[0:00:00] This BBC podcast is supported by ads outside the UK. Host Thanksgiving with industry leading quality standards from Whole Foods Market. There are no antibiotics in their meat department ever. So carve up turkey and spiral cut ham with peace of mind. Shop organic sides like stuffing mix and cranberry sauce from 365 brand. And guess what? All their delectable pies are made with no high fructose corn syrup.
[0:00:28] Gourmet Thanksgiving flavors with the quality you want. Only at Whole Foods Market.
[0:00:36] At Lowe's this Veterans Day and every day, military members, veterans, and their spouses get automatic MyLowe's Rewards Silver status benefits. Plus, 10% off eligible purchases with no annual limit. It's one way we honor and give back to those who have served and still do. Learn more and verify now at Lowes.com slash military. 10% discount can't be combined with another offer. Exclusions, terms, and conditions apply. Loyalty programs subject to terms and conditions. Details at Lowes.com slash terms. Subject to change.
[0:01:06] Hello, today's guest will be the 200th person I've interviewed on The Life Scientific, and it's someone who's been on my list for a while. Demis Hassabis, the brains behind some of the most sophisticated artificial intelligence systems on the planet.
[0:01:22] Demis knows more about how machines think than anyone I know, and his company, DeepMind, was sold to Google for £400 million. He retired from international chess championships at the age of 12, worked in the computer games industry for many years, and later studied computer science and neuroscience. Using the things he learnt from all these different areas of interest, he pioneered a radical new approach to creating artificial intelligence. His computer systems are not...
[0:01:51] His AI system AlphaZero defeated the best chess computer in the world, having learnt how to play chess from scratch in just four hours. Now, Demis wants to use AI to help humanity. And he hopes one day humans and AI working together will make a Nobel Prize winning discovery. Demis Hassabis, welcome to The Life Scientific. Thanks very much for having me. Now, am I right in saying that...
[0:02:19] Your aim has always been to model what's called artificial general intelligence, to create machines that can think about, well, about anything. Yeah, our mission at DeepMind has always been to try and understand natural intelligence and then try to recreate that artificially. So that's the first aspect of what we do. And then our idea was and our vision was that if you created intelligence in a machine in this general way, then you should be able to use that general purpose intelligence.
[0:02:49] for almost any other task, you know, to help solve many other difficult tasks in science and medicine that we would like to, as a society, to understand better. So not just, you know, driving a car or looking at a medical scan, but thinking. Yes, exactly. Thinking to be general in the way that the human mind is.
[0:03:07] Okay, well, we're not there yet. I'm going to come to that and talk about that a little later. It all started, I feel, with your early love of chess. That's right. So when I was, I started playing chess when I was around four years old. So I was very young. Gasp from listeners, four years old. And, you know, I think I don't actually remember, but my father tells me I saw him playing chess against my uncle and I asked them both, could I play? And after a couple of weeks, I was managing to beat them. And I mean, they're not very strong chess players, but it was still surprising. And the thing was, I started... Hang on.
[0:03:36] At the age of four, after a couple of weeks, you could beat your father. Yeah. So I think he decided, well, we'd better take him to a chess club. And it all kind of kicked off from there. We went to our local chess club and I started winning tournaments there. And it was sort of my first career, really. And I was playing for the England junior teams and captaining the England junior teams and so on.
[0:03:54] And at that time, I wanted to become a professional chess player. And when you do that, and you're going to training camps and so on, and you're reading lots of chess books, you're trying and competing. You're trying to improve your decision-making process. That's what you're trying to do when you're a chess player. Like, why did I make that mistake? How can I avoid doing that in the future? How can I come up with better plans? And that, for me at least, made me start introspecting about how is my mind coming up with these ideas, coming up with these plans? What is that process?
[0:04:24] Obviously, I was trying to improve that process. And then I discovered computers and programming. And so for a while, they were two parallel tracks. When did you encounter your first computer? So I got my first computer when I was around eight years old, and it was a Sinclair Spectrum. I remember that. Like a lot of British kids had at the time. And then I remember...
