[0:00:00] Self-driving cars, mobile assistants, playing the top songs by Smash Mouth, fraud detectors that protect your bank account, and face detection on your social media network. These all rely on artificial intelligence, or AI, a term that once sounded like some futuristic fantasy, but is now part of everyday life. In other words, the robots are here, folks, and they're just getting started. When the history of AI is written,
[0:00:29] By robots, no doubt. One prominent figure in their story will be Demis Hassabis and his company, DeepMind. When we set up the company, the way we articulated our mission was as a simple two-step process. Step one, solve intelligence. Step two, use it to solve everything else. And after that, we thought probably it's best not to have a step three. A slightly more accessible way to think of DeepMind is...
[0:00:56] It's kind of an Apollo program for artificial intelligence. One thing they did a few years ago was to create a computer program able to beat top-ranked players in the ancient game of Go, a game far more complex than chess. That victory made big news in 2015 and 16 and was declared a new milestone in the development of AI, or machine learning.
[0:01:23] You'll find out why on this episode of What It Takes, a podcast about passion, vision, and perseverance from the Academy of Achievement. I'm Alice Winkler. At a main, this child is gifted. And I heard that enough that I started to believe it. If you have the opportunity, not a perfect opportunity, and you don't take it, you may never have another chance. It all was so clear. It was just like the picture started to...
[0:01:50] There was no way in which a lie could prevail over the truth, darkness over light, death over life. Every day I wake up and decide, today I'm going to love my life. Decide. My advice is, if they're going to break your leg once when you go in that place, stay out of there. And then along come these differential experiences that you don't look for, you don't plan for, but boy, you better not miss them.
[0:02:24] Demis Hassabis was born in 1976, around the time the first gaming consoles were born, an important historical coincidence and a harbinger of things to come. But long before Hassabis became a star video and computer game developer, or a neuroscientist, or a leading thinker about artificial intelligence and consciousness and the universe, because why not, he was a chess prodigy.
[0:02:53] At 13, in fact, he was the number two chess player in the world in his age group. And chess is where this conversation began between Washington Post journalist Mary Jordan and Damas Hassabis for the Academy of Achievement in March of 2018. Do you remember when you first said, you know, I'm pretty good at chess?
[0:03:16] Yeah, it's hard for me to, you know, chess is something like I've always known how to do. So I learned when I was four years old and I don't even remember learning. Apparently, I've been told after I saw my dad and my uncle play a game of chess and they were just, you know, normal amateur players. And apparently I saw them playing and I asked to play. And they thought, well, we'll just, you know, humor me and just teach me the rules. And then a couple of weeks later, I was, you know, beating both of them.
[0:03:45] Then my dad thought, well, better, maybe we should take him to a chess club. And they just sort of went from there, really. So I've always, and then from all my youth, I was always captain of my age group for the national team and usually playing with much older kids. And so I've sort of had the chess part of me as being a big core part of my upbringing and who I am. How did that change you? I mean, how does chess right now play into your daily life?
[0:04:12] Yeah, I think chess, actually a number of things that I learned when I was young, you know, it wasn't so much the chess itself, it was the skills that you were training while you were playing chess. So problem solving, you know, imagination, creative thinking, strategic thinking, all of these things, if you play chess to a higher level, what you're really honing is those meta skills. And I realized from quite young that...
[0:04:36] It's one of the things I like chess was this was a great practice arena to get good, sort of like a mental gym, if you like, to get really good at those other skills, which you could then translate to all sorts of other subjects like science or business. Hassabis was super self-directed for a kid whose parents were a bit bohemian, as he calls them. His father never had a set job. He ran a toy store at one point. He was a singer-songwriter. He was a teacher.
[0:05:04] And he bought and sold houses. The family moved 10 times before Demas was 12 years old. But it all felt normal to him. He learned to roll with things, to make new friends easily. And as he told Mary Jordan, to follow his passions. So you're 13 years old. You're the second best player in the whole world. And you decide to do something different. Why did you not stick with chess? What were you thinking at 13?
[0:05:32] Well, so I had a bit of an epiphany around that age because I was playing, you know, I was more or less a professional chess player. So I was going to school, but I was taking weeks off school to go around the world playing chess and all my summer holidays and Christmas holidays and so on. So it was pretty full on. And I was, you know, the rest of the time I was studying chess. And I remember thinking that
[0:05:54] To become world champion, which was my plan at that point, then your whole life would have to be dedicated to nothing but chess and learning opening moves and theory and all this kind of stuff. And I started realizing the more I was specializing in chess, the less it was about these meta skills that would be useful for other things, but the more specific that knowledge would be to just chess. So it wouldn't be useful for other things you might want to do.
