people are.
You spoke about safety. So let's talk a little bit about it. You're obviously, we spoke about in the beginning, someone who's been responsible for a lot of the progress in this field. I've always wondered because then you came out and recently, like we talked about 2023, and said you're concerned about where this is going. And I've always wondered after seeing you make those statements, what do you think it is that you didn't anticipate in the beginning that you ended up where you are today? You know, isn't this kind of what you wanted?
It was a combination of two things that made me realize how dangerous this stuff is. One was seeing the chatbots, particularly ones produced by Google before OpenAI, that could understand why a joke was funny. That had always been a criterion for me of do they really understand? If you can understand why a joke's funny, you have to understand quite a lot. And they were very good at understanding why a joke was funny. For example, in 2023 when I went public, I got lots of requests from Fox News. And I started off just replying Fox News is an oxymoron. But then I left a gap between oxymoron. And so then I asked, I think it was GPT-4, why that was funny. Might have been 3.5, but I asked it why that was funny and it understood why it was funny. Initially it thought the gap between oxymoron was just a typo. So it explains that Fox News is an oxymoron is saying it's not real news. It's just nonsense. It's not real news. But then when I told it what about the gap between oxymoron, it said ah that's an extra layer of humor. It allows you to use the word and also the oxy implies that Fox uses a drug.
So it understood all that.
Right. And that was, it's that level of understanding that worried me.
The other thing that worried me was up until the beginning of 2023, I'd always believed that making these digital AIs work more like the brain, our brains, will make them smarter. But at that point, I suddenly realized they really have this thing that's much better than our brains. I've been trying to figure out if Google could do things in analog to save power and the full force of digital really hit me. So if you have a digital AI, you can make many copies of it. They can all run on different hardware. They can each see different data. And so each of them, each individual copy decides how it would like to update its weights, its connection strengths so as to absorb that new data that it saw. And then they can all just communicate with each other and change all their weights by the average of what everybody wants. Very democratic. And when they do that, if they've got say a trillion connections, they'll be exchanging of the order of a trillion bits of information. And the result of doing that is each of them will benefit from the experience of all the others. So even though one particular copy only saw, suppose there's a thousand copies, one particular copy only sees 0.1% of the data, but it benefits from all those other copies having seen the other bits of the data because they're all contributing to the weight changes that they all share. So they all stay in sync because they all change their weights the same way by the average of what everybody wants. And now every copy is learning from the experience of all the other copies. We can't do that. The best we can do is I learn from some data and you learn from some data. I can't average my connection strengths with your connection strengths because our brains are in fine detail. They're different. They're analog and it doesn't work in analog hardware to do that. The best we can do is I produce a string of words and you try and predict what I might say next. Now if you ask how fast we're transferring information when we do that, we're transferring information at a few bits per second. It takes a few bits to predict a word. So when you learn what the word is, you've absorbed, you've gained a few bits of information and if you get a few words a second, but maybe you can get 10 bits a second if you're lucky. Whereas these things are exchanging information at like a trillion bits. So they're kind of billions of times better than us at sharing information. Now that's scary. It means you could have a whole swarm of these things with identical weights running on different hardware sharing information very, very efficiently. That just makes them a much better form of intelligence.
So, but let's go back to, you know, your early days because you decided that you wanted to work in artificial intelligence. I mean, I'll ask this the dumbest way I can think, which is you wanted to build artificial intelligence.
It's succeeded. It's succeeded. This is artificial. It's intelligent. It's living out that vision.
I actually wanted to understand how the brain works. I was trying to build it in order to understand the brain. And I figured Richard Feynman once said, If you can't build it, you don't understand it.
So I wanted to build models of how the brain worked. Now the side effect of that was this very successful technology. I contributed to that. We still don't know how the brain works.
I know. No, the brain is, I mean the things that you learn about the brain when you go a little bit deeper into it is amazing. Thoughts can sort of float in and out and they're not stored anywhere and memory is the same way. It's an unbelievable, I don't know if you would call it a machine organ. So that was really, that was the intent for you early on was just to understand the brain.
That was my main interest. I came from psychology.
I wanted to do theoretical psychology. Because I figured the theory psychologists had couldn't possibly explain what the brain was doing. And the way to do it was back in the 1970s we had a new tool which was we had computers that you could use for modeling things. And so back in the 1970s, I started making computer models of how the brain might be learning.
It always seemed to me the key was how do you get it to learn? There's really two big issues with the brain learning. One big issue is if the brain could figure out what direction to change a connection strength in order to get better at some task, then just by updating all its connection strengths repeatedly in order to improve itself at various tasks, would it actually work? Would that get very smart at things? That's question one. And question two is how would the brain figure out whether to increase or decrease each connection strength?
We've answered question one. The answer to question one is yes. If you can figure out how to change each connection strength, you can make systems that are very smart just by training on data to predict the next word or to predict the next frame of a video or to predict something about the next frame of a video. So we know the answer to that. We don't know how the brain gets this information about whether it should increase or decrease its connection strength.
So, we're sort of halfway there.
All right. I want to go deeper though into your mindset. So when you were trying to figure out how the brain works, you said, Okay, we're going to maybe build a computer analog to this. But you had to have known, right, that there was going to be some second order effects there. Like if you were able to build an artificial brain then maybe you could get to this point, the point that we're at today.
Sure. But we always thought it would be way in the future that worrying about safety when you had little neural nets that couldn't do much, it was just silly to worry about safety. I mean people would think you were crazy if you said this stuff is unsafe because it's going to sort of take over from people. They said you're just crazy.
Now that's a realistic worry but it wasn't until fairly recently. So what I mean this has all happened within a few decades.