[0:04:45] Going down to, actually it was Foyle's bookstore with my father and looking at the rather small selection of computer books there were at the time. And then just selecting out some basic programming books and going home and starting to implement that on the spectrum. And it was a very magical discovery in some sense for me that these tools existed. That in some way extended the power of your mind. At least that's what I felt from it. And I gather it was another visit to Foyle's bookshop where you actually found a book by David Levy about...
[0:05:14] Computer chess, bringing the two passions together. Yes, you know, my two passions started coming together, I guess, in my early teenage years. And I remember very clearly it was this one book. It was called The Chess Computer Handbook. And the reason I was interested in that was in order to improve my chess, what we'd ended up buying was one of those old school chess computers that was a physical board.
[0:05:36] I don't know if you remember those, where you actually press down the chess pieces on the physical board. And then it had little LED lights to tell you where the computer was thinking. And it was a special piece of kit, hardware, that was specifically just a chess computer. So I was using that to train against and learn openings and so on. So it was natural for me to start thinking, oh, that's interesting. How do computers play chess? And then I was fascinated how they were programmed. So I got this book, The Chess Computer Handbook.
[0:06:05] And it really explained very clearly these handcrafted rules and how the machine would use that to evaluate positions. And it was actually a beautifully written book, and it was quite short. And it didn't seem that difficult. So I thought, oh, well, I'll implement it on my computer that I had at that point, which was a Commodore Amiga, a bit more powerful than the Spectrum. I thought, oh, well, I'll have a go at writing one of these programs. It doesn't seem that hard. So this is about 12 years old. But I realized that I couldn't write a chess program on my Amiga because it would have needed too much memory. So instead, I picked a slightly simpler program.
[0:06:34] Yes. It's another strategy board game.
[0:06:50] Designing it, I played against my kid brother and it managed to beat him. And I remember being very proud about that, even though he was only five. So he probably didn't know how to play very well. So it's the computer that beat him, not you. That's right. But did you somehow feel that it was you who was winning? Yeah, well, it was a really weird feeling. I remember being quite proud of the fact that this thing I created was able to then do something.
[0:07:14] On its own, that was then useful. So I had that kind of pride, which was a different sort of one from the pride I was used to by winning a chess game directly. But I actually found it equally satisfying, interestingly.
[0:07:25] It's around about this time, I think, at the age of 12, that you sort of, is it fair to say you fell out of love with chess or you certainly gave up competitively? Yes, it was a gradual process. You know, over around that time, my kind of love for computers and my passion for programming and AI grew. And when that happened, you know, it was kind of connected with some epiphanies I had playing in big chess tournaments about...
[0:07:50] Maybe this was too narrow a pursuit to dedicate one's whole life to. I just had this intuitive feeling that although I love chess and I still love it as a competitive game, and it's incredible for training the mind, actually, I think it's like one of the best. I mean, I think it should be in school curriculum because I think it's just as useful for training your mind as doing maths or something else. It teaches problem solving and imagination, lots of things like that.
[0:08:15] And dealing with pressure, actually. And I just decided there was... I love programming. I started getting into physics, actually. It was my favourite subject at school. So I could just feel there were many things out there that were exciting pursuits and passions. Well, Demis Hassabis, you got your A-levels at your comprehensive school in North London at the age of 16. But you were, of course, then too young to go to university. You wanted to go to Cambridge. What did you do with yourself until you were old enough? So I wanted to go straight to Cambridge, but they...
[0:08:45] Quite correctly said that I wasn't old enough and that I should wait. So what I did instead was actually during my A levels, I entered a programming competition, a kind of national programming competition that was in a games magazine. And I came second in this competition. And interestingly, if you won it, you won a job.
[0:09:02] At then probably Europe's best game development house, it was called Bullfrog Productions, very famous at the time, and it was based in Guildford. But I came second, but I rang up the CEO, the owner, the founder of the company, Peter Molyneux, very famous game designer anyway, and said, you know, I came second, can I just come down for some work experience?