[0:06:19] And I kind of had this thought when I was in a big tournament actually in Liechtenstein near Switzerland. I remember...
[0:06:26] Playing a big match, it was like a 10-hour match against, I think it was the Danish champion at the time, the adult champion. And I lost the match after 10, 12 hours. I remember thinking, was that really a good use of all that brain power? And I went for a long walk in some beautiful field. At least I remember that in my mind, in the mountains. And I remember thinking, you know, maybe this isn't, you know, there's this whole room full of amazingly bright people and they're using their minds to, you know...
[0:06:55] Basically compete each other and try and win. And what was really, what was the, you know, maybe they should all be using their minds to solve, you know, cancer or something. And I was just sort of thinking maybe we could hone, you know, all of these bright minds. Maybe there's a better use of that time and energy that would be better for the world. And so I made this decision actually at that moment that.
[0:07:15] I would move away from chess and start exploring other areas and passions of mine and use those meta skills as my foundations for getting into these other things. But I just felt, although I love chess and I still love chess now, I felt it was too narrow a thing to dedicate your entire life to.
[0:07:31] So what do you see yourself now dedicating your life to? So I dedicate my life to, you know, it's artificial intelligence is obviously what we do at DeepMind. But really the underlying reason I work on artificial intelligence is I want to understand how the universe works. And I want to, you know, it's the deepest scientific question. I want to try and answer the most deep questions that we have. You know, what is consciousness? What are we doing here? What's the universe about?
[0:07:57] Do you have any early answers to what we're doing here? I have some theories, but they're sort of not well-formed yet enough. I'm still working on them. When I was a kid, my favorite subject was physics. And usually, if you're interested in the big questions about the universe, you would become a physicist. That's the normal.
[0:08:17] You know, root for that. And it was my, you know, it's my favorite subject all the way up to university. But then I read this book called Dreams of a Final Theory by Steven Weinberg, who's a Nobel Prize winner. And, you know, it must have been in my late teens. And this is all about physicists struggling to find, you know, a final theory of everything, like how everything works. And, you know, it's what string theory is about and so on. And I realized that they actually...
[0:08:42] I hadn't got that far yet and some of the things they were tackling were incredibly difficult and no one seemed to be making progress. So it felt to me like maybe what was needed was more intelligence and maybe something like AI as a sort of smart tool to help scientists could be something that would allow us to make those kinds of breakthroughs and find the kind of theories that we were after. So it felt to me like working on AI would be more fruitful and then using that as a tool to help solve things in physics and other sciences.
[0:09:12] So let's talk about some breakthroughs. What excites you most about breakthroughs kind of in the near future that we're going to see because of AI?
[0:09:23] Well, the thing I'm most excited about is I think we're on the cusp now of a new era in scientific discovery where we'll be able to use AI and machine learning to help us in areas of science and medicine to make sense of enormous amounts of data that we're generating from our experiments and find new insights and new structure in that vast amount of data. And I think that will lead to new scientific theories and new scientific breakthroughs. Can you give us some examples of how AI is going to make us healthier?
[0:09:53] Yeah, so the obvious thing that's already happening today is to do with radiography. So the low-hanging fruit in some sense is making sense of imaging data, whether that's a PET scan of cancerous tissue or what we've worked directly on, which is retina scans and detecting macular degeneration very, very early on. And there aren't enough radiographers, and often the diagnosis from these scans is one of the bottlenecks you get in treatment. And in a lot of these cases...
[0:10:22] So I think that's going to result in much more efficient diagnosis and much more efficient treatment. So I think that's just the beginning. And then the next steps will be...
[0:10:49] Things like drug discovery, where you've got some target, some virus or some bacteria, and you need to generate some new compound that will target that. And I think, again, we're on the cusp of being able to use things like machine learning to discover new types of drugs. It's so hard to understand exactly how a machine can do something like drug discovery. So can you just walk us through an example?
[0:11:16] Yeah, it's quite simple really. What happens is, you know, if you think about...
[0:11:20] It's a massive combinatorial problem, right? You've got lots and lots of possibilities. So, you know, you've got all these molecules that you could put together to create a new compound, which would create a new drug. And then you've got this complex new, you know, virus that you've just discovered and you need to build a compound that will attach to it, right? And so, really, the problem is that, you know, we know quite a lot about physics, we know a lot about chemistry, we know a lot about biology, but the combinations of possibilities is huge, right? It's astronomical, the number of compounds you could combine.