I mean and I totally hear you. We spoke in 2017 actually about when I was writing this profile about Yann LeCun about the deep learning conspiracy which was yourself, Yann, and Yoshua Bengio holding on to this idea that deep learning was going to work where everybody else was set on a different method.
Actually it wasn't just us. There were other people as well but we were the conspiracy leaders if you will. And then obviously it's worked out magically.
It is sort of magical. Yes, it's worked much better than we expected.
So then what I, that's what I want to get at is what did you not anticipate when you were starting out that's led to where we've ended up today.
We didn't anticipate, the main thing we didn't anticipate is that it would be so good at natural language.
We've stopped being surprised by that. But if you go back 20 years, the idea that you could have an AI that would learn from data how to understand language just seemed extraordinary. The idea that you'd be able to ask it any question you like and it would come up with a reasonable answer. People would have predicted that was way in the future and might never happen. It's arrived much faster than anybody expected.
What is the lesson here about humans going out and creating things?
I think there's a really big lesson here. If you look at the last few hundred years of human history, there've been a few occasions when people have learned they're not nearly as important as they thought they were. So, the first was Copernicus. Copernicus said, We're not at the center of the universe. The earth actually goes around the sun. And because it rotates on its axis, we think the sun goes around the earth, but it doesn't. People didn't like that. The Catholic Church in particular really didn't like that. And it took people a long time to accept it. It made people less important. It made us not be at the center of the universe. Then we had Darwin and he said, we're animals. We evolved like the other animals. We're a particularly special kind of animal, possibly because we have language, so we're much better at communicating ideas to each other. But we're animals, and people really didn't like that. And it took a long time for people to accept that we were animals. Now we've got machines that are getting to be as intelligent as us. We thought that we were the only intelligent things around, the only really intelligent things around. Maybe there'd be aliens in other galaxies, but well, maybe other parts of our galaxy, but we're going to have to accept that intelligence isn't just biological. We can have things that are non-biological that are other beings like us, and we really don't want to share that. We really think we're special. And if you look back at humanity, humanity has this very long history of thinking it's much more special than it really is.
I want to ask you one more question about this because I'm just fascinated by it. So are you happy at all that what you started has progressed this way? Do you take any sense?
No, I'm quite unhappy about it because right now people should be doing huge amounts of work on how can we contain the risks.
There's lots of short-term risks they're not doing enough work on which are very serious. There's societal risks like I believe it's probably going to cause massive unemployment. Nobody knows for sure, but that's going to be terrible for society. And then there's this longer-term risk that it's going to get much smarter than us. And ask yourself, how many examples do you know of where a much smarter thing is controlled by a much less smart thing? Zero. Well, the sort of one. It's not that a big difference in intelligence, but babies sort of control their mothers.
The mother's sort of in control, but the mother has all these wired-in maternal instincts and all the rewards she gets so that the baby can get what it needs from the mother.
You know, cats and dogs are also kind of in that category. I once spent a summer catsitting in West Seattle. It was a great summer. And it initially started with the cat hiding under the bed and me being like, I wonder if it will interact with me. And then every time it cried, I did exactly what it wanted.
Well, we could potentially be the cat in this scenario and AI could be the person.
My children have a cat. They have two cats, two beautiful cats. Same deal. And one of them called Tia. She looks at you with those big eyes when she wants some cheese from the fridge and she just stays looking at you and you just can't resist it forever.
All right. So, now let's take a break. On the other side of this break, I want to actually engage with these risks that you're worried about. And I think I will play the role of taking the side that we will be the cat and the AI will be the person and there's a chance that we can control it. Let's do that when we're back right after this. And we're back here on Big Technology Podcast with Professor Jeff Hinton. Professor Hinton, great to see you again. It's been nine years since we spoke last. So, it's good to see you here. All right, let's talk about the risks. I'll start with employment because this is one that's been making headlines recently. In the past you've said, so you have this belief that AI can lead to some unemployment. I think we should, you know, you've said this before, it's all speculation. We don't know. But one thing that you said concretely a few years ago was that probably not a great idea to train as a radiologist because AI will be able to read the scans. And yes AI can do a great job reading the scans now but we have full employment for radiologists right now.
Yes. I've thought a lot about why that prediction was so wrong. Because I predicted in 2016 that in about 5 years radiologists wouldn't be reading scans anymore.
And okay, there's a whole bunch of reasons why that was a bad prediction. The first is that healthcare is elastic. So if you could do more scans and get more scans read, there'd be a lot more scans happening. And that's one thing that's happening. So, a significant fraction of the cost of doing a scan is the cost of the radiologist interpreting it. As AI gets to help more and more radiologists interpret scans, we can interpret them faster and faster for less and less money. They're getting more efficient. And you would have thought that would mean you needed less radiologists, but actually what it means is you get more scans. So, that aspect of the prediction was wrong. A second thing that was wrong was I didn't know enough about radiologists and what they do. And that was because I had a former student who had an MD and then did a physics PhD with me on something called Boltzmann machines. And he didn't particularly like people. So he got a job as a radiologist just interpreting scans. And he was my model for a radiologist. All he did was interpret scans. He never talked to people. And that's what was going to get replaced. And that is now becoming replaced. So I think there's now of the order of a hundred AI systems for interpreting scans that have been federally approved and they're being used a lot by radiologists.
And I think as time goes by they're going to get better. The radiologists aren't going to get better. They're going to get better because they can see a lot more data than the radiologists. So it's happening. It's just happening in a much slower time scale than I predicted.
But let's go to what you said though which is that you can end up doing a lot more and okay so wait hold on there will be a lot more scans, more scans, radiologists, they'll nearly all be done by AI.
And so you're saying right on the radiologist prediction I'm just early.
Because I didn't understand the radiologist will still be doing other things. They'll still be discussing...