[0:09:19] And when I got there, their doorstep, you know, they were pretty shocked because I was a graduate. Yeah. And, you know, they were barely able to legally employ me. But I proved myself over one of the summers. And then I asked them whether I could come back for my, you know, my gap year and program professionally, effectively. And so we created a game called Theme Park, which was the main game I programmed. And the idea there was that you created your own Disney World and then you could design it however you liked.
[0:09:48] Thousands of little people would come into your theme park and enjoy your rides and so on. And the better that you designed the theme park, the more they would be willing to pay for things like the Coke and the French fries and so on and the balloons in your shops. And for its time, this is back in 93, 94, it was one of the first games of its kind, these management simulation games they're called, that were built on top of this AI emergent simulation. And it was incredibly popular and successful. It was.
[0:10:18] It sold more than 10 million copies. So it was one of the most successful games of its time. And people loved interacting with this simulation.
[0:10:26] And I gather at the time, you know, you're only what, 15, 16. You'd moved away from home and you were living in a hostel. You're living with this group of programmers. Yes, that's right. So I initially I was staying in the YMCA youth hostel in Guildford. And, you know, I was getting paid 200 pounds a week, I think, in a brown paper envelope and cash. And I was losing money by getting the train up and down from North London where I lived. So I had to.
[0:10:48] And in the end, me and the people working on some of these games ended up moving into Peter Molyneux's rather large kind of mansion, which became a kind of coding camp. And we would just be 24-7.
[0:11:03] Programming, sometimes, you know, two nights with no sleep, you know, three days straight, no sleeping for two nights, and just socializing and programming and creating all in one big mix, you know, and it was a lot of fun and very informative.
[0:11:20] Do your parents know about this lifestyle? I mean, how do they feel about you? No, so I think around this time, probably around 15 or 16, my parents, to be honest, kind of lost track of what I was doing. So both my parents are not into computers at all. So I would say they're both on the artistic and creative side of things. So yeah, once I finished my schooling, I think they basically left me to it. So having worked then for Bullfrog for a year and a half, you were finally...
[0:11:48] Old enough to go to university, you went to Cambridge to study computer science and having got your degree in computer science, you went to work for a computer video game company, Lionhead. You then left to set up your own computer game company with one of your Cambridge friends, David Silver. So this was Alexia Studios. But after a decade working away from academia, basically the whole of your 20s, you then went back to university to do a PhD, this time in neuroscience at UCL.
[0:12:17] What prompted that change of heart to go back to academic study? Well, when I was working on computer games for almost 15 years, I suppose, at that time, the computer game industry was probably the most cutting-edge technical place you could be.
[0:12:32] But around the mid-2000s, that wasn't the case anymore. It became about franchises and marketing and other things. So I actually decided that it was time to move fields. And what better place to look at than the brain? I mean, obviously, the brain, the human brain, is a marvel of the universe. It's the only existence proof we have, perhaps in the entire universe, that general intelligence is possible. And so it makes sense to me that if you only have this one data point of what intelligence is, that it's worth spending the time to...
[0:13:01] And you weren't just looking at how the brain is wired and then try and get a computer to model that. You were trying to understand the nature of memory and imagination in the brain, which is a pretty ambitious thing to do. That wasn't what other neuroscientists probably were thinking about at that point.
[0:13:22] That's right. It's an important distinction to think about when we talk about getting inspiration from the brain, how the brain works. I'm really meaning at a systems level. So, you know, we're not trying to reverse engineer exactly how the brain works in terms of its low-level synapses and details. Because probably in a computer, you're going to implement that in a different way. I mean, it's a very different system. One's in silicon, the other one's carbon-based. So, you know, there's no reason to think the implementation details will be the same. But the principles, the algorithms, the architectures...
[0:13:51] I felt that those principles...