[0:11:50] And you would need to test that each one laboriously. Maybe you'd have some theories about things that could work, but you wouldn't know for sure.
[0:11:57] Right, as a scientist. And what the kinds of systems we're building, like AlphaGo and AlphaZero, our new systems, is you can think of them as optimization programs. You give them some goal to optimize, whether that's efficacy with a drug or winning a game, and they figure out for themselves the right search path through this massive combinatorial space. So it's just sort of, and they do that by having, be able to experience many, look at much more data than any human.
[0:12:26] These systems can find patterns that are there in the data that we as just as human scientists or human experts or human medical practitioners wouldn't have the bandwidth to do.
[0:12:42] I told you we'd return to the subject of that computer victory over the humans in the game of Go, and here's the perfect moment. Demis Hassabis elaborated on the significance of those Go matches in a talk he gave at the Academy of Achievement Summit in London a few months before he sat for this interview we've been listening to. Here's an excerpt of his talk.
[0:13:08] Now, for those of you who don't know, Go is the game they play in Asia, in Japan and China and Korea, instead of chess in the West. But it's much deeper and a more complex game than chess. In fact, I would say it's probably the most complex game mankind has ever devised. And one simple illustration of that is that there are more possible
[0:13:28] Configurations of the board than there are atoms in the observable universe. So 10 to the power 170 possible variations. So what that means is that the solutions to this game can't be brute forced. So even if you collected together all the world's compute power and ran it for a million years, that still wouldn't be enough compute power, computation time to solve this game.
[0:13:51] So you might ask, well, how do human players deal with this complexity, these brilliant grandmasters in Asia? And actually, they deal with it by using their intuition. So while chess, I would say, is a calculation game, so you calculate out explicitly what you're going to do and what you're planning to do, in Go, you kind of feel your way to the right move.
[0:14:15] And in fact, in Asia, it's considered to be like an art form rather than just a mere game. So in fact, Confucius wrote about Go as one of the four great arts to be mastered by any true scholar. So words like intuition and creativity are not normally associated with computers. And this is why Go has been one of the holy grails and grand challenges for artificial intelligence for the last 20 years, ever since Deep Blue beat Garry Kasparov at chess.
[0:14:42] And the way we cracked this is by mimicking human intuition with the use of artificial neural networks. And what we did was, instead of programming the solution to the game directly, like Deep Blue was for chess, so what they did there and IBM did there is program a set of rules that the computer dumbly executed at very fast speeds in order to decide what to do. Instead of that, what we built was a learning system that played the game itself.
[0:15:11] And the way it did this is by learning incrementally from its own mistakes, using a process called reinforcement learning, which is actually the way the brain learns. And so we did this, we developed this program for a couple of years, and then we challenged some of the best...
[0:15:29] players in the world in Korea and China in two huge matches that were watched by over 200 million people around the world. And we won those matches and this was proclaimed to be a decade ahead of its time by both the artificial intelligence community and the Go community.
[0:15:45] But the important thing was actually not the fact that we just won the matches. It was actually how AlphaGo played, the types of moves that it played. And what we found was through playing these matches is that AlphaGo created some truly innovative and what were described afterwards as creative and beautiful moves in order to win these matches.
[0:16:05] And in so doing, it overturned actually thousands of years of received wisdom in the game. It did some moves that were so shocking that none of the professionals could believe these types of moves were possible. And it sort of overturned what was thought to be the right ideas in this game. In fact, one of the commentators said afterwards, watching the match, it was like seeing games from a thousand years in the future. Actually, you know, I love that sort of comment.
[0:16:29] So after we did these matches, of course, the point of building AlphaGo was not just to win at the game of Go. It was actually to build general purpose learning systems that could potentially be applied to almost any domain. And I think that's the promise of this new generation of artificial intelligence technology, is the power of these learning systems to potentially go beyond human knowledge and actually discover new knowledge to help society.
[0:16:59] And that's what my dream is and why I've worked my whole career in artificial intelligence is the dream of AI-assisted science. The idea of having smart AI-powered tools that help experts make faster breakthroughs in areas of science and medicine and human understanding. Before we go back to our interview with Demis Hassabis, I want to take a couple of minutes to play you excerpts from a couple of earlier pioneers in the field of AI, Marvin Minsky and Ray Kurzweil.