[0:13:58] could potentially be very useful to guide our research in artificial intelligence. So, yes, I picked the areas of the cognitive abilities that I thought were fundamental to intelligence, things like memory and imagination, but that we had at the time no idea how to implement in machines. Well, during your PhD, under the supervision, I should say, of Eleanor Maguire, you published some hugely influential papers on the connection between memory and imagination. For example, you showed...
[0:14:26] For the first time, patients who have damage to their hippocampus, which we know causes amnesia, they were also unable to imagine themselves in new experiences. Yes, we decided to test the patients with damage to their hippocampus on imagination, and no one had thought to do that. And it turns out they can't imagine either. So the headlines after that, and it was a very influential paper, were, you know, hippocampal patients are stuck in the present. Because in some sense they are. They can't...
[0:14:53] And of course, that's a key thing you might want for artificial intelligence because...
[0:15:07] So presumably this function of imagination evolved so that you can plan for hypotheticals, plan into the future, think about whether they would be desirable futures without having to do it in the real world, right? Because there's costs associated to that, especially if it's something dangerous. You want to plan it in your mind first and think through, would that be a good outcome before you have to actually try it out in the real world, which potentially could be very costly or expensive.
[0:15:32] Well, I mean, it sounds like a glittering academic career could have beckoned for you, having done this very good work during your PhD. But actually, not long after you finished, you again left academia, this time to set up a new AI company, DeepMind. This is together with Mustafa Suleiman and Shane Legg. Your mission statement was to solve intelligence and use it to solve everything else. Hardly lacking in ambition. But what does it actually mean then?
[0:16:00] Although it sounds a little bit far-fetched on first sort of reading, in a way it's sort of logical because it's...
[0:16:07] Human general intelligence that has created the wonders of modern civilization around us. That same type of brain that was evolved for, you know, hunter-gathering 10,000 or 20,000 years ago, it was flexible enough to invent, you know, iPhones and televisions and radio and all the modern things that are sort of unbelievable that we have now, that we take for granted, actually. And so humans are unlocked because our intelligence was general enough, you know, more general than, say, a monkey.
[0:16:36] The hardware was there in place. We had opposable thumbs and other things, but it was the mind. And so if we can recreate at least some aspects of that capability and then have that working alongside as a smart tool to help human experts, then I think that could really unlock human potential and sort of act as a multiplier onto human ingenuity.
[0:17:00] Well, at DeepMind, you were pioneering a technique, and you're still working on it now, a technique called deep learning. This is the deep in DeepMind. And also what's called enforcement learning. So we used two...
[0:17:15] And these are not sort of physical things? No, they're loosely based on the idea of neurons and synapses in the brain. But they're much more simplified compared to these nodes are much simpler than real neurons.
[0:17:44] And deep learning refers to neural networks that have many, many layers. So traditionally, neural network technology has been around since the dawn of AI, since the 50s and 60s. But for many 50 years, people were only using three or four layers of nodes, so quite shallow networks. And deep learning came along in the mid-2000s as a way of training very, very large neural networks, maybe with dozens and dozens of layers. Still not as complex as the brain, probably, the human brain, but very, very large.
[0:18:13] Now, the thing is, those types of deep learning systems are much more complicated to train, but they result in much more powerful systems that are able to model very complex data.
[0:18:26] So that's where deep learning is. And then once you have something like that, a model of the data or the environment that you're in, you can then use it to that model to make decisions. And that's where the second technique comes in, reinforcement learning, which is about reward learning. So you learn what are the right actions to take from a kind of process of trial and error. And you work out, the system works out which actions are useful, which ones are harmful for its goal. This is generally what people...
[0:18:56] Mean when they talk about machine learning. Yes. It's not something that's hardwired. The program can adapt and evolve. So the difference between machine learning and traditional AI is that traditional AI, you as the programmers or the creators, program directly in, in kind of logic statements, the solutions to problems. With machine learning, what you do instead is you build these architectures like neural networks and you expose it to data or experience directly, much like humans get exposed to experience. And then it...
[0:19:24] The algorithm itself updates itself in order to get better at whatever task that you've given it. And your first really big success, which got the world to sit up and really take notice, was this algorithm called AlphaGo, which uses this reinforcement learning idea. Tell me about AlphaGo. What did it do? AlphaGo was our program to play the ancient game of Go. And for those of you who don't know what Go is...
[0:19:54] It's a very esoteric game. In fact, in Asia, it's considered an art form, not just a game. So it doesn't lend itself to the type of expert systems, the kind of handcrafted systems that were used by IBM and others in the 90s to conquer chess and beat the world champion at chess. So famously, Deep Blue was the IBM computer that beat Kasparov. So Deep Blue beat Kasparov in 1997. And you might be thinking, well, that's 20 years ago. We have much more powerful computers now. How come that just didn't scale up to go? And the reason is because...
[0:20:23] Actually, no matter, even if you used all the computers in the world for a million years, it wouldn't be enough to brute force a solution to Go because of the number of possibilities. So we had to come up with something much smarter. And what AlphaGo was, was our program to learn for itself from first principles by playing millions of games against itself of Go. It figured out its own rules. It teaches itself. It's not like Bootstrap. It pulls itself up from the Bootstrap. From random, from a kind of random soup, if you like, you know, from first principles, what are the right motifs?
[0:20:52] What are the right ideas? What are the right evaluations about Go positions? It figured that out for itself. We just built the architecture and set it running. And then after actually only a few hours, once we optimized it, you know, within a day, it would go from random play to better than world champion. And it's quite fascinating to see that in real time almost happening in front of your eyes on the servers that we use. And of course, you did famously put it to the test, the equivalent of IBM's computer beating Garry Kasparov.
[0:21:21] You took on the world's top.
[0:21:24] Go player. That's right. So in 2016, we had our huge match in Seoul, in South Korea, against this legendary Go player called Lee Sedol, acknowledged to be the kind of greatest Go player of the past decade. And it was a huge deal. I mean, 200 million people watched the match live. It's incredible. And basically, yeah, we managed to win the match 4-1. So that was a pleasant surprise. But even more important than winning was...
[0:21:53] Actually, the unexpected thing was the style that AlphaGo played in. There's the famous Move 37. That's right. It was such a surprising move. The commentators that I was watching at the time, we were obviously there live, literally almost fell off their chairs. And the reason is, you know, it played this Move 37 in what would be regarded as a totally bad position. It was played on the fifth line instead of the fourth line. And any, you know, amateur Go player will tell you in that position, you should never play on the fifth line.
[0:22:22] It's like one of the
[0:22:24] Proverbs of Go. So it went against one of the oldest sort of proverbs in the book. So people thought it had made a mistake. Yeah, and not just made a mistake, it was just a ridiculous move. So it wasn't luck that it put it there. This was planned? Well, it was predicted, let's put it that way. So it obviously decided that although it looked locally as a bad move, in the grand scheme of things, this was in the right position. And so later on, so this game has been analysed to death by the millions of Go players afterwards, and it's regarded now...
[0:22:53] Now there's an unbelievable classic move that will probably go down in Go history. And people are learning now. The grandmasters of Go are learning from this AI. Yeah, so there's books, lots of books now written about AlphaGo style. But also, I think, as it was described to me by one of the Go players that helped us, Fan Wei, that it unlocked his mind. It freed his mind from the shackles of tradition. That's how he phrased it to me. I mean, when Lee Sedol was defeated, you were there in Seoul witnessing it.
[0:23:21] How did it make you feel? There was relief
[0:23:23] There was joy and amazement in the way that AlphaGo had played. And then on the other hand, the games player in me. So, you know, I continue, after chess, I continue to play many other games. You were some sort of five times world champion games player. Yes, all-round games player. This sort of thing called the Mind Sports Olympia. It's a bit like a decathlete rather than a, you know, 100-meter sprinter. So, you know, I love playing all kinds of games, including Go. And so the games player in me felt some pain for him because, and also, he was such a...
[0:23:53] Such a noble, Lisa Dole was incredible, noble, almost like samurai or warrior spirit he had.