[0:17:32] People who helped create a roadmap to the future that Demis Hassabis is now on. They're a good reminder of how far we've come, how fast. Marvin Minsky was one of the first people to contemplate the very idea of AI in the 1950s. And his work at Harvard and at MIT actually helped bring about the personal computer and the internet. Here he is speaking to the Academy of Achievement.
[0:18:01] Well, I'd like to tell you about artificial intelligence. What that is, is it's a new science. Since computers appeared in about 1950, what also appeared were new ideas about processes. Ideas that I think are new under the sun in the sense that before 1950, we had a lot of mathematics, we had theories of psychology, but we had no way to describe knowledge and processes. Philosophers were very muddled. They had logic.
[0:18:31] Thank you. Thank you. Thank you.
[0:18:59] In 30 years, it's going to be the most important industry in the world. Computers are going to slowly become intelligent. It may take 300 years. And that's a small amount of time. You know it's 300 years from Newton to Einstein. And there was no moment in that period of 300 years when there weren't great discoveries made every year, every five years. When Marvin Minsky died in 2016 at the age of 88, an international prize was created in his name.
[0:19:27] And who was the first winner? Let's see. That would be Demis Hassabis. Okay, but next I'm going to play you a little Ray Kurzweil. Kurzweil is a computer scientist, an inventor, and a futurist. And he was a student of Marvin Minsky's.
[0:19:44] Kurzweil developed the first optical character reader, you know, the scanner that turns written text into digital text. Also, the first reading machine for the blind that turns text into speech, the first commercial speech recognition machine, and the first music synthesizer that can replicate the sound of actual, or I guess I should say, traditional instruments.
[0:20:11] Some of the ideas in his books on machines and the future sound a little out there, but many of the predictions he made in his 1990 classic book, The Age of Intelligent Machines, have already come true. So keep that in mind while you listen to this excerpt from the interview Ray Kurzweil did with the Academy of Achievement in 2000. I mean, I think we will have machines that have the complexity, richness, and subtlety of human.
[0:20:39] I mean right now it's very easy to dismiss machines because they're still a million times simpler than human beings so they don't have all the endearing qualities. If you have a virtual character that claims to be conscious and have feelings so quickly reach the limits of their performance and they lack the subtle cues that we associate with really having consciousness and emotion that it's not an effective impression. But we will achieve machines in a few decades that have the complexity and richness of
[0:21:09] In many ways, they'll be derivative of human thinking or even copies of human thinking in a different medium, and they'll have the richness of behavior. And when they claim to be having feelings and being conscious, people will believe them. If we don't believe them, they'll get mad at us. But there'll be philosophers who come along and say, yes, that entity is a really great simulation of a human, and it really is very convincing.
[0:21:34] But it doesn't squirt neurotransmitters, so it can't be conscious. And there's no way to really settle that argument. There's no machine you can put it in, and the light will go on that, yes, this entity is conscious. It becomes fundamentally a deeply philosophical issue. But I think the reality will be that we will accept the consciousness of these entities, and there won't be a clear distinction. We're going to have human minds, brains, that have non-biological thinking processes.
[0:22:03] Cursweil also said in that interview back in 2000, to anticipate the future, you have to invent it. In 2017, when Demis Hassabis was named one of the world's
[0:22:31] 100 Most Influential People by Time Magazine. It was Ray Kurzweil who described the importance of Demis Hassabis. Quote, he believes as I do, Kurzweil wrote, that AI will help solve humanity's grand challenges, alleviating poverty, curing disease, and improving the environment. You can go learn a lot more about Ray Kurzweil and Marvin Minsky.
[0:22:58] And listen to their interviews at achievement.org. But getting back here to Demis Hassabis, our interviewer, Mary Jordan, pointed out that machines have a huge advantage when it comes to speed and efficiency. And she wanted to know what Hassabis himself does to increase his human efficiency.
[0:23:18] So, yeah, I'm pretty obsessive about being efficient because I think time is so short and life is so short. And there's so many interesting and amazing things to do and experience. So I don't sleep very much. You know, I've always been a nocturnal person. So I normally sleep at sort of three or four in the morning. And, you know, I only have about five or six hours sleep, which is probably not enough, actually, but as brain studies tell us. But there's just so many things I need to pack into the day. So the way I normally split it is...
[0:23:47] During the daytime, I manage the company and have meetings and do administrative things and organizational things. And then when I get home, I start a sort of second day of work after dinner around about 10 p.m. where I start thinking about all my creative things. So that's when I would read books or research papers or write research papers or come up with ideas about algorithms. I would do that in the small hours of the morning. And I've always found...