[0:24:00] And so I felt his pain as well because he was expecting to win. But it was an incredible once-in-a-lifetime experience. And overall, it was amazing. And what pleased me is he came out of it. It was great that he won a match, which he played his own brilliant move, which is also going down in history, move 78, which confused the machine. AI's can lose as well. Yes, and he won an incredible, brilliant match as well, brilliancy. And then he went on to say it was the best experience of his life. So I felt in the end...
[0:24:29] The whole thing worked out very well. And, of course, you didn't stop there because your next move was to generalise AlphaGo's intelligence and you developed the programme AlphaZero. How is that different from AlphaGo? So there's a bit of a lineage of systems. So AlphaGo was the first programme that was able to play Go, but we'd always had in mind a general game-playing system that could actually play any two-player board game.
[0:24:55] Perfect information board game. You just give it the rules and it just learns. You just give it the rules and nothing else and it will just learn itself. And you could give it any rules to any game.
[0:25:04] And then we ultimately ended up with AlphaZero. So the word zero in the name refers to using zero domain knowledge. So other than the rules, it's literally from first principles. And then over just a few hours, it can learn any game to better than world champion level. Within a few hours, it learned to play chess and defeat the best dedicated chess-playing supercomputer. Never mind humans. Humans were long down the road being defeated. Yes.
[0:25:34] So that was the incredible thing. So of course, me being a chess player, the first thing I wanted to try AlphaZero on was chess. The ultimate testing ground, if you like, for AI. And again, like in Go, it actually invented a kind of new style of playing chess, which astounded all my chess friends in the chess world because everyone thinks that chess has been fully explored.
[0:25:56] Because obviously we've had chess computers for a long time now, and the best chess players have explored every variation they can think of. And yet, AlphaZero was still able to discover a kind of new style.
[0:26:06] I think it suddenly occurs to me, I have to say at this point, in case listeners are imagining AlphaGo or AlphaZero as some sort of humanoid robot sitting across the table from the human player. These are just, you know, computers. These are just programs within a computer. Yes, we should explain that they're basically running on big servers, data centers. Right. And we're running, you know, usually about a thousand of them at once that are playing in huge virtual tournaments against each other.
[0:26:32] And I also think that many listeners, while impressed with what these AIs can already do in terms of making us look like amateurs when playing them at games, are also wondering what the real broader benefits of AI will be. So using AI systems to play games is very fun. And of course, it has a special place for me as a games player. But the main reason we use games in virtual environments is that we want to train...
[0:26:58] General systems that one day we feel will be powerful enough to generalise to real world problems, like in science and healthcare. And so it's very exciting now because in the last couple of years, we now have powerful enough systems that we think we can start having real impact on real world challenges. And I know that you and DeepMind are very involved in a number of projects. Tell me about the work you're doing with Moorfields Eye Hospital.
[0:27:24] So a doctor from Moorfields called Piers Keane, he came to us when he saw, I think, an article about AlphaGo and Wired. And he said, oh, I've got this problem. We have these retina scans. There's way too many of them that need diagnosis. Then we have doctors and radiographers to scan. And so we collaborated with a bunch of the amazing, you know, the world's best eye doctors at Moorfields. And we created the system that was as good as...
[0:27:48] The best doctors there at diagnosing whether there's eye disease. And the really extra cool thing about the system was it could project back onto the retina scan.
[0:27:59] Where on the image was the part of the eye that his diagnosis was relying on? And that allows it, when the doctor then comes to look at that prediction and the eye scan, it can see what did the AI system base its diagnosis on and kind of overrule that or not. So it's the first step towards interpretability, which is very important if we want to use AI systems in the real world on safety-critical problems. And it's a good example of how AI and doctors work.
[0:28:29] Together, rather than replacing doctors. Yes, this is absolutely about complementing what doctors do. So the idea is not that the AI system, you'd rely on that on diagnosis, because there's a lot of nuances, you know, the experience of a real top-level eye doctor. But what it allows them to do is triage very efficiently the key cases. And of course, there are many other examples where AI can be used to examine and study scans. All sorts of scans.