[0:24:17] Myself, best creatively in the small hours of the morning when everything's quiet in the world.
[0:24:22] Is there anything else you do to spur your creativity? Yeah, I use music a lot actually. So when I'm focusing, I'll listen to music, different music depending on what mood I'm trying to create. Classical, rock? Yeah, I like actually everything from classical to, you know, drum and bass. It just depends on what I'm trying to do. So if I'm programming, it will be something more like, you know, something with a beat and just more...
[0:24:48] Let's say you have a deadline and you really, really are thinking hard. What music do you put on? It would probably be some kind of, you know...
[0:25:15] I think it's called liquid drum and bass, actually. It's the actual sub-genre of drum and bass. So it's got a beat, but it's got melody as well. But it has to be interesting. If I have to focus, it can't have words, because then your brain starts listening into the words. So I don't have any lyrics. I can't listen to anything that has lyrics in it. It has to be just instrumental. And then I think probably what's happening on a neuroscience level, I think there's been some studies about this, is it gets you into a kind of alpha state.
[0:25:45] Because this type of modern music is quite repetitive. So once you've heard it a few times, it's almost in the background. And I think it's just getting your brain into some kind of rhythm, which I find better for focusing. So it's very to do with getting you into the right kind of zone. And I've learned a lot about myself. This is something I'd recommend young people to do is learn.
[0:26:11] Learn about how they work best is one of the most important things and what triggers you into certain creative states and then maximize that. And the other thing I use is I really like film. So that's one of the main things I do to relax is I love watching great films and I find them very inspirational. And then once I've watched a great film, I can usually re-trigger that mood or that emotion just by watching a small clip on YouTube of that film. And I can pretty much evoke the whole of that.
[0:26:40] She's a replicant, isn't she? I'm impressed. How many questions does it usually take to spy on?
[0:27:09] I don't get it, Tyrell. How many questions? 20, 30 cross-referenced. It took more than a hundred for Rachel, didn't it? She doesn't know. She's beginning to suspect, I think. Suspect? How can it not know what it is? Commerce is our goal here at Tyrell. More human than human is our motto. Rachel is an experiment, nothing more. So, movies and music.
[0:27:35] But Demis Hassabis didn't mention whether he also gets in the zone by using the running track and the napping pods, the free massages, and the cooking school, all available at DeepMind's headquarters ever since the company was bought by Google in 2014 and moved into their London offices. But Mary Jordan did want to know specifically about the music room at DeepMind.
[0:28:01] And he proudly answered that there's actually a deep mind band and they're very good.
[0:28:07] The thing we have here is we have a lot of people who have hidden talents in many domains. So they're all like brilliant scientists and engineers. But often in growing up, they'll have hidden talents. Like there'll be concert pianists on the side or songwriters on the side. And we have actually a very talented house band that gets together and plays at all of our Christmas parties and other things. So I think they use the music room mostly for jamming sessions. So we know that you're a chess player. We know that you're a neuroscientist. We know that you have a company.
[0:28:36] We know you think big. Is there some other hidden talent that you have that we might not know about? One reason that I think I've been successful in creating an unusual company like DeepMind is that my passion, I've been lucky in life and I've been able to follow many passions that I've had. Initially it was games, then it was programming, then it's AI and neuroscience. And I think I've managed to build a...
[0:29:05] World that I'm in or career that utilizes all of those interests and I can sort of indulge in all those passions as part of my job. But the extra one that I have a passion for which is sort of tangential to those things is organizational design. And I don't know why that is because it's quite different. It's probably connected to my obsession with efficiency where, you know, companies are really, you know, obviously they're made of, you know, groups of individuals. And so how you organize them and what kind of culture you have will...
[0:29:34] And I think that's led to DeepMind having quite a unique culture. And I think that's just as big an effect on the output we have as our research ideas.
[0:30:03] How many PhDs do you have in this building? So we have over 400 PhDs just on these two floors. So I think it's the biggest collection anywhere in the world of brain power on this topic that's ever been collected. And what is your kind of short-term goal to do with all that brain power?
[0:30:20] Well, the goal is to make as fast progress as possible towards building human level AI. And then the idea is that as we build these technologies, we'll be able to apply them usefully for the benefit of the world in all sorts of sectors like healthcare and science. So I think in a way, you can think of AI as a kind of meta solution to all of the other problems that we would like to have solutions to as a society. Human level AI.
[0:30:48] You know, it's very exciting, as Bill Gates said, you know, hey, we might have longer vacations because robots are going to be doing a lot of things we are. But other people, it's very unsettling. Where do you fall on that about the need kind of, you know, that people like to think that they can do things that a robot can't? Sure. Well, I think, you know, I'm in between on those camps, actually. So in the sense of...
[0:31:13] You know, obviously we work on AI and I spent my whole career on it because I think it's going to be one of the most amazing, empowering technologies for humanity ever invented.
[0:31:20] I think it's, you know, I think of it as a smart tool that will unleash the true potential of human ingenuity by sort of working, you know, whilst being able to draw on these kinds of smart tools to help us, as I mentioned, in science and medicine. But, you know, there's going to be a lot of disruption too, like with any new powerful technology. And AI might be one of the most disruptive of all. And, you know, we've always been cognizant of that from the start of DeepMind and even before. And we've always thought about ethics.
[0:31:49] And we've always been thinking about the ethics of this from the beginning. So, you know,
[0:32:19] We believe that stewarded correctly, this technology should be for the benefit of everyone. And I think we're going to see some amazing breakthroughs that I think, frankly, society needs from climate change to Alzheimer's, where we're not making enough progress as a society on these kind of very, very pressing problems. And I think if we could do science better and we had more intellectual horsepower behind it, we could make better, faster solutions.
[0:32:45] Let's just talk about climate change for a second. So how can you advance AI to help climate change? There's so many ways. You could invent some new, much more efficient renewable energy.
[0:32:57] You know, you could maybe solve fusion. You could try and create a new material that would capture carbon cheaply out of the atmosphere. You know, the sky's the limit, really, in terms of if you have that kind of optimization technique in a box that you can call upon and have it present you with alternatives. You could give it a goal function. You could specify a goal clearly to it, you know, some metric, which I think you probably could with something like climate.
[0:33:25] I think that you could really make progress with new materials or new solutions to those problems. Another way that it could help with climate is you could do better modeling of the climate. So predictions could be more accurate about what's happening. You could analyze satellite imagery in real time so that you could see if icebergs were breaking up and you could see if deforestation was happening where it wasn't supposed to be happening. I think NGOs, we've talked to a lot of NGOs about this.
[0:33:54] You know, international agreements, but how do you actually track that on the ground when there's such vast amounts of space going on? So I think there's multiple ways, actually, on a number of levels that AI could help with something like climate. So we already have driverless cars. We see that in medicine, even for radiology, that a computer actually can be more correct than a human. So people are worried about jobs. What is left for humans?
[0:34:21] Yeah, I think there's many things left.
[0:34:25] There's all the creative pursuits. There's also all the empathetic jobs. So I make this joke sometimes that, you know, I think it's semi-serious that at the moment, you know, let's take healthcare. We pay doctors a lot more than nurses as a society. And, you know, in the future, in 10, 20 years' time, I could imagine that, you know, a doctor that doesn't have some kind of AI database that they're using to keep them up to speed with the latest developments, you know, would be, you know, not a doctor you'd want to see.
[0:34:54] But on the other hand, you know, if you're ill in hospital and you want to be looked after, you know, I think a nurse and a human touch would be a lot better than, you know, having a robot. So I feel like those kinds of jobs like teaching and other things that maybe society undervalues currently will become more valued. So I can imagine one day, I mean, maybe nurses will be paid more than doctors and maybe that's a good thing.
[0:35:18] People don't quite get if all these jobs are mechanized and robots are doing them, how will they make money in the future? Will there be enough jobs that actually earn them a living?
[0:35:27] Well, but this has happened before many times. I mean, the industrial revolution did that. So we don't, you know, spin wheels in weaving houses anymore. They went on to be, I guess, designers and fashion designers or something, right? So there was, there's always been, this is nothing new. There's always been an evolution or new technologies come in. So that makes some jobs more efficient. So there's some disruption and then it opens up new jobs that we can't even imagine today. I think it's going to be the same. Is there anything a robot...
[0:35:55] Is not going to be able to do? Well, right now, the robots can't do anything, basically. But in the coming years, when you're looking out and you're talking about human level... Well, that's an interesting philosophical question. I mean, you've got to decide...
[0:36:09] This is where neuroscience comes in. What is the brain? Is it just the physical biology or is there something more? And from the neuroscience discovery so far, it looks like that you can describe the brain in mechanistic terms. And if that's true, then most of the functions that the brain has, you should be able to mimic in some way with a computer. And this is what Alan Turing was saying.