[0:28:58] So we're looking at mammographies and other things now. Okay, Demis Hassabis, let's talk a little bit about the public's perceptions and, I guess, fears of AI. In terms of the fears, you know, I think it's valid, but I think here AI is no different to any other powerful technology that was invented in the past.
[0:29:16] I think the technology itself is neutral. The question is, how does society decide to deploy it? So I think it's really about a more geopolitical question, really, of how the AI systems get used, what they get used for, how we design them, and then how we spread the benefits of those systems. You, of course, Demis, are uniquely well-placed to advise us all on how to make the best use of AI, how to avoid some of the more worrying scenarios that a lot of people fear.
[0:29:45] Do you feel a sense of responsibility to help society make the smart decisions about
[0:29:50] I do feel a responsibility of that. And I do think scientists have a responsibility if they're working on cutting edge research that can have big impacts in the world to think through the consequences of that ahead of time. And what we try and do is you try and understand solutions, not just look at the performance or the outcome. You actually want to understand the process. And that allows you to project forwards or imagine forwards what the consequences of what you do might be. I mean, I've recently been arguing.
[0:30:20] I think it's good for decision makers, government, other people to have at least a basic understanding of...
[0:30:48] What these AI systems do, what they're capable of. Also, what isn't AI? I mean, the other thing that's happened is AI has become, I mean, it's incredible for me to see because 10 years ago, almost nobody was talking about AI when we started DeepMind. It was almost impossible to raise any money to get going. And now it seems like everybody's talking about AI. But a lot of what people are branding as AI and companies are branding as AI is not AI. It's just basic stats or basic maths, actually. So I think there's a lot of misinformation out there. So it would be good for people to understand.
[0:31:18] Of course, a lot of people...
[0:31:42] Say they're terrified that AI will one day reach the so-called singularity, where every sci-fi buff knows this, computers achieve full artificial general intelligence and become smarter than us. Does this concern you? Do you think it'll happen one day, and does it concern you? Well, I don't really think about it in those terms. I think, at least in the next phase of development, the next near term, it's going to be complementary tools that are...
[0:32:10] Better in some dimensions, but much, much worse than human experts in many other dimensions. And so there'll be a great, I think, almost a new renaissance of discovery. It's what I'm hoping we're going to see in the next few decades, where we use these smart tools to almost enhance our capabilities, like we've always used tools to discover new things. Also understanding our place in the universe, you know, physics, you know, biology, DNA, all of these things, genetics, they all involve creating.
[0:32:39] Huge amounts of data that we as human scientists have to somehow find the insights in and understand. There's probably all sorts of insights, both scientific and medical, that could be found in data we have right now, but we just don't know what it is. We can't see the patterns. We can't see the patterns. Right. And that's what I think...
[0:32:58] Systems like AlphaZero, which found patterns in the game of Go that were very complex, could one day be able to do. So you could imagine in the future, maybe we'll have even better tools that enhance our own minds so that we become more intelligent and more capable and therefore stay at the forefront of innovations long into the future, way past the singularity. Exciting times. Demis Hassabis, thank you very much for sharing your life scientific. Thank you. And thank you for listening.
[0:33:28] My guest on the 200th edition of The Life Scientific was Demis Hassabis. The producer was Anna Buckley. And if you're a fan of The Life Scientific, you might be interested to know that the second book of the series, The Life Scientific Inventors, will be published on November the 14th and is available to pre-order now. My name is Shannon Maldonado. I'm the founder of Yowie, a gift shop from the lens of artists and handmade objects.
[0:33:53] It was the first dinosaur ever to be scientifically named. And he came up with an animal that was about 40 foot long.
[0:34:23] A profound discovery that led to paleontology as we know it today. Impossible monsters, these giant, weird reptilian creatures. This was like a thunderbolt. Join me, Misha Glennie, and my three guests for In Our Time on this episode, Megalosaurus. Listen on bbc.com or wherever you get your podcasts.