[0:36:38] What do you think? Is there going to be a machine that can write a play, that can sing a song?
[0:37:02] Well, I don't think, you know, the other question is what do we mean by that? So I think we'll have machines that can write things that are coherent English, but would they be a great novel? I doubt it because I was having this debate actually with somebody, a famous writer.
[0:37:19] We were talking about this exact topic and my view is that a great novel is also partly because you know the person who's written it was human and they went through some things and you can connect to the human condition that they're describing. And I think that's important for a novel to be great. You know, I don't think it's just the technical grammatical components. I mean, that's the reason why I think, you know, we bother to say for some stories this is based on a true story. Why would we say that unless for some reason that makes us...
[0:37:47] You know, it makes it more emotive that we know somebody else has gone through that, actually gone through that in real life. And I think there's an element of that. I don't know if it would be as meaningful if you knew that it was written by a machine. You know, they're just machines, right? Like cars, you know, and planes. They let us extend the human capability, but the human at the heart of it still has to be educated and soulful.
[0:38:12] And it's still providing the kind of emotional input to impetus to these tools. These tools would be nothing without humans. Should we adjust what we teach students in school and universities because of the coming huge changes? I don't think this is just to do with AI. I think this is also just to do with technology in general. You look at the internet and mobile, how fast things have changed. And just going back to the jobs questions, there are jobs exist today that you couldn't have imagined 20 years ago.
[0:38:39] You know, being a YouTube content creator or, you know, some certain types of game designers, other things like that. There's all these kind of whole new classes of jobs that have come up. And I think that will happen again, enabled by all the new technologies of which AI is one. So what I would recommend to students today is that rather than focus on, you know,
[0:39:02] Jobs for life, like I'm going to be an accountant and that's going to be 50 years and I'm done, so I'll just train in that. I think it'd be better to learn general skills that you could then adapt to any new type of sector that you've got to learn about. So that's things like learning to learn. So what I would encourage schools to teach is not specifically so much the rote learning or the specific subject you're learning, but actually...
[0:39:32] The ability to learn.
[0:39:34] If that makes sense. So how do you learn best? What makes you learn best? How can you accelerate your own learning? No matter what subject you're learning, what motivates you? What are you passionate about? What hours of the day do you work best on? These are all the things that you're never really taught at school. And actually the most important things is how to get the best out of yourself. So it's subject independent, but it's just the ability to learn. And that's something that would be useful no matter which way the world goes. Because I think the world's becoming unpredictable and we don't know what new technology is going to come out.
[0:40:03] And things are changing very fast. So I think in that type of environment, you want to be as adaptable as possible. Well, people are very excited about living longer and having all the help that AI can bring medicine. But there are also people that are unsettled by it. And they say, we learned how to make fire before we made the fire extinguisher. Is there anything in this amazing power that is being unleashed by AI that you're a little bit worried about?
[0:40:31] Well, you know, it's like any powerful technology. The technology is neutral in of itself. It depends on how humans are going to use it. We can use, you know, all of the history of science has been like that, from nuclear to biology. You know, there are many good and bad ways to use these technologies. And we just have to make sure that we build AI in a safe way, in a controllable way. And I see that as like a tool. And then the second question is making sure you deploy it for the good of the world rather than the bad.
[0:41:00] One of the big questions that people are talking about at the Pentagon is how this is going to change war. How do you think it will change war? Will it be just robots fighting each other and weaponized drones? Well, I'm against using any of this kind of technology for any kind of military, so that's not something I'm going to get involved with, or we will do at DeepMind.
[0:41:22] You know, I can't speak for what other organisations will do. My view is that, and we've signed open letters and UN letters to this effect, that there should always be humans in the loop for any kind of warfare or weapon system. Again, as I mentioned earlier, you know, we are at the forefront of having these ethical debates and bringing that into light about how do we want to use these systems and deploy these systems. For example, not use them for war.
[0:41:48] Just one other thing on that is that people are worried about an arms race. There's been a lot of talk about that and that China sees it as a Sputnik moment because they think that U.S. is...
[0:42:03] More advanced than China, and so they're putting a lot of focus and money into that. Is there an AI arms race going on? I wouldn't say so. I think that on the research front, the researchers know each other pretty well and go to the same conferences, and there's a spirit of sharing information. So, for example, we publish all of our work in big journals like Nature and Science.
[0:42:27] There's definitely a competition in terms of being the best commercially and trying to get the biggest benefit out of these technologies. But I wouldn't say it's a classical arms race, no. The only race Demis Hassabis is committed to is the one to understand the human mind and to improve humankind. And there are a couple of ways his company is starting to do that, building on their success with AlphaGo. For one...
[0:42:55] They've started a project with Alzheimer's researchers. The goal is to figure out how and why strings of amino acids fold to become 3D proteins. Scientists believe that misfolded proteins may be behind Alzheimer's. Also, Parkinson's, cystic fibrosis, Huntington's, and other diseases. So understanding the folding mechanism.
[0:43:22] is a first step toward coming up with treatments. Demis Hassabis is also trying to work with astronomers to use machine learning for the discovery of new planets and new particles. One thing Mary Jordan wanted to know was where Hassabis sees human beings fitting into the universe.
[0:43:41] I think humans occupy actually a very special place in the universe. As far as we know, we haven't discovered any alien life, so we're the only sort of fully conscious beings in the universe, and I think that's very key. I mean, we have to figure out what consciousness is, but I think it was Carl Sagan that said, you know, we're like waking up the universe, right? So the universe is kind of, you know, it's huge, but it doesn't have any self-awareness, but, you know, that's what we bring to the universe.
[0:44:10] I think humans have an extraordinary kind of unique part in the universe from that perspective. And that conversation led to one about God. Because when you're talking to someone who spends so much time thinking about the nature of the universe, well, it seems worth asking. Hassabis said he couldn't really give an answer in the time remaining. But then he added. You know, I'm kind of open-minded about these things. And if you can...
[0:44:40] What you know as a scientist, actually, is that if you're a proper scientist, of course, the scientific method is one of the greatest inventions mankind has ever had, that we've ever had as humanity. And it works brilliantly, the scientific method, and that's what has allowed us to have the Enlightenment and modern civilization. But I think as a scientist, what you realize is the more you find out, actually,
[0:45:03] The bigger questions there are unresolved. You sort of go deeper, and then there's more things you don't know. And in physics, for example, we don't know what dark matter is and dark energy now. And there's theories about it, but we don't really know.
[0:45:19] And I think there's always more questions. Like, you know, there's the Big Bang. We think that's how the universe started, but what was before the Big Bang? So I'm quite open-minded to the fact that there may be many aspects of the physical universe that we just don't understand. And that when we have maybe something like AI helping us, we will better get a truer understanding of what's going on here. And then we'll be able to answer the questions about how this, you know, what the purpose of the universe was, and was it designed, and how did it come about?
[0:45:49] Much better. You've had such an extraordinary career, but can you just tell us about one aha moment? Like one moment you were like, wow. I've probably had quite a lot of aha moments. I mean, I guess one of the really formative ones was when I first got my first...
[0:46:09] I was about eight years old. It was a ZX Spectrum. And I remember starting to learn to program on it and sort of learning from books. And I programmed some.
[0:46:21] One of my first programs I can remember was a program to play this game called Reversi, or Othello. It's called Othello in the UK, but it's called Reversi, I think, in the US. And it played it reasonably well, and it could beat My Kid Brother. And I remember thinking, this is amazing that you can create a program that...
[0:46:43] You know, you can go to sleep afterwards and it can carry on number crunching. And then you can wake up the next day and it will solve something for you that you, you know, you hadn't, you know, you wanted to know the answer to when you went to sleep. So it seemed, from again, I guess I was even then obsessed with efficiency, is that that seems incredible, like a way to enhance your mind, right? So suddenly you can outsource problems to this computer and it can solve certain types of things for you. To me, that just seemed like magic and a magical extension of the mind.
[0:47:13] And from then I was sort of hooked on computers and programming and then ultimately AI. Demis Hassabis is a neuroscientist and the founder of DeepMind. He spoke with the Washington Post's Mary Jordan for the Academy of Achievement in 2018. Give another listen to this podcast in about 20 years and I'm confident you'll be marveling at all that has come to pass in that time. That's about as much of a prediction as you'll get out of me.
[0:47:49] Oh, except for our next episode. I can predict that with some accuracy. It'll be Kazuo Ishiguro, the author of Remains of the Day and Never Let Me Go, two novels I'm certain that could never have been written by a robot. I'm Alice Winkler, and this is what it takes from the Academy of Achievement. Funding for What It Takes comes from the Katherine B. Reynolds Foundation. Thanks to them, and thanks to you for listening